Matrix Finite Element Methods in Dynamics

D
Mass Matrix
Diagonalization
D1
Appendix D: MASS MATRIX DIAGONALIZATION
TABLE OF CONTENTS
Page
D.1
D.2
D.3
D.4
HRZ Lumping
. . . . . . . . .
Lobatto Mass Lumping . . . . . .
Nonconforming Velocity Shape Functions
Congruential Mass Transformation
.
D2
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
.
.
.
D3
D4
D6
D6
D.1
HRZ LUMPING
The construction of the consistent mass matrix (CMM) is fully defined by the choice of kinetic
energy functional and shape functions. No significant procedural deviation is possible, other than
possibly using reduced integration to obtain a singular matrix. On the other hand, the construction of
a diagonally lumped mass matrix (DLMM) is not so clear cut, except for simple elements in which
the lumping is uniquely defined by conservation and symmetry considerations. A consequence of
this ambiguity is that various methods have been proposed in the literature, ranging from heuristic
through algorithmic. A good discussion of mass diagonalization schemes starting from the CMM
can be found in the textbook by Cook et al. [148]. Its use in explicit DTI is well covered in [74].
This Appendix gives a quick overview of proven methods, as well as a promising but as yet untried
one.
D.1. HRZ Lumping
This scheme is acronymed after the authors of [363]. It produces a DLMM given the CMM. Let
m e denote the total element mass. The procedure is as follows.
1.
For each coordinate direction, select the DOFs that contribute to motion in that direction. From
this set, separate translational DOF and rotational DOF subsets.
2.
Add up the CMM diagonal entries pertaining to the translational DOF subset only. Call the
sum S.
3.
Apportion m e to DLMM entries of both subsets on dividing the CMM diagonal entries by S.
4.
Repeat for all coordinate directions.
The see HRZ in action, consider the three-node prismatic bar with CMM given by (17.2). Only
one direction (x) is involved and all DOFs are translational. Excluding the factor A/30, which
does not affect the results, the diagonal entries are 4, 4 and 16, which add up to S = 24. Apportion
the total element mass A to nodes with weights 4/S = 1/6, 4/S = 1/6 and 16/S = 2/3. The
result is the DLMM (17.3).
Next consider the 2-node Bernoulli-Euler plane beam element. Again only one direction (y) is
involved but now there are translational and rotational freedoms. Excluding the factor A/420,
the diagonal entries of the CMM (?), are 156, 42 , 156 and 42 . Add the translational DOF
entries: S = 156+156 = 312. Apportion the element mass A to the four DOFs with weights
156/312 = 1/2, 42 /312 = 2 /78, 156/312 = 1/2 and 42 /312 = 2 /78. The result is the
DLMM (18.4) with = 1/78.
The procedure is heuristic but widely used on account of three advantages: easy to explain and
implement, applicable to any element as long as a CMM is available, and retaining nonnegativity.
The last attribute is particularly important: it means that the DLMM is physically admissible,
precluding numerical instability headaches. As a general assessment, it gives reasonable results if
the element has only translational freedoms. If there are rotational freedoms the results can be poor
compared to customized templates.
D3
Table D.1.
One-Dimensional Lobatto Integration Rules
Points
Abscissas i [1, 1]
Weights wi
2
3
4
1 = 1 = 2
1 = 1 = 3 , 2 = 0
1 = 1 = 4 , 2 = 1/ 5 = 3
w1 = w2 = 1
w1 = w3 = 13 , w2 = 43
w1 = w4 = 16 , w2 = w3 =
5
6
Common names: Trapezoidal rule and Simpsons rule for p =

2, 3, respectively.
In the p = 4 rule, interior points are not thirdpoints, since 1/ 5 0.447213596 = 13 .
Lobatto rules with 5 p 10, rarely important in FEM work, are tabulated in [2, Table 25.6].
Table D.2.
Points
Abscissas i [1, 1]
2
3
4
5
One-Dimensional Newton-Cotes Integration Rules

Weights wi
Same as 2-point Lobatto; see Table D.1

Same as 3-point Lobatto; see Table D.1
1 = 1 = 4 , 2 = 1/3 = 3
1 = 1 = 5 , 2 = 1/2 = 4 , 3 = 0
w1 = w4 = 14 , w2 = w3 = 34
7
w1 = w5 = 90
, w2 = w3 = 32
, w3 =
90
12
90
Common names for p = 4, 5: Simpsons 3/8 rule and Booles rule, respectively.
Additional NC formulas may be found in [2, Table 25.4]. For p > 5 they have negative weights.
D.2. Lobatto Mass Lumping

A DLMM with n eF diagonal entries m i is formally equivalent to a numerical integration formula
with n eF points for the element kinetic energy:
Te =
n e
i=1
m i Ti ,
where
Ti = 12 u i2
(D.1)
Assume the element is one-dimensional (1D), possesses only translational DOF, and that its geometry is described by the natural coordinate that varies from 1 through 1 at the end nodes. Then
(D.1) can be placed in correspondence with the so-called Lobatto quadrature in numerical analysis.
(Also called Radau quadrature by some authors, e.g. [130]; however the handbook [2, p. 888] says
that Lobatto and Radau rules are slightly different.)
A Lobatto rule is a 1D Gaussian quadrature formula in which the endpoints of the interval [1, 1]
are sample points. If the formula has p 2 abscissas, only p 2 of those are free. Abscissas are
symmetric about the origin = 0 and all weights are positive. The general form is
1
1
f ( ) d = w1 f (1) + w p f (1) +
p1
wi f (i ).
(D.2)
i=2
The rules for p = 2, 3, 4 are collected in Table D.1. Comparing (D.1) with (D.2) clearly indicates
that if the nodes of a 1D element are placed at the Lobatto abscissas, the diagonal masses m i are
simply the weights wi . This correspondence was first observed in [285], and further explored in
[463,464]. For the type of elements noted, the equivalence works well for p = 2, 3. For p = 4
a minor difficulty arises: the interior Lobatto points are not at the thirdpoints, as can be seen in
D4
D.2
(a)
(b)
LOBATTO MASS LUMPING
(c)
Figure D.1. A pair of degree-2, 3-point Gauss quadrature rules for the six-node plane stress triangle with
constant metric: (a) node configuration; (b) the 3-interior point rule; (c) the 3-midpoint rule, which is a Lobatto
rule for this node configuration. All weights are 1/3. Lines within triangle mark triangular natural coordinates
(a.k.a. barycentric coordinates) of constant value, to illustrate constant metric.
(a)
(b)
(c)
Figure D.2. As in Figure D.1, but triangle has now curved sides and variable metric.
Table D.1. If the element nodes are collocated there, one must switch to the Simpson 3/8 rule,
which is a Newton-Cotes formula listed in Table D.2. and adjust diagonal masses accordingly.
As a generalization to multiple dimensions, for conciseness we call FEM Lobatto quadrature one
in which the DOF-endowed element nodes are sample points of an integration rule. (Sample points
at other than nodal locations are precluded.) If so, the equivalence with (D.1) still holds. But one
quickly runs into difficulties:
Negative Masses. If one insists in higher order accuracy, weights of 2D and 3D Lobatto rules are
not necessarily positive, a feature noted in [203]. The subject is studied in detail in [285]. This
shortcoming can be alleviated, however, by accepting lower accuracy, or sticking to product rules in
suitable geometries. For example, applying a product 1D Lobatto rule over each side of a triangle
or quadrilateral. Of course a more flexible alternative is provided by templates, because these allow
the stiffness to be concurrently adjusted.
Rotational Freedoms. If the element has rotational DOF, Lobatto rules do not exist. Any attempt
to extend (D.2) to node rotations inevitably leads to translation-rotation coupling.
Varying Properties. If the element is nonhomogeneous or has varying properties (for instance, a
tapered bar element, or a plate of variable thickness) the construction of accurate Lobatto rules runs
into additional difficulties, for the problem effectively becomes the construction of a quadrature
formula with non-unity kernel.
As a general assessment, Lobatto mass lumping is useful when the diagonalization problem happens
to fit a Gaussian quadrature rule with element nodes as sample points and nonnegative weights.
Formulas of that type were developed for multidimensional domains of simple geometry during the
1950s and 60s. They are can be found in handbooks such as [706,707], along with many other rules.
D5
As noted above, an obvious hindrance is the emergence of negative weights as the rule degree gets
higher. This feature excludes those from contention except under extreme caution, whereas zero
weights are less deadly. Rules useful for FEM work are compiled in [244], as well as Appendix I
of [253], for seven element geometries.
The six-node plane stress triangle, shown in Figure D.1(a), illuminates obstacles typically encountered in multiple space dimensions. The total element mass is m e = A h, in which A denotes the
plane area and h the plate thickness, assumed uniform. There are two 3-point Gauss quadrature
rules of degree 2 for a constant metric triangle, shown in Figure D.1(b,c), which is extracted from
[255]. The midpoint rule, illustrated in Figure D.1(c), is also a Lobatto rule for this element, but
the 3-interior-point rule pictured in Figure D.1(b) is not.
Using the midpoint rule to build the DLMM results in three masses of m e /3 collocated at the
midpoints, while all corner masses vanish [203]. The HRZ scheme leads to the same result. This
DLMM has rank 6 and rank deficiency 6. To attain full rank one must take some mass from the
midpoints and move it to the corners: not a well defined process. An heuristic way out would be
to apply the Simpson rule line lumping along the three edges. This results in m e /9 at corners and
2m e /9 at midpoints but the degree drops to 1. To retain accuracy, a simultaneous change of the
stiffness matrix could be tried within the template framework.
For a curved-side six-node triangle with variable metric, a case illustrated in Figure D.2, node
and sample points remain at the same location in terms of natural coordinates, but local Jacobian
determinants enter the formula.
D.3. Nonconforming Velocity Shape Functions
This is a variational technique based on assuming velocity shape functions (VSF) that differ from
the usual displacement shape functions (DSF). To produce a diagonal mass matrix, the VSF must
satisfy additional mass orthogonality conditions that effectively decouples each VSF with respect
to all others in the kinetic energy integral. This can be practically realized by making each VSF
vanish at all points of a Gauss integration rule except one. Which rule? That appropriate to the
correct integration of the kinetic energy over the element.
Rather than explaining the technique further, the interested reader may want to study the examples
provided in Appendix V.
D.4. Congruential Mass Transformation
A congruential mass transformation, or CMT, is a general framework than can be applied to
transform a given source mass matrix into a target one. In particular all model reduction techniques
mentioned in H.3. Here it is specialized to the following case of importance in diagonalization:
(i)
The source mass matrix M S is nondiagonal and positive definite (PD); for example a CMM.
(ii) The target mass matrix MT is diagonal and nonnegative (that is, zero diagonal entries are
permitted)
Both matrices have order n D O F . The congruential transformation that converts source to target is
MT = HT M S H.
D6
(D.3)
D.4
CONGRUENTIAL MASS TRANSFORMATION
If H is nonsingular, the inverse mapping is M S = GT MT G, in which G = H1 . We will say

that M S and MT are congruentially linked through H. Even if M S and MT are both given and
parameter-free, there are generally many H matrices that satisfy (D.3). In fact the number of
solutions typically grows exponentially with n D O F .
One particular form, however, is unique under conditions (i)-(ii). Perform the Cholesky factorization
M S = L S LTS = L S U S , where L S is lower triangular and U S = LTS is upper triangular. If M S is PD,
1/2
this factorization is unique and both L S and U S are nonsingular [797]. Let MT be the principal
square root of MT , obtained by taking the positive square root of each diagonal entry. By inspection
1/2
H = U S MT ,
HT = MT L1
S .
1/2
(D.4)
This will be called the Cholesky form of H, and identified by subscript CF if necessary. Since
the inverse of a nonsingular upper triangular matrix is also lower triangular, and scaling by the
1/2
diagonal matrix MT does not alter that configuration, HT and H are lower and upper triangular,
respectively.
As an example, the CMM and DLMM of the two-node prismatic bar given in (?) are linked by the
Cholesky form

1
3 1
1.22474 0.707107
=
.
(D.5)
HC F =
2
0
1.41421
2 0
For the CMM and DLMM of the three-node prismatic bar studied in ?, the Cholesky form is

1.180303 0.288675 0.816497
5/2 1/(2 3) 2/3
(D.6)
HC F =
0
1.154701 0.816497 .
2/3 =
0
2/ 3
3/2
0
0
1.224745
0
0
The Cholesky form of (D.3) is unique and easy to obtain, but does not link naturally to the algebraic
Riccati equation mentioned below. For that purpose finding a symmetric H is more convenient.
Those will be identified by subscript Sy if necessary. Symmetric forms are not unique; in fact
typically one generally finds 2n D O F different solutions. It is rather easy, however, to extract a
principal solution.
For the two-node prismatic bar, the transformation (D.3) from CMM to DLMM with symmetric H
has 22 = 4 solutions. The only one with positive eigenvalues is

1.366025 0.366025
1 1 + 3 1 3
H Sy = 2
=
.
(D.7)
0.366025 1.366025
1 3 1+ 3
For the 3-node pristamic bar, one gets 23 = 8 solutions. The only one with all eigenvalues positive
is (only given numerically, as its analytical expression is complicated):

1.2051889
0.2051889 0.1472036
(D.8)
H Sy = 0.2051889
1.2051889 0.1472036 .
0.1472036 0.1472036 1.1518024
The determination of H in (D.3) is related to the quadratic matrix equation XT AX = B, where
M S A and MT B are data and H X the unknown. If H X is symmetric so XT = X,
the equation X AX = B is a specialization of the algebraic Riccati equation extensively studied in
optimal control systems [7,433]. Hopefully this interdisciplinary resource could be eventually be
applied to devise robust mass diagonalization schemes using a matrix function library [359]. But
as of now, templates remain the most practical method to find optimal diagonalizations.
D7
A Short History
of Mass Matrices
H1
Appendix H: A SHORT HISTORY OF MASS MATRICES
TABLE OF CONTENTS
Page
H.1
H.2
H.3
H.4
H.5
H.6
H.7
H.8
H.9
H.10
Pre-FEM Work
. . . . . .
Consistent Mass Matrices Appear
Dynamic Model Reduction . .
Selective Mass Scaling
. . . .
Singular Mass Matrices
. . .
Frequency Dependent Matrices .
Templates
. . . . . . . .
Multidimensional Elements
. .
Connection To Molecular Physics
Conclusions and Future Work
.
. .
.
. .
.
. .
.
. .
.
.
.
H2
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
H3
H3
H4
H4
H5
H5
H6
H6
H7
H7
H.2
CONSISTENT MASS MATRICES APPEAR
This Appendix summarizes previous developments on various topics addressed by this paper.
H.1. Pre-FEM Work
The first appearance of a structural mass matrix in a journal article occurs in two papers by Duncan
and Collar [190,191], which appeared in the mid 1930s. The authors were members of the world
famous aeroelasticity team at the National Physics Laboratory in Teddington (UK), led by Frazier.
As narrated in [238] those two papers represent the birth of Matrix Structural Analysis (MSA).
Befitting the overdesigned aircraft structures of the time, the focus is on dynamics and vibrations
rather than on statics. In [190] the mass matrix is called inertia matrix and denoted by [m]. The
first example [190, p. 869] displays the 3 3 diagonal mass of a triple pendulum. In the 1938 book
[282], which collects that early work plus intermediate papers, the notation changes to A.
Diagonally lumped mass matrices (DLMM) were strongly preferred in early publications. In fact
they dominate all pre-1963 work. Three reasons may be aduced for the preference:
The three-century astronomical heritage of Newtonian Mechanics. For unperturbed orbit

calculations, celestial bodies were idealized as point masses regardless of actual size.
Computational simplicity in vibration analysis and explicit direct time integartion (DTI).
Direct lumping gives an obvious way to account for nonstructural masses in simple discrete
models of the spring-dashpot-point-mass variety. For example, in a multistory building stick
model wherein each floor is treated as one DOF in lateral sway under earthquake or wind
action, it is natural to take the entire mass of the floor (including furniture, insulation, etc.)
and assign it to that freedom.
Nondiagonal (but not consistent) masses pop up ocassionally in pre-1960 aircraft matrix analysis
e.g. wing oscillations in [282,10.11] as a result of measurements. As such they necessarily
account for nonstructural masses due to fuel, avionics equipment, etc.
H.2. Consistent Mass Matrices Appear
The formulation of the consistent mass matrix (CMM) in structural mechanics by Archer [34,35]
was a major advance. Most of the CMM derived in Chapters 1719 appear in those papers. The
underlying idea, however, is older. In fact, Irons and Ahmad observe [397] that consistent masses
had been used in acoustics for over two decades before Archers papers; see e.g., the textbook [808].
The CMM at the system (master) level follow directly from the Lagrange dynamic equations
established in the late XVIII Century [430]. If T is the kinetic energy of a FEM-discretized
structural system occupying volume V , and u i (xi ) the velocity field defined by the nodal velocities
the master CMM is simply the Hessian of T with respect to nodal velocities:
collected in u,

2T
1
.
(H.1)
u i u i d V, u i = u i (u),
M=
T =2
u u
V
as happens in linear structural dynamics. Two key
This matrix is constant if T is quadratic in u,
decisions had to be reached, however, before this idea was applicable to FEM.
Localization: (H.1) is applied element by element, and the master M assembled through the standard
steps of the Direct Stiffness Method (DSM).
H3
Consistent Interpolation: the interpolation of velocities and displacements is made identical. That
is, velocity shape functions (VSF) and displacement shape functions (DSF) coincide.
These in turn had to wait until three major ingredients became slowly established during the 1960s:
(i) the DSM of Turner [759,761], (ii) the concept of shape functions progressively evolving in early
FEM publications [197,203,393,394,484], and (iii) the FEM connection to Rayleigh-Ritz. The last
one was critical. It was established in Meloshs thesis work [484,485]. The link to dynamics was
closed with Archers contributions, and CMM became a staple of FEM. But only a loose staple.
Problems persisted:
(a) Nonstructural masses are not naturally handled by CMM. In vehicle systems such as ships or
aircraft, the structural mass is only a fraction (10 to 20%) of the total.
(b) It is inefficient in explicit DTI, since M is never diagonal.
(c) It may not give the best results compared to other alternatives. Why? If K results from a
conforming displacement interpolation, pairing it with the CMM is a form of the conventional
Rayleigh-Ritz, and thus guaranteed to provide upper bounds on natural frequencies. This is
not necessarily a good thing. In practice it is observed that errors increase rapidly as one moves
up the frequency spectrum. If the response is strongly influenced by intermediate and high
frequencies, as in contact-impact and wave propagation dynamics, the CMM may give poor
results.
(d) For elements derived outside the assumed-displacement framework, the displacement shape
functions may be either unknown or altogether missing.
Problem (a) can be addressed by rigid mass elements accounting for inertia (and possibly gravity
or centrifugal forces) but no stiffness. Nodes of these elements are linked to structural (elastic)
nodes by multifreedom kinematic constraints. This is more of an implementation issue than a
research topic, although numerical difficulties typical of multibody dynamics may crop up.
Problems (b,c,d) can be attacked by parametrization. MacNeal was the first to observe [20,452,461]
that averaging the DLMM and CMM of Bar2 produced better results than using either alone. This
idea was further studied by Belytschko and Mullen [71] using Fourier analysis; they also studied
the CMM and SLMM of Bar3 but not their parametrizations. Krieg and Key [422] had emphasized
that in transient analysis by DTI the introduction of a time discretization operator brings new
compensation phenomena, and consequently the time integrator and the mass matrix should not
be chosen separately. The template approach addresses this problem by allowing and encouraging
customizing of the mass and stiffness to the problem at hand.
H.3. Dynamic Model Reduction
Concurrently with advances in variational mass lumping leading to the CMM, the 1960s and early
1970 were a fertile time for the development of reduced order dynamic models based on Component
Mode Synthesis (CMS), a name coined by Hurty [386,387]. These were motivated by the high cost
of dynamic and vibration analysis in the computers of the time, and blended well with the emerging
use of substructuring methods in aerospace engineering, summarized in [596].
Seminal publications of the period include [157,454,649], which originated the widely used CraigBampton and MacNeal-Rubin CMS methods. There is abundant literature since. Good textbooklevel descriptions are provided in [158,305].
H4
H.6
FREQUENCY DEPENDENT MATRICES
H.4. Selective Mass Scaling

The Selective Mass Scaling (SMS) method, proposed in the mid 2000s [522,523], has attracted
attention for rapid-transient simulations involving contact-impact, such as vehicle crash or explosions. Those are typically treated by explicit direct time integration (DTI), in which ephemeral
high frequencies produced by transient shocks may require extremely small timesteps for stability,
as well as producing significant spurious noise (pollution). In this approach, a diagonally lumped
mass matrix (DLMM) is augmented by a scaled version of the stiffness matrix. The underlyting
idea is to knock down the high frequency of the mesh modes, as quantitatively shown in ?
The derivation of the SMS variant for Bar3 in ? shows that for the three-node bar, SMS can be
presented as a subset of the general mass template. Nevertheless, it deserves consideration on its
own because of two attractive features:
It involves only one free parameter, which may be adjusted during the response simulation
process. This makes it especially suitable for multidimensional elements.
It does not depends on knowledge of element shape functions. In fact it may be used without
knowing the source of the FEM matrices, an attractive feature for some commercial codes.
These are counteracted, however, by two disadvantages:
Loss of low-frequency accuracy in the acoustic branch. Consequently SMS is not recommended for conventional structural dynamics and vibration analysis.
Adding the stiffness term necessarily makes M nondiagonal, complicating explicit DTI. If
the stiffness contribution is relatively small, however, DLMM diagonal dominance might be
retained, which permits the use of iterative schemes [547].
The method has been further explored in several recent publications, some of which focus on the
use of singular mass matrices as well as connection with parametrized variational principles (PVP)
[413,627,740,741]. The variational connection outlined in Appendix V, in which VSF and DSF are
independent, has the potential to link to that recent line of research as well as earlier work cited
there.
H.5. Singular Mass Matrices
This approach to RHFP has been primarily developed with an applied mathematics flavor and with
multibody dynamics as focus [834,762,763,764,765]. The key idea is to get the optical branch (or
branches) out of the way at low frequencies to increase the acoustoptical gap. It can be readily
subsumed under templates through spectral parametrization, as worked out for the Bar3 element in
23.1.11. While the BSSM instance constructed there shows promise in meeting both LFCF and
RHFP customization goals, more numerical experimentation will be required to substantiate that
promise.
H.6. Frequency Dependent Matrices

Making mass and stiffness frequency dependent (FD) was proposed by Przemieniecki [596], who
expanded both Me () and Ke () as Taylor series in 2 . An indirect derivation scheme, which
preceeds [596], consists of starting from dynamic transfer matrices and convert them to mass and
H5
stiffness by partial inversion. The procedure is described in detail in [572]. It is restricted to 1D

elements.
The idea was subsequently pursued by other authors. Pilkey [581,582] derived such matrices by
using exact solutions of the unforced EOM as shape functions. For the two-node bar, those matrices
are not instances of the FDMS template presented in ? because the baseline mass matrix for zero
frequency is the CMM.
Since such exact solutions are available only for a limited number of 1D models, the approach is
hardly extendible beyond prismatic bars and beams. More general 1D elements have been handled
by numerical ODE integration [438,838] on the way to transfer matrices.
The approach can be generalized to the template context by making free parameters frequency
dependent, as illustrated in 22.1.12 for Bar2. As noted there, this extension might be of interest
for problems dominated by a driving frequency, such as some electronic and optical devices. For
more general use, keeping the parameters frequency independent is far more practical. In multiple
dimensions it merges with boundary integral and spectral element methods in elastodynamics. These
are specialized topics beyond the scope of this historical review; for recent work and references,
see [438,683,686,806].
H.7. Templates
The template approach originally evolved in the late 1980s and early 1990s to construct highperformance stiffness matrices [228,229]. Its roots can be traced back to the Free Formulation
of Bergan [80,82,86,87,216], in which the stiffness matrix was decomposed into basic and higher
order components. A historical account is provided in a tutorial chapter [243]. The general concept
of template as parametrized forms of FEM matrix equations is discussed in [235,241,246].
Mass templates in the form presented here were first described in [236,237] for a Bernoulli-Euler
plane beam analyzed with Fourier methods. The study addressed mass-stiffness (MS) pairs. The
idea was extended to other elements in [248].
The Bar3 stiffness template (23.4) was first stated in [228], in which the only free parameter has a
slightly different definition. The Timoshenko beam model first appeared in [735]; see also [738].
The symbolic derivation scheme used for the Timoshenko beam EOM in (24.14) is due to Flaggs
[263]; see also [551,553,554]. The optimal static stiffness matrix (24.22) for the Ti-beam element
appeared originally in [758]. A detailed derivation may be found in [596,5.6]. It is an instance of
a template given in [245].
Two powerful customization techniques used regularly for templates are Fourier analysis and modified differential equations (MoDE). Fourier analysis are limited to separable systems but are straightforward to apply, requiring only undergraduate mathematics. (As tutorials for applied Fourier analysis, Hammings textbooks [335,334] are highly recommended.) MoDE methods, first published
in correct form in 1974 [783] are less restrictive but more demanding on two fronts: mathematical
ability and support of a computer algebra system (CAS). Processing power limitations presently
restrict MoDE to two-dimensional elements and regular meshes. In the present exposition, only
Fourier methods are used since those are likely to be more familiar to potential readers.
The selection of a priori constraint criteria for template free parameters is not yet on firm ground.
For example: is conservation of angular momentum useful in mass templates? The answer seem
to depend on element type and complexity.
H6
H.10
CONCLUSIONS AND FUTURE WORK
H.8.
Multidimensional Elements
.
Stiffness templates for 2D and 3D structural elements have been considered since a modest beginning
in the 1980s, as Free Formulation plate elements that incorporated a scaling parameter for the higher
order stiffness [87,216]. A significant number has been developed since; see references in H.7.
The development of mass matrix templates for multidimensional elements has lagged because of
four complications:
Mesh directionality effects that require angular averaging
Additional dependence on elastic material properties
Multiple plane wave types (pressure and shear waves in the case of an isotropic material)
The free parameter explosion as matrices get larger
The first study of this nature for a nontrivial 2D element has appeared as a 2012 thesis [320],
which dealt with triangular membrane elements with and without corner node drilling freedoms.
An example extracted from this thesis is presented in 25.2.
H.9. Connection To Molecular Physics
DLMM results for regular lattices of structural elements have counterparts in a very different area:
molecular physics. More precisely, the wave mechanics of crystalline solids created in the XX
Century by particle mechanicians; e.g., [101,601,832]. In crystal models, lattice nodes are occupied
by molecules interacting with adjacent ones. Thus the element dimension acquires the physical
meaning of molecular gap. Both acoustic and optical branches have physical significance.
In such models masses are always lumped at molecule locations, and atoms vibrate as harmonic
oscillators in the potential well of the force fields of their neighbors. Dispersion curves govern
energy transmission. In a linear atomic chain, the dimensionless wavenumber range [, ],
or [0, 2], is called the first Brillouin zone [108,402]. The happy connection of mass templates
to periodic materials may be of interest as FEM and related discretization methods are extended
into multiscale applications of crystal, micro- and nano-mechanics, and phononics [45,186].
H.10. Conclusions and Future Work
It is clear from the material presented in Chapters 21ff that mass matrix customization by templates
can be effective in structural dynamics. The examples of 22.1.9, 23.1.14 and 24.2.8 illustrates
two typical advantages:
Orders of magnitude improvements in frequency accuracy can be achieved for the same computational effort.
The space discretization need not be changed at all. Only the template free parameters need
to be adjusted by supplying the appropiate signature.
These should be attractive to engineers for practical FEM computations. The last one is particularly
important, since redoing a structural dynamics model not amenable to mesh generation may take a
significant portion of a design and analysis process.
Would availability of customized templates eliminate the need for h and p adaptivity? Certainly
not. Elements have performance limits, so such refinement schemes cannot be ruled out. It should
H7
be noted, however, that mesh adaptivity is less effective in dynamics, particularly in problems with
rapid transients and shocks. Irregular meshes and high order elements are notorious sources of HF
pollution, and adaptivity can make things worse by exacerbating nonphysical dispersion.
As regards future work, the most ambitious plan is extension to multiple space dimensions. Additional challenges emerge there:
Directionality. This means that the dynamic accuracy of the FEM model, as compared to
the continuum, depends on the direction of plane wave propagation. This is a phenomenon
missing in 1D. Integration averaging over propagation angles necessarily appears as another
component of the Fourier analysis.
Material property dependence. Optimal free parameters become dependent on additional

elastic material properties missing from the 1D treatment. For example if the material is
isotropic Poissons ratio appears. Actually this is not unique to mass templates, but affects
stiffness templates as well.
Multiple plane wave types. In isotropic 2D and 3D continua, one needs to consider pressure
(P-waves) and shear (S-waves). Plainly this impacts customization. If the medium is nonisotropic, more complicated wave types may have to be considered.
Parameter explosion. This can be expected to hinder symbolic calculations. At first sight it
seems inevitable given the rapidly increasing size of the element matrices. Growth could be
controlled, however, by making use of a priori reduction techniques such as those mentioned
in 21.3.1, 21.3.2, and 21.3.4.
The first three additional ingredients are illustrated for a simple 2D element (three-node linear
triangle) in 25.2. The number of parameters is kept to one therein by assuming a linear combination
of the CMM and DLMM.
Less ambitious research thrusts may focus on extending the results presented here for 1D elements.
For example:
Assess the performance of different template variant construction approaches to reduce highfrequency pollution caused in direct time integration. Three were described for the Bar3
element: singular mass, selective mass scaling and constant optical branch. As of this writing,
no comparison with numerical experiments is available.
Find out whether the impressive gains in accuracy observed in LFF-customized templates
survive in irregular meshes and/or heterogeneous element mixtures.
Unfinished business remains for the Timoshenko beam. First, the unexpected deterioration
in vibration accuracy as the coefficients and increase is presently unexplained. Could
the deterioration be arrested using MS template pairs? Prior experience with the BE-beam,
referenced in H.7, shows only modest improvements, but such continuum model is comparatively well behaved. Second, the relative performnace of the various template instances listed
in Tables 24.1 and 24.2 for direct time integration (DTI) remains to be assessed.
H8
Implementation
of Bar3 Template
in Mathematica
I1
Appendix I: IMPLEMENTATION OF BAR3 TEMPLATE IN MATHEMATICA
TABLE OF CONTENTS
Page
I.1
I.2
I.3
I.4
Auxiliary Modules . . . . . . . . . . . . . . . . . . .
I.1.1
Name To Signature Mapper . . . . . . . . . . . .
I.1.2
Mass Template Variant Parameter Mapper . . . . . . . .
Element Level Modules . . . . . . . . . . . . . . . . .
I.2.1
Element Mass and Stiffness Modules
. . . . . . . . .
Assembly Level Modules
. . . . . . . . . . . . . . . .
I.3.1
Lattice Master Mass and Stiffness Modules
. . . . . . .
I.3.2
Lattice Patch Modules . . . . . . . . . . . . . .
Dispersion Analysis and Display Modules . . . . . . . . . . .
I.4.1
Characteristic Equation Module . . . . . . . . . . .
I.4.2
Dispersion Branches And Taylor Series . . . . . . . . .
I.4.3
Dispersion Diagram Plotting . . . . . . . . . . . .
I2
I3
I3
I4
I5
I5
I6
I7
I8
I8
I8
I9
I11
I.1
AUXILIARY MODULES
This Appendix presents the computer implementation of the mass-stiffness template pair for the
three-node bar element, abbreviated to Bar3. It is written in the Mathematica language. Although
the element is admittedly simple it is not trivial. In fact the implementation illustrates the use of
template variants to simplify customization. Why Mathematica?. As observed in the Introduction,
use of a CAS is essential for template development because analytical derivations soon exceed
human endurance. Once the development phase is completed, a production version in a compiled
language can be easily produced. But the CAS version should not be discarded.
Bar3CharFreq
Forms the characteristic
equation for plane wave
propagation over a regular
Bar3 patch, and returns
characteristic frequencies
inlined
from
Bar3Dispersion
Given a template, finds its
characteristic frequencies
(using inlined results from
Bar3CharFreq) and their
Taylor series about =0
Bar3DispersionPlot
Given a template instance,
plot its DDD and/or
DGVD over a specified
range. Several display
options available
Bar3MassPatch
Extracts the 2 x 5 patch
equation mass matrix
from a 2-element lattice
patch for subsequent
Fourier analysis
Bar3MassLattice
Assembles the master
mass matrix of a regular,
homogeneous,Bar3
FEM-discretized lattice
with arbitrary # of elements
Bar3StiffLattice
Assembles the master
stiffness matrix of a regular,
homogeneous, Bar3
FEM-discretized lattice
with arbitrary # of elements
Bar3StiffPatch
Extracts the 2 x 5 patch
equation stiffness matrix
from a 2-element lattice
patch for subsequent
Fourier analysis
Bar3TempSignature
Given a template instance
mnemonic name (such as
CMM for Consistent Mass
Matrix), returns template
variant id and signature
Bar3ElemMassTemp
Returns mass matrix of
Bar3 element as template
instance. Six template
variants implemented:
GEN through COBB
Bar3ElemStiffTemp
Returns stiffness matrix of
Bar3 element as template
instance. Same form for
all template variants
Bar3MassVarParMap
Given a target mass
template variant, return a
rule to replace the general
template parameters
by those of the variant
Figure I.1. Organization of Bar3 template analysis modules presented in this Appendix.
The hierarchical organization of the modules presented in this Appendix is shown in Figure I.1. The
bottom-up description that follows starts from the lowest level of that chart, going up and traversing
against the arrows.
I.1. Auxiliary Modules
The two outside modules at the lowest level of the chart of Figure I.1 provide auxiliary services to
modules at all levels.
I.1.1. Name To Signature Mapper
Auxiliary module Bar3TempSignature, listed at the top of Figure I.2, maps an abbreviated template instance name to its full signature definition. It is invoked by
tsign=Bar3TempSignature[name]
(I.1)
The only argument is name: a character string of 3 or 4 letters that abbreviates a template instance.
Examples: "CMM" for the consistent mass matrix or "SLMM" for Simpson-lumped mass matrix.
I3
Table I.1. Bar3 Template Signature List Specification

Template
variant
Ref.
eqn.
# of Signature format, identified as

pars tsign in Mathematica code
Mass conservation
constraint
General, i pars
(23.2)
{ "GEN",{ },{ 1 , 2 , 3 , 4 } } 21 +2 +23 +44 = 0
General, i pars
(23.7)
{ "GEX",{ },{ 1 , 2 , 3 } }
preimposed
Lumped
(23.26)
{ "LUM",{ },{ L1 , L2 } }
2 L1 + L2 = 0
Spectral
(23.34)
{ "SPE",{ },{ S1 , S2 } }
preimposed
Selective
Mass Scaling
(23.44)
{ "SMS",{ },{ L1 , L2 , K } }
2 L1 + L2 = 0
{ "COBA",{ 1 },{ A } }
or { "COBB",{ 1 },{ B } }
preimposed
Constant Optical (23.47,23.49)

Branch
When doing symbolic work, the mass conservation constraint is not always preimposed in
some template variants, as that may complicate intermediate expressions
For this variant, two families: COBA and COBB, are implemented. Cf. 23.1.13
Names currently implemented can be gathered by examining the code. Some of these are also
listed in Table 23.1.
The function returns tsign as template signature. This is a list of the form
{ tvar,{ kpars },{ mpars } }
(I.2)
Here tvar is a character string that identifies template variants, kpars a list of stiffness parameters,
and mpars a list of mass parameters. Configuration details for this data structure are given in
Table I.1. If name is not recognized, a warning message is printed and "CMM" is assumed.
Example: Bar3TempSignature["SLMM"] returns { "LUM",{ 1 },{ 0,0 } } as function value.
I.1.2. Mass Template Variant Parameter Mapper
Auxiliary module Bar3MasVarParMap, listed at the bottom of Figure I.2, returns a replacement rule
that maps the four parameters of the Bar3 general mass template (23.2) to those of a variant. The
latter is called the target form. The rule is used to specialize results such as dispersion equations;
cf. the link drawn in Figure I.1. It is invoked by
rule=Bar3MasVarParMap[gmpars,tsign]
(I.3)
The arguments are:

gmpars
A list of symbols used for the free parameters of of the Bar3 mass template (23.2).
Normally the parameters are labeled { 1 ,2 ,3 ,4 }. Those symbols will appear
in the left side of the replacement rule.
tsign
Signature of the target form.
Items in gmpars must be individual symbols, while those in tsign may be symbolic or numeric
(see examples below). Note that in tsign is used if the target pertains to the SMS variant.
I4
I.2
ELEMENT LEVEL MODULES
Bar3TempSignature[name_]:=Module[{a1,a2,,L1,S1,S2,K,
modname="Bar3TempSignature"},
If [name=="CMM", Return[{"GEN",{1},{0,0,0,0}}]];
If [name=="SLMM", Return[{"LUM",{1},{0,0}}]];
If [name=="BLCD", Return[{"GEN",{1},{2,8,2,-4}/3}]];
If [name=="BLFM", a1=375^(1/4); a2=Sqrt[15];
Return[{"GEN",{1},{91-12*a1-7*a2,64-16*a2,
61-12*a1-a2,-92+12*a1+8*a2}/6}]];
If [name=="BLFD", =3/(4*(Sqrt[3]-1)); L1=5*(2-Sqrt[3]);
Return[{"LUM",{},{L1,-2*L1}}]];
If [name=="BSSM", =(5+Sqrt[10])/12; S1=3*(5+Sqrt[10])/2;
Return[{"SPE",{},{S1,0}}]];
If [name=="SMS1", K=1/24; =1/(1-12*K); Return[{"SMS",{},{0,0,K}}]];
If [name=="SMS2", K=1/2;
=1; Return[{"SMS",{},{0,0,K}}]];
If [name=="SMS3", K=2;
=1; Return[{"SMS",{},{0,0,K}}]];
If [name=="COB0", Return[{"COBA",{1},{-5/3}}]];
If [name=="COB1", Return[{"COBA",{1},{0}}]];
If [name=="COB2", Return[{"COBB",{1},{0}}]];
Print[modname,": illegal template name ",name," CMM assumed"];
Return[{"GEN",{0},{0,0,0,0}}]];
Bar3MassVarParMap[gmpars_,tsign_]:=Module[{tvar,,mpars,1,2,3,4,
1,2,3,4,1,2,3,L1,L2,S1,S2,K,A,B,s,d,r,kw,rep={}},
{1,2,3,4}=gmpars; {tvar,{},mpars}=tsign; kw=ToUpperCase[tvar];
If [kw=="GEN", {1,2,3,4}=mpars; rep={1->1,2->2,3->3,4->4}];
If [kw=="GEX", {1,2,3}=mpars; r=Sqrt[30]*Sqrt[1-3]-2*1; s=1+2;
d=1-2; rep={1->s-4,2->14-4*1-4*r,3->d+1,4->r-2}];
If [kw=="LUM", {L1,L2}=mpars; rep={1->L1+1,2->L2+4,3->1,4->-2}];
If [kw=="SPE", {S1,S2}=mpars; rep={1->(S1+S2-2)/3,2->(4*S2-38)/3,
3->(13-S1+S2)/3,4->(4-2*S2)/3}];
If [kw=="SMS", {L1,L2,K}=mpars; rep={1->1+L1+10*(4*+3)*K,
2->4+L2+160**K,3->1+10*(4*-3)*K,4->-2-80**K}];
If [kw=="COBA",{A}=mpars; rep={1->11-5*A/2,2->-2*(3+5*A),
3->6-5*A/2,4->-7+5*A}];
If [kw=="COBB",{B}=mpars; rep={1->8/3-5*B/2-5*B^2/72,2->32/3,
3->8/3-5*B/6+(5*B^2)/72,4->(5*B-16)/3}];
Return[rep]];
Figure I.2. Two auxiliary modules. Bar3TempSignature maps a template name to its signature.
Bar3MassVarParMap returns a replacement rule that maps the general mass template to a target variant.
The function returns

rule
Replacement rule. If the target is not recognized, the empty list { } is returned.
Example 1. Let gmpars={ 1 ,2 ,3 ,4 } and tsign={ "LUM",{ },{ L1 , L2 } }. The module

call Bar3MasVarParMap[tsign,gmpars] returns { 1 -> L1 +1,2 -> L2 +4,3 ->1,4 ->-2 },
a rule that maps the general template (23.2) to the lumped mass variant as per (23.25).
Example 2. Same gmpars as above but now tsign={ "LUM",{ },{ 0,0 } }. The call returns
{ 1 ->1,2 ->4,3 ->1,4 ->-2 }, a replacement rule that produces the SLMM instance.
I.2. Element Level Modules
This section describes element modules that return mass and stiffness templates.
I.2.1. Element Mass and Stiffness Modules
The Bar3 element mass and stiffness template modules are called Bar3ElemMassTemp and
Bar3ElemStiffTemp, respectively. They are listed in Figure I.3. The call sequences are
Me=Bar3ElemMassTemp[Le,rho,A,tsign,numer]
I5
(I.4)

Bar3MassTemp[Le_,_,A_,tsign_,numer_]:=Module[
{me=*A*Le,1,2,3,4,L1,L2,S1,S2,K,A,B,MKe,1,2,3,
m1,m2,m3,tvar,,mpars,kw,varOK,MeZ,Me,modname="Bar3MassTemp"},
{tvar,{},mpars}=tsign; varOK=False;
kw=ToUpperCase[tvar]; MeZ=Table[0,{3},{3}];
If [kw=="GEN", varOK=True; {1,2,3,4}=mpars;
Me=me*{{4+1,-1+3,2+4},{-1+3,4+1,2+4},
{2+4,2+4,16+2}}/30];
If [kw=="GEX", varOK=True; {1,2,3}=mpars;
m1=1+2; m2=1-2; m3=Sqrt[30]*Sqrt[1-3]-2*1;
Me=me*{{m1,m2,m3},{m2,m1,m3},{m3,m3,30-4*1-4*m3}}/30];
If [kw=="LUM", varOK=True; {L1,L2}=mpars;
Me=me*{{5+L1,0,0},{0,5+L1,0},{0,0,20+L2}}/30];
If [kw=="SPE", varOK=True; {S1,S2}=mpars;
Me=me*{{10+S1+S2, 10-S1+S2, 10-2*S2},
{10-S1+S2, 10+S1+S2, 10-2*S2},
{10-2*S2,
10-2*S2,
10+4*S2}}/90];
If [kw=="SMS", varOK=True; {L1,L2,K}=mpars;
MKe=K*me*Bar3StiffTemp[1,1,1,tsign,numer];
Me=me*{{5+L1,0,0},{0,5+L1,0},{0,0,20+L2}}/30+MKe];
If [kw=="COBA", varOK=True; {A}=mpars;
Me=me*{{ 6-A,
2-A,-2+2*A},
{ 2-A,
6-A,-2+2*A},
{-2+2*A,-2+2*A, 4-4*A}}/12];
If [kw=="COBB", varOK=True; {B}=mpars;
Me=me*{{96-36*B-B^2, 24-12*B+B^2, -48+24*B},
{24-12*B+B^2, 96-36*B-B^2, -48+24*B},
{-48+24*B,
-48+24*B,
384}}/432];
If [!varOK, Print[modname,": bad template var ",tvar,
", zero matrix returned"]; Return[MeZ]];
If [!numer, Me=Simplify[Me]];
Return[Me]];
Bar3StiffTemp[Le_,Em_,A_,tsign_,numer_]:=Module[
{ke=Em*A/Le,tvar,,mpars,Keb,Keh,Ke},
{tvar,{},mpars}=tsign;
Keb=ke*{{1,-1,0},{-1,1,0},{0,0,0}};
Keh=(4/3)**ke*{{1,1,-2},{1,1,-2},{-2,-2,4}};
Ke=Keb+Keh; If[!numer, Ke=Simplify[Ke]];
Return[Ke]];
Figure I.3. Bar3 element mass and stiffness template modules.
Ke=Bar3ElemStiffTemp[Le,Em,A,tsign,numer]
(I.5)
The arguments are:

Le
Element length
Em,A,rho Elastic modulus, cross section area, and mass density, respectively, of bar
tsign
Template signature. See Table I.1 for configuration details.
numer
Logical flag. If True, process in floating point. If False, process symbolically.
As function values the modules return

Me
3 3 element mass matrix
Ke
3 3 element stiffness matrix
I6
I.3
ASSEMBLY LEVEL MODULES
Bar3MassTempLattice[numele_,Le_,_,A_,tdef_,numer_]:=Module[
{tsign=tdef,e,i,j,ii,jj,eft,Me,M,numnod=2*numele+1},
If [Head[tdef]==String, tsign=Bar3TempSignature[tdef]];
M=Table[0,{numnod},{numnod}];
Me=Bar3MassTemp[Le,,A,tsign,numer];
For [e=1,e<=numele,e++, eft={1,3,2}+(e-1)*{2,2,2};
For [i=1,i<=3,i++, ii=eft[[i]];
For [j=1,j<=3,j++, jj=eft[[j]];
M[[ii,jj]]+=Me[[i,j]]
];
];
]; If [!numer, M=Simplify[M]];
Return[M] ];
Bar3StiffTempLattice[numele_,Le_,Em_,A_,tdef_,numer_]:=Module[
{tsign=tdef,e,i,j,ii,jj,eft,Ke,K,numnod=2*numele+1},
K=Table[0,{numnod},{numnod}];
Ke=Bar3StiffTemp[Le,Em,A,tsign,numer];
For [e=1,e<=numele,e++, eft={1,3,2}+(e-1)*{2,2,2};
For [i=1,i<=3,i++, ii=eft[[i]];
For [j=1,j<=3,j++, jj=eft[[j]];
K[[ii,jj]]+=Ke[[i,j]]
];
];
]; If [!numer, K=Simplify[K]];
Return[K] ];
Figure I.4. Bar3 master mass and stiffness assembler modules.

Bar3StiffTempPatch[Le_,Em_,A_,tdef_,numer_]:=Module[
{e,i,j,ii,jj,K},
K=Bar3StiffTempLattice[2,Le,Em,A,tdef,numer];
Return[{K[[2]],K[[3]]}] ];
Bar3MassTempPatch[Le_,_,A_,tdef_, numer_]:=Module[
{e,i,j,ii,jj,M},
M=Bar3MassTempLattice[2,Le,,A,tdef,numer];
Return[{M[[2]],M[[3]]}] ];
Figure I.5. Bar3 mass and stiffness patch extraction modules.
I.3. Assembly Level Modules

This section covers modules that work at the assembly (master) level. These are the midlevel four
pictured in Figure I.1.
I.3.1. Lattice Master Mass and Stiffness Modules
Modules Bar3ElemMassTempLattice and Bar3StiffTempLattice, listed in Figure I.4, assemble the master mass and stiffness matrices, respectively, of a prismatic homogeneous bar member
discretized as a regular lattice with a given number of elements. Since all elements are identical,
only one call to the appropriate element-level module is made. The returning matrix is reused in
the merge loop. The call sequences are similar:
M=Bar3MassTempLattice[numele,Le,rho,A,tdef,numer]
K=Bar3StiffTempLattice[numele,Le,Em,A,tdef,numer]
I7
(I.6)
(I.7)
The arguments are:

numele
Number of elements in lattice. The number of freedoms is numdof=2*numele+1
Le
Element length. Total member length will be Le*numele
Em,A,rho Elastic modulus, cross section area, and mass density, respectively, of bar
tdef
Template definition argument. Two possibilities:

If a list, tdef is taken to be the template signature tsign configured as shown in
Table I.1, and thus passed directly to the element module
If a character string (for example: "CMM"), tdef is interpreted as a template instance
abbreviation and Bar3TempSignature called as per (I.3) to build tsign, which is
then passed to the element modules
numer
Logical flag; see I.2.1

M
Master mass matrix of order ndofndof
Master stiffness matrix of order ndofndof
I.3.2. Lattice Patch Modules

Modules Bar3ElemMassPatch and Bar3StiffPatch, listed in Figure I.5, return the assembled
mass and stiffness matrices, respectively, of a patch of two identical Bar3 elements. This is done
by calling Bar3MassTempLattice and Bar3StiffTempLattice with numele=2 and returning
only the second and third equations. The call sequences are similar:
M=Bar3MassTempPatch[Le,rho,A,tdef,numer]
K=Bar3StiffTemppatch[Le,Em,A,tdef,numer]
(I.8)
(I.9)
The arguments are identical to those for the lattice master mass and stiffness modules, respectively,
described in I.3.1, except that numele is not supplied.
Mp
Patch mass equations as a coefficient matrix of order 2 5; see (?)(?)
Kp
Patch stiffness equations as a coefficient matrix of order 2 5; see (?)(?)
I.4. Dispersion Analysis and Display Modules

This section describe modules that produce and display dispersion diagrams. Those are the top
three shown in Figure I.1.
I.4.1. Characteristic Equation Module
Module Bar3CharFreq, listed in Figure I.6, forms the characteristic equation of a plane wave
propagating over a regular Bar3 lattice patch and solves it for the two characteristic frequencies.
The calling sequence is
{ detCm,
2aco,
2opt }=Bar3CharFreq[wavars,tdef,numer]
I8
(I.10)
I.4
DISPERSION ANALYSIS AND DISPLAY MODULES
Bar3CharFreq[wavars_,tdef_,numer_]:=Module[
{, ,,2, ,Bc,Bm,Kp,Mp,up,fm,fc,mfac,cfac,Cm,detCm,
sol,21,22,2aco,2opt}, {, ,,2, ,Bc,Bm}=wavars;
PlaneWave[k_,_,B_,x_,t_]:=B*Exp[I*(k*x-*t)];
Mp=Bar3MassTempPatch [1,1,1,tdef,numer];
Kp=Bar3StiffTempPatch[1,1,1,tdef,numer];
up={PlaneWave[,,Bc,-1, ],PlaneWave[,,Bm,-1/2, ],
PlaneWave[,,Bc, 0, ],PlaneWave[,,Bm, 1/2, ],
PlaneWave[,,Bc, 1, ]};
{fm,fc}=Simplify[ExpToTrig[(Kp-2*Mp).up]];
mfac=(Cos[/2+* ]-I*Sin[/2+* ]);
cfac=(Cos[* ]-I*Sin[* ]);
{fm,fc}=Simplify[{fm/mfac,fc/cfac}];
Cm={{Coefficient[fm,Bm],Coefficient[fm,Bc]},
{Coefficient[fc,Bm],Coefficient[fc,Bc]}};
detCm=Simplify[Det[Cm]];
sol=Simplify[Solve[detCm==0,2]];
2aco=21=Simplify[2/.sol[[1]]];
2opt=22=Simplify[2/.sol[[2]]];
{210,220}=Simplify[Limit[{21,22},->0]];
If [220==0, 2aco=22; 2opt=21];
Return[{detCm,2aco,2opt}]];
Figure I.6. Bar3 characteristic equation module.
The arguments are:

wavars
A list of symbols representing plane wave dispersion analysis variables, configured

as the list { , ,
,
2, ,Bc,Bm }, in which
Dimensionless wavenumber = k
Dimensionless space coordinate = x/ [1, 1] over patch.
Dimensionless characteristic frequency /c0
Dimensionless characteristic squared frequency
Dimensionless time = tc0 /
Bc,Bm Corner and midpoint wave component amplitudes, respectively

These must be individual symbols. No numbers or expressions should be in this list,
because they are internally used as variables. For instance, entering
*
or
^2
for
2 will cause errors.
tdef
Template definition argument; see I.3.1
numer
Logical flag; see I.2.1
The module returns the list { detCm,

2aco,
2opt }, in which
detCm
Determinant of characteristic matrix Cm as a function of and

2
2aco

a2 of acoustic branch (AB), expressed as function of
2opt

a2 of optical branch (OB), expressed as function of
The last two expressions:

2aco and
2opt, collectively define the dimensionless dispersion
diagram (DDD) for the template specified by tdef.
I9

Bar3Dispersion[v_,tdef_,{ma_,mo_},slevel_]:=Module[
{=v,kw,kw3,tsign=tdef,tvar,,mpars,1,2,3,4,L1,L2,
S1,S2,K,A,B,P,Q,R,c1,c2,c3,c4,c5,c6,c7,c8,c9,
assume,parmap,rep,2aco,2opt,2acos=Null,2opts=Null},
{tvar,{},mpars}=tsign; kw=ToUpperCase[tvar]; kw3=StringTake[kw,3];
parmap=MemberQ[{"GEX","LUM","SPE","SMS"},kw3]; assume=>=0&&>=0;
If [kw3=="COB", 2aco=12*(1-Cos[])/(5+Cos[]);
If [kw=="COBA", A=mpars[[1]]; 2opt=16/(1-A)];
If [kw=="COBB", B=mpars[[1]]; 2opt=432/(36-12*B-B^2)];
If [ma>=0, 2acos=Series[2aco,{,0,ma}]];
If [mo>=0, 2opts=Series[2opt,{,0,mo}]];
Return[{2aco,2opt,2acos,2opts}]];
If [kw=="GEN", {1,2,3,4}=mpars];
c1=4**(40+4*1+2+4*4); c2=16+2; c3=4*(4+4);
c4=4*(5+3+4); c5=60+4*2+1*c2; c6=16*3-c3;
c7=4**(2+c4); c8=c6+2*(3-1)-20; c9=c7-3*c2;
{c1,c2,c3,c4,c5,c6,c7,c8,c9}=Simplify[{c1,c2,c3,c4,c5,c6,c7,c8,c9}];
P=c1+3*c2+c9*Cos[]; R=c5-4*4-4^2+c8*Cos[];
Q=192**(Cos[]-1)*(c5-c3+c8*Cos[])+(c1+3*c2+c9*Cos[])^2;
If [parmap, rep=Bar3MassVarParMap[{1,2,3,4},tsign];
{P,Q,R}={P,Q,R}/.rep];
If [slevel==1, {P,Q,R}=Simplify[{P,Q,R},assume]];
If [slevel>1, {P,Q,R}=FullSimplify[{P,Q,R},assume]];
2aco=5*(P-Sqrt[Q])/R; 2opt=5*(P+Sqrt[Q])/R;
If [slevel==1, {2aco,2opt}=Simplify[{2aco,2opt},assume]];
If [slevel>1, {2aco,2opt}=FullSimplify[{2aco,2opt},assume]];
If [ma>=0, 2acos=Series[2aco,{,0,ma}]];
If [mo>=0, 2opts=Series[2opt,{,0,mo}]];
If [slevel==1,{2acos,2opts}=Simplify[{2acos,2opts},assume]];
If [slevel>1, {2acos,2opts}=FullSimplify[{2acos,2opts},assume]];
Return[{2aco,2opt,2acos,2opts}]];
Figure I.7. Bar3 dispersion module that returns dimensionless characteristic squared frequencies as
function of dimensionless wavenumber, and their Taylor series up to given order about = 0
I.4.2. Dispersion Branches And Taylor Series

Given a Bar3 template (or instance) definition, module Bar3Dispersion, listed in Figure I.7,
returns the dimensionless characteristic squared frequencies
a2 and
2o of the AB and OB as
function of the dimensionless wavenumber . This module was built by inlining symbolic results
produced by Bar3CharFreq with the goal of speeding up direct retrieval of those expressions. In
addition, this module can compute and return their Taylor series expansions about = 0 up to
specified orders. The calling sequence is
{
2aco,
2opt,
2acos,
2opts }= Bar3Dispersion[,tdef,{ ma,mo },slevel] (I.11)
The arguments are:
Dimensionless wavenumber
tdef
ma
If ma 0, return acoustic branch Taylor series (ABTS):

a2 expanded in about
= 0, up to and including order ma. If a negative integer, return Null.
mo
If mo 0, return optical branch Taylor series (OBTS):

2o expanded in about
= 0, up to and including order mo. If a negative integer, return Null.
slevel
Simplification level for output results: an integer 0, 1 or 2.

I10
I.4
DISPERSION ANALYSIS AND DISPLAY MODULES
Bar3DispersionPlot[_,tdef_,plotwhat_,range_,DVrange_,
imgsiz_,title_]:=Module[{v,tsign=tdef,tvar,kpars,mpars,
cat,2aco,2opt,2acos,2opts,min,max,Drange,Vrange,
style,pD=False,pV=False,dfun=$DisplayFunction},
{tvar,kpars,mpars}=tsign;
cat=tvar<>" "<>ToString[N[kpars]]<>ToString[N[mpars]];
If [plotwhat=="D"||plotwhat=="DV", pD=True];
If [plotwhat=="V"||plotwhat=="DV", pV=True];
{min,max}=range; {Drange,Vrange}=DVrange;
style={{AbsoluteThickness[1.80],RGBColor[1,0,0]},
{AbsoluteThickness[1.80],RGBColor[0,0,1]},
{AbsoluteThickness[1.80],RGBColor[0,0,0]}}; v=;
{2aco,2opt,2acos,2opts}=Bar3Dispersion[v,tsign,{-1,-1},1];
{2aco,2opt}={2aco,2opt}/.v->;
If [title!=" ", Print[title]];
If [$VersionNumber>=6.0, dfun=Print];
If [pD, Plot[{Sqrt[2aco],Sqrt[2opt],},{,min,max},
PlotStyle->style, Frame->True, PlotRange->Drange,
ImageSize->imgsiz, DisplayFunction->dfun,
PlotLabel->"DDD for "<>cat]];
caco=D[Sqrt[2aco],]; copt=D[Sqrt[2opt],];
If [pV, Plot[{caco,copt,1},{,min,max},
PlotStyle->style, Frame->True, PlotRange->Vrange,
ImageSize->imgsiz, DisplayFunction->dfun,
PlotLabel->"DGVD for "<>cat]];
Return[]];
Figure I.8. Bar3 dispersion diagram plotting module.
0: (or negative): no simplifications

1: ordinary simplification using the Simplify function
2: more exhaustive simplification using the FullSimplify function. Note: this
level should be used with caution. Reason: full simplification may return unexpected
weird results with terms involving Abs, or conditional expressions.
The module returns the list {
2aco,
2opt,
2acos,
2opts }, in which
2aco
AB dimensionless characteristic squared frequency

a2 expressed as function of
2opt
OB dimensionless characteristic squared frequency

2o expressed as function of
2acos
ABTS about = 0 up to and including order ma. If ma < 0, Null is returned.
2ocos
OBTS about = 0 up to and inclusing order ma. If mo < 0, Null is returned.
I.4.3. Dispersion Diagram Plotting

Given a Bar3 template instance (that is, with all-numeric signature) module Bar3DispersionPlot,
listed in Figure I.8, can plot its DDD, which includes the acoustic and optical branches returned
by Bar3Dispersion. It may also plot its DGVD, which is the ratio c = c/c0 of the FEM plane
wave speed c to that of the continuum wave speed c0 . The module has been used to produce all
Bar3 dispersion plots of this paper.
The call sequence is
Bar3DispersionPlot[,tdef,plotwhat,range,DVrange,imgsiz,title]
I11
(I.12)
The arguments are:
tdef
plotwhat A character string of the form "D", "V", or "DV". If the letter D appears, plot the
DDD (the so-called D-plot). If the letter V appears, plot the DGVD (the so-called
V-plot). If none of those strings is given, no plot is produced.
range
A 2-item list { min,max } that specifies the (horizontal) plot range for . It is used
for both DDD and DGVD plots. Usual range is { 0,2 Pi }.
DVrange
A list of the form { Drange,Vrange }. Drange is in turn a two-item list:

{
min,
max } that specifies the DDD plot range for
. Vrange has a similar configuration: { cmin, cmax } and specifies the DGVD range for c = c/c0 . Both
lists must be supplied even if only one plot is requested. Common specifications are
{ 0,8 } for Drange and { -2,2 } for Vrange.
imgsiz
Width of plot in points. Normally set to 300 to 400.
title
An optional character string to be printed before the plot. If " " no title appears.
The module does not return a value. Its output is the plot image object written to the Mathematica
default display function. (Its name changed in Version 6.0 from $DisplayFunction to Print.)
I12
Mass Templates in a
Variational Framework
V1
Appendix V: MASS TEMPLATES IN A VARIATIONAL FRAMEWORK
TABLE OF CONTENTS
Page
V.1
V.2
V.3
Variationally Derived Bar2 Mass Template

. . . . . . . . . .
Variationally Derived Bar3 Mass Template
. . . . . . . . .
A Comment on the Variational Formulations of Elastodynamics
. .
V2
V3
V4
V6
V.1
VARIATIONALLY DERIVED BAR2 MASS TEMPLATE
A question that may be interest to FEM theoreticians: can any mass template be produced by a
conventional variational framework? By conventional is meant based on shape functions injected
in the kinetic energy. More precisely: velocities are interpolated over the element from nodal
velocities using velocity shape functions (VSF), and the element kinetic energy T e evaluated by
integration. The mass matrix follows as the Hessian of T e with respect to nodal velocities, as per
(16.5). In short, a variationally derived mass matrix (VDMM). For practical template construction
and customization, the variational interpretation is superfluous, since templates can be expediently
postulated and algebraically customized. The reformulation may be worthwhile, however, for
mathematical investigations, as well as linkage to work conducted by other researchers.
Presently it is unknown whether the template-to-VDMM connection for arbitrary elements can be
established. It has been only investigated for the two simplest bar elements: Bar2 and Bar3. In
both cases, the general template was considered. The findings may be summarized as follows:
(1) VSF that reproduce the general template as a VDMM can be found. They are not unique.
(2) For any template instance that deviates from the CMM, the VSF do not coincide with the
displacement shape functions (DSF) used in the derivation of the element stiffness.
(3) VSF that deviate from the DSF are noninterpolatory and nonconforming with respect to nodal
velocities computed from the displacements by time differentiation. They do not necessarily
satisfy the unit-sum condition (also called partition of unity in the literature). A uniform
velocity field, however, must produce the exact kinetic energy.
Two simple elements are analyzed below.
V.1. Variationally Derived Bar2 Mass Template
We investigate whether the general one-parameter Bar2 mass template (22.3) can be produced as a
VDMM. The velocity field derived from the axial displacement u e (x, t) is u e (x, t) = d e (x, t)/dt.
Evaluation at the nodes yields the nodal velocities u 1 and u 2 , collected in u e = [ u 1 u 2 ]T . Let
N1 ( ) = (1 )/2 and N2 ( ) = (1 + )/2 denote the well known displacement shape functions
(DS)F of Bar2, being the usual iso-P natural coordinate. The element velocity interpolation is
taken to be
v e ( ) = u 1 Nv1 ( ) + u 2 Nv2 ( ),
(V.1)
in which the velocity shape functions (VSF) Nv1 and Nv2 are linked to the DSF through the linear
map
Nv1 ( ) = (1 + 12 1 ) N1 ( ) + 12 2 N2 ( ),
Nv2 ( ) = 12 2 N2 ( ) + (1 + 12 1 ) N1 ( ).
(V.2)
In (V.2), 1 and 2 are functions of the template parameter (but not of ), representing the deviations
of the VSF from the DSF. Note that prismatic bar symmetry is built-in: Nv1 ( ) = Nv2 ( ). The
1
2
associated kinetic energy T e is A (/2) 1 v e ( ) d , which can be evaluated either analytically
or through 2-point Gauss integration. Taking its Hessian with respect to u e gives a mass matrix
denoted by Me below. As for the Bar2 template, it is preferable to use the alternative form Me of
(22.4) rather than Me of (22.3) because solutions are simpler. Summarizing, the two matrices to
V3
(a)
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv1()
(b)
VSF: Bar2 mass template

instance with =0 (CMM)
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv2()
0.5
0.5
Nv1()
(c)

instance with =1/2 (BLFM)
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv2()
0.5
0.5
Nv1()
Nv2()
Abcissas of 2-Point Gauss Rule

instance with =1 (DLMM)
0.5
0.5
Figure V.1. Velocity shape functions (VSF) that produce the general Bar2 mass template (22.2) in a
variational framework, for three instances: (a) =0 (CMM); (b) = 1/2 (BLFM) and (c) = 1 (DLMM).
be matched are

Me
1
24
11
21

12
,
22
Me
1
12
3+
A
3
11 = 22 = 2(4 + 41 + 22 + 12 + 1 2 + 22 ),
12 = 21 = 4 + 41 + 82 +
= 1 + 2,
12
+ 41 2 +

3
,
3+
(V.3)
22 ,
= 12 ( 1),
in which (22.4) is reproduced for convenience.
On equating Me = Me we get four solutions: {1 = 3 , 2 = 1+ }, {1 =
1 , 2 = 1+ }, {1 = 3+ , 2 = 1 }, and {1 = 1 + , 2 = 1 }.
Only the last one reduces the VSF to DSF when = 0 or = 1. Inserting it into (V.2) and
simplifying yields

Nv1 ( ) = 12 (1 ) = 12 (1 1 + 2), Nv2 ( ) = 12 (1 + ) = 12 (1 + 1 + 2).

(V.4)
These VSF satisfy the conservation condition Nv1 + Nv2 = 1 for any and . They are plotted in
Figure V.1 for three instances: = 0 (CMM) = 12 (BLFM), and 1 (DLMM). If = 0, the VSF
depart from the DSF, and are plainly nonconforming.
display a distinguishing geometric feature: each VSF

The two VSF for = 1, namely (1 3)/2,
vanishes at one of the sample points = 1/ 3 of the 2-point Gauss rule; see Figure V.1(c). This
1
effectively orthogonalizes them in the sense that the kinetic energy cross integral 1 Nv1 Nv2 d
is zero. The result is the diagonal mass matrix MeL of (22.1).
Comparing the results (V.4) with the ansatz (V.2), plainly the latter was too elaborate. Little harm
is done, however, for this simple element. For more complicated ones, such as the Bar3 studied
next, a recursive adjustment is recommended using an interactive CAS.
V.2. Variationally Derived Bar3 Mass Template
Next we find whether the general mass template for Bar3 can be derived variationally. The well
known displacement shape functions are N1 ( ) = ( 1)/2, N2 ( ) = ( + 1)/2, and N3 ( ) =
V4
V.2
VARIATIONALLY DERIVED BAR3 MASS TEMPLATE
1 2 . The velocity interpolation is assumed to be

v e ( ) = u 1 Nv1 ( ) + u 2 Nv2 ( ) + u 3 Nv3 ( ),
(V.5)
in which
Nv1 ( ) = N1 ( ) 12 (1 2 ), Nv2 ( ) = N2 ( )+ 12 (1 +2 ), Nv3 ( ) = N3 ( )+(3+4 2 .
(V.6)
Here 1 through 4 are functions of the template parameters to be determined. The VSF ansatz
(V.6) was obtained after some simplifying initial computations. Note that prismatic bar symmetry
is preimposed: Nv1 ( ) = Nv2 ( ) and Nv3 ( ) = Nv3 ( ). The associated kinetic energy T e can
be evaluated either analytically or from 3-point Gauss integration. Taking its Hessian with respect
to u e gives the mass matrix

11 12 13
21 22 23 ,
31 32 33
= 2(2 + 51 + 512 + 32 + 322 ),
A
Me =
30
11 = 22
12 = 21 = 1 101 1012 + 62 + 622 ,

13 = 23 = 31 = 32 = (1 + 22 )(2 + 53 + 34 ),
(V.7)
33 = 2(8 + 1532 + 44 + 342 + 103 (2 + 4 )).

It is convenient to match Me to the 3-parameter, -form of mass matrix template (23.7) instead of
against (23.2). Matching entries gives 8 solutions, of which the one that yields 1 = 2 = 3 =
4 = 0 for the CMM (1 = 5/2, 2 = 3/2, 3 = 2/3) is picked:
1 = 1 1/2, 2 = 2 1/2, 3 = 33 /2 1, 4 = 1 22 + 3 54 /2, (V.8)
in which 1 = 2 /10, 2 = 1 /6, 3 = 3 /1 , and 4 = 5(1 3 /1 ). Except for the

CMM, these VSF do not verify the strong (pointwise) unit sum condition Nv1 + Nv2 + Nv3 = 1 for
each , but do satisfy the more lenient element mass conservation constraint
1
2
(Nv1 + Nv2 + Nv3 )2 d = 1.
(V.9)
In terms of the i , (V.9) is 1222 + 1532 + 103 (3 + 4 ) + 4 (10 + 34 ) + 42 (5 + 53 + 34 ) = 0.

The VSF produced by (V.8) are plotted in Figure V.2 for nine Bar3 mass instances, as labeled
therein. Except for the CMM they depart from the DSF, and are nonconforming. Some mass
matrix properties can be discerned visually:
For the diagonally lumped instances SLMMand BLFD shown in Figure V.2(b,e), two VSF
vanish at each of the sample points {0, 3/5} of the 3-point Gauss rule. Those points are
marked in the Figure. This feature effectively energy-orthogonalizes the VSF in the sense of
1
kinetic energy, since all cross integrals 1 Nvi Nv j d for i = j vanish. As a result, diagonal
mass matrices are produced.
V5
(a)
VSF: Bar3 mass template instance CMM
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv1() Nv2()
(d)
0.5
0.5
VSF: Bar3 mass template instance BLFM
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv1() Nv2()
0.5
0.5
Nv1()
Nv2()
0.5
0.5
0.5
0.5
0.5
0.5
0.5
Nv2()
0.5
Nv1() Nv2()
0.5
Nv2()
Nv1()
Nv3()
1
(i)
0.5
VSF: Bar3 mass template instance BSSM
0.5
0.5
VSF: Bar3 mass template instance COB1
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv3()
(f)
VSF: Bar3 mass template instance COB0
Nv1()
Nv3()
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv1() Nv2()
VSF: Bar3 mass template instance LCDM
Nv3()
(h)
VSF: Bar3 mass template instance BLFD
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv3()
(e)
VSF: Bar3 mass template instance SMS2
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv1() Nv2()
(c)
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv3()
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv3()
(g)
VSF: Bar3 mass template instance SLMM
1.5
1.25
1
0.75
0.5
0.25
0
0.25
Nv3()
(b)
Nv1()
Nv2()
Nv3()
0.5
0.5
Figure V.2. Velocity shape functions (VSF) that produce the general Bar3 mass template
(23.7) in a variational framework, for the nine labeled instances.
The VSF for the singular mass instance BSSM shown in Figure V.2(f), clearly displays linear
dependence among the VSF.
Aside from those special cases, it is difficult to draw general conclusions from a glance at Figure V.2
as to performance. For example, why does the VSF in (d) provides the best low frequency matching?
Shapes for say, (a) through (e) look quite similar (once youve seen one parabola ...). The obvious
conclusion: Fourier analysis is a much sharper tool in dynamics.
V.3. A Comment on the Variational Formulations of Elastodynamics
The use of VSF that differ from DSF dates back to the early days of FEM. It was done, for
example, in [203] for the HCT plate bending element, following suggestions by R. W. Clough.
(The consistent mass of that tricubic macroelement was quite complicated for hand derivations in
1966.) The idea can be incorporated into the well-known stationary-action variational principle
(VP) of elastodynamics, called Hamilton-Kirchhoff by Gurtin [321, p. 225], by weakening the
temporal kinematic link.
V6
V.3
A COMMENT ON THE VARIATIONAL FORMULATIONS OF ELASTODYNAMICS
That minor generalization of the primal VP of elastodynamics should not be confused with the use
of dual (also called complementary or reciprocal) forms. Research in that subject took off with
Toupins formulation [748] of a dual form of Hamiltons principle for a system of mass particles
with interaction impulses as unknown variables. For corrections and evolution into continua see
[196,718] and references therein. FEM applications to vibrations and dynamics emerged during the
early 1970s; see e.g., [277,300,717], but have stagnated since. Reason: impulse DOF are foreign
to the DSM, which dominates general purpose codes.
V7
Overview
of Dynamical
Systems
11
Chapter 1: OVERVIEW OF DYNAMICAL SYSTEMS
TABLE OF CONTENTS
Page
1.1.
1.2.
1.3.
1.4.
1.5.
1.6.
Scope
Dynamics versus Statics and Quasi-Statics
Overcoming Mustiness
Terminology and Notation
Systems Theory Terminology
Open Systems and Hierarchical Decomposition
12
13
13
14
15
16
18
1.2
DYNAMICS VERSUS STATICS AND QUASI-STATICS
1.1. Scope
This is a book about dynamics. Nowadays this term has acquired several meanings. It is used here
in the traditional sense:
The study of the relationship between motion and the forces affecting motion
This is in fact meaning (1a) in the American Heritage Dictionary of the English Language. As
such, it pertains to the science of Mechanics.1 . The corresponding adjectives are dynamic2 and its
equivalent dynamical.3
But even the traditional definition is far too broad in two respects. First, Mechanics embodies a
wide range of scales that span from cosmological through atomic and sub-atomic. Second, the
study can focus on three aspects: theoretical, applied and computational. Our focus is restricted to
a particular subset:
Classical Mechanics, which obeys Newtons laws. This allows the use of continuum (field)
models as well as certain lumped idealizations (point masses) that can be derived directly
from such laws.
Computational Mechanics, which relies on model-based simulation on digital computers. Of

the various discretization methods, our focus will be on the Finite Element Method (FEM).
The main applications of this subset are to Solid and Structural Mechanics. Although on first sight
this appears to be a bit limited in scope, many modeling and computational tools covered here
are application independent in the sense discussed in the IFEM book [?, Chapter 7]. Hence the
inclusion of qualifiers such as Structural Mechanics in the book title would be too confining.
1.2. Dynamics versus Statics and Quasi-Statics
Dynamic models in Classical Mechanics possess a common feature: the appearance of inertial
effects modeled by Newtons second law, which states that inertia forces are proportional to mass
times accelerations. Since accelerations are time derivatives of displacements, which characterize
the motion, the formulation inevitably leads to differential equations in time, whose solutions exhibit
time dependence. The converse is not true: time dependence does not necessarily require a dynamic
model, as discussed below.
At the other extreme lies statics. This is the equilibrium mechanics of stationary bodies. The
corresponding adjective is static, which means motionless, at rest, quiescent. Quantities associated
with stationary bodies do not vary with time, whence modeling is greatly simplified.
1
The term dynamics has nowadays acquired a generalized meaning beyond Mechanics, as illustrated by definition (2) of
that Dictionary: The physical and moral forces that produce motion and changes in any field or system.
Etymology: French dynamique, from Greek dunamikos: powerful, from dunamis, power, from dunashai, to be able. Its
use in terms of active, energetic, vigorous, forceful, and the like, is comparatively recent: 1856 (from Emerson).
The variant Dynamical tends to be often used in a more abstract sense. For example, MathWorld defines dynamical
system as a means of describing how a state evolves into another over the course of time. This is followed by
Technically, a dynamical system is a smooth action of the reals or integers into another object (usually a manifold).
This gobbledygook brings to mind one tongue-in-cheek comment about Lambs Hydrodynamics: one can read the whole
book without realizing that water is wet.
13
In between dynamics and statics lies the world of quasi-static scenarios, in which quantities vary
with time but do so slowly that inertial and damping effects can be ignored.4 For example one
may imagine situations such as a roof progressively burdened by falling snow before collapse, the
gradual filling of a dam over a decade, or the construction of a tunnel. Or foundation settlements:
think of the Pisa tower before leaning was stopped.5
By contrast dynamic analysis is appropriate when the variation of displacements with time is so rapid
that inertial effects cannot be ignored. There are numerous practical examples: earthquakes, rocket
launches, vehicle crashes, explosive forming, air blasts, underground explosions, rotating machinery, airplane flutter, dancing robots. The structural accelerations, which are second derivatives with
respect to time, must be kept in the governing equations.
Damping effects, which are usually associated with velocities (the first temporal derivatives of
displacements), may be also part of a dynamic model. Passive damping effects are often neglected,
however, since they tend to take energy out of a system and thus reduce the response amplitude.
Hence ignoring such kind of damping may lead to conservative designs.
Developments in the AFEM [?], IFEM [?], MFEMS [?], and NFEM [?] books pertain to statics
and quasi-statics. In this book, inertial effects will be always included, whereas damping effects
are occasionally considered.
Remark 1.1. Quasi-static behavior should not be confused with steady-state. The latter describes the response
to certain kinds of forced excitation (usually periodic) once effects of initial conditions disappear over the
course of time. The opposite of steady-state is transient, as in transient response. Those two qualifiers
pertain only to dynamic systems.
1.3. Overcoming Mustiness

Face it: Classical Mechanics smells musty. Its heyday was attained during the Victorian and
Edwardian eras: the world of Upstairs, Downstairs, just before relativity theory emerged. Reading
Euler, Lagrange or Hamilton one can imagine horse carriages, powdered wigs, feather pens and
blotting paper. But despite cowebs, it is not an obsolete subject. Far from it. But it needs to be
spruced up with modern language and tools. Three pick-me-ups are used in this book.
Systems Nomenclature. General System Theory (GST) emerged as a discipline by 1968 [775].
Certainly not musty: GST is still a vigorous topic that has brought about a fresh and holistic
approach to old questions. Lipstick on a pig? Perhaps, but terms like open system and system
environment are shiny lipstick that connect well with related topics such as control.
Linear Algebra. Matrices are not exactly spring chickens. They were invented by Cayley and
Sylvester in the mid XIX century. For over a century they were concealed behind suffocating
tracts on determinants such as [501]. Linear and matrix algebra come to the expository forefront,
however, once digital computers appeared in the early 1950s, and the baby boomer generation was
4
Quasi-statics includes statics as limit when the motion (or the time) is frozen. It does not include dynamics as the opposite
limit because inertial and damping effects cannot be recovered from quasi-static models. Thus the name is apt since
there is no such a thing as quasi-dynamics.
The quasi-static assumption can be done during design if dynamic effects can be accounted for through appropiate safety
factors. For many types of structures (e.g., buildings, bridges, offshore towers) these are specified in building codes.
This saves analysis time when dynamic effects are inherently nondeterministic, as in traffic, winds or water wave effects.
14
1.4 TERMINOLOGY AND NOTATION

Table 1.1. Some Terms Frequently Used in Dynamics
Term
Definition
Body
Mass
Position
Point
Particle
Any bounded aggregate of matter.

A measure of the resistance of a body to acceleration.
A location in space.
A geometric object devoid of any properties except position.
A body whose spatial extent and internal configuration are
irrelevant in a specific context.
A finite mass assigned to a position. Its spatial extent and
internal configuration are irrelevant.
A punctiform particle with a finite mass. (Abbreviation: PMP)
Point-Mass
Point-Mass Particle
Space and time are not defined as they are considered primitives.
trained in their use. Linear Algebra is not only the natural language of numerical computation but
that of the Finite Element Method. And it blends smoothly with the vector notation that is now the
bread and butter of college Physics.
Finite Elements. The Finite Element Method (FEM), as used today, came out in 1956 [758] as an
offspring of three interlaced developments: matrix structural analysis, energy methods, and digital
computers [238]. Although FEM has reached middle age, it retains sufficient vigor to freshen up
dynamics.
There is a surprising formal equivalence between certain widely-used dynamic FEM models and
point-mass particle dynamics. This one-to-one mapping is discussed in Chapter 2. Suffices to
say that decorating portions of the old musty stuff with the lipstick of modern tools makes the pig
tolerably attractive.
1.4. Terminology and Notation
Some definitions of terms used frequently in this book are collected in Table 1.1 for further reference.
Of those the definitions of mass6 and particle7 are the ones that have changed most over history.
Although particle and point-mass are often viewed as synonymous, they are not. The term particle is
more versatile, especially in molecular and atomic physics. Thus saying particle dynamics means
little until one specifies what kind of particles and effects we are talking about. On the other hand,
6
From Latin massa, a large irregular lump of something. Newton spoke of quantity of matter and defined mass indirectly
via proportions. The related term inertia is less specific, as it informally conveys inertness, inactivity, or sluggishness. In
Mechanics inertia is the property of matter by which it retains its state of rest, or its velocity along a straight line, unless
acted upon by an external force. This is actually Newtons First Law of Motion; cf. 3.2.1.
From Latin particula, a small part. As noted by Truesdell in a critical essay on the axiomatics of classical mechanics
[754, p. 512], the intended meaning has varied from author to author over the centuries. The roughly equivalent term
corpusculum, which means small body, is used by Newton and his contemporaries. In modern Physics, particle
generally implies portions of matter considered at molecular or atomic scales.
15
Table 1.2. Terminology from General Systems Theory

Term
Definition
GST System
GST State
An entity that maintains its existence through the interaction of its parts.
The relevant properties, values or characteristics of
a system component, or of the entire system.
A change in the state of the system, or part of a system.
A system event that initiates other events.
A set of entities and their relevant properties that are not part of the system,
but such that a change on any of which can produce a change in the system.
A system that is influenced by entities outside the system.
A self-contained system that is not affected by entities outside the system.
A system whose state changes over time. It may be open or closed.
GST Event
GST Behavior
GST Environment
GST Open System
GST Closed System
GST Dynamic System
By maintains its existence the definition of GST System emphasizes self-regulation processes.
point-mass particle dynamics or PMP dynamics is precise: it means Newtonian mechanics

applied to particles of finite mass. Its essentials have not changed over the past 4 centuries.
As implied by the book title, matrix notation will be heavily used. Both matrices and vectors are
identified by bold symbols. These symbols are lower case for vectors and upper case for matrices.
Occasionally an arrow will be placed above a vector symbol to emphasize that it is a field.
Real time will be always denoted by t. Derivatives with respect to t are abbreviated by superposed
dots. For example, if u(t) is a scalar motion, the associated velocity and acceleration are compactly
written
d2 u
du(t)
.
(1.1)
,
u
u
dt
dt 2
Sometimes we will denote velocity by v and acceleration by a when appropriate. Note that partial
differentiation with respect to t is not necessary when working with individual point-mass particles.
1.5. Systems Theory Terminology
The name General Systems Theory (GST), was introduced by von Bertalanffy in the late 1960
[775].8 Some GST definitions are collected in Table 1.2, taken from [4,659]. The contrast with
those in Table 1.1 is obvious. The GST definitions are nebulous: no images are conveyed. This
is a consequence of ambitious goals: GST was originally intended to apply conjointly to areas as
diverse as physical, climatological, biological, and even cultural systems, with initial emphasis on
self-regulatory closed systems.9
This kind of generality has a silver lining: it illuminates features that connect different disciplines
as long as they are not too discordant. For example: structural dynamics and control or, to make
8
GST development was helped by the success of operations research and control theory for servomechanisms during
World War II and its aftermath.
A contemporaneous but shorter-lived overreach: catastrophe theory, hampered by ridiculous claims from its founders, is
mercifully gone. GST is still around, but has wisely evolved toward concrete applications.
16
1.5
SYSTEMS THEORY TERMINOLOGY
Table 1.3. Systems-Related Terminology for Physical Systems

Term
Definition
System
Configuration
State
State Variables
Dynamic System
Response
Event
Behavior
Environment
Open System
Closed System
Interaction
A functionally related set of components regarded as a physical entity.

The relative disposition or arrangement of system components.
The condition of the system as regards its form, structure or constitution.
A set of variables that uniquely characterizes the system state.
A system whose state changes over time. It may be open or closed.
The value of the state variables as a function of time.
A change in the state variables produced by an agent.
A pattern of events.
A set of entities that do not belong to the system, but can influence its behavior.
A system that is influenced by entities outside the system (its environment).
A system that is not affected by entities outside the system.
The mutual effect of a system component, or group of such components,
on other components.
The action agents through which effects are transmitted between system
components, or between environment entities and system components.
Forces that act between system components.
Forces that act between environment entities and system components.
The set of all actions that can influence the state of a system, or component.
The external forces acting on the system, or component, as function of time.
The set of all quantities that characterize the state of a system, or component.
The response of the system, or system component.
A system, or system component, which may be viewed solely as a transformer
from input to output without knowledge of its internal workings.
A system, or system component, which is described in sufficient level of
detail to fully derive the transformation from input to output.
A system, or system component, which is described at a level of detail
intermediate between that of a white box and a black box.
Forces
Internal Forces
External Forces
Input (general)
Input (restricted)
Output (general)
Output (restricted)
Black Box
White Box
Grey Box
Several terms are adaptations to physical systems of corresponding ones in Table 1.2.
it a bit more general: multiphysics. Table 1.3 takes up several useful concepts suggested by GST,
and redefines them concretely for the dynamics of physical systems.10
Several changes, as well as a panoply of new terms, may be noticed. As regards system, the GST
phrase maintains its existence is gone: a physical system, or a component thereof, may cease to
exist through catastrophic events such as collision. The definition of state variables, missing in
Table 1.2, is important, since those will be the primary unknowns to be carried along and solved
for to get the system response. The notion of force, also missing there, is indirectly introduced as
an agent or carrier of change.11
The distinction between the system proper and its environment is an important contribution of
10
Meaning systems considered in the physical sciences: mechanics, thermomechanics, optics, electromagnetics, and so
on, at all observable scales.
11
The question of what is force is omitted since, as Joda says, it has no obvious answer.
17
;;;;;
;;;;;
;;;;;;;;;;
(a)
(b)
DYNAMIC
SYSTEM
Input: Forces
DYNAMIC
SYSTEM
Input: Forces
Output: Motion
ENVIRONMENT
Output: Motion
ENVIRONMENT
Figure 1.1. Open dynamic system that exchanges input and output with its environment: (a) black box
view; (b) grey box view. Double arrow symbols in the latter stand for component interaction effects.
GST. It allows a clear distinction between closed systems, in which the dynamic model includes
everything, and open systems, in which the environment is viewed as separate: environment entities
influence the system response but not the other way around. The distinction is crucial in practical applications since it allows engineers and scientists to focus on important interactions while
neglecting unimportant feedback.
The open vs. closed separation also offers an unambiguous way of classifying forces into external
and internal. Many expositions still struggle with this particular decomposition.
Note that definitions of input and output span two levels: general and specific. In a general
sense everything coming into the system from outside (its environment) is input, while everything
going out is output. The specific sense restricts these to external forces and state variable response,
respectively. This duality is necessary to accomodate the unsettled use of such terms in the literature.
Finally the definition of white, grey and black boxes responds primarily to te use of block diagrams
to show system schematics. The distinction is important in verification and validation testing.
1.6. Open Systems and Hierarchical Decomposition
All dynamical systems considered in this book are open. The system proper is distinguished from
its environment. An input-output relationship is established, as schematically shown in Figure 1.1.
In this figure, (a) depicts the system as a black box, which conceals internal details. On the other
hand, (b) gives some system details and thus may qualify as a grey box.
This view of open system can be continued hierarchically. Suppose the dynamic system is decomposed into first-level components. Each of these may be considered as an open subsystem, with
the rest of the system as its environment. A first-level component may in turn be decomposed into
second-level system components, which may also be viewed as open sub-subsystems, with the rest
of the subsystem as environment. And so on. This process is continued into as many levels as
necessary until further decomposition is deemed unnecessary.
This divide and conquer, hierarchical multistage decomposition is natural for many engineering
systems. Obviously it is not restricted to dynamics. It allows a problem to be decomposed into
subproblems, subproblems into sub-subproblems, etc. Often the lowest component level is populated with black boxes, meaning that only their input-output transformer relationships need to be
defined, while internal details of the transformer are of no relevance.
The Finite Element Method provides a systematic way to do multilevel decomposition using the
concept of superelements. For static problems this technique is described in Chapter 10 of [?].
18
PMP & LMDFE

Systems:
Kinematics
21
22
Chapter 2: PMP & LMDFE SYSTEMS: KINEMATICS
TABLE OF CONTENTS
Page
2.1.
Equivalence of Dynamical Systems

2.1.1. PMP-LMDFE Equivalence . . . . . . .
2.1.2. Generalization: More General Interaction Forces
2.1.3. Generalization: Continuum Models . . . .
2.2. Configurational Properties
2.2.1. Particle Coordinates . . . . . . . . . .
2.2.2. Velocities and Accelerations from Coordinates
2.2.3. Initial, Current and Final Configurations . . .
2.2.4. Initial Conditions . . . . . . . . . .
2.3. Motion Properties
2.3.1. Displacements
. . . . . . . . . . .
2.3.2. Velocities and Accelerations From Displacements
2.3.3. Kinematic Constraints . . . . . . . . .
2.3.4. Response and Trajectories . . . . . . .
2.3.5. Possible and Actual Displacements . . . . .
2.3.6. Virtual Displacements
. . . . . . . .
2.4. Mass Distribution Properties
2.4.1. Center of mass
. . . . . . . . . . .
2.4.2. Linear Moment . . . . . . . . . . .
2.4.3. Quadratic Mass Moment . . . . . . . .
2.4.4. Mass Moment of Inertia . . . . . . . .
2.4.5. Mass Product of Inertia
. . . . . . . .
2.4.6. Inertia Tensor
. . . . . . . . . . .
2.4.7. Inertia Tensor Transformations . . . . . .
22
. . . . . .
. . . . .
. . . . . .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
. .
. .
. .
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
. .
. .
. .
. .
. .
. .
.
. .
.
. .
.
. .
.
.
.
.
.
.
.
.
23
23
25
26
27
27
27
27
28
28
28
29
29
210
210
211
211
212
212
212
212
213
213
214
23
2.1
EQUIVALENCE OF DYNAMICAL SYSTEMS
2.1. Equivalence of Dynamical Systems

Classical dynamics reached maturity in the last half of the XIX Century. Over its 300+ year development, this discipline has been primarily concerned with point-mass particle or PMP systems.1
Such systems are composed by interacting particles of finite mass that evolve in time as per the
action of external and internal forces.
But the primary concern of this book are dynamic models constructed through the Finite Element
method (FEM). Why should we worry about PMP systems?
2.1.1. PMP-LMDFE Equivalence
The surprising answer is
Point-Mass Particle (PMP) systems with short-range interaction forces and

Lumped-Mass-Discretized Finite Element (LMDFE) dynamic models are
computationally equivalent.
The system classes referred to above may be easier to grasp by looking at Figure 2.1. Instances are
pictured as two-dimensional models for easier visualization. The model pictured in Figure 2.1(a)
is a classical PMP system: a set of point-mass particles interacting with their neighbors (particle
self interactions are not pictured to reduce clutter).2 The model shown in Figure 2.1(b) is a typical
FEM mesh that discretizes a circular domain with 3-node triangles, in which the distributed mass is
lumped at the nodes. This model only has translational freedoms, which are the node displacements.
The presence of rotational freedoms is discussed in the next subsection.
What does computational equivalence mean? That these seemingly disparate models produce
semidiscrete equations of motion (EOM) in one-to-one correspondence. For example, the matrix
EOM for linear, undamped dynamic models from either source can be written
M u + K u = f,
(2.1)
in which M, and K are the mass and stiffness matrices, respectively, u is a vector of displacements
from a reference position, and f the corresponding vector of known forces. Although the EOM
(2.1) for PMP and LMDFE models are constructed in completely different ways, once that is done
the systems have exactly the same form.
Table 2.1 shows an item-by-item correspondence. Note that there is none for one component of
the LMDFE system: elements. This means that element-by-element operations, which for those
models are the basis for forming and updating the semidiscrete EOM, have no PMP counterpart.
But if abstraction is made on how such operations are carried out (or are viewed as a black box)
the equivalence holds.
1
Recall that a PMP was defined in Chapter 1 as: a particle that possesses only two essential attributes: position in space
xi and finite mass m i . All other properties, such as spatial extent and internal structure, are irrelevant.
Node-to-node FEM interactions forces are not necessarily node colinear (that is, act along the line that join the two nodes)
if elements model a flexible medium. They are depicted that way in Figure 2.1 for visual simplicity.
23
(a)
24
(b)
Figure 2.1. Two dynamical models that are computationally equivalent in a formal sense: (a) a PMP
system with short-range interaction forces, with particle self interactions not drawn to reduce clutter;
(b) a Lumped-Mass-Discretized Finite Element (LMDFE) model built of 3-node triangles.
Why is the equivalence important? Knowledge reuse. Scientists from Newton through Poincare
developed the theory and practice of PMP systems over three centuries. The fruits of that labor
should not be allowed to go through waste just becase FEM and computers changed the way we do
dynamic simulations.
It is true that many of the classical expositions look cluttered and obscure in modern eyes. For
example, Euler, Lagrange and Hamilton published their masterpieces way before matrix and vectors
were invented, and lacked present knowledge of numerical methods and computation3 But their
discoveries and results endure. There is no point in reinventing the wheel and letting all that fine
work go to waste. What we can do is to present relevant excerpts of their results in a clearer and
more compact notation, and reinterpret them as necessary in light of current modeling tools and
computational methods.
Remark 2.1. Note that the formal equivalence holds only between two spatially discrete models. It completely
bypasses the sticky question of continuum model equivalence, which is briefly discussed in 2.1.3.
Remark 2.2. Both models in 2.1 have only translational degrees of freedom. An immediate generaliztion
is to allow for more general freedoms, such as, for instance, nodal rotations in FEM. On the PMP side, this
equivalence is easily covered by the dynamic of rotating bodies, which in turn is a special case of the general
description of Lagrangian dynamical systems by generalized coordinates.
In fact, Lagrange boasts in his famous book [227] about not having a single figure or diagram, and there are no numerical
examples. (Contrast with Euler and the Bernoullis, who favored a balanced mix of theories and applications.) This lack
of practical exercitation led Lagrange and others hands-off purists into various errors, some of which propagated for
several generations. Most errors were fixed, however, over the course of time.
24
25
2.1
EQUIVALENCE OF DYNAMICAL SYSTEMS
Table 2.1. Correspondence Between PMP and LMDFE Dynamical System Models
LMDFE system
PMP system
Node
Lumped node mass
Stiffness and damping forces
Applied node forces
Node position
Node displacements
Node velocities
Node accelerations
Element
Particle
Point mass
Interaction forces
Applied particle forces
Particle position
Particle displacements
Particle velocities
Particle accelerations
No counterpart
2.1.2. Generalization: More General Interaction Forces

Each of the dynamic models illustrated in Figure 2.1 obeys certain restrictions:
The PMP system in Figure 2.1(a) assumes short range interaction forces. More precisely, each
particle interacts only with its neighbors as well as itself. What is behind this restriction? In
the LMDFE model, the node-by-node interaction extends only over one element because of
the local support feature of FEM.
The LMDFE system in Figure 2.1(b) assumes that the mass of the discretized domain is lumped
at the nodes. This restriction is dictated by the definition of point-mass particle.
Can these restrictions be removed while mantaining computational equivalence? The answer is
yes, but equivalence must be understood in a less formal sense. To make the discussion concrete,
reference to the linear matrix EOM (2.1) will be convenient.
Long range interaction. In PMP systems, allowing each particle interact with all others, as well as
with itself, poses no problem. The effect on the EOM (2.1) is that the stiffness matrix K becomes
fully populated. In conventional FEM models, K is sparse on account of the local support of element
interpolation. There is no conceptual difficulty, however, in imagining LMDFE models based on
a global interpolation (across elements) that couples all nodal freedoms. Although such models
would be computationally inefficient, there are no laws that preclude them.
Generalized mass discretization. A lumped-mass FEM discretization scheme produces a diagonal
mass matrix M in (2.1). But more general mass discretizations, abbreviated by GMDFE, are in
common use in FEM, and lead to a nondiagonal M. An example is consistent mass lumping.
Accounting for this more encompassing approach has only one side effect on the EOM (2.1): the
mass matrix M becomes nondiagonal. Such a change is practically important only as regard computational efficiency of certain procedures, for example explicit time integration. Corresponding
PMP systems are less common because they would involve particle-to-particle interaction forces
that depend on accelerations.4
4
Some controversy exists in this regard. In his well known book on Analytical Dynamics, Pars [298] states that accelerationdependent forces would violate the tenets of Newtonian dynamics. This claim has been shown to be fallacious in a recent
paper [419].
25
26

No generally known
equivalence
Limit continuum models
Source continuum models
Finite element
discretization
Continuification
Point-Mass Particle (PMP)

systems with long-range
interaction
Informally
equivalent
Generalized
mass
discretization
Generalized
interaction
range
Point-Mass Particle (PMP)
systems with short-range
interaction
Generally-mass-discretized
finite element
(GMDFE) models
Formally
equivalent
Models
illustrated
in Figure 2.1
Lumped-mass-discretized
finite element
(LMDFE) models
Figure 2.2. Flowchart showing two generalizations levels beyond the equivalent
dynamic models of Figure 2.1.
In summary, lifting both restrictions has only side effects on the configuration of the matrices K
and M in the EOM (2.1). We can say that informal equivalence persists, since those changes have
consequences on computational efficiency but not on methodology. See Figure 2.2 for a diagramatic
representation of those extensions.
2.1.3. Generalization: Continuum Models
A more challenging extension is to continuum models, governed by field equations. This can be
done in two conceptually different ways:
Passing to the continuous limit in a PMP system.
Going back to the source continuum model from which a FEM discretization was derived.
Although some forms of continuum-discrete equivalence can be proven between specific systems,
the general case is still the matter of much debate and controversy in the physical sciences, as
recently reviewed in [232]. Thus the outer loop of Figure 2.2 is not yet closed. Since that kind
of correspondence is beyond the scope of this book, nothing more will be said about the topic.
The following sections summarize well known kinematic properties of PMP systems. The description largely follow standard expositions of Classical Dynamics. Some deviations are made,
however, when there is explicit need to link up with the finite element models that are the main target
of the book. In compliance with this objective, matrix and vector notation will be emphasized.
26
27
2.2
CONFIGURATIONAL PROPERTIES
2.2. Configurational Properties

The properties described here pertain to the composition of the system and its time evolution.
The PMP system consists of N particles of mass m i in 3D space, in which i denotes the particle
index. As regards terminology, for a equivalent FEM model, use the correspondences annotated in
Table 2.1; for example particle node, point mass lumped node mass, etc.
2.2.1. Particle Coordinates
The position coordinates of the i th particle are taken with respect to a fixed Rectangular Cartesian
Coordinate (RCC) reference frame O x yz with origin at O and axes {x, y, z}. The coordinates of
the i th point mass are {xi , yi , z i }.5 The coordinates are grouped into the 3-vector
xi = [ xi
yi
z i ]T .
(2.2)
All particle coordinates are grouped into a 3 N matrix form as

X = [ x1
x2
. . . xN ] =
x1
y1
z1
x2
y2
z2
...
...
...
xN
yN
zN

.
(2.3)
For some derivations it is useful to have all particle coordinates arranged as a 3N -vector configured
as
(2.4)
x = [ x1 y1 z 1 x2 y2 z 2 . . . x N y N z N ]T .
In a dynamical system, particle positions evolve in time, so xi = xi (t), etc. For all particles one
writes X = X(t) or x = x(t).
2.2.2. Velocities and Accelerations from Coordinates
Particle velocities and accelerations are obtained by taking derivatives of the position coordinates
with respect to time, viz.,
d 2 x(t)
dx(t)
.
(2.5)
,
x (t) =
x (t) =
dt
dt 2
These can be alternatively expressed in terms of particle displacements taken with respect to a fixed
reference configuration, as described later in 2.3.2.
2.2.3. Initial, Current and Final Configurations
Dynamic system simulations start at an initial time t0 , which is usually taken to be t0 = 0, and end
at a final time t f . A generic time t considered within the interval [t0 , t f ] is called the current time.
Positions occupied by the particles at initial, current and final times are called initial, current and
final coordinates, respectively.
The particle arrangements at t0 , t and t f are collectively called initial, current and final configuration,
respectively. Configurations will be labeled by calligraphic letters: C0 , Ct and C f , respectively, to
5
Note that if the axes were labeled {x1 , x2 , x3 }, an additional subscript would be needed, cluttering formulas.
27
28
avoid confusion with symbols such as C0 , which express function continuity order.6 The subscript
t in Ct may be dropped for brevity.
It is often convenient for brevity to exhibit the time as a subscript of the position coordinate symbols
(2.3) or (2.4); for example
X0 = X(0),
Xt = X(t),
X f = X(t f ),
x0 = x(0),
xt = x(t),
x f = x(t f ).
(2.6)
2.2.4. Initial Conditions

For a dynamic system the initial conditions specified at C0 include
Initial Position. The positions x0 = x(t0 ) of the particles at are given.
Initial Velocities. The velocities x 0 = x(t0 ) of the particles are given.
The former may be substituted by initial displacements, while the latter may be substituted by
initial momenta. If the motion is subject to kinematic constraints, as discussed in 2.3.3, the initial
conditions are assumed to satisfy those restrictions.
2.3. Motion Properties
This section introduces terminology associated with specific aspects of the kinematics of motion.
Displacements are given a key role comparable to that of particle coordinates. This balanced
viewpoint is motivated by finite element applications; see Remark 2.1 at the end of 2.3.2.
2.3.1. Displacements
Displacements measure change in particle positions. The vector symbol for displacements is u,
with its O x yz components denoted by u x , u y and u z . Suppose that the i th particle moves from
xi (t1 ) to x(t2 ). The corresponding displacement vector is

ui (t2 , t1 ) = xi (t2 ) xi (t1 ) =

u xi (t2 ) u xi (t1 )
u yi (t2 ) u yi (t1 ) .
u zi (t2 ) u zi (t1 )
(2.7)
If t1 = t0 , the initial time, and t2 = t, the current time, the notation is simplified by dropping the
arguments:
ui = ui (t) = ui (t, t0 ) = xi (t) xi (t0 ).
(2.8)
This ui is called the total displacement of the particle, or simply its displacement unless a distinction
need to be made. For the set of all particles a 3 N displacement matrix arranged like X in (2.3)
is rarely used. More convenient is to have all displacements arranged as a 3N -vector configured as
u = [ u x1
6
u y1
u z1
u x1
u y1
u z1
. . . ux N
uyN
u z N ]T ,
(2.9)
State and configuration should not be confused. The former term is quantitative: it refers to values of the state variables,
which is a set that fully describes the system and its response to inputs. The latter term is qualitative: it refers to the
spatial arrangement of the particles that make up the system. Emphasis is on form visualization or, informally, how the
system is set up.
28
29
2.3 MOTION PROPERTIES
in direct correspondence to the position coordinate vector displayed in (2.4). The omission of
arguments in (2.9) implies change from the initial configuration C0 to to the current one C. For
other time intervals, arguments may be used as necessary for clarity. Thus
u = u(t) = u(t, t0 ) = x(t) x(t0 ) = xt x0 ,
u(t2 , t1 ) = x(t2 ) x(t1 ).
(2.10)
For equivalent FEM models, u denotes the node displacement vector; cf. Table 2.1. In this case
the entries in u are ordered node-by-node.
2.3.2. Velocities and Accelerations From Displacements
As described in 2.2.2, particle velocities and accelerations result on taking time derivatives of the
position coordinates; cf. (2.5). Since the initial configuration is fixed, they may be also obtained
by taking time derivatives of total displacements:
u(t)
=
d(x(t)x(t0 ))
du(t)
=
= x (t),
dt
dt
u(t)
=
d 2 u(t)
d 2 (x(t)x(t0 ))
=
= x (t).
dt 2
dt 2
(2.11)
Again for brevity time arguments may be dropped, or replaced by appropriate subscripts. For
0 ), etc.
example, x = x t = x (t), u 0 = u(t
Remark 2.3. In many textbooks and monographs on Classical Mechanics, total displacements are conspicuous
for their absence. One may look in vain for symbols such as (2.7) through (2.10). More likely is to find xi
for incremental displacements and xi for virtual displacements (or xi and xi if the exposition uses vectors).
In sharp contrast, node displacements are the key state variables in the Direct Stiffness Method (DSM) of
FEM, and play a fundamental role therein. Since FEM dynamic models are our ultimate objective, both
displacements and position coordinates will be used in a balanced fashion.
2.3.3. Kinematic Constraints

Kinematic constraints are restrictions on the motion of a dynamic system. Although those restrictions are studied in greater detail in Chapter 5, we introduce here basic definitions required to discuss
the distinction between possible and virtual displacements in 2.3.6. Two possible constraint forms
are of interest. Holonomic constraints involve particle coordinates and possibly time:
ci (x, t) = 0,
i = 1, . . . m h ,
(2.12)
A holonomic constraint that does not depend on time: ci (x) = 0, is called rheonomic. It is
called scleronomic otherwise. Nonholonomic constraints are those that involve particle coordinates,
particle velocities, and possibly time, and that cannot be integrated into a holonomic form. In this
subsection we will consider only the subclass that is linear in velocities:
di (x, t) x + gi (x, t) = 0,
i = 1, . . . m n ,
(2.13)
The total number of constraints, m = m h + m n , must be less than the total number of degrees of
freedom n DOF of a N -particle system, which is n DOF = 3N in 3D and n DOF = 2N in 2D. (Note
that either m h or m n may be zero.) Differentiating (2.12) with respect to time yields
ci (x, t)
ci (x, t)
x +
,
x
t
29
i = 1, . . . m h ,
(2.14)
210
The nonholonomic constraints (2.13) and the differentiated holonomic constraints (2.14) can be
combined in the compact matrix form
A x = g,
(2.15)
in which A = A(x, t) is a m n DOF matrix and g = g(x, t) is a n DOF -vector. Multiplying both
sides by dt gives the differential form
A dx = g dt.
(2.16)
A x = 0.
(2.17)
The homogeneous version of (2.15) is

Form (2.17) results if all holonomic constraints are rheonomic (that is, do not depend on time), and
the velocity-independent terms gi of all nonholonomic constraints vanish.
Note that the compact matrix forms (2.15) through (2.17) only apply if the nonholonomic constraints
(2.13) are linear in the velocities, or if all constraints are holonomic, that is, m n = 0. A matrix
form that accomodates nonlinear-in-velocities nonholonomic constraints is worked out later.
Example 2.1. Suppose that a single particle moving in 3D space and referenced to the O x yz frame is subjected
to the two constraints

x1 + y1 + z 1 =
1
2
B t 2,
x1 x1 + y1 y1 + z 1 z 1 = 3 C,
(2.18)
in which x1 , y1 and z 1 denote the coordinates of the particle while B and C are coefficients that restore the
proper physical dimensions. Time differentiation of the first (holonomic) constraint gives x1 + y1 + z 1 = B t.
Combining with the second (nonholonomic) constraints yields
1
x1
1
y1
1
z1

x1
y1
z 1
Bt
,
3C
(2.19)
which befits the matrix form (2.15). Here A is a 2 3 matrix.
2.3.4. Response and Trajectories

The vector function x(t) for t0 t t f that satisfies the equations of motion, initial conditions
and kinematic constraints is called the response history, or simply response, of the PMP system.
By analogy the associated functions u(t) = x(t) x(0), u(t)

= x (t), and u(t)
= x (t), are called
the displacement, velocity and acceleration response, respectively, of the system.
If the focus is on an individual particle, say the i th one, the locus of its position history xi (t) is called
the trajectory or orbit of the particle. By extension, x(t) is sometimes called the system trajectory,
especially in control system applications.
2.3.5. Possible and Actual Displacements
The system is assumed to be in configuration Ct at current time t. We now take a deeper look
at motions in that neighborhood. The particle displacements at Ct are ut = u(t) and u t = u(t),
respectively. These are assumed to satisfy the EOM as well as any kinematic constraints such as
is called the actual velocity at Ct .
(2.12) and (2.13). Under these assumptions, u t = u(t)
210
211
2.4
MASS DISTRIBUTION PROPERTIES
Consider the displacement change between the current configuration at t and a possible one taken
at an infinitesimal time increment t + dt:
du P = x(t+dt) x(t),
or
du P = u(t+dt) u(t).
(2.20)
By possible configuration is meant one that satisfies the kinematic constraints grouped in (2.15),
at t + dt. If so du P is called a possible displacement increment or simply possible displacement.7
Notice the omission of the time arguments in du P for brevity.
If a possible displacement satisfies the equations of motion at Ct , in addition to the constraints, it is
called a actual displacement increment or simply actual displacement. If the actual velocity u t (t)
at Ct is unique, the actual displacement du = u t dt is also unique.
2.3.6. Virtual Displacements
The vector of virtual displacements, denoted by u, plays a central role in analytical mechanics. If
there are no restrictions on the motion, any set of displacements relative to that at Ct qualifies. In
this case, u coalesces with the possible displacements defined by (2.20), whence u du P .
The definition becomes more subtle when there are kinematic constraints, which are collectively
described by (2.15). Several equivalent definitions are possible; three of which are listed next.
Algebraic Definition. Consider the homogeneous version (2.17) of (2.15). Any n DOF -vector that
satisfies this homogeneous equation is a virtual displacement vector. More formally: multiply both
sides of A x = 0 by dt, and call u = x dt. Thus
A u = 0.
(2.21)
Slightly different track favored by some authors: take the differential version (2.16), set g = 0 and
replace dx by u.
Definition From Possible Displacements. Consider two possible (and distinct) displacements du1P
and du2P taken fromthe same configurationCt . Their difference is a virtual displacement:
u = du2P du1P .
(2.22)
Equivalence with the previous definition is readily established; see page 133 of [390].
Variational Calculus Definition. In the variational framework of PMP dynamics, either x or u is
considered the state variable vector, whereas time t is the independent variable. Then x or u is
the conventional state variation taken at frozen time t.
Remark 2.4. Readers should be warned that many expositions have conceptual glitches as regards virtual
displacements. Some define it as a displacement compatible with the constraints, which is wrong unless
(2.15) is homogeneous to begin with. The correct assertion is a displacement compatible with the homogenized
constraints. Even more common is the statement that virtual displacements are infinitesimal. That is nonsense.
By the algebraic definition A u = 0. Therefore A (c u) = 0 where c is an arbitrary factor, whence the virtual
displacement order-of-magnitude is irrelevant: microns, kilometers or light-years make no difference.
To exemplify how the aforemetioned glitches appear even in reputable expositions, consider the following
statement in [347, p. 50]: A virtual displacement is an arbitrary, instantaneous, infinitesimal change of the
position of the system compatible with the conditions of constraint. Score: two out of four.
7
Other names for du P in the literature are real displacements, admissible displacements and feasible displacements. Some
expositions replace the differential symbol d by the increment symbol ; this makes no essential difference.
211
212
2.4. Mass Distribution Properties

This section collects properties associated with the distribution of masses in PMP systems and
equivalent LMDFE models. As noted in Remark 2.5, these can be extended without difficulty to
continuum mass distributions.
2.4.1. Center of mass
The center of mass of a PMP system is the point located at

i m i xi

m
x
1
1
i
i
=
m i xi =
(2.23)
xC = i
m i yi ,
m tot i
m tot i m z
i mi
i i
i

in which i ranges from 1 through N , and m tot = i m i is the total mass. For a continuum body,
see Remark 2.5.
2.4.2. Linear Moment
The linear mass moment of an N -particle PMP system with respect to a point Q of coordinates
x Q = [ x Q y Q z Q ]T is the 3-vector

i m i xi Q
pQ =
m i (xi x Q ) = i m i yi Q ,
(2.24)
i
i m i zi Q
in which i ranges from 1 through N , xi Q = xi x Q , yi Q = yi y Q , and z i Q = z i z Q . If point
Q is the center of mass C, the three components of p Q pC vanish. For a continuum body, see
Remark 2.5.
2.4.3. Quadratic Mass Moment
The quadratic mass moment of an N -particle PMP system with respect to a point Q of coordinates
x Q = [ x Q y Q z Q ]T is the 3 3 symmetric matrix

m
x
x
m
x
y
i
i
i
Q
i
Q
i
Q
i
Q
i
i
i m i xi Q z i Q

, (2.25)
PQ =
m i (xi x Q ) (xi x Q )T = i m i yi Q xi Q
i m i yi Q yi Q
i m i yi Q z i Q

i
i m i z i Q xi Q
i m i z i Q yi Q
i m i zi Q zi Q
in which i ranges from 1 through N , xi Q = xi x Q , yi Q = yi y Q , and z i Q = z i z Q . Taking Q
to be the center of mass minimizes P Q in some norms. For a continuum body, see Remark 2.5.
Remark 2.5. For a continuum body of density and volume
, the foregoing definitions apply if summa-
tions are replaced by integrations over the volume. More precisely, (2.23), (2.24), and (2.25) are replaced,
respectively, by

x dV
,
xC =
dV
(xx Q ) d V

p Q =
(yy Q ) d V ,

(zz Q ) d V
(xx
)
(xx
)
d
V
(xx Q ) (yy Q ) d V
(xx Q ) (zz Q ) d V
Q
Q

P Q =
(yy Q ) (xx Q ) d V
(yy Q ) (yy Q ) d V
(yy Q ) (zz Q ) d V .

(zz Q ) (xx Q ) d V
(zz Q ) (yy Q ) d V
212
(zz Q ) (zz Q ) d V
(2.26)
213
2.4
(a)
dai m
i
MASS DISTRIBUTION PROPERTIES
(b)
(c)
plane p
dpi
plane q
d
mi qi
xi
x
axis a
O
yi
mi
zi
Figure 2.3. Moments and product of inertia of an individual particle of mass m i : (a) mass moment of inertia
with respect to a line L at distance di ; (b) mass product of inertia with respect to two orthogonal planes at
distances di1 and di2 , respectively; (c) mass moments and products of inertia with respect to a RCC frame
Ox yz.
2.4.4. Mass Moment of Inertia

The mass moment of inertia,8 is closely related to the quadratic mass moment defined in 2.4.3
but is taken with respect to an axis instead of a point. It is a measure of the resistance of a massendowed object to rotation. It plays roughly the same role in rotational dynamics as mass does
in translational dynamics. It connects related quantities such as angular momentum and angular
velocity, as well as torque and angular acceleration. There is an associated quantity called the mass
product of inertia, which completes the formation of the moment of inertia tensor, and appears in
transformation equations. In the sequel the mass qualifier is often implied and may be omitted for
brevity; for example moment of inertia written instead of the more precise mass moment of inertia.
The mass moment of inertia of a particle of mass m i with respect to a given axis a is Ia = m i dai2 ,
where dai is the distance from the mass point to a (the perpendicular distance from a to the mass
point). See Figure 2.3(a). For a PMP system of N particles the moment of inertia is obtained by
summation:

m i dai2 , i = 1, . . . N .
(2.27)
Ia =
i
2.4.5. Mass Product of Inertia

Connected with moments of inertia is a quantity called the mass product of inertia. This is not very
useful by itself, but completes the formation of the moment of inertia tensor described in the next
subsection, and appears in its transformation equations.
The mass product of inertia of a particle of mass m i with respect to a pair of given perpendicular
planes p and q is m i d pi dqi , in which d pi and dqi are signed distances from the mass point to the
planes. See Figure 2.3(b). For a PMP system of N particles, the product of inertia is obtained by
summation:

m i d pi dqi , i = 1, . . . N .
(2.28)
I pq =
i
Also called rotational inertia, polar mass moment of inertia and angular mass
213
214
2.4.6. Inertia Tensor

To study properties when rotation axes are not known in advance, it is convenient to introduce the
system mass inertia tensor with respect to a RCC frame Ox yz. This tensor is fully defined by a
system mass inertia matrix. This is a symmetric 3 3 matrix formed by grouping mass moments
and products of inertia as

Ix x
Ix y Ix z
(2.29)
I O x yz = I yx
I yy I yz
I yx Izy I yz
in which
Ix x =
m i (yi2 + z i2 ),
Ix y = I yx =
I yy =
m i (z i2 + xi2 ),
m i xi yi ,
I yz = Izy =
Izz =
m i (xi2 + yi2 ),
m i yi z i ,
Izx = Ix z =
m i z i xi .
(2.30)
Here xi . yi and z i are coordinates of the i th particle of mass m i ; see Figure 2.3(c).
2.4.7. Inertia Tensor Transformations
Consider an axis a passing through the RCC frame origin O, having direction cosines ax , a y and az
with respect to x, y and z, respectively. Define the unit-vector t = [ ax a y az ]T . Then the moment
of inertia with respect to a is given by
Ia = tT I O x yz t.
(2.31)
Next consider a totated RCC system {a, b, c} with origin at O Construct the transformation matrix

ax bx cx
T = ay by cy
az bz cz
whose columns store the direction cosides of {a, b, c}, respectively, with respect to {x, y, z}. The
transformed inertia tensor is
I Oabc = TT I O x yz T.
The associated eigenvalue problem is Ix yz c j = I j c j . The eigenvalues I j , j = 1, 2, 3, are the
principal moments of inertia. Since Ix yz is symmetric real, the three eigenvalues are positive. It can
be shown that Ix yz is also nonnegative, whence the eigenvalues are nonnegative. They are typically
ordered as
(2.32)
I1 I2 I3 0.
The associated eigenvectors form the principal inertia directions.
(TBC)
214
PMP & LMDFE

Systems:
Kinetics
21
22
Chapter 2: PMP & LMDFE SYSTEMS: KINETICS
TABLE OF CONTENTS
Page
2.1.
2.2.
2.3.
2.4.
Introduction
Newton Laws of Motion
2.2.1. First Law . . . . . . . . . . . .
2.2.2. Second Law
. . . . . . . . . .
2.2.3. Third Law
. . . . . . . . . . .
Forces
2.3.1. Concept of Force . . . . . . . . .
2.3.2. Force Decomposition . . . . . . . .
2.3.3. Action-Reaction (AR) Force Decomposition
2.3.4. Internal-External Force Decomposition . .
Particle Forces
2.4.1. External and Internal Forces . . . . .
2.4.2. Resultant Force and Moment
. . . . .
22
. . . . . . .
. . . . . . .
. . . . . . .
. .
.
.
.
.
. .
.
. .
. .
.
. .
.
.
. .
.
. .
.
.
.
.
. . . . . . .
. . . . . . .
23
23
23
23
24
24
25
25
26
26
27
27
27
23
2.2
NEWTON LAWS OF MOTION
2.1. Introduction
This Chapter continues the study of point-mass particle (PMP) systems by going over their kinetics.
This is the study of the effect of forces upon the motion. Modeling such effects using Newtons
laws results in the equations of motion (EOM) of the system.
The concept of work is briefly introduced in this Chapter, since it is necessary to support some
criterion for force decomposition. The concept and applications of work and energy principles are
elaborated in following Chapters.
2.2. Newton Laws of Motion
The laws of motion enuntiated by Newton in his Principia [269] will be stated in axiomatic form,
as derivable from observation.
2.2.1. First Law
Original Statement. Lex I: Corpus omne perseverare in status suo quiescendivei movendi uniformiter in directum, nisi quatenus a viribus impressis cogito statum illum mutare.
Literal English Translation. An object at rest will remain at rest unless acted upon by an external
and unbalanced force. An object in motion will remain in motion unless acted upon by an external
and unbalanced force.
Modern Version. A body remains in a state of rest or of uniform rectilinear motion unless compelled
to change its state by acting forces.
This is essentially Galileos law of inertia, enunciated by him long before Newton: A body moving
on a level surface will continue in the same direction at a constant speed unless disturbed. The
concept was further developed by Descartes, whereas the term inertia was introduced by Kepler.
However Newton recognized the fundamental important of the concept and stated it as the First
Law of Motion.
In a definition preceding the statement of this law, Newton introduces the body quantity of motion
as the measure of the same, arising from the velocity and the quantity of matter conjunctly [347,
p. 3]. By quantity of matter Newton meant the mass of the body. The product of mass and
velocity is what we now call momentum. Since the mass is a scalar whereas velocity is a vector,
the momentum is also a vector:
p = m v = m u.
(2.1)
Of course vectors have not been invented at Newtons time.1 Using the modern notation (2.1) we
can restate the First Law succintly as: if no forces act on the body, the momentum p is conserved:
p = C,
(2.2)
in which C is independent of time.

1
Vector analysis was simultaneously created by the American mathematician J. W. Gibbs and the English electrical
engineer O. Heaviside in the late XIX Century; cf. [76]. The notation of Gibbs is that primarily in use today. Vector
analysis and matrix algebra were not linked until the appearance of digital computers in the 1950w. This synergistic
marriage was delayed by the unseemly preoccupation of algebrists with determinants for over a century.
23
24
2.2.2. Second Law

Original Statement. Lex II: Mutationem motus proportionalem vi motrici impressate, et fieri
secumdum lineam rectam qua vis illa imprimatur.
Literal English Translation. The alteration of motion is ever proportional to the motive force
impressed, and is made in the right line in which the force is impressed.
Modern Version. The change in motion is proportional to the acting force, and takes place in the
direction of the straight line along which the force acts.
Using again the modern concept of momentum (2.1), Newtons change in motion can be briefly
stated as: the change with time of the momentum: p = dp/dt is proportional to the acting force
on the body, denoted by f. Thus the second law can be written
p = f.
(2.3)
This may be labeled as the law of momentum. Evidently if f = 0, p is constant in time, thus
recovering the conservation law (2.2). If the mass m is constant in time, (2.3) reduces to the more
familiar mass times acceleration expression
m v = m u = f.
(2.4)
This form is called Newtons law of acceleration in the literature. Although more restrictive, (2.4)
has a wide application range, since time-varying mass problems are rare in Classical Mechanics.2
2.2.3. Third Law
Original Statement. Lex III. Actionem contrariam semper et qualem esse reactionem: sive
corporus duorum actiones in se mutuo semper esse quales et in partes contrarias dirigi.
Literal English Translation. All forces occur in pairs, and these two forces are equal in magnitude
and of opposite directions.
Modern Version. For every action there is an equal and opposite reaction.
This is the principle of action and reaction. It says that forces are always paired in nature. Example
from [347, p. 6]: the falling stone attracts the Earth just as strongly as the Earth attracts the stone.
This law makes it possible to move from the mechanics of single mass points to that of systems,
whether at rest (statics) or moving (dynamics).
2.3. Forces
The term force appears in each of the three laws introduced above. In [384, p. 527], Truesdell cites
Hamel as writing in the concept of force lies the chief difficulty in the whole of mechanics. These
difficulties motivated Hertz [190] to construct a Mechanics without the force concept. His attempt,
however, was not successful within the framework of Classical Mechanics.
So what is force? The dictionary is no help. In fact, definitions pertinent to mechanical forces given
by the American Heritage Dictionary are:
2
They are more important in relativistic mechanics, because particle mass depends on its velocity relative to light speed.
24
25
2.3
1.
Capacity to do work or cause physical change; strength; power.
2.
Power made operative against resistance; exertion.
3.
The use of such power or exertion.
FORCES
These three statements confuse power and force, a serious mistake (as defined later, power is work
done in unit time). The capacity to do work is energy, which has dimensions of force times
motion. Hence the foregoing definitions are technically useless.
2.3.1. Concept of Force
As suggested by Sommerfeld [347], in Classical Mechanics force can be concretely defined as the
right hand side of the law of momentum (2.3). It is true that the left side still contains the mass,
which so far has only been defined qualitately as quantity of matter. But some properties of forces
can now be stated.
Consequently force is
First, since the momentum defined in (2.1) is a vector, so is its time rate p.
a vector: it has direction and magnitude. Second, the rule of composition of forces appears as a
corollary to the laws of motion: two forces applied to the same mass point compound to act like
the diagonal of the parallelogram formed by their vectors. If the forces are denoted by f1 and f2 ,
that parallelogram law states that the law of momentum (2.3) is simply
p = f1 + f2 = f.
(2.5)
Thus forces add as vectors to provide a resultant force. The above equation expresses succintly the
principle of superposition of forces.3
2.3.2. Force Decomposition
The three Newtonian laws of motion, plus the principle (2.5), are sufficient to develop all of Classical
Mechanics for PMP systems. Because of the importance of the force superposition principle, some
authors elevate its status to that of a Fourth Law of Motion. Without taking sides in this issue, the
crucial role of (2.5) lies in the possibility of decomposing forces.
Force decomposition is crucial to the analysis of general PMP systems. The total force acting
on each particle is in general a composition of effects that originate from different sources. For
example, some forces may depend on position, some on displacements, others on velocities, and
others on accelerations. Some forces may be prescribed as data, while others, called reactions,
depend on kinematic constraints4 and are unknown until the response is obtained. In summary, force
decomposition is essential to the construction of the correct equations of motion, their boundary
and initial conditions, as well as their solution.
It turns out that for complex dynamic systems there are several possible decompositions. Obvious
question: which one should be used? Quick generic answer: whatever it works. More precisely,
a decomposition that allows solvable equations of motion (EOM) to be formulated and solved in a
3
This is a precise restatement in vector notation of Galileos fuzzier Principle of Superposition: If a body is subjected
to two physical influences that are independent of each other, it responds to each without modifying its response to the
other. Galileo used this principle to study the motion of projectiles by superposing vertical and horizontal motions.
Kinematic constraints were introduced in 2.3.3.
25
26
systematic way. The qualifier systematic is important. Often ad-hoc decompositions allow a case
by case investigation of specific problems, but are not extendible beyond those.
Decompositions based on physical intuition do exist. A common one in older textbooks is: contact
forces (e.g. collision) versus action-at-a-distance forces (e.g. gravity). The distinction is easy to
understand and explain but largely useless aside of some special problems.
Several proven top-level force decompositions are introduced in the following sections.
2.3.3. Action-Reaction (AR) Force Decomposition
A useful top-level force decomposition givide forces into action and reaction:

Applied forces
Action forces
Interaction forces
Forces
(2.6)

Reaction forces E-reactions

C-reactions
Reaction forces are those associated with kinematic constraints. They are the forces required to
exactly enforce the constraint if the kinematic devices on which it depends are removed. All
remaining forces are action forces. If the dynamic system is unconstrained, all forces are of action
type.
Reaction forces are subclassified into E-reactions and C-reactions. E-reactions are associated with
the environment of the dynamic systemwhile C-reactions are associated with constraints internal to
the system They are also called external reactions and internal reactions, respectively.
Reaction forces are in subclassified into applied forces and interaction forces. Applied forces are
those specified directly on a particle from physical effects external to the dynamic system. For
example: gravity, fluid pressure, electromagnetic fields. Interaction forces are particle-to-particle
effects, which typically depend on relative displacements, velocities or accelerations of the linked
particles.
Why is this decomposition useful? It helps to select unknowns. If all or part of the reaction forces
are carried along in the EOM, the result is the Lagrange multiplier method, which is more fully
described in the next Chapter.
(TBC)
26
PMP Dynamical
Systems:
Matrix EOM Formulation
41
42
Chapter 4: PMP DYNAMICAL SYSTEMS: MATRIX EOM FORMULATION
TABLE OF CONTENTS
Page
4.1.
4.2.
4.3.
4.4.
4.
Matrix Equations of Motion

4.1.1. Approaches to EOM Formulation . . . .
EOM Derivation by Force Equilibrium
4.2.1. PMP Dynamic System Example . . . . .
4.2.2. Equilibrium Equations
. . . . . . .
4.2.3. Matrix Form . . . . . . . . . . .
4.2.4. Matrix Symmetry Check . . . . . . .
4.2.5. Center of Mass Motions Check . . . . .
4.2.6. PMP Equations of Motion for Numerical Data
EOM Derivation by the Finite Element Method
4.3.1. Element-Level Calculations . . . . . .
4.3.2. Assembly of Master Mass Matrix . . . .
4.3.3. The FEM Equations of Motion . . . . .
Comparison Between Formulation Approaches
Notes and Bibliography
. . . . . . . . . . . . . . .
42
. . . . . . .
. .
. .
. .
. .
. .
.
. .
. .
. .
. .
. .
. .
. .
. . .
. .
. . .
. .
. . .
. . . . . .
. . . . . . .
. . . . . .
. . . . . . .
43
43
43
43
45
46
46
47
47
48
48
49
49
410
411
43
4.2
EOM DERIVATION BY FORCE EQUILIBRIUM
4.1. Matrix Equations of Motion

This Chapter describe how to construct the dynamic equations of motion (EOM) in matrix form.
Several simplifying assumptions are made for clarity of exposition:
The dynamical model is a point-mass particle (PMP) system, with point masses linked by
collinear interaction forces.
Motions from the reference configuration are small in the sense that the initial geometry can be
used throughout
Interaction forces depend linearly on displacements and may be idealized as extenssional
springs. Damping effects are ignored although they are briefly covered in other Chapters.
The last two assumptions ensure that the EOM are linear in displacements and accelerations.
The first assumption (dynamic model is PMP) is partially lifted when considering the equivalence
between two ways of constructing the matrix EOM: by equilibrium or via FEM, as outlined next.
4.1.1. Approaches to EOM Formulation
Sections 4.2 and 4.3 cover two approaches to the formulation of matrix EOM:
Equilibrium Method. Writing down force balance expressions for each point-mass particle via Free
Body Diagrams (FBD) that account for dynamic effects modeled as forces.
Finite Element Method. Assembling element-level EOM provided by a FEM discretization.
The two methods are quite different in philosophy as well as procedure. The equilibrium method
reflects the direct application of Newtons laws to free bodies isolated from the rest of the system
by appropriate external and internal forces. Technically those are known as Free Body Diagrams
or FBD. By contrast, FEM equations are generally derived in a variational framework, such as the
Principle of Virtual Work, Lagranges equations, or Hamiltons principle.
Nonetheless, for the PMP systems considered in this Chapter, the two methods produce exactly
the same matrix EOM. Why then bother to go over both? The answer is that each has distinct
advantages as well as shortcomings. Those are contrasted in 4.4.
4.2. EOM Derivation by Force Equilibrium
The force equilibrium approach to deriving matrix EOM will be illustrated in the two-dimensional
PMP system pictured in Figure 4.1. A reader familiar with IFEM may note its similarity with the
example truss used in [106, Chapters 23]. That is not accidental: the stiffness matrix equations
derived there will be reused for part of the EOM derivation via FEM carried out in 4.3.
4.2.1. PMP Dynamic System Example
The two-dimensional PMP system of Figure 4.1(a) is referred to a fixed RCC frame {x, y}, with
origin as shown in the Figure. The geometry is defined by the reference position of the three point
masses, labeled 1,2 and 3.
43
fy3 , uy3
m3 = 4
(a)
3(10,10)
45
k(3) = 20
(3)
y
1(0,0)
45o
m1 = 5 x
(b)
(1)
k(1) = 10
(2)
44
fx3 , ux3
45o
k(2) = 5
(3)
2(10,0)
fx1 , ux1
(2)
(1)
45o
fx2 , ux2
2
fy2 , uy2
1
fy1 , uy1
m2 = 3
External forces:
fx3ext=2H(t), fy3ext=H(t),
others zero. Here
H(t) is the Heaviside
unit-step function
Figure 4.1. Example PMP system to illustrate derivation of matrix EOM by force equilibrium:
(a) reference geometry and physical properties; (b) kinematic DOF and associate forces.
Point masses are connected by linearly elastic extensional springs, which are labelled as (1), (2)
and (3) for convenience.1 The springs possess extensional stiffnesses k (1) , k (2) and k (3) with values
indicated as in the Figure. The system only moves on the {x, y} plane. Consequently it has 6
kinematic degrees of freedom (DOF), which are taken to be the point mass displacements in the
coordinate directions. These DOF are collected in a system displacement 6-vector with a mass-bymass arrangement:
(4.1)
u = [ u x1 u y1 u x2 u y2 u x3 u y3 ]T .
The velocity and acceleration 6-vectors are configured accordingly as
u = [ u x1
u y1
u x2
u y2
u x3
u y3 ]T ,
u = [ u x1
u y1
u x2
u y2
u x3
u y3 ]T .
f x3
f y3 ]T
(4.2)
The 6-vector of generic forces associated to (4.1) is

f = [ f x1
f y1
f x2
f y2
(4.3)
Unlike displacements, several kinds of forces may be considered: external, internal, interaction,
constraint (also called reaction), inertial, effective, residual, and so on. When there is need to distinguish force type, the generic symbols of (4.3) are adorned with appropriate 3-letter superscripts.
For example,
ext
ext
ext
ext
ext
ext T
f ext = [ f x1
f y1
f x2
f y2
f x3
f y3
] ,
(4.4)
int
int
int
int
int
int T
f int = [ f x1
f y1
f x2
f y2
f x3
f y3
] ,
denote vectors of external and internal forces, respectively, for the example system. To distinguish
internal from interaction, the latter is identified with superscript iac, as in f iac . For consistency
with NFEM [107], residual forces may be denoted by r instead of f r es to reduce clutter.
1
This follows the element-identification convention of IFEM [106, Chapter 2], since in 4.3 the springs of Figure 4.1 are
shown to be equivalent to bar elements in a truss FEM model. Enclosing parentheses distinguish these from point-mass
labels, while avoiding confusion with exponents when placed as superscripts of spring properties.
44
45
4.2
fy3ext
(a)
3
f s(3)
(3)
f s = k(3) (ux3 cos
45 o + uy3 sin
o
m3
(b)
(2)
1
m1
f s(1)
ext
fs
f s(3)
f s(1)
(1)
f s = k(1)(ux2 ux1 )
..
m 3 u y3
f s(2)= k(2)(uy3 uy2 )
ux2 cos 45 uy2 sin 45 )
2
m2
..
m 1 u x1
fx3ext
f s(2)
f s(3)
f s(2)
45o
o
..
m 3u x3
ext
fx1
fy1 f (3)
s
f s(1)
1 m u..y1
1
..
m 2 u x2
f s(1)
ext
fy2
f s(2)
ext
fx2
2
..
m 2 u y2
Figure 4.2. Auxiliary diagrams in equilibrium EOM derivation for PMP system of Figure 4.1: (a) springinteraction forces in terms of displacements; (b) point-mass FBD diagrams. External, spring-interaction
and inertial forces are pictured in black, blue and red, respectively, for visualization clarity.
Derivations are carried out first for a floating PMP system, which has no motion constraints. In
Chapter 5 the EOM will be modified to account for kinematic constraints of various types.
4.2.2. Equilibrium Equations
Begin by disconnecting the point-masses. Replace the springs by interaction forces f s(1) , f s(2)
and f s(3) . These are depicted in Figure 4.2(a), in which their expressions in terms of point-mass
displacements is listed. The + sense of the interaction forces is governed by an easy-to-remember
convention: assume that each spring is in tension. To complete the Free Body Diagrams (FBD)
shown in Figure 4.2(b), two more force sets are added:
ext
ext
The external forces f x1
through f y3
.
ine
ine
The inertial forces2 f x1
= m 1 u x1 through f y3
= m 3 u y3 . Their displayed sense is dictated
by a simple rule: the pertinent acceleration is assumed positive. Since masses are positive,
these forces point in the opposite sense of a + acceleration. See Figure 4.2(b).
The three force sets: external, interaction and inertial, are pictured in different colors in Figure 4.2(b)
for easier visualization. Next, force equilibrium conditions for each degree of freedom (DOF)
are written down. Interaction and inertia forces are expressed in terms of displacements and
accelerations. For example, force balance of the motion of mass 1 along x gives

ext
ine
+ f s(1) + f s(3) cos 45 + f x1
f x1 = f x1

ext
= f x1
+ k (1) (u x2 u x1 ) + k (3) u x3 cos 45 + u y3 sin 45
(4.5)

u x1 cos 45 u y1 sin 45 cos 45 m 1 u x1

ext
= f x1
+ k (1) (u x2 u x1 ) + 12 k (3) u x3 +u y3 u x1 u y1 m 1 u x1 = 0.
2
Also known as DAlembert forces, pseudo forces, or effective forces in the literature.
45
46
At this point it is convenient to revert the sign of the LHS of (4.5) so as to make the inertia term
m 1 u x1 positive. While doing this, common displacements are collected, and the external force
term passed to the RHS. The rearranged expression is
ext
m 1 u x1 + (k (1) + 12 k (3) ) u x1 k (1) u x2 + 12 k (3) u y1 12 k (3) u x3 12 k (3) u y3 = f x1
.
(4.6)
Following exactly the same pattern, five more balance equations are obtained. The end result is six
equilibrium equations, one for each DOF.
4.2.3. Matrix Form
Once the equilibrium equations for all DOF are obtained and appropriate arranged as discussed
above, several steps are carried out to put them in matrix form:
The external forces, which are givens, are kept in the RHS and arranged as a vector fext ,
configured as per the first of (4.4).
LHS terms that depend on accelerations are organized as a diagonal master mass matrix M
postmultiplied by the acceleration vector u configured as in the second of (4.2).
LHS terms that depend on displacements are collected and organized as a master stiffness matrix
K postmultiplied by the displacement vector u configured as in (4.1).
The resulting matrix EOM have the compact form
M u + K u = f ext .
For the example system the mass and stiffness matrices are
m1 0
0
0
0
0
0
0
0
0 m1 0
0
0
0 m2 0
0
M=
= diag [ m1, m1, m2, m2, m3, m3 ].
0
0
0 m2 0
0
0
0
0
0 m3 0
0
0
0
0
0 m3
(1) 1 (3)
1 (3)
k + 2k
k
k (1)
0
12 k (3)
12 k (3)
2
1 (3)
1 (3)
k
k
0
0
12 k (3)
12 k (3)
2
2
(1)
(1)
k
0
k
0
0
0
K=
(2)
(2)
0
0
0
k
0
k
1
1
1
1
(3)
(3)
(3)
(3)
k
k
0
0
k
k
2
2
2
2
1 (3)
1 (3)
1 (3)
1 (3)
(2)
(2)
2k
2k
0
k
k
k + 2k
2
(4.7)
(4.8)
(4.9)
If the RHS of (4.7) consists entirely of external (applied, given) forces, we will often drop the
superscript, and write
M u + K u = f,
being tacitly understood that f contains only given forces.
46
(4.10)
47
4.2
4.2.4. Matrix Symmetry Check

Observe that both M and K in (4.8) and (4.9) are symmetric. For M that is not a surprise, since
real diagonal matrices are necessarily symmetric. Can that be expected for K? Yes: it is not an
accident. It previews the equivalence of this approach with the Finite Element Method, as covered in
4.3. Since FEM equations are automatically symmetric when derived in a variational framework,
equivalence implies symmetry. Should K come out unsymmetric, two fixes can be applied:
Reverse sign of individual equations (which become matrix rows) as necessary so that diagonal
entries of K and M are positive. If they are not of the same sign, look for derivation errors.
If diagonal signs check out but unsymmetry persists, try row scaling by appropriate factors.
4.2.5. Center of Mass Motions Check
After verifying symmetry, reduction to center-of-mass motion provides another useful EOM check.
Assume that x and y translations are rigid by setting u x1 = u x2 = u x3 = u x and u y1 = u y2 =
u y3 = u y for all t. Hence u x1 = u x2 = u x3 = u x and u y1 = u y2 = u y3 = u y . Replace these into
vectors u and u of (4.10). Add equations 1, 3, and 5, factoring u x and u x , and add equations 2, 4,
and 6, factoring u y and u y . The result is
M u x = Fx ,
M u y = F y ,
ext
ext
ext
+ f x2
+ f x3
in which M = m 1 + m 2 + m 3 is the total mass of the system, whereas Fx = f x1
ext
ext
ext
and F y = f y1
+ f y2
+ f y3
are the resultants of the x and y external forces, respectively. These
are the well known equations for the translational motion of the center of mass, which is located
at xC = 10(m 2 + m 3 )/M and yC = 10m 3 /M, as may be easily verified. A similar check can be
performed for an infinitesimal rotation about the center of mass: the result is IC = TC , where
IC is the mass moment about C and TC the torque of the external forces about C.
In these reductions the contribution of the stiffness term K u cancels out. This results from the
rigid-body property of the stiffness matrix: K u R = 0 if u R is a rigid body motion. A direct
verification can be done by extracting the eigenvalues of K and checking that its rank is 3 [106,
Chapter 20], but symbolic eigenvalue verification becomes unwieldy for larger systems.
4.2.6. PMP Equations of Motion for Numerical Data
Numerical values for point masses and spring constants given in Figure 4.1(a) are: m 1 = 5,
m 2 = 3, m 3 = 4, k (1) = 10, k (2) = 5, and k (3) = 20. For the external forces, Figure 4.1(b) states
ext
ext
ext
ext
ext
ext
= f y1
= f x2
= f y2
= 0, f x3
= 2H (t), and f y3
= H (t), in which H (t) is the
that f x1
Heaviside unit-step function. Replacing into the foregoing matrix expression yields
5
0
0
0
0
5
0
0
0
0
0
0
3
0
0
0
0
0
0
3
0
0
0
0
0
0
4
0

0
u x1
20
0 u y1 10

0 u x2 10
+
0 u y2 0

u x3
0
10
u y3
4
10
10 10 0
10
0
0
0
10 0
0
0
5
10 0
0
10 0 5
47
10
10
0
0
10
10
10
u x1
0
10 u y1 0
0 u x2 0
5 u y2 0
u x3
10
2H (t)
u y3
15
H (t)
(4.11)
fy3 , uy3
3
(3)
(3)
=
2 2,
(3)
y
fx1 , ux1
1
fy1 , uy1
45o
External forces:
fx3ext=2H(t), fy3ext=H(t),
others zero. Here
H(t) is the Heaviside
unit-step function
(2)
E (2) = 50, A(2) = 1,

(2)
L = 10, (2) = 1/5
fx2 , ux2
x (1)
(1)
E (1) =
(1)
fx3 , ux3
45o
E = 100, A
(3)
2, (3) = 3/20
L = 10
48
50, A = 2,
L = 10, (1) = 1/5
2
fy2 , uy2
Figure 4.3. The three-member example truss of IFEM [106, Chapters 2-3] reproduced for
convenience. The dynamic model is (for now) floating: it has no supports and thus may
move freely on the {x, y} plane. It is subject to the external forces indicated in the box.
This is a linear system of six second-order ODEs in time. The problem specification is closed by
providing 12 initial conditions: six displacements and six velocities at a start time, which is often
taken to be t = 0. For example, the rest condition at t = 0 is specified by
u x1 (0) = u y1 (0) = . . . , u y3 (0) = 0,
u x1 (0) = u y1 (0) = . . . , u y3 (0) = 0.
(4.12)
A rest initial condition would be appropriate for a structure hit by a transient event at t = 0, e.g.,
earthquake, explosion or impact.
4.3. EOM Derivation by the Finite Element Method
The FEM model for the example truss of IFEM [106, Chapters 23] is reproduced in Figure 4.3.
One additional property is now activated: the mass density of the three elements (truss members).
This is denoted by e .
4.3.1. Element-Level Calculations
As typical of FEM, computations proceed element by element. In this case two element-level
matrices are generated: stiffness and mass. The element stiffness matrices are exactly those derived
in IFEM [106, Chapters 23]. The reader is referred there for details.
The mass matrices are produced by a nodal lumping scheme that proceeds as follows. The total
mass of element e is m e = e Ae L e , in which L e and Ae are the length and cross section area,
respectively. Assign one half of this to each end node, in both the x and y directions. As a result
the so-called lumped mass matrix of the truss (bar) element is
1 e
m
0
0
0
2
1 e
0
m
0
0
2
= 1 e A e L e I4 ,
(4.13)
Me =
1
e
0
0
m
0 2
2
0
48
1 e
m
2
49
4.3
EOM DERIVATION BY THE FINITE ELEMENT METHOD
(3)
m3 =3
(a)
m3 =1+3=4
(2)
m3 =1
(b)
(3)
(2)
(3)
m1 =3
1
(2)
(1)
(1)
m1 =2
(1)
m2 =2
m2 =1
m1 =3+2=5
m2 =2+1=3
Figure 4.4. Mass lumping procedure for the example truss of Figure ?: (a) element-level
lumping; (b) addition at nodes to form master mass matrix.
in which I4 denotes the identity matrix of order 4. What happens to Me if the axes are rotated by
an angle to be {x,
y }? It becomes
e = (Te )T Me Te = e Ae L e (Te )T I4 Te = e Ae L e (Te )T Te ,
M
where Te is a 44 transformation matrix very similar to that used in [106, 2.8]. But (Te )T Te = I4
e = Me for any . It follows that Me is invariant with respect to
because Te is orthogonal, and M
rotation of axes. Consequence: the machinery for deriving FEM equations in a local system and
converting to global coordinates, as done for the truss element stiffness matrix, can be bypassed for
the mass matrix.
4.3.2. Assembly of Master Mass Matrix
The assembly of the master mass matrix follows exactly the same set of rules used for the master
stiffness matrix. Since Me is diagonal, so will be the master mass matrix. A simple hand calculation
proceed by inspecting which elements contribute to a given node. For the example truss:
Node 1
gets half of the mass of elements (1) and (3): m 1 = 12 m (1) + 12 m (3) .
Node 2
Node 3
These values are assigned to both x and y directions. Consequently the master mass matrix is
M=
1
2
diag [ m (1) +m (3) , m (1) +m (3) , m (1) +m (2) , m (1) +m (2) , m (2) +m (3) , m (2) +m (3) ]
= diag [ m 1 , m 1 , m 2 , m 2 , m 3 , m 3 ] = diag [ 5, 5, 3, 3, 4, 4 ].
(4.14)
This two-step construction of the master mass matrix is schematized in Figure 4.4. The node masses
are m 1 = 5, m 2 = 3 and m 3 = 4.
Remark 4.1. In later Chapters we will see that the lumped mass matrix is not the only choice; in fact there is an
infinite number of possible element mass matrices. Those other than the lumped one are no longer diagonal.
Consequently the assembly process becomes more involved, but can be carried out by the same rules followed
for the master stiffness matrix.
49
410
4.3.3. The FEM Equations of Motion

Combining the master mass matrix derived above with the master stiffness derived in IFEM [106,
Chapter 3] and the external forces specified in Figure 4.3, we obtain

20 10 10 0 10 10
u x1
0
u x1
5 0 0 0 0 0
0
0 10 10 u y1 0
0 5 0 0 0 0 u y1 10 10

10 0
0
0 u x2 0
0 0 3 0 0 0 u x2 10 0
+
=

0
0
5
0 5 u y2 0
0 0 0 3 0 0 u y2 0

u x3
u x3
10 10 0
0 10 10
2H (t)
0 0 0 0 4 0
u y3
u y3
10 10 0 5 10 15
H (t)
0 0 0 0 0 4
(4.15)
Comparing(4.15) with (4.11) shows these matrix EOM to be identical. Is this a fluke? No. It is a
consequence of two modeling choices.
The mass matrices are the same because of the lumping choice in FEM: conservation conditions
forces diagonal matrix entries to agree. If another choice had been made, the FEM mass matrix
would be nondiagonal and clearly different from that of the PMP model.
Stiffness matrices coalesce because we have set the spring constants of the PMP model of Figure 4.1
to be
(4.16)
k e = E e Ae /L e , e = (1), (2), (3),
where E e , Ae and L e are member properties of the FEM truss model of Figure 4.3. As shown in
Mechanics of Materials textbooks, as well as IFEM [106, Chapter 2], (4.16) provides the equivalent
axial stiffness of a truss (bar) element.
Given that the same EOM are obtained, downstream tasks such as
Application of constraint conditions
Modal analysis
Direct time integration
Internal force recovery
need not distinguish between the two derivation methods. Those topics are covered in the following
Chapters.
4.4. Comparison Between Formulation Approaches
Having gone through the two EOM formulation methods, it is appropriate to summarize their
relative advantages and shortcomings.
The key advantages of the force equilibrium method are simplicity and physical transparency. These
emanate from the use of just one modeling tool: Newtons laws, as well as basic linear algebra. This
makes it the method of choice in undergraduate instruction. Students in those courses have been
exposed to those laws, as well as trained in static FBD and basic linear algebra through introductory
lower-division courses, but have not yet encountered the more advanced mathematical machinery
used in FEM.
410
411
4. Notes and Bibliography
For modeling dynamic systems in practical projects in science or engineering, the equilibrium
method is recommended in the following scenarios:
Preliminary design. The nature of the method encourages the use of highly simplified dynamic
models. For example, an experienced engineer designing a car suspension would likely lump
the car and the wheels as link-connected point masses, thus bypassing the overkill ingrained in
using FEM models too early. Likewise, seismic design of a multistory building often starts with
a point-mass stick model. Keeping-it-simple can significantly reduce design flowthrough,
while avoiding costly modifications.
Particle model fits problem best. There are some scientific applications where PMP models are
proper and natural. The classical example is Astronomy, which is precisely where Newtonian
mechanics emerged as system of the world. The key advantage is the ability to discard ab initio
internal particle structural details, which may not be known anyway. This enforces retention of
key effects, such as inertia and gravitational interaction in that application. A FEM model of
the Solar System at astronomical scales (or the trajectory of a space vehicle) would be not only
absurd but a total waste of time.
The key advantages of FEM are generality and automation. Once system details need to be included,
the equilibrium force method rapidly loses its attractive simplicity. FEM is naturally adapted to
distributed systems modeled by field (continuum) theories within a variational framework. Its
computer implementation favors abstraction and modularity, reducing formulation errors while
concealing internal processing details.
Summarizing: in system design that involves dynamics, the two approaches form a natural hierarchy.
The PMP approach may capture gross effects during preliminary design, while FEM can zero in
details in subsequent stages. At the other extreme: verification and validation of existing (in situ)
systems, FEM is typically the preferred choice.
(To be completed)
411
PMP Dynamical
Systems:
Kinematic Constraints
51
Chapter 5: PMP DYNAMICAL SYSTEMS: KINEMATIC CONSTRAINTS
52
TABLE OF CONTENTS
Page
5.1.
5.2.
5.3.
5.4.
5.5.
5.6.
5.7.
5.8.
5.
Overview
Analytically-Related Constraint Classification
5.2.1. Algebraic, Differential or Integral . . . . . . .
5.2.2. Stationary or Time Dependent
. . . . . . .
5.2.3. Equality or Inequality
. . . . . . . . . .
Implementation-Related Constraint Classification
5.3.1. Single or Multiple Freedoms . . . . . . . .
5.3.2. Homogeneous or Nonhomogeneous . . . . . .
5.3.3. Linear or Nonlinear . . . . . . . . . . .
Unified Treatment By Constraint Forces
Holonomic SFC Examples
5.5.1. Scleronomic, Homogeneous SFC . . . . . . .
5.5.2. Rheonomic, Nonhomogeneous SFC
. . . . .
5.5.3. Pseudo-Nonholonomic SFC
. . . . . . . .
Holonomic, Homogeneous MFC Examples
5.6.1. Scleronomic, Homogeneous MFC by Master-Slave
5.6.2. Rheonomic, Nonhomogeneous MFC by Master-Slave
Penalty Function Augmentation Methods for MFC
Lagrange Multiplier Adjunction Methods for MFC
. . . . . . . . . . . . . . . . . .
52
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
53
53
54
54
55
55
55
55
57
57
58
58
59
510
510
510
511
512
512
512
53
5.2
ANALYTICALLY-RELATED CONSTRAINT CLASSIFICATION
5.1. Overview
Kinematic constraints are restrictions on the motion of a dynamic system.1 If there are no constraints, the system is said to be unconstrained or unrestrained. If an unconstrained model represents
a structure or solid body, it is also called free-free or floating.2 The PMP example system treated
in the previous Chapter is of this type. Familiar examples: an airplane or bird in flight, a planet,
satellite or rocket in orbit.
Constrained systems are more common, however, especially in Civil and Mechanical Engineering.
For functional reasons, such as stability, control, safety, or operational usability, motions may be
restricted by devices such as building foundations, bridge supports, member joints, failsafe stops,
sliding guides, etc. This Chapter explains how to impose some practically important constraint
types into a PMP dynamical system, once the unconstrained semidiscrete EOM are set up by one
of the methods covered in the previous Chapter. Although the example is admittedly simplistic,
most techniques can be extended to the more complex systems treated in later Chapters.
The introduction of kinematic constraints into a dynamical system comprises three stages:
Idealization: stating physical restrictions on the motion as mathematical expressions.
Implementation: modifying unconstrained EOM so that the idealized constraints are verified.
Solution: accounting for constraints in the solution procedure if and as necessary.
The first step results in a mathematical statement that may include displacements and/or velocities
(involvement of accelerations is relatively rare), as well as possibly time and associated forces. In
simple cases the idealization may be obvious from inspection. For instance, suppose that point
mass 2 of the PMP example system of the previous Chapter, illustrated in Figure 4.1, can only
move along x. Then evidently u y2 = 0. Such local or single-DOF constraint forms are often called
spatial boundary conditions in the literature.3
More complicated scenarios lead to constraints that link multiple DOF. These are discussed in [106,
Chapters 89] for the static case. But the spectrum of kinematic constraints in dynamics is far
richer, because they may display time dependence and/or involve temporal state derivatives such as
velocities. Accordingly, constraint implementation techniques span a wider range, and may bring
about additional modeling as well as computational difficulties. To prevent such troubles it helps
to beware of the different types of constraint that may occur in practice. Hence a categorization is
in order.
For convenient reference, the classifications outlined in 5.2 and 5.3 are summarized in Tables 5.1
and 5.2, respectively.
1
The appropriate definition for constraint, taken from the American Heritage Dictionary, is something that restricts,
limits or regulates. Synonyms: coercion, control, limitation, prevention, regulation, restriction, restraint, suppression.
The qualifier kinematic introduces the flavor of physical motion. The term constraint is relatively recent; other
names found in the older literature are: side conditions and auxiliary conditions.
Qualifier free-free has the connotation of exempt from external authority, interference or restrictions, whereas qualifier
floating is used in the sense of having little or no attachment to a particular place.
In Mathematics, the term boundary condition technically applies to situations where the boundary of a region can be
readily identified, as in boundary-value problems. But occasionally the meaning is informally extended to encompass
constraints of local type, whether pertaining to a boundary or not.
53
54
5.2. Analytically-Related Constraint Classification

In Classical Mechanics4 constraints were primarily classified according to mathematical features
relevant to the use of analytical solution techniques. These include: (i) algebraic, differential or
integral form; (ii) time dependence or independence, and (iii) equality or inequality. The following
definitions assume that the idealized expression is of equality type since inequality constraints are
beyond the scope of the book; see 5.2.3 for more details.
5.2.1. Algebraic, Differential or Integral
An equality kinematic constraint is holonomic5 if it can be expressed in the algebraic form
f k (u, t) = 0,
(5.1)
in which k denotes a constraint index, u is the vector of state variables (in PMP systems, point-mass
displacements). Holonomic constraints are often associated with conservative systems.
An equality kinematic constraint is nonholonomic if it is expressed as a differential or algebraic
differential form that includes state variable velocities collected in vector u:
t) = 0,
f k (u, u,
(5.2)
and is not reducible to (5.1) by either integration, or differentiation followed by elimination. [Constraints originally stated as (5.2) but reducible to (5.1) are sometimes called pseudo-holonomic
or semiholonomic.] These are often associated with nonconservative systems (those exhibiting
dissipative effects such as friction), as well as rolling contact or robotic motions.
An equality constraint is isoparametric if it is expressed as a definite integral in time; for example
t2
t) dt = C,
f k (u, u,
(5.3)
t1
where C is a constant. Such conditions are important in contexts that involve conservation of some
quantity such as path length or energy, but will not be treated here.
An obvious generalization of (5.2) is an equality differential form that includes accelerations:
u,
t) = 0, and is not reducible to either (5.1) or (5.2). But as previously noted such forms
f k (u, u,
are comparatively rare.6
4
Classical refers to a period extending roughly from Newton to the advent of digital computers in the early 1950s. For
major advances in the treatment of kinematic constraints during this period, see Notes and Bibliography.
The distinction between holonomic and nonholonomic constraints was first emphasized by Hertz in 1894 [381], and further
studied by Ferrers [375], Routh [414] and other scientists of the time. For a historical account of these developments,
see [390].
One reason for their rarity is that selected accelerations can often be eliminated using the original EOM.
54
55
5.3
IMPLEMENTATION-RELATED CONSTRAINT CLASSIFICATION
5.2.2. Stationary or Time Dependent

Here are two more Victorian-era adjectives for impressing your Facebook friends. The holonomic
constraint (5.1) is called scleronomic if it is stationary (that is, time independent), thus reducing to
f k (u) = 0. It is rheonomic otherwise.7 Such a distinction is not relevant for the nonholonomic
form (5.2), which depends at least implicitly on time because it contains temporal derivatives.
This classification is vacuous for the static case, in which only the scleronomic holonomic form
f k (u) = 0 occurs since both time and velocities are absent.
5.2.3. Equality or Inequality
Note the qualifier equality used for the definitions in 5.2.1. If the equals sign in (5.1) or (5.2) is
replaced by > or , the expression becomes an inequality constraint. In Mechanics these occur
primarily in contact and impact problems as well as optimization. Those forms are beyond the scope
of this book, and will not be considered further. Computational handling of inequality constraints
normally demands application-specific tools, such as search algorithms for contact-impact problems
and sequential quadratic programming (SQP) methods in optimization. Those highly specialized
topics are covered in advanced textbooks and monographs.
5.3. Implementation-Related Constraint Classification
The classification discussed in 5.2 was important in the pre-computer era, where long and laborious dynamic calculations were carried out by hand.8 As computational dynamics evolved in the
computer era, certain constraint features previously ignored or neglected were found to have significant impact on the computer implementation as well as in organizing time-stepping calculations.
These are collected in Table 5.2 and summarized in the following subsections.
5.3.1. Single or Multiple Freedoms
If the constraint involves just one state variable, it is called Single Freedom Constraint or SFC. if it
links more than one, it is a Multiple Freedom Constraint, or MFC. The distinction was introduced
for static FEM analysis in IFEM [106, Chapters 89], and is also important in dynamics.
Suppose that for the example PMP system we specify a holonomic SFC such as u y2 = 0, or
u y2 = 2 + 12 t 2 . The accelerations u y2 are computed, and together with u y2 replaced into the matrix
EOM. Known data is transferred to the RHS and the modified EOM partitioned to exhibit the
remaining, unconstrained DOF. This reduction technique is illustrated with examples in 5.5.
How about u y2 = 4u y2 ? Some preprocessing is required. It is integrated to yield u y2 = exp(t/4) +
C y2 , with C y2 determined by initial conditions. Then one proceeds as in the previous case. If
integration is difficult or impossible, treat as a MFC.
For a holonomic MFC, the three implementation methods described in IFEM [106, Chapter 89], and collected in Table 5.3, are still applicable with some extensions illustrated later. For a
nonholonomic MFC, only the Lagrange multiplier method, used with extreme care, offers hope.
7
Roots come from Greek sclero: hard or solid, and rheo: flowing.
For example, the mathematician E. W. Brown (18661938) spent most of his professional life doing accurate calculations
of the Moon orbit, reaching 1500 perturbation terms. A laptop can reproduce his 1908 tables in a few seconds.
55
56
Table 5.1. Constraint Classification Related to Analytical Issues

Type
Definition
Comments
Holonomic
Nonholonomic
Isoparametric
Algebraic form (5.1)

Differential form (5.2)
Integral form (5.3)
If not reducible to (5.1)

Not treated here
Scleronomic
Rheonomic
Holonomic form (5.1), with no t dependence

Holonomic form (5.1), with explicit t dependence
Equality
Inequality
= separates LHS and RHS

or > separates LHS and RHS
Not treated here
Table 5.2. Constraint Classification Related to Implementation Issues

Type
Description
Comments
Single freedom (SFC)

Multiple freedom (MFC)
One DOF in (5.1) or (5.2)

Multiple DOF in (5.1) or (5.2)
Tougher to implement
Homogeneous
Nonhomogeneous
Homogeneous polynomial in state variables

Not meeting previous definition
Linear
Nonlinear
State variables appear linearly

State variables appear nonlinearly
No RHS modification
Not treated here
Table 5.3. Constraint Implementation Methods

Method
Variants
Procedure
Elimination
Direct
Master-slave
Substitute, then partition

Congruential transformation on stiffness & mass
Penalty function
Standard (one pass)

Augmented Lagrangian
Inject penalty elements of fixed weights

Inject penalty elements but iterate on weights
Lagrange multiplier
Standard
Regularized
Constraint forces are made system unknowns

Similar to Augmented Lagrangian
5.3.2. Homogeneous or Nonhomogeneous

A scleronomic holonomic constraint is homogeneous if it is expressible as a homogeneous polynomial in the state variables. Otherwise it is nonhomogeneous. For the PMP example system, u y2 = 0
is homogeneous, and so is (u x2 u x1 )n = 0 for any integer exponent n > 0. On the other hand
u x2 = t, u x2 u x1 = 2, and u 2x2 = 1 + u 2x1 , are nonhomogeneous. Rheonomic (time dependent)
holonomic constraints, as well as nonholonomic ones, are always considered nonhomogeneous.
56
57
5.4
UNIFIED TREATMENT BY CONSTRAINT FORCES
;;;;;;
;;;;;;
;;;;;;
f con
Figure 5.1. Remove the bridge (log) and replace its effect on the elephant by constraint forces
fcon on the legs. The elephant stays happy: nothing happens. This illustrates Newtons Third Law
of action and reaction, whence constraint forces are often called reaction forces by engineers.
The significance of this feature in computer implementation is that, if the constraint is nonhomogeneous, applied force terms must be modified and converted to effective forces if elimination or
penalty function methods are used. No such modifications are needed in the homogeneous case.
5.3.3. Linear or Nonlinear
A holonomic constraint is linear if any displacements present in the constraint appear linearly.
Otherwise the constraint is nonlinear. The dependence on time may be arbitrary. For instance
u x1 2u y1 = t 2 is linear but (u x2 u x1 )2 + (u y2 u y1 )2 = 0 is not.
Likewise, a nonholonomic constraint is linear if any displacements and velocities present in the
constraint appear linearly. The dependence on time may be arbitrary.
The presence of nonlinear constraints affects computer implementation if the time marching solution
procedure iterates at each step, as is often the case with implicit time integrators. If so, such
constraints must be appropriately linearized.
Nonlinear constraints are not treated in this book.
For convenient reference, Table 5.3 collects constraint implementation methods. Their basic features are described in IFEM [106, Chapters 3, 8 and 9]. Modifications required for the dynamic
case can be learned from the examples that follow.
5.4. Unified Treatment By Constraint Forces
Looking at the Tables 5.1 through 5.3 may discourage the reader. There are 12 constraint classes
listed, along with 3 implementation methods with 2 variants each. But there is no cause for alarm.
Only a few of the simplest constraints are those overwhelmingly found in practical engineering and
physics problems.
In previous chapters the unifying role of forces in unconstrained dynamic systems has been emphasized. Their role is equally strong once constraints are introduced. The basic rule is
Any kinematic constraint may be replaced by a system of forces
(5.4)
See cartoon in Figure 5.1. Unsurpringly, these forces are called constraint forces. Since this recipe
is nothing more than Newtons Third Law of action and reaction under another name, they are often
57
fy3 , uy3
3
58
fx3 , ux3
45o
(3)
(2)
fy1 , uy1
;;
45o
(1)
2 fy2 , uy2
fx2 , ux2
;;
fx1 , ux1 1
Figure 5.2. Support conditions for Example worked out in 5.5.1.
called reaction forces by engineers, particularly when they happen at a foundation or support. From
a mathematical standpoint, constraint forces can always be identified with Lagrange multipliers,
which explains the power and generality of this methodology.
The examples that follow will illustrate the rule (5.4) in action.
5.5. Holonomic SFC Examples
We begin our set of examples with the simplest kinematic constraints. These are holonomic, single
freedom constraints (SFC). Three variations on this theme are considered in this Section.
5.5.1. Scleronomic, Homogeneous SFC
The floating example truss studied in Chapter 4 (see Figure 4.3) is now supported as shown in
Figure 5.2. Node 1 is fixed whereas node 2 can only moved along x. Thus displacements u x1 , u y1 ,
and u y2 vanish and so do the associated accelerations u x1 , u y1 , and u y2 . We can state the constraints
as
u x1 = u y1 = u y2 = 0,
u x1 = u y1 = u y2 = 0.
(5.5)
In the classification covered previously, these are scleronomic (time-independent), holonomic, and
homogeneous SFC. To implement, insert (5.5) into the unsconstrained EOM (4.15) derived in the
previous Chapter to get
0
0
0
0
0
0
0
0
4
0
0
10
0
0
u x2
0
10
u x3 +
0
0
u y3
0
0
4
0
10
10
0
0
10
10

10
0
10
u x2

0
0
u x3 =
+
5
0
u y3

10
2H (t)
15
H (t)
con
f x1
con
f y1
0
ext
+f con . (5.6)
con = f
f y2
0
0
in which fcon collects unknown constraint (support reaction) forces. System (5.6) is obtained from
the matrix EOM (4.15) by striking out columns 1, 2 and 4 of M and K. (Those can be deleted
58
59
5.5
HOLONOMIC SFC EXAMPLES
because they multiply zero entries.) Now (5.6) partitions naturally into two subsets. Keeping rows
3, 5 and 6 produces the reduced system

10
0
0
u x2
0
3 0 0
u x2
(5.7)
u x3 + 0
u x3 = 2H (t)
10 10
0 4 0
u y3
u y3
0 10 15
H (t)
0 0 4
which has no constraint forces. This system provides three linear ODE for three unknowns, and
may be solved once the initial conditions are specified. The other subset consists of rows 1, 2 and
4. Because accelerations drop out (note the 3 null rows in the reduced M), we get an algebraic
system:
con

f x1
10 10 10
u x2
con
=
(5.8)
f con = f y1
u x3 ,
0
10 10
con
f y2
u y3
0
0
5
This is called the reaction recovery system because the constraint (reaction) forces may be computed,
if so desired, once u x2 , u x3 and u y3 are obtained by solving (5.7) in time. Note that (5.8) is time
independent, and merely reflects static force equilibrium.
5.5.2. Rheonomic, Nonhomogeneous SFC
The truss support conditions of the foregoing example are modified by making u x1 , u y1 , and u y2
specified functions of time, which are denoted by gx1 , g y1 , and g y2 , respectively:
u x1 = gx1 (t), u y1 = g y1 (t), u y2 = g y2 (t),
u x1 = g x1 (t), u y1 = g y1 (t), u y2 = g y2 (t). (5.9)
These are rheonomic (time-dependent), holonomic, and nonhomogeneous SFC. As before insert
(5.14) into (4.15). But now contributions from columns 1, 2 and 4 do not vanish, and are transferred
to the RHS before those columns are removed:
10 10 10
0 0 0
0 0 0
0 10 10
u x2
u x2
0
0
10
3 0 0
(5.10)
u x3 = f ext + f,tra + f con .
u x3 +
0
0
5
0
0
0
y3
y3
0
10 10
0 4 0
0 10 15
0 0 4
in which
f ext
0
0
0
=
,
0
2H (t)
H (t)
f tra
5g x1 20gx1 10g y1
5g y1 10gx1 10g y1
10gx1
=
,
3g y2 5g y2
10gx1 + 10g y1
10gx1 + 10g y1 + 5g y2
f con
con
f x1
con
f y1
0
con ,
f y2
0
0
denote external (applied), transfer and constraint forces, respectively. Only f con carries unknowns.
Keeping rows 3, 5 and 6 provides the reduced EOM system

10
0
0
u x2
10gx1
3 0 0
u x2
(5.11)
10 10
u x3 + 0
u x3 =
2H (t) + 10gx1 + 10g y1
0 4 0
u y3
u y3
H (t) + 10gx1 + 10g y1 + 5g y2
0 10 15
0 0 4
59
510
This RHS is a known function of time; so (5.11) can be directly solved once the initial conditions
are given. The other subset, which consists of rows 1, 2 and 4, gives the reaction recovery system
con

f x1
10 10 10
u x2
5g x1 + 20gx1 + 10g y1
con
=
(5.12)
f y1
u x3 + 5g y1 + 10gx1 + 10g y1 .
0
10 10
con
f y2
u y3
3g y2 + 5g y2
0
0
5
This can be used to recover the constraint (support reaction) forces once u x2 , u x3 and u y3 are
obtained by solving (?) in time.
5.5.3. Pseudo-Nonholonomic SFC
The truss support conditions are specified in terms of time-dependent velocities, and time differentated to get accelerations:
u x1 = 1 + t,
u y1 = 2t,
u y2 = 3t 2 ,
u x1 = 1,
u y1 = 2,
u y2 = 6t.
(5.13)
These look like nonholonomic constraints, but actually can be integrated in time to get displacements:
u x1 = t + 12 t 2 + C x1 , u x1 = t 2 + C y1 , u y2 = t 3 + C y2 .
(5.14)
The constants of integration can be determined from the initial conditions. For example suppose
that at t = 0 we have u x1 (0) = u y1 (0) = u y2 (0) = 0. Then C x1 = C y1 = C y2 = 0, and we are
back to the rheonomic, holonomic SFC case treated in 5.5.2.
5.6. Holonomic, Homogeneous MFC Examples
The examples in this Section pertain to a more complicated case: a set of holonomic homogeneous
constraints, at least one of which involves multiple freedoms (MFC). The appearance of MFC calls
for more powerful implementation methodologies that straightforward elimination.
5.6.1. Scleronomic, Homogeneous MFC by Master-Slave
The example truss is subject to three MFC:
u y1 = u x1 ,
u y2 = u x2 ,
u y3 = u x3 .
(5.15)
Interpretation for infinitesimal displacements: the truss rotates about the midpoint of element
(3) halfway between nodes 1 and 3. Differentiate (5.23) twice in time to establish acceleration
constraints:
u y1 = u x1 , u y2 = u x2 , u y3 = u x3 .
(5.16)
To apply the master-slave elimination method, pick 3 slave displacement DOF, say the y components
u y1 , u y2 and u y3 , which will be eliminated.9 The x components: u x1 , u x2 and u x3 are the master
DOF. Write down the DOF transformation
u x1
u x1
1 0 0
1 0 0
u y1 1 0 0
u y1 1 0 0

u x1

u x1
u x2 0 1 0
u x2 0 1 0
(5.17)
=
u x2 ,
=
u x2 .
u y2 0 1 0
u y2 0 1 0
u
u
x3
x3
u x3
u x3
0 0 1
0 0 1
u y3
u y3
0 0 1
0 0 1
9
For details of this method in the static case, see IFEM [106, Chapter 8].
510
511
5.6
HOLONOMIC, HOMOGENEOUS MFC EXAMPLES
In compact form
u = T u,
u = T u
(5.18)
which may be used for initial conditions as noted below. Substitute

Likewise for velocities: u = T u,
u and u in the unconstrained matrix EOM M u + Ku = f and premultiply both sides by TT to get
the transformed EOM
(5.19)
TT M T u + TT K T u = TT f.
or
u + K
u = f.
M
(5.20)
in which

= TT M T =
M
8 0 0
10 10 0
0
= TT K T = 10 15 5 , f = TT f =
. (5.21)
0 6 0 , K
0
0 0 10
0 5 5
2H (t)
has rank 2 and nullity 1. The initial conditions u(0) = u0 and

The transformed stiffness matrix M
= u 0 = T u0 and u(0)
= v 0 = T v0 . Once the reduced matrix
u(0)
= v0 transform as per u(0)
EOM (5.26) is solved in time for u = u(t),

the slave displacements, velocities and accelerations
may be recovered from the constraints (5.23); or, equivalently, from the master/slave transformation
equations.
5.6.2. Rheonomic, Nonhomogeneous MFC by Master-Slave
The example truss is subject again to three MFC but now they are nonhomogeneous and rheonomic:
u y1 = u x1 + g1 (t),
u y2 = u x2 + g2 (t),
u y3 = u x3 + g3 (t).
(5.22)
Here gi (t), i = 1, 2, 3 are arbitrary but twice differentiable functions of time, which may reduce
to constants. If all three functions vanish (5.23) coalesces with the homogeneous case (5.23).
Differentiating (5.23) twice in time gives the acceleration constraints:
u y1 = u x1 + g 1 (t),
u y2 = u x2 + g 2 (t),
u y3 = u x3 + g 3 (t).
(5.23)
As in the previous example, the slave displacements are u y1 , u y2 and u y3 , while u x1 , u x2 and u x3
are the master DOF. The master-slave transformations may be written in the compact matrix form
u = T u + g,
u = T u + g ,
(5.24)
in which T, u and u are the same as in (5.17), and

g = [ g1
g2
g3 ]T ,
g = [ g 1
g 2
g 3 ]T ,
(5.25)
The transformed matrix EOM are

TT M T u + TT K T u = TT (f Kg)
(5.26)
u + K
u = f.
M
(5.27)
or
511
anf K
are the same as above, but f is different:
in which M

4g 1
f =
.
5g2 + 5g3 3g 2
H (t) 5g2 + 5g3 + 5g 3
512
(5.28)
Examples on nonholonomic MFC will not be given here, since they are non treatable by master-slave
methods unless they integrate to holonomic constraints.
5.7. Penalty Function Augmentation Methods for MFC
General recommendation: be careful in using penalty function methods to treat dynamic constraints.
To justify this statement one needs to resort to variational techniques, which are the topic of next
Chapter.
5.8. Lagrange Multiplier Adjunction Methods for MFC
These is the most powerful technique available for the treatment of constraints. But explanation of
use is deferred until next Chapter, which introduces variational forms.
Constraints were not important in early days of particle dynamics since the main application was astronomical:
the movement of celestial bodies. Few kinematic constraints appear there except collitions. (Curiously the
same thing appears at the other end of the scale: atomic physics and quantum mechanics.) Collitions were
studied in the XVIII Century, still using particle models. Euler treated isoparametric constraints using the
calculus of variations that he invented. The first investigator that studied constraints in some generality was
Lagrange [389], who developed his method of multipliers as a powerful technique that has been heavily used
to date.
In his Principles of Mechanics book, Hertz was the first to emphasize a clear dictinction between holonomic
and nonholonomic constraints. The importance of the latter had increased during the XIX Century as the
Industrial Revolution focused attention on rolling motions. But the treatment of those constraints opened up
a controversy that persists to the current day, complete with incorrect and misleading statements in even the
most reputed textbooks and monographs. This unfortunate saga is well documented in the Hertz biography
by Lutzen [390, Chapter 23].
Within the FEM context, development in this topic have been minimal and tyipcally application dependent.
512
Dynamics &
Vibration
Overview
91
Chapter 9: DYNAMICS & VIBRATION OVERVIEW
TABLE OF CONTENTS
Page

9.1
Introduction
. . . . . . . . . . . . . . . . . . . . .
9.2
Semidiscrete Equations of Motion . . . . . . . . . . . . .
9.2.1
Vibrations as Equilibrium Disturbance . . . . . . . . .
9.2.2
Undamped Free Vibrations
. . . . . . . . . . . .
9.2.3
The Vibration Eigenproblem
. . . . . . . . . . . .
9.2.4
Eigensystem Properties
. . . . . . . . . . . . .
9.3
Solving the Vibration Eigenproblem
. . . . . . . . . . . .
9.3.1
Determinant Roots . . . . . . . . . . . . . . .
9.3.2
Reduction to the Standard Eigenproblem
. . . . . . . .
9.3.3
Unsymmetric Reduction
. . . . . . . . . . . . .
9.3.4
Symmetry Preserving Reduction . . . . . . . . . . .
9.
. . . . . . . . . . . . . . . . . . . . . .
9.
References . . . . . . . . . . . . . . . . . . . . . .
9.
Exercises . . . . . . . . . . . . . . . . . . . . . .
92
93
93
94
95
96
96
97
97
97
97
98
98
99
910
9.2
SEMIDISCRETE EQUATIONS OF MOTION
9.1. Introduction
Developments in the IFEM-AFEM-NFEM book sequence pertain to static analysis, in which results
are independent of time. This kind of analysis applies also to quasi-static scenarios, in which the
state varies with time but does so slowly that inertial and damping effects can be ignored. For
example one may imagine situations such as a roof progressively burdened by falling snow before
collapse, the filling of a dam, or the construction of a tunnel. Or foundation settlements: think of
the Pisa tower before leaning stopped. The quasi-static assumption is commonly used in design
even for loads that vary in a faster time scale. For example, vehicles travelling over a bridge or
wind effects on buildings.1
By contrast dynamic analysis is appropriate when the variation of displacements with time is so
rapid that inertial effects cannot be ignored. There are numerous practical examples: earthquakes,
rocket launches, vehicle crashes, explosive forming, air blasts, underground explosions, rotating
machinery, airplane flutter. The structural accelerations, which are second derivatives with respect
to time, must be kept in the governing equations. Damping effects, which are associated with
velocities (the first temporal derivatives of displacements), may be also included. However, passive
damping effects are often neglected as they tend to take energy out of a system and thus reduce the
response amplitude.
Dynamic analysis may be performed in the time domain or the frequency domain. The latter is
restricted in scope in that it applies to linear structural models, or to linearized fluctuations about an
equilibrium state. The frequency domain embodies naturally the analysis of free vibrations, which
is the focus of the present Chapter.
Remark 9.1. Mathematically, a dynamical system consists of a phase space together with an evolution law.
(J.C. Yoccoz). A major goal of the theory is to understand the long term behaviour of the system.
9.2. Semidiscrete Equations of Motion

The essence of structural analysis is mastering forces. In the development of FEM, this was
understood by the pioneers of the first generation, as narrated in 1.7.1. With the victory of
the Direct Stiffness Method (DSM) by 1970, displacements came to the foreground as primary
computational variables because they scale well into complicated systems.
To understand dynamic analysis, that dual role must be kept in mind. Displacements become even
more important as computational variables. After all, velocities and accelerations are temporal
derivatives of displacements. There is no easy way to do the job with forces only, since dynamics
is about motion. On the other hand, the fundamental governing equations of structural dynamics
are force balance statements. They are elaborate versions of Newtonian mechanics.
This Newtonian viewpoint is illustrated in Table 9.1 for several modeling scenarios that span
statics, dynamics and vibrations. For notational simplicity it is assumed that the structure has been
discretized in space, for example by the FEM. The right column shows the vector form of the
governing equations as force balance statements. The table defines nomenclature.
1
The quasi-static assumption can be done during design if dynamic effects can be accounted for through appropiate safety
factors. For many types of structures (e.g., buildings, bridges, offshore towers) these are specified in building codes.
This saves time when dynamic effects are inherently nondeterministic, as in traffic, winds or wave effects.
93
Table 9.1. Discrete Structural Mechanics Expressed as Force Balance Statements

Case Problem type
I
General nonlinear dynamics
Governing force balance equations

u,
t) = f(u, u,
u,
p(u,
t)
exter nal
internal
II
General nonlinear statics
p(u) = f(u)

exter nal
internal
III
Flexible structure nonlinear dynamics
u,
t) + pd (u, u,
t) + pe (u, t) = f(u, t)
p (u, u,
i

damping
elastic
iner tial
IV Flexible structure linear dynamics

Linear elastostatics
exter nal
damping
elastic
static equilibrium
damping
elastic
periodic
+ C u(t)
+ K u(t) = 0
u(t)
M

iner tial
IX Undamped free vibrations
exter nal
+ C u(t)
+ K u(t) = f p (t)
u(t)
M

iner tial
VIII Damped free vibrations
elastic
= f(u)
M(u)
d(t) + C(u)
d(t) + K(u)
d(t) + p(u)

iner tial
VII Damped forced vibrations
damping
K u =
f

elastic
VI Dynamic perturbations
exter nal
+ C u(t)
+ K u(t) = f(t)
u(t)
M

iner tial

damping
elastic
+ K u(t) = 0
u(t)
M

iner tial
elastic
Symbol u is array of total displacement DOFs; d in case VI is a linearized perturbation of u.

Symbol t denotes time. Superposed dots abbreviate time derivatives: u = du/dt, u = d 2 u/dt 2 , etc.
The history u = u(t) is called the response of the system. This term is extendible to nonlinear statics.
Initial force effects f I may be accommodated in forced cases by taking f = f I when u = 0.
When the model is time dependent, the relations shown in the right column of Table 9.1 are called
semidiscrete equations of motion. The qualifier semidiscrete says that the time dimension has
not been discretized: t is still a continuous variable. This legalizes the use of time differentiation,
abbreviated by superposed dots, to bring in velocities and accelerations.
This table may be scanned top down by starting with the most general case I: nonlinear structural
dynamics, branching down to more restricted but specific forms. Along the way one finds in case
V an old friend: the DSM master equations K u = f for linear elastostatics, treated in previous
Chapters. The last case IX: undamped free vibrations, is that treated in this and next two Chapters.
Some brief comments are made as regards damped and forced vibrations.
9.2.1. Vibrations as Equilibrium Disturbance
An elastic structure is placed in motion through some short-term disturbance, for example an
impulse. Remove the disturbance. If wave propagation effects are ignored and the structure remains
94
9.2
SEMIDISCRETE EQUATIONS OF MOTION
elastic, it will keep on oscillating in a combination of time-periodic patterns called vibration modes.
Associated with each vibration mode is a characteristic time called vibration period. The inverse of
a period, normalized by appropriate scaling factors, is called a vibration frequency. The structure
is said to be vibrating, or more precisely undergoing free vibrations. In the absence of damping
mechanisms an elastic structure will vibrate forever. The presence of even minute amounts of
viscous damping, however, will cause a gradual decrease in the amplitude of the oscillations. These
will eventually cease.2
If the disturbances are sufficiently small to warrant linearization, this scenario fits case VI of
Table 9.1, therein labeled dynamic perturbations. Its main application is the investigation of
dynamic stability of equilibrium configurations. If the perturbation d(t) is unbounded under some
initial conditions, that equilibrium configuration3 is said to be dynamically unstable.
The analysis of case VI does not belong to an introductory course because it requires advanced
mathematical tools. Moreover it often involve nondeterministic (stochastic) effects. Cases VII
through IX are more tractable in an introductory course. In these, fluctuations are linearized about
an undeformed and unstressed state defined by u = 0. Thus d (the perturbed displacement)
becomes simply u (the total displacement). Matrices M, C and K are called the mass, damping
and stiffness matrices, respectively. These matrices are independent of u since they are evaluated
at the undeformed state u = 0. Two scenarios are of interest in practice:
1.
Forced Vibrations. The system is subjected to a time dependent force f(t). The response u(t)
+ C u(t)
+ K u(t) = f(t) of case IV.
is determined from the linear dynamics equation: M u(t)
Of particular interest in resonance studies in when f(t) is periodic in time, which is case VII.
2.
Free Vibrations. The external force is zero for t > 0. The response u(t) is determined from
initial conditions. If damping is viscous and light, the undamped model gives conservative
answers and is much easier to handle numerically. Consequently the model of case IX is that
generally adopted during design studies.
9.2.2. Undamped Free Vibrations

From the foregoing discussion it follows that case IX: undamped free vibrations is of paramount
importance in design. The governing equation is
+ K u(t) = 0.
M u(t)
(9.1)
This expresses a force balance4 in the following sense: in the absence of external loads the internal
The only ingredient beyond the
elastic forces K u balance the negative of the inertial forces Mu.
by now familiar K is the mass matrix M. The size of these matrices will be denoted by n f , the
number of degrees of freedom upon application of support conditions.
2
Mathematically a damped oscillation also goes on forever. Eventually, however, the motion amplitude reaches a molecular
scale level at which a macroscopic idealization does not apply. At such point the oscillations in the physical structure
can be considered to have ceased.
Usually obtained through a nonlinear static analysis. This kind of study, called dynamic stability analysis, is covered
under Nonlinear Finite Element Methods.
Where is f = ma? To pass to internal forces change the sign of f : f int + ma = ku + ma = 0. Replace by matrices
and vectors and you have (9.1).
95
Equation (9.1) is linear and homogeneous. Its general solution is a linear combination of exponentials. Under matrix definiteness conditions discussed later the exponentials can be expressed as a
combination of trigonometric functions: sines and cosines of argument t. A compact
representation of such functions is obtained by using the exponential form e jt , where j = 1:

u(t) =
vi e ji t .
(9.2)
i
Here i is the i th circular frequency, expressed in radians per second, and vi = 0 the corresponding
vibration mode shape, which is independent of t.
9.2.3. The Vibration Eigenproblem
Replacing u(t) = v e jt in (9.1) segregates the time dependence to the exponential: (2 M +
K) v e jt = 0. Since e jt is not identically zero, it can be dropped leaving the algebraic condition:
(2 M + K) v = 0.
(9.3)
Because v cannot be the null vector, this equation is an algebraic eigenvalue problem in 2 . The
eigenvalues i = i2 are the roots of the characteristic polynomial be indexed by i:
det(K i2 M) = 0.
(9.4)
Dropping the index i this eigenproblem is usually written as

K v = 2 Mv.
(9.5)
If M and K satisfy some mild conditions, solutions of (9.5) are denoted by i and vi . This are
called the vibration frequencies or eigenfrequencies, and the it vibration modes or eigenmodes,
respectively. The set of all i is called the frequency spectrum or simply spectrum.
9.2.4. Eigensystem Properties
Both stiffness K and mass M are symmetric matrices. In addition M is nonnegative. Nothing
more can be assumed in general. For example, if K incorporates Lagrangian multipliers from the
treatment of a MFC, as explained in Chapter 10, it will be indefinite.
If M is positive definite, the following properties hold.
1.
There are n f squared vibration frequencies i2 , which are roots of the characteristic polynomial
(9.4). These are not necessarily distinct. A root of (9.6) that appears m times is said to have
multiplicity m.5
2.
All roots i2 of (9.6) are real. The corresponding eigenmodes vi have real entries.

2
If K is nonnegative, i 0 and the frequencies i = + i2 are also real and nonnegative.
Furthermore, if K is positive definite, all i2 > 0 and consequently + i > 0.
3.
If M is nonnegative, care must be exercised; this case is discussed in an Exercise. If M is indefinite

(which should never happen in structures) all of the foregoing properties are lost.
5
For example, a free-free (fully unsupported) structure has n R zero frequencies, where n R is the number of rigid body
modes.
96
9.3
SOLVING THE VIBRATION EIGENPROBLEM
Example 9.1. This illustrates the weird things that can happen if M is indefinite. Consider
K=
1
,
2
M=
0
1
1
,
1+
(9.6)
where and vary from 1 to 1. Then

M1 K =
1 1
2 1
The eigenvalues are

2 + +
(1 )
.
(1 + + )
(9.7)
[4 4 + (1 + )2 ]
.
(9.8)
4 2
These are complex if the radicand is negative. But that is not all. If 0 one eigenvalue goes to . If
= 0, A = M1 K is a 2 2 Jordan block and one eigenvector is lost.
2
=
1,2
9.3. Solving the Vibration Eigenproblem

In what follows we often denote i = i2 to agree more closely with the conventional notation for
the algebraic eigenproblem.
9.3.1. Determinant Roots
Mathematically the i2 are the roots of the characteristic equation (9.4). The simple minded approach
is to expand the determinant to get the characteristic polynomial P(i2 ) and get their roots:
det(K i2 M) = P(i2 ) = 0.
(9.9)
This approach is deprecated by numerical analysts. It seems as welcome as anthrax. Indeed for
numerical floating point computations of large systems it risks numerical overflow; moreover the
roots of the characteristic polynomial can be very ill-conditioned with respect to coefficients.
For small systems and using either exact or symbolic computation there is nothing wrong with this
if the roots can be expressed exactly in terms of the coefficients, as in the above example.
9.3.2. Reduction to the Standard Eigenproblem
The standard algebraic eigenproblem has the form
Ax = x.
(9.10)
Most library routines included in packages such as Matlab and Mathematica are designed to solve
this eigenproblem. If A is symmetric the eigenvalues i are real; moreover there exist a complete
system of eigenvectors xi . If these are normalied to length one: ||xi ||2 = 1 they satisfy the
orthonormality conditions

1 if i = j
T
xi x j = i j =
,
xiT Ax j = i ,
(9.11)
0 if i = j
where i j is the Kronecker delta. If the xi are collected as columns of a matrix X, the foregoing
conditions can be expressed as XT X = I and XT KX = = diagi .
97
9.3.3. Unsymmetric Reduction

If M is nonsingular, a simple way to reduce Kv = 2 Mv to standard form is to premultiply both
sides by M1 whence
M1 Kv = 2 v
Ax = x,
with
A = M1 K,
= 2 ,
x = v.
(9.12)
The fastest way to form A is by solving MA = K for A. One nice feature of (9.12) is that the
eigenvectors need not be backtransformed, as happens in symmetry-preserving methods.
As in the case of the characteristic polynomial, this is deprecated by numerical analysts, also not so
vehemently. Their objection is that A is not generally symmetric even if K and M are. So Ax = x
has to be submitted to an unsymmetric eigensolver. Thus risks contaminating the spectrum with
complex numbers. Plus, it is slower.
The writers experience is that (9.12) works perfectly fine for small systems. If tiny imaginary
components appear, they are set to zero and life goes on.
9.3.4. Symmetry Preserving Reduction
It is possible to retain symmetry by proceeding as follows. Decompose the mass matrix as
M = LLT
(9.13)
This is the Cholesky decomposition, which can be carried out to completion if M is positive definite.
Then
(9.14)
A = L1 KLT .
The demonstraion is in one of the Exercises. The symmetric eigenproblem can be handled by
standard library routines, which give back all the eigenvalues and eigenvectors. The square root
of the eigenvalues give the vibration frequencies and the vibration modes are recovered from the
relation Lvi = xi , which can be handled by standard library routines.
The literature on dynamics and vibrations of structures is quite large. It is sufficient to cite here titles that
incorporate modern analysis methods: Clough and Penzien [142], Geradin and Rixen [305], Meirovich [479,?]
and Wilson [804].
Several books in matrix methods and FEM books contain at least an introductory treatment of dynamics.
Citable textbooks include Bathe [54], Cook, Malkus and Plesha [148], Hughes [385]. Despite their age,
Przemieniecki [596] remains a useful source of mass matrices, while Pestel and Leckie [572] contains a
catalog of transfer matrices (an early-1960 method suitable for small computers but restricted to 1D models).
As regards books on linear algebra matrix theory and matrix calculus see the Bibliography cited in Appendix
A. The most elegant coverage is that of Strang [699]. Two comprehensive references on matrix computations
in general are Golub and VanLoan [311] and Stewart [693]. The former is more up to date as regard recent
literature. Bellman [?] contains more advanced material. Stewart and Sun [694] cover the sensitivity analysis
of standard and generalized eigenproblems.
There are comprehensive books that treat the algebraic eigenproblem. Wilkinsons masterpiece [796] is dated
in several subjects, particularly the generalized eigenproblem and the treatment of large eigenproblems. But
it is still unsurpassed as the bible of backward error analysis. More up to date in methods is Parlett [542],
which is however restricted to the symmetric eigenproblem.
98
9.
As regards source code for matrix computations, the Handbook compilation of Algol 60 procedures by Wilkinson and Reisch [797] is elegant and still useful as template for other languages. Half of the handbook deals with
eigenvalue problems. By contrast, the description of Fortran EISPACK code [291] suffers from the inherent
ugliness and unreadability of Fortran IV.
And of course there is Numerical Recipes in various flavors. To borrow from the immortal words of Winston
Churchill, never have so few wasted the time of so many.
References
Referenced items moved to Appendix R.
99
Homework Exercises for Chapter 9 - Dynamics & Vibration Overview

EXERCISE 9.1 [A:15]. A 3-element model of a bar in 1D gives
2
1
M=
0
0
1
4
1
0
0
1
4
1
0
0
,
1
2
1 1 0 0
1 2 1 0
K=
.
0 1 2 1
0 0 1 1
(E9.1)
Solve the vibration eigenproblem and show the natural frequencies and associated vibration modes. Normalize
the latter so that VT MV = I (mass normalized eigenvectors).
EXERCISE 9.2 [A:25]. In (E9.1) replace the (4,4) mass entry by 2 and the (4,4) stiffness entry by 1/2.
Using Matlab or Mathematica, solve the eigenproblem for varing from 0to4 in 0.5 increments. Discuss
what happens to the frequencies and vibration modes as goes to 2 and beyond. Explain.
EXERCISE 9.3 [D:20]. Eigenvectors can be scaled by arbitrary nonzero factors. Discuss 4 ways in which
the eigenvectors vi of Kvi = i2 Mvi can be normalized, and what assumptions are necessary in each case.
910
16
Mass Matrix
Construction
Overview
161
Chapter 16: MASS MATRIX CONSTRUCTION OVERVIEW
TABLE OF CONTENTS
Page
16.1
16.2
16.3
16.4
16.5
16.6
16.7
Introduction
. . . . . . . . . . . . . . . . . . . . . 163
Mass Matrix Construction Steps
. . . . . . . . . . . . .
163
Mass Matrix Construction Methods . . . . . . . . . . . . . 163
Element Mass Matrix Construction
. . . . . . . . . . . .
166
16.4.1
Direct Mass Lumping
. . . . . . . . . . . . . . 166
16.4.2
Variational Mass Lumping
. . . . . . . . . . . .
167
Mass Matrix Properties . . . . . . . . . . . . . . . . . 167
Rank and Numerical Integration
. . . . . . . . . . . . .
168
Globalization . . . . . . . . . . . . . . . . . . . . . 169
16.7.1
Directional Invariance . . . . . . . . . . . . . .
169
16.7.2
Failure of Repetition Rule
. . . . . . . . . . . . . 1610
162
16.3
MASS MATRIX CONSTRUCTION METHODS
16.1. Introduction
This Part focuses on the construction of the mass matrix for FEM dynamic analysis of structures.
Chapters 16 through 20 present and illustrate the two standard methods used to accomplish that goal:
direct mass lumping and variational mass lumping. Those methods gave been used for decades, are
well understood by now, and are implemented in production FEM programs. Chapters 21ff deal with
a more general and advanced technique: templates. This approach produces parametrized mass
matrices that include the standard ones as special instances. This newer scheme is still undergoing
development.
For the convenience of the reader, Table 16.1 collects acronyms often used in this Part.
16.2. Mass Matrix Construction Steps
In Part II the master (system) mass matrix of a structural FEM model emerged as the discrete
operator that converts nodal accelerations to inertial nodal forces:
f I = M u.
(16.1)
This relation expresses Newtons second law for a discrete dynamic system with masses constant
in time.1 In the framework of the Direct Stiffness Method (DSM), the construction of M is done
through three steps:
Step 1: Localization: Form the mass matrix of each element, using a local frame if convenient.
Step 2: Globalization: Transform element mass matrices to global coordinates if necessary.
Step 3: Assembly: Merge the globalized element mass matrices to form M.
In practice these operations are carried out concurrently within an element-by-element loop. On
loop exit, the master mass matrix is complete.
Readers familiar with the Direct Stiffness Method (DSM) for FEM static analysis [255] may notice
that the formation of M through the preceding steps largely parallels that of the master stiffness
matrix K. In particular, merging of element mass matrices into the master mass matrix follows
exactly the same techniques. Consequence: assemblers for K and M, before application of boundary
conditions, can be made identical except for obvious indexing shortcuts in the case of diagonal
mass matrices. This procedural uniformity is one of the strengths of DSM.
A notable difference with the stiffness matrix is the possibility of using a diagonally lumped mass
matrix (DLMM) based on the direct mass lumping scheme described below. A master DLMM can
be stored as a vector. If all entries are nonzero, it is easily inverted in place, since the inverse of a
diagonal matrix is diagonal. Plainly using a DLMM entails significant advantages in computations
that involve M1 ; for example explicit time integration [74,766] as well as symmetric eigenproblem
solution [542]. Those benefits are counteracted by some negative features discussed later.
1
For the relatively rare cases in which the mass varies with time, the law must be used in the original form stated by
Newton: the time derivative of momentum equals the inertial force.
163
Table 16.1. Acronyms Used in Part III

Acronym
Stands for
AB
ABTS
BLCD
BLFD
BLFM
CMM
CMS
CMT
COB
COF
DDD
DGVD
DIMM
DLMM
DML
DSF
DTI
DCF
DWN
FFB
FPMM
HF
LCD
LF
LFF
LLMM
MOF
NCT
OB
OBTS
SDAV
SFB
SLMM
SMS
TML
VDMM
VML
VSF
Acoustic branch in DDD: has physical meaning in continuum models

AB Taylor series in DWN , centered at = 0
Best linear combination (LFF sense) of the CMM and a selected DLMM
Best possible DLMM (LFF sense); acronym also applies to MS pair with this mass
Best possible FPMM (LFF sense); acronym also applies to MS pair with this mass
Consistent mass matrix: a special VDMM in which VSM and DSF coalesce
Component Mode Synthesis: model reduction framework for structural dynamics
Congruential (also spelled congruent) mass transformation
Constant optical branch: OB frequency is independent of wavenumber
Cutoff frequency: OB frequency at zero wavenumber (lowest one if multiple OB)
Dimensionless dispersion diagram: DCF vs. DWN
Dimensionless group velocity diagram: c = c/c0 vs. DWN
Directionally invariant mass matrix: repeats with respect to any RCC frame
Diagonally lumped mass matrix; qualifier diagonally is often omitted
Direct mass lumping
Displacement shape functions to interpolate displacements over element
Direct time integration of EOM
Dimensionless circular frequency, always denoted by
Dimensionless wavenumber, always denoted by
Flexural frequency branch in Bernoulli-Euler or Timoshenko beam models
Fully populated mass matrix (at element level); includes CMM as special case
High frequency: short wavelength, small DWN, typically > 1
Mass matrix obtained as linear combination of the CMM and a selected DLMM
Low frequency: long wavelength, small DWN, typically < 1
Low frequency fitting of AB to that of continuum
Lobatto lumped mass matrix: a DLMM based on a Lobatto quadrature rule
Maximum overall frequency: largest frequency in DDD over Brillouin zone
Non-continuum term: a term in the ABTS that is not present in the continuum
Optical branch (or branches) in DDD: no physical meaning in continuum models
OB Taylor series in DWN , centered at = 0
Structural dynamics and vibration applications: low frequency range important
Shear frequency branch in the Timoshenko beam model
Simpson lumped mass matrix: a LLMM based on Simpsons 3-pt quadrature rule
Selective mass scaling: modifying a mass matrix by adding a scaled stiffness
Template mass lumping
Variational derived mass matrix: Hessian of discretized kinetic energy
Variational mass lumping
Velocity shape functions to interpolate velocities and produce a VDMM
164
16.3
MASS MATRIX CONSTRUCTION METHODS
16.3. Mass Matrix Construction Methods

Structural elements based on continuum models have distributed mass characterized by the material
density. Associated inertia forces are body forces: forces per unit volume triggered by motioninduced accelerations. The FEM discretization transforms those distributed forces to nodal (point)
forces f I , which are linked to nodal accelerations by M as per (16.1).
The discretization process is generically called mass lumping or simply lumping.2 A mass matrix
that satisfied certain behavioral constraints, such as nonnegativity and mass conservation, will be
called admissible.
As noted in 16.2 the construction of M starts at the element level. Three methods to construct a
element mass matrix can be distinguished:
Direct Mass Lumping (DML). The total mass of the element is distributed to the nodes so that a
diagonally lumped mass matrix (DLMM) is produced. The physical interpretation of this particular
matrix configuration is that no inertial interaction occurs between the lumped masses.
Variational Mass Lumping (VML). The kinetic energy of the element is expressed in terms of the
degrees of freedom (DOF) using an interpolation scheme for the nodal velocities. The Hessian of the
kinetic energy taken with respect to the DOF produces a variational mass matrix, which is generally
non-diagonal. If the element stiffness is constructed using displacement shape functions, and these
are used to interpolate velocities, the so-called consistent mass matrix or CMM is obtained.
Template Mass Lumping (TML). This approach aims to produce an element mass matrix Me that
has free parameters. Their range is restricted so that Me satisfies admissibility conditions. This is
called a mass matrix template. Setting parameters to specific numeric values produces a template
instance or simply instance. If the template happens to include all admissible mass matrices, it is
called a general template.
The union of the first two methods: direct and variational mass lumping, will be collectively
called the standard approach to the construction of mass matrices. Since both CMM and DLMM
configurations are by construction admissible, they are simply instances of a general template. Why
then bother to make a distinction? Practical reasons:
Both direct and variational lumping have been used in structural FEM for a fairly long time.3
They are straightforward to explain and implement, and are well understood after decades of
experience. Furthermore, they produce specific matrices, avoiding parameters.
The template approach is more recent and there is less experience with it. The presence of free
parameters makes derivations far more involved and usually require the help of a computer
algebra system (CAS) to arrive at useful results in reasonable time. Most of the derivations
to date concern one-dimensional elements. Extension to multidimensional elements poses
challenges outlined later.
In English the verb to lump conveys the following meanings: combine, put together, group, bunch, aggregate, aglomerate, unite, pool, merge, collect, throw together, consider together. Its use for mass discretization comes from the direct
lumping process through which an object of finite extent is idealized as a point mass. The idea can be traced back to
centuries of orbital computations: celestial objects such as planets were idealized as point masses.
DLMM since the 1930s and CMM since the 1960s, as narrated in Appendix H.
165
(a)
Element total mass is m = A

1
m1 =
me
2
m2 =
(b)
x
ux1
me
m1
ux2
m2
massless connector
Figure 16.1. Direct mass lumping for two-node prismatic bar element: (a) lumping element
mass to end nodes; (b) endowing the element with 2 translational degrees of freedom.
Because of the newness, dimensionality restrictions and computational demands imposed by templates, the derivation of mass matrices is restricted to the time-tested standard approach in this and
following chapters.
16.4. Element Mass Matrix Construction
The master mass matrix is built up from element contributions, and we start at that level. The
construction of the mass matrix of individual elements with distributed mass density can be carried
out through any of the three methods outlined above.4 By now both direct and variational mass
lumping enjoy extensive coverage in the structural dynamics literature at the textbook level; see
e.g., [142,158,305,596,685], and references therein. They are implemented in all general purpose
FEM codes.
16.4.1. Direct Mass Lumping
This is the simplest procedure. The total mass of element e is directly apportioned to nodal
freedoms, ignoring any cross coupling. The goal is to build a diagonally lumped mass matrix or
DLMM, denoted here by MeL .
As the simplest example, consider a 2-node prismatic bar element with length , cross section area
A, and mass density , which can only move in the axial direction x, as shown in Figure 16.1(a).
We often denote this element as Bar2 in the sequel. The total element mass is m e = A. This is
divided into two equal parts and assigned to each end node. The element is endowed with the two
freedoms shown in Figure 16.1(b). Thus

1 0
e
1
= 12 m e I2 ,
(16.2)
M L = 2 A
0 1
in which m e = A is the element mass and I2 denotes the 2 2 identity matrix. As sketched in
Figure 16.1, we have effectively replaced the continuum bar with a dumbbell: two masses separated
by a massless connector.
This process conserves the translational kinetic energy or, equivalently, the linear momentum. To
check this property for the bar example, take the constant x-velocity vector u e = v [ 1 1 ]T . The
kinetic energy of the element is T e = 12 (u e )T MeL u e = 12 A v 2 = 12 m e v 2 . Thus the linear
4
Beyond the element level, methods to produce the master mass matrix coalesce.
166
16.5
MASS MATRIX PROPERTIES
momentum p e = T e /v = m e v is preserved. When applied to simple elements that can rotate,

however, the direct lumping process generally doe not preserve angular momentum.
Historical motivations for direct lumping are noted in H.1. Most crucial, it covers naturally
the case where concentrated (point) masses are natural part of model building. For example, in
aircraft engineering it is common to idealize nonstructural masses (fuel, cargo, engines, etc.) as
concentrated at given locations. (Such point masses in general have rotational freedoms; rotational
inertia lumping is then part of the process.)
16.4.2. Variational Mass Lumping
The second standard procedure is based on a variational formulation. This is done by taking the
kinetic energy as part of the governing functional. The kinetic energy of an element of mass density
that occupies the domain e and moves with velocity field ve is

T =
e
1
2
e
(ve )T ve de .
(16.3)
Following the conventional FEM philosophy, the element velocity field is interpolated using shape
functions: ve = Nv u e , in which u e are node DOF velocities and Nv a shape function matrix. (For
1D elements, Nv is a row vector.). Inserting into (16.3) and taking the node velocities out of the
integral yields

def
e
1 e T
T = 2 (u )
(Nv )T Nv d u e = 12 (u e )T Me u e ,
(16.4)
e
whence the element mass matrix follows as the Hessian of T e :

2T e
M = e e =
u u
e
(Nev )T Nv d.
(16.5)
If the same shape functions used in the derivation of the stiffness matrix are chosen, that is, Nev = Ne ,
(16.5) is called the consistent mass matrix or CMM. It is denoted here by MCe . A better name
for (16.5) would be stiffness-consistent mass matrix. The shorter sobricket has the unfortunate
implication that other choices are inconsistent, which is far from the truth. In fact, the consistent
mass is not necessarily the best performer, a topic elaborated in Chapters that deal with templates.
The shorter name is, however, by now ingrained in the FEM literature.
For the Bar2 element moving along x, pictured in Figure 16.1(a), the well known stiffness shape
functions are N1 = 1 (xx1 )/ = (1 )/2 and N2 = (xx2 )/ = (1 + )/2, in which
= 2(x x1 )/ 1 is the isoparametric natural coordinate that varies from 1 at node 1 to +1 at
node 2. With d x = 12 d , the consistent mass is easily obtained as

1
=
A (N ) N d x =
A
.
[ 1 1 + ] d = m
2
0
1
(16.6)
It can be verified that this mass matrix preserves linear momentum along x. If allowed to move in
the x y plane, as considered in 16.7, it also preserves angular momentum about z.
MCe
e T
+1
1
1+
167
1
6
2
1
16.5. Mass Matrix Properties

Mass matrices must comply with conditions that can be used for verification and debugging at the
element level. They are: matrix symmetry, physical symmetries, conservation and positivity.
Matrix Symmetry. This means (Me )T = Me , which is easy to check. For a variationally derived
mass matrix this follows directly from the definition (16.5), whereas for a DLMM is automatic.
Physical Symmetries. Also called geometric or fabrication symmetries. They are dictated by the
physical configuration. For example, the CMM or DLMM of the prismatic Bar2 element must be
symmetric about the antidiagonal: M11 = M22 . To see this, flip the end nodes: the element remains
the same and so does the mass matrix.5
Conservation. At a minimum, total element mass must be preserved (we are talking about classical mechanics here; in relativistic mechanics mass and energy can be exchanged). This is easily
verified by applying a uniform translational velocity and checking that linear momentum is conserved. Higher order conditions, such as conservation of angular momentum, are optional and not
necessarily desirable.
Positivity. For any nonzero velocity field defined by the node values u e
= 0, (u e )T Me u e 0. That
is, Me must be nonnegative. Unlike the previous three conditions, this constraint is nonlinear in the
mass matrix entries. It can be checked in two ways: through the eigenvalues of Me , or the sequence
of principal minors. The second technique is more practical if the entries of Me are symbolic.
A stricter form of the last condition requires that Me be positive definite: (u e )T Me u e > 0 for any
u e
= 0. This is physically reassuring because one half of that form is the kinetic energy associated
with the velocity field defined by u e . In a continuum T can vanish only for zero velocities (a rest
state). But allowing T e = 0 for some nonzero u e makes life easier in some situations; e.g., elements
with rotational or multiplier freedoms, or in the rapid-transient applications noted in H.4.
The u e for which T e = 0 collectively form the null space of Me . Because of the conservation
requirement, a rigid velocity field (that is, the time derivative u eR of a rigid body mode ueR ) cannot
be in the mass matrix null space, as it would imply zero total mass. This scenario is dual to that
of the element stiffness matrix. For the latter, Ke ueR = 0 because a rigid body motion produces no
strain energy. Thus ueR must be in the null space of the stiffness matrix.
16.6. Rank and Numerical Integration
Suppose the element has a total of n eF freedoms. A mass matrix Me is called rank sufficient or of
e
= n eF . Because of the positivity requirement, a rank-sufficient mass matrix
full rank if its rank is r M
must be positive definite. Such matrices are preferred from a numerical stability standpoint.
e
e
e
< n eF the mass is called rank deficient by d M
= n eF r M
. Equivalently Me is
If Me has rank r M
e
times singular. For a numerical matrix the rank is easily computed by taking its eigenvalues
dM
and looking at how many of them are zero. The null space can be extracted by functions such as
NullSpace in Mathematica without the need of computing eigenvalues.
The computation of Me by the variational formulation (16.5) is often done using Gauss numerical
quadrature. Each Gauss points adds n D to the rank, where n D is the row dimension of the shape
5
The antisymmetry property would not generally hold if the element is not prismatic.
168
16.7
GLOBALIZATION
function matrix Ne , up to a maximum of n eF . For most elements n D is the same as element spatial
dimensionality; that is, n D = 1, 2 and 3 for 1, 2 and 3 dimensions, respectively. This property can
be used to pick the minimum Gauss integration rule that makes Me positive definite.
16.7. Globalization
Like their stiffness counterparts, mass matrices are often developed in a local or element frame.
Should globalization be necessary before merge, a congruent transformation is applied:
Te .
Me = (Te )T M
e
(16.7)
is the element mass referred to a local frame xi (a.k.a. element frame), whereas Te is
Here M
the local-to-global displacement transformation matrix. The recipe (16.7) follows readily from the
Principle of Virtual Work, or equivalently the invariance of the first variation of the element kinetic
energy:
e
e ue = (u e )T (Te )T M
e Te u e = (u e )T Me u e = T e .
(16.8)
T e = (u )T M
e
Matrix Te is in principle the same used for the stiffness globalization. Some procedural differences,
however, must be noted. For stiffness matrices Te is often rectangular if the local stiffness has lower
dimensionality. For example, two-node bar, shaft and spar elements have 2 2 local stiffnesses.
Globalization to 2D and 3D involves application of 2 4 and 2 6 transformation matrices,
respectively. This works fine because the local element has zero stiffness in some directions, and
those zero rows and columns may be omitted at the local level.
In contrast to stiffnesses, translational masses never vanish. One way to realize this is to think of
an element moving in a translational rigid motion u R with acceleration u R . According to Newtons
second law, f R = m e u R , where m e is the element translational mass. Regardless of direction, this
inertia force cannot vanish.
Conclusion: all translational masses must be retained in the local mass matrix. A two-node
prismatic bar, moving in the {x, y} plane as in Figure 16.2, furnishes a simple illustration. With
the element freedoms arranged as ue = [ u x1 u x2 u y1 u y2 ]T , the local mass matrix constructed by
variationally consistent and diagonalized lumping are, respectively,
2
e
1
C = 1 me
M
6
0
0
1
2
0
0
0
0
2
1
0
0
,
1
2
1
e
0
L = 1 me
M
2
0
0
0
1
0
0
0
0
1
0
0
0
=
0
1
1
2
m e I4 ,
(16.9)
in which m e = A is the total element mass. For 3D, repeat the diagonal block once more.
16.7.1. Directional Invariance
For the case illustratedin in Figure 16.2 the local-to-global freedom transformation u e = Te ue is

c
u x1
u x2 0
=
u y1
s
u y2
0
0
c
0
s
s
0
c
0
0
u x1
s u x2
,
u y1
0
u y2
c
in which
169
c = cos ,
s = sin .
(16.10)

_
uy2
uy2
ux2
Element RCC frame

_
y
_
uy1
uy1
x
Global RCC frame
m1
ux2
m2
ux1
ux1
massless connector
Figure 16.2. Bar2 element with diagonally lumped mass moving in 2D.
Ce and
Now apply (16.7) to either MCe or MeL of (16.9) using (16.10). The result is MCe = M
eL : no change. We say that these mass matrices repeat. Verification for the DLMM is
MeL = M
eL Te = 1 m e (Te )T I4 Te = 1 m e (Te )T Te = 1 m e I4 . For
easy because Te is orthogonal: (Te )T M
2
2
2
the CMM, however, repetition is not obvious. It can be shown to hold by expressing MCe and Te in
2 2 partitioned form
e
M
MC =
0

0
cI2
e
, T =
sI2
M

sI2
,
cI2
=
with M

1
6
2
1

1
.
2
(16.11)
Carrying out the congruent transformation in block form gives

(cs cs)M
M
(c2 + s 2 )M
=
= (T ) MC T =
2
2
(c + s )M
0
(cs cs)M
MCe
e T

0
e
= MC .
M
(16.12)
A mass matrix that repeats upon transformation to any global frame is called a directionally invariant
are irrelevant to the result (16.12).
mass matrix, or DIMM. Note that the contents and order of M
Hence the following generalization follows. If upon rearranging the element DOF so that they are
grouped node by node:
e has a repeating block diagonal form, and
(i) M
(ii) Te is configured as the block form shown above,
then local and global matrices will coalesce. For (ii) to hold, it is sufficient that all nodal DOF be
translational and be referred to the same coordinate system. The same conclusion is easily extended
to 3D, and to any arrangement of the element freedoms. This repetition rule can be summarized as
follows:
A local mass matrix is DIMM if all element DOFs are translational and all of
them are referred to the same global RCC system.
(16.13)
This property should be taken advantage of to skip superfluous local-to-global transformations. That
operation may cost more than forming the local mass matrix. If the rule fails on actual computation,
something (mass matrix or transformation) is wrong and must be fixed.
1610
16.7
GLOBALIZATION
16.7.2. Failure of Repetition Rule

e is not a DIMM. This occurs under the following
The repetition rule can be expected to fail if M
scenarios:
1.
The element has non-translational freedoms; for example node rotations, or displacement
derivatives. (Occasionally the rule may work, but that should not be taken for granted.)
2.
The mass blocks are different in content and/or size. This occurs if different continuum models
are used in different local directions. Examples are furnished by beam-column elements, shell
elements, and elements with curved sides or faces.
3.
Nodes are referred to different coordinate frames in the global system. This can happen if
certain nodes are referred to special frames to facilitate the application of boundary conditions.
1611
17
Standard Mass
Matrices
for Bar Elements
171
Chapter 17: STANDARD MASS MATRICES FOR BAR ELEMENTS
TABLE OF CONTENTS
Page
17.1
17.2
The Three-Node Bar

. . . . . . . . . . .
17.1.1
The Three-Node Bar With Equidistant Nodes
17.1.2
The Three-Node Bar With Offset Midnode .
17.1.3
The Three-Node Bar With Variable Area . .
The Four-Node Bar
. . . . . . . . . . .
172
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. . .
. .
. . .
. .
. . .
173
173
174
174
174
17.1
(a)
(b)
Midnode 3
@ center
,E,A = const
1
= 1
(e)
3
Center
,E,A = const
(e)
1
2 x
=1
=0
THE THREE-NODE BAR
Midnode 3 offset
from center by
2 x
3
= 1
=0
=1
Isoparametric natural
coordinate
coordinate
= Le
= Le
Figure 17.1. Three-node prismatic bar element with (a) centered midnode; (b) offset midnode.
To further illustrate the use of the standard mass lumping methods, this Chapter derives the consistent
mass matrix (CMM) and diagonally lumped mass matrix (DLMM) for two bar elements: Bar3 (3
nodes), and Bar4 (4 nodes).1 Since all examples pertain to the element level, overbars to distinguigh
local and global frames are omitted for brevity.
17.1. The Three-Node Bar
The three-node bar element, henceforth abbreviated to Bar3, is shown in Figure 17.1(a). It is
prismatic with length , area A, and uniform mass density , Midnode 3 is at the center. The DOFs
are arranged ue = [ u x1 u x2 u x3 ]T . The three shape functions, collected in the shape function
matrix, are
Ne = [ N1 ( ) N2 ( ) N3 ( ) ] = [ (1 )/2
(1 + )/2
1 2 ] .
(17.1)
in which is the isoparametric natural coordinate pictured in Figure 17.1(a).

17.1.1. The Three-Node Bar With Equidistant Nodes
Midnode 3 is at the center. The consistent mass follows as
MCe M M
me
e T e
= A
(N ) N J d =
30
1
4
1
2
1
4
2

2
2 ,
16
(17.2)
in which the Jacobian J = d x/d = /2, and m e = A .

To produce a DLMM, the total mass of the element is divided into three parts: A, A, and
(1 2) A, which are assigned to nodes 1, 2 and 3, respectively. See Figure 17.2. As discussed
below the standard choice is = 1/6. Consequently 2/3 of the total mass goes to the midpoint,
and what is left to the corners, giving

MeS L M M = 16 m e
1
1
0
0
0
1
0

0
0 .
4
Matrices for the Bar2 (2 nodes) element were derived in the previous Chapter.
173
(17.3)
Chapter 17: STANDARD MASS MATRICES FOR BAR ELEMENTS
Element total mass is me = A

3
1
(a)
m1 =
2 x
m2 =
me
m3 =
massless connector
(b)
ux1
me
m1
ux2
ux3
m3
m2
me
Figure 17.2. Diagonal mass lumping for 3-node bar element (a) lumping as per Simpsons
integration rule; (b) assigning element freedoms.
The 1:1:4 allocation happens to be Simpsons rule for integration, whence the label SLMM. This
meshes in with the interpretation of diagonal mass lumping as a Lobatto integration rule, a topic
discussed in D.2. Both (17.2) and (17.3) are DIMM, and may be used as 3 3 building for blocks
to expand the element to 2D or 3D space. The repetition rule (16.13) holds.
17.1.2. The Three-Node Bar With Offset Midnode
To be done
17.1.3. The Three-Node Bar With Variable Area
To be done
17.2. The Four-Node Bar
To be done
174
18
Standard Mass
Matrices For Plane
Beam Elements
181
Chapter 18: STANDARD MASS MATRICES FOR PLANE BEAM ELEMENTS
TABLE OF CONTENTS
Page
18.1
18.2
The Plane BE Beam

. . . . . . . . . . . . . . . . . .
The Plane Timoshenko Beam
. . . . . . . . . . . . . .
182
183
184
18.1
THE PLANE BE BEAM
18.1. The Plane BE Beam

The two-node plane BE-beam element has length , cross section area A and uniform mass density
. Only the translational inertia due to the lateral motion of the beam is considered in the kinetic

x)
2 d x of the element, whereas its rotational inertia is ignored. The freedoms
energy T = 12 0 v(
are arranged as ue = [ v1 1 v2 2 ]T . The natural coordinate varies from = 1 at node 1
(x = 0) to = +1 at node 2 (x = ), whence d x/d = 12 and d/d x = 2/. The well known
cubic shape functions in terms of are collected in the shape function matrix
Ne = [ 14 (1 )2 (2 + )
1
(1
8
)2 (1 + )
1
(1
4
+ )2 (2 )
18 (1 + )2 (1 ) ]
(18.1)
The CMM obtained by analytical integration is

MCe M M = A
156
22
e
m
22
42
J (Ne )T Ne d =
54
13
420
13 32
in which the Jacobian J = d x/d = /2 and m e =

integration rules of 1, 2 and 3 points are
16 4 16 4
86 13 22
2
2
4
4
13 22 5
C1
, C2
16 4 16 4
22 5 86
2
2
4 4
5 2 13
54
13
156
22
13
32
.
22
42
(18.2)
A . The mass matrices obtained with Gauss
5
444 62
2

62 112
, C3
13
156 38
2
2
38 92
38
92
,
62
112
(18.3)
in which C1 = m e /64, C2 = m e /216 and C3 = m e /1200. Their eigenvalue analysis shows that all
three are singular, with rank 1, 2 and 3, respectively. The result for 4 and more points agrees with
(18.2), which has full rank. The main purpose of this example is to illustrate the rank property stated
in 16.6: each Gauss point adds one to the rank up to 4, since the problem is one-dimensional.
156
38
444
62
The matrix (18.2) conserves linear and angular momentum. So do the reduced-integration mass
matrices (18.3) if the number of Gauss points is 2 or greater.
To get a diagonally lumped mass matrix is trickier. Obviously the translational nodal masses must
be the same as that of a bar: 12 A. See Figure 18.1. But there is no easy road on rotational masses.
To accommodate these variations, it is convenient to leave the latter parametrized as follows
1
0
0
0
2
0
0 2 0
eL = m e
(18.4)
M
, 0.
1
0
0
0
2
0
0
0 2
Here is a nonnegative parameter, typically between 0 and 1/100. The choice of has been argued
in the FEM literature over several decades, but the whole discussion is largely futile. Matching
the angular momentum of the beam element gyrating about its midpoint gives = 1/24. This
violates the positivity condition. It follows that the best possible as opposed to possible best
is zero. This choice gives, however, a singular mass matrix. This is undesirable in scenarios
where a mass-inverse appears.
183
Chapter 18: STANDARD MASS MATRICES FOR PLANE BEAM ELEMENTS
Total mass A
A
xA
Figure 18.1. Direct mass lumping for two-node plane BE beam element.
This result can be readily understood physically. The m e /2 translational end node masses grossly
overestimate (in fact, by a factor of 3) the angular momentum of the element. Hence adding any
rotational lumped mass only makes things worse.
18.2. The Plane Timoshenko Beam
The Timoshenko beam (Ti-beam) incorporates two refinements over the Bernoulli-Euler (BE)
model:
1.
For both statics and dynamics: plane sections remain plane but not necessarily normal to the
deflected midsurface. See Figure 24.4 for the kinematics. This assumption allows the averaged
shear distortion to be included in both strain and kinetic energies.
2.
In dynamics: the rotary inertia is included in the kinetic energy.
This model is more important for dynamics and vibration than BE, and indispensable for rapid
transient and wave propagation analysis. More specifically, the BE beam has infinite phase velocity,
because the EOM is parabolic, and thus becomes useless for high-fidelity wave propagation.
According to the second assumption, the kinetic energy of the Ti-beam element is given by

T =
1
2

2 d x.
A v(x)
2 + I R (x)
(18.5)
Here I R is the second moment of inertia to be used in the computation of the rotary inertia and
= v + is the cross-section rotation angle shown in Figure 24.4; = V /(G As ) being the
section-averaged shear distortion. The element DOF are ordered ue = [ v1 1 v1 2 ]T . The
lateral displacement interpolation is
e
e
v( ) = v1 Nv1
( ) + v1 Nve 1 ( ) + v2 Nv2
( ) + v2 Nve 2 ( ),
2x
1,
(18.6)
in which cubic interpolation functions are used. A complication over BE is that the rotational
freedoms are 1 and 2 but the interpolation (18.6) is in terms of the neutral surface end slopes:
v1 = (dv/d x)1 = 1 and v2 = (dv/d x)2 = 2 . From a kinmatic analysis we can derive
the relation

v1

2
1+ 2
1
v1
1

=
(18.7)
,

1+

v2
v2
1 +

2
2
2
184
18.2
THE PLANE TIMOSHENKO BEAM
in which the dimensionless parameter = 12E I /(G As 2 ) characterizes the ratio of bending and
shear rigidities. The end slopes of (18.7) are replaced into (18.6), the interpolation for obtained,
and v and inserted into the kinetic energy (18.5).
After lengthy algebra the CMM emerges as the sum of two contributions:
MCe M M = MCe T + MCe R =
13 7
11
11
1
9
13
3
1
+ + 13 2 ( 210
+ 120
+ 24
2 )
+ 3 + 16 2
( 420
+ 40
+ 24
2 )
35 10
70 10
1
1
1
13
3
1
1
1
1
( 105
+ 60
+ 120
2 )2 ( 420
+ 40
+ 24
2 ) ( 140
+ 60
+ 120
2 )2
CT
7
1
11
11
1
13
2
2
+ + 3
( 210 + 120 + 24 )
35 10
symmetric
+ CR
1
( 10
12 )
65
2
( 15
+ 16 + 13 2 )2
1
( 10
+ 12 )
6
5
1
1
1
( 105
+ 60
+ 120
2 )2
1
( 10
12 )
1
( 30
+ 16 16 2 )2
.
1
1
( 10 + 2 )
2
+ 16 + 13 2 )2
( 15
symmetric
(18.8)
in which C T = A /(1 + ) = m /(1 + ) and C R = I R /((1 + ) ). Matrices MC T and
MC R account for translational and rotary inertia, respectively. Caveat: the I in = 12E I /(G As 2 )
is the second moment of inertia that enters in the elastic flexural elastic rigidity. If the beam is
homogeneous I R = I , but that is not necessarily the case if, as sometimes happens, the beam has
nonstructural attachments that contribute rotary inertia.
2
The scale factor of MCe R can be further transformed to facilitate parametric studies by introducing
r R2 = I R /A as cross-section gyration radius and = r R / as element slenderness ratio. Then
C R = I R /((1 + )2 ) = A 2 /(1 + )2 = m e 2 /(1 + )2 . If = 0 and = 0, MCe R
vanishes and MCe T in (18.8) reduces to (18.2).
A DLMM can be obtained through the HRZ scheme explained in D.1. The optimal lumped mass
is derived in 24.2.5 via templates.
185
19
Standard Mass
Matrices for Plane
Stress Elements
191
Chapter 19: STANDARD MASS MATRICES FOR PLANE STRESS ELEMENTS
TABLE OF CONTENTS
Page
19.1
19.2
The Plane Stress Linear Triangle . . . . . . . . . . . . . .

The Plane Stress Bilinear Quadrilateral . . . . . . . . . . .
192
193
193
19.2
THE PLANE STRESS BILINEAR QUADRILATERAL
Here we pass to two dimensional elements in a plane stress state, also called membrane elements
in the literature.
19.1. The Plane Stress Linear Triangle
We consider the three-node linear displacement triangle to model a plate in plane stress. The element
will be identified as Trig3 in the sequel. Its formulation using triangular natural coordinates i is
available online [255]. For the following Me derivations, the plate is assumed to have constant
mass density , area A, and uniform thickness h. The motion is restricted to the {x, y} plane.
The six DOFs are arranged as ue = [ u x1 u y1 u x2 u y2 u x3 u y3 ]T . The CMM is obtained using
the well known displacement shape functions (DSF), which are simply the triangular coordinates
i . Accordingly the shape function matrix is
N = 1
0
e
0
1
2
0
0
2
3
0

0
.
3
(19.1)
Expanding (Ne )T Ne gives a 6 6 matrix quadratic

in the triangular
coordinates. This can be area

integrated with formulas exemplified by e 12 d = A/3, e 1 2 d = A/6, etc. The result
is
2
1 1 0 1 2 0 1 3 0
0 1 1 0 1 2 0 1 3
0

me
1
2 1 0 2 2 0 2 3 0
= h
d =
2 1 0 2 2 0 2 3
12 0
e 0
3 1 0 3 2 0 3 3 0
1
0 3 1 0 3 2 0 3 3
0
MCe M M
0
2
0
1
0
1
1
0
2
0
1
0
0
1
0
2
0
1
1
0
1
0
2
0
0
1
0
(19.2)
1
0
2
in which m e = Ah. This computation may be done by numerical integration, using Gauss rules.
Since the order of Me is 6, and each Gauss point adds two (the number of space dimensions) to
the rank, a rule with 3 or more points is required to reach full rank, as can be verified by simple
numerical experiments.
The diagonally lumped mass matrix is constructed by taking the total element mass element, which
is Ah, dividing it by 3 and assigning those to the corner nodes. See Figure 19.1. This process
produces a diagonal matrix:
MeDL M M =
Ah
me
diag [ 1 1 1 1 1 1 ] =
I6 .
3
3
(19.3)
If this element is used in three dimensions (for example as membrane component of a shell element),
it is necessary to insert the normal-to-the-plate z mass components in either (19.2) or (19.3).
According to the invariance rule (16.13) the globalization process is trivial because MCe or MeL
becomes RBD on grouping the element DOFs by component. Plainly the local element mass
matrix repeats in the global frame.
193
Chapter 19: STANDARD MASS MATRICES FOR PLANE STRESS ELEMENTS
Ah
Total mass Ah
one third goes
to each node
y
x
Ah
Ah
massless
wireframe
Figure 19.1. Diagonal mass lumping for the Trig3 element in plane stress.
19.2. The Plane Stress Bilinear Quadrilateral

We finally consider the 4-node, 8 DOF bilinear quadrilateral modeling a plate in plane stress.
The element is identified as Quad4 in the sequel. It is assumed homogeneous with density
and constant thickness h. It moves in the x, y plane. The nodal displacement vector is ue =
[ u x1 u y1 u x2 . . . u y4 ]. The shape functions and appropriate Gauss quadrature rules are described
in [255].
The integration is carried out numerically using a p p Gauss product rule, with p variable. Testing
the mass matrix module on a rectangular element of dimensions a and b in the x and y directions,
respectively, returns the following CMMs for the 11 and 22 Gauss rules:
1 0 1 0 1 0 1 0
4 0 2 0 1 0 2 0
0 1 0 1 0 1 0 1
0 4 0 2 0 1 0 2
1 0 1 0 1 0 1 0
2 0 4 0 2 0 1 0
0 1 0 1 0 1 0 1
0 2 0 4 0 2 0 1
MCe M M1 = C1
, MCe M M2 = C2
,
1 0 1 0 1 0 1 0
1 0 2 0 4 0 2 0
0 1 0 1 0 1 0 1
0 1 0 2 0 4 0 2
1 0 1 0 1 0 1 0
2 0 1 0 2 0 4 0
0 1 0 1 0 1 0 1
0 2 0 1 0 2 0 4
(19.4)
in which C1 = abh/32 = m e /32 and C2 = abh/72 = m e /72. The mass given by 1-point
integration has rank 2 and 6 zero eigenvalues, and thus it is rank-deficient by 6. The mass given by
the 22 rule is rank-sufficient and positive definite. Either matrix repeats on globalization. Using
rules with 3 or more points returns the same matrix. The DLMM is obtained by assigning one
fourth of the total element mass m e = abh to each freedom.
For a quadrilateral of general geometry, use of 2 2 Gauss quadrature rule is recommended, as it
provides full mass matrix ran
194
21
Mass Matrix
Templates:
General Description
211
Chapter 21: MASS MATRIX TEMPLATES: GENERAL DESCRIPTION
TABLE OF CONTENTS
Page
21.1
21.2
21.3
Templates: A Tool for Mass Matrix Customization

. . . . . . . 213
Is Customization Worth The Trouble?
. . . . . . . . . . .
213
Mass Parametrization Techniques
. . . . . . . . . . . . . 214
21.3.1
Matrix-Weighted Parametrization
. . . . . . . . . .
216
21.3.2
Spectral Parametrization . . . . . . . . . . . . . . 217
21.3.3
Entry-Weighted Parametrization . . . . . . . . . . .
218
21.3.4
Multilevel Parametrization . . . . . . . . . . . . . 218
21.3.5
Selective Mass Scaling . . . . . . . . . . . . . .
218
21.3.6
Singular Mass Matrices . . . . . . . . . . . . . . 219
21.3.7
Constant Optical Branch Variant
. . . . . . . . . .
219
21.3.8
Mass-Stiffness Template Pairs . . . . . . . . . . . . 219
21.3.9
Frequency Dependent Templates
. . . . . . . . . . 2110
212
21.2
IS CUSTOMIZATION WORTH THE TROUBLE?
21.1. Templates: A Tool for Mass Matrix Customization

The present Chapter provides a general description of template mass lumping. This is a general approach through which customized mass matrices can be constructed for specific structural
elements.The qualifier customized is defined more precisely later.
The standard procedures for constructing FEM mass matrices are well known. They are presented in
Chapter 1620. They lead to consistent and diagonally-lumped forms, respectively. Conventional
forms of those models are denoted by MC and M L , respectively, with additional subscripts or
superscripts as necessary or convenient. Abbreviations CMM and DLMM, respectively, are also
used. Collectively those two models take care of many engineering applications in structural
dynamics. Occasionally, however, they fall short. The gap can be filled with a more general
approach that relies on templates. These are algebraic forms that carry free parameters. The set of
parameters is called the template signature. When given numerical values, the signature uniquely
characterizes a mass matrix instance. Templates are described in this and the next 5 chapters.
The template approach has the virtue of generating a set of mass matrices that satisfy certain a priori
constraints; for example symmetry, nonnegativity, invariance and linear momentum conservation.
A mass matrix that satisfies those will be called admissible. In particular, the diagonally-lumped
and consistent mass matrices should be obtained as instances. Thus those standard models are not
excluded. Availability of free parameters, however, allows the mass matrix to be customized to
special requirements.
Several customization scenarios are listed in Table 21.1, along with their acronyms. The last
one: reduction of directional anisotropy in wave propagation, is not applicable to one-dimensional
elements and therefore not treated in this paper.
The versatility of application will be evident from the examples. It will be also seen that optimizing
templates for one scenario generally does not help with others, and in fact may make things worse.
Thus, ability to adapt the mass matrix to particular needs as well as problem regions is an important
virtue. Note that mesh and freedom configuration need not be modified in any way; only template
signatures are adjusted.
An attractive feature of templates for FEM programming is that each custom mass matrix need
not be coded and tested individually. It is sufficient to implement the template as a single elementlevel module, with free parameters as arguments. (Alternatively, useful instances may be identified
by predefined mnemonic character strings, and converted to numerical signatures internally.) The
signature is adjusted according to goals and needs. In particular the same module should be able
to produce the conventional DLMM and CMM models as instances. This can provide valuable
crosschecking with other programs while doing benchmarks.
In problems characterized by rapid transients, such as contact-impact and fragmentation, templates
allow a flexible customization: reduced high-frequency pollution in elements in or near shock
regions while maintaining low-frequency continuum fit away from such regions. In these scenarios,
signatures may evolve in time.
21.2. Is Customization Worth The Trouble?
The ability to customize a mass matrix is not free of development costs. The presence of free
parameters makes template derivations considerably more complicated than those based on the
213
Table 21.1
Acronym
LFCF
AMC
RHFP
MSTS
RDAW
Template Customization Scenarios
Customization
Low-frequency continuum fit: matching acoustic branch (AB) to continuum model
Angular momentum (= rotary inertia) conservation: useful for transverse motions.
Reduced high-frequency pollution (spurious noise) in direct time integration (DTI)
Maximum stable time step in conditionally stable direct time integration (DTI)
Reduced directional anisotropy in wave propagation (not relevant to 1D meshes)
two standard procedures described in Chapter 16. Reason: everything must be carried along
symbolically: geometry, material and fabrication properties, in addition to the free parameters.
Consequence: hand computations rapidly become unfeasible, even for fairly simple 1D elements.
Help from a computer algebra system (CAS) is needed to get timely results. A key issue is: when
is this additional work justified? Two specific cases may be mentioned.
One is high fidelity systems. Dynamic analysis covers a wide range of applications. There is
a subclass that calls for a level of simulation accuracy beyond that customary in engineering
analysis. Examples are deployment of precision space structures, resonance analysis of machinery
or equipment, adaptive active control systems, medical imaging, phononics (wave guidance at
molecular level), vehicle signature detection, radiation loss in layered circuits, and molecular- and
crystal-level simulations in micro- and nano-mechanics.
In static structural analysis an error of 20% or 30% in peak stresses is not cause for alarm such
discrepancies are usually covered adequately by safety factors. But a similar error in frequency
analysis or impedance response of a high fidelity system can be disastrous. Achieving acceptable
precision with a fine mesh, however, can be expensive. Model adaptivity comes to the rescue in
statics. This approach is less effective in dynamics, however, on account of the time dimension and
the fact that irregular meshes are prone to develop numerical pollution. Customized elements may
provide a practical solution: achieving adequate accuracy with a coarse regular mesh.
Another possibility is that the stiffness matrix comes from a method that avoids displacement shape
functions. For example, assumed-stress or strain elements. [Or, it could simply be an array of
numbers provided by a black-box program, with no documentation explaining its source.] If this
happens the concept of consistent mass matrix, in which velocity shape functions (VSF) are taken
to coincide with displacement shape functions (DSF), loses the comfortable variational meaning
outlined in 16.4.2. An expedient way out is to choose an element with similar geometry and
freedom configuration derived with DSF and take those as VSF. But which element to pick? If time
allows, constructing and customizing a template avoids uncritically rolling the dice.
21.3. Mass Parametrization Techniques
There are several ways to parametrize mass matrices. Techniques found effective in practice are
summarized below. Most of them are illustrated in the worked out examples of ensuing sections.
It is often advantageous to have several template expressions for the same element configuration.
For example, to study the subset of diagonally lumped mass matrices (DLMM) it may be convenient
to streamline the general form to one that produces only such matrices. Likewise for singular mass
matrices. In that case we speak of template variants. These may overlap totally or partially: the
214
21.3
Table 21.2
Acronym
AB
ABTS
BLCD
BLFD
BLFM
CMM
CMS
CMT
COB
COF
DDD
DGVD
DIMM
DOF
DLMM
DSF
DSM
DTI
DCF
DWN
EOM
FEM
FFB
FPMM
HF
LCD
LF
LFF
LLMM
MOF
MSA
NCT
NND
PD
PVP
OB
OBTS
RCC
SDAV
SF
SFB
SLMM
SMS
VDMM
VP
VSF
MASS PARAMETRIZATION TECHNIQUES
Acronyms Used in Paper
Stands for
Acoustic branch in DDD: has physical meaning in continuum models
AB Taylor series in DWN , centered at = 0
Best linear combination (LFF sense) of the CMM and a selected DLMM
Best possible DLMM (LFF sense); acronym also applies to MS pair with this mass
Best possible FPMM (LFF sense); acronym also applies to MS pair with this mass
Consistent mass matrix: a special VDMM in which VSM and DSF coalesce
Component Mode Synthesis: model reduction framework for structural dynamics
Congruential (also spelled congruent) mass transformation
Constant optical branch: OB frequency is independent of wavenumber
Cutoff frequency: OB frequency at zero wavenumber (lowest one if multiple OB)
Dimensionless dispersion diagram: DCF vs. DWN
Dimensionless group velocity diagram: c = c/c0 vs. DWN
Directionally invariant mass matrix: repeats with respect to any RCC frame
Degree(s) of freedom
Diagonally lumped mass matrix; qualifier diagonally is often omitted
Displacement shape functions to interpolate displacements over element
Direct Stiffness Method: the most widely used FEM implementation
Direct time integration of EOM
Dimensionless circular frequency, always denoted by
Dimensionless wavenumber, always denoted by
Equations of motion
Finite Element Method
Flexural frequency branch in Bernoulli-Euler or Timoshenko beam models
Fully populated mass matrix (at element level); includes CMM as special case
High frequency: short wavelength, small DWN, typically > 1
Mass matrix obtained as linear combination of the CMM and a selected DLMM
Low frequency: long wavelength, small DWN, typically < 1
Low frequency fitting of AB to that of continuum
Lobatto lumped mass matrix: a DLMM based on a Lobatto quadrature rule
Maximum overall frequency: largest frequency in DDD over Brillouin zone
Matrix Structural Analysis: invented by Duncan and Frazier at NPL (1934)
Non-continuum term: a term in the ABTS that is not present in the continuum
Nonnegative definite; a qualifier reserved for symmetric real matrices
Positive definite; a qualifier reserved for symmetric real matrices
Parametrized variational principle
Optical branch (or branches) in DDD: no physical meaning in continuum models
OB Taylor series in DWN , centered at = 0
Rectangular Cartesian Coordinate: qualifier to frame, system, axes, etc.
Structural dynamics and vibration applications: low frequency range important
Shape function
Shear frequency branch in the Timoshenko beam model
Simpson lumped mass matrix: a LLMM based on Simpsons 3-pt quadrature rule
Selective mass scaling: modifying a mass matrix by adding a scaled stiffness
Variational derived mass matrix: Hessian of discretized kinetic energy
Variational principle
Velocity shape functions to interpolate velocities and produce a VDMM
215
Table 21.3
Template Related Nomenclature
Term or abbreviation Meaning

Template
Signature
Instance
Subset
Family
Variant
Admissible
MS template
FD template
FDM template
FDS template
FDMS template
EW template
ML template
MW template
SP template
An algebraic expression for a FEM matrix that contains free parameters.

So far used to construct stiffness and mass matrices of linear FEM models
The set of free parameters that uniquely defines a template
Matrix (or matrices) obtained by setting the signature to numeric values
Generic term for template specialization: includes families and variants
A template subset in which some free parameters are linked by constraints
A template subset that introduces free parameters from scratch
(the subset may be the original template if reparametrized)
Qualifier applied to instances that satisfy predefined conditions such
as positiveness, element mass conservation, and fabrication symmetries
Mass-stiffness pair template: both M and K have free parameters
Frequency-dependent template: free parameters may depend on frequency
Frequency-dependent mass template
Frequency-dependent stiffness template
Frequency-dependent mass-stiffness template
Entry weighted parametrization of a template; see 21.3.3
Multilevel parametrization of a template; see 21.3.4
Matrix weighted parametrization of a template; see 21.3.1
Spectral parametrization of a template; see 21.3.2
DLMM variant is plainly a subset of the general mass template. The key difference between a
template subset and a variant is that the latter redefines free parameters from scratch.
For the readers convenience, acronyms often used in this paper are listed in Table 21.2. A set of
definitions and abbreviations pertaining to templates are collected in Table 21.3.
Notational conventions for mathematical expressions that appear in this paper are summarized in
Table 21.4. Specific conventions used for free template parameters are given in Table 21.5.
21.3.1. Matrix-Weighted Parametrization
A matrix-weighted (MW) mass template for element e is a linear combination of (k + 1) component
mass matrices, k 1 of which are weighted by parameters i , (i = 1, . . . k):
def
Me = Me0 + 1 Me1 + . . . k Mek .
(21.1)
Here Me0 is the baseline mass matrix. This should be an admissible mass matrix on its own if
1 = . . . k = 0. The simplest instance of (21.1) is a linear combination of the consistent mass
matrix (CMM) and a diagonally-lumped mass matrix (DLMM):
def
Me = (1 )MCe + MeL .
(21.2)
This can be reformatted as (21.1) by writing Me = MCe +(MeL MCe ) = Me0 + Me1 . Here k = 1,
the baseline is Me0 MCe , 1 and Me1 is the mass deviator MeL MCe . The specialization
(21.2) is often abbreviated to linear combination of consistent and diagonally lumped masses, with
acronym LCD; cf. Table 21.2. The rationale behind (21.2) is that the CMM typically overestimates
216
21.3
Table 21.4
General Notational Conventions For Mathematical Expressions
Letter symbol
Used for
Examples
UC bold
LC bold
US roman
SS LC roman
SS LC roman
DS UC roman
Greek letters
Superposed dot
Prime
Matrices
Vectors
Scalar coefficients or functions
Subscripted variants of scalar coefficients
Vector entries conforming with vector symbol
Matrix entries conforming with matrix symbol
Dimensionless quantities except as noted below
Temporal differentiation
1D spatial differentiation, usually with respect to x
K, M
u, u
u(x, t)
a, b, Q,
c1 , c2
u i : entries of u
K i j : entries of K
, ,
u d 2 u(t)/dt 2
v (x) dv(x)/d x
UC:uppercase; LC:lowercase; US:unsubcripted; SS:single subscripted; DS:double subscripted

Exemption made for well established symbols; e.g. : frequency or : mass density
Table 21.5
Symbol
i
i
i
i , i
Notational Conventions For Template Parameters
Used for
Free parameters in basic stiffness matrix template (not used in this paper)
Free parameters in higher order stiffness matrix template
Original free parameters in mass template. Additional letter subscripts may
be appended as appropriate to distinguish template families or variants
Alternative notations for mass template parameters. Often derived from the
original i to streamline expressions, or to identify families or variants
The subscript index is suppressed if only one parameter appears; e.g. , .
natural frequencies while a DLMM usually underestimates them. Thus a linear combination has a
good chance of improving low-frequency accuracy for some [0, 1].
A MW mass template represents a tradeoff. It cuts down on the number of free parameters. Such
a reduction is essential for 2D and 3D elements. It makes it easier to satisfy conservation and
nonnegativity conditions through appropriate choice of the Mie . On the minus side it generally
spans only a subspace of admissible mass matrices.
21.3.2. Spectral Parametrization
A spectrally parametrized (SP) mass template has the form
def
Me = HT D H,
D = diag [ c0 0 c1 1 . . . ck k ] .
(21.3)
in which H is a generally full matrix. Parameters 0 . . . k appear as entries of the diagonal matrix
D . Scaling coefficients ci may be introduced for convenience so the i are dimensionless. Often
the values of 0 and/or 1 are preset from conservation conditions.
Configuration (21.3) occurs naturally when Me is constructed first in generalized coordinates,
followed by congruential transformation to physical coordinates via H. If the generalized mass is
derived using mass-orthogonal functions (for example, Legendre polynomials in 1D elements), the
unparametrized generalized mass D = diag [ c0 c1 . . . ck ] is diagonal. Parametrization is effected
by scaling its entries. As noted, some entries may be left fixed to satisfy a priori constraints.
217
Expanding (21.3) and collecting matrices that multiply each i leads to a matrix weighted combination form (21.1) in which each Mie is a rank-one matrix. The analogy with the spectral representation
theorem of symmetric matrices is obvious. But in practice it is usually better to work directly with
the congruent representation (21.3).
As remarked later in 21.3.6, SP is especially convenient for constructing singular mass matrices
under customization scenario RHFP of Table 21.1.
21.3.3. Entry-Weighted Parametrization
An entry-weighted (EW) mass template applies free parameters directly to each entry of the mass
matrix, except for a priori constraints on symmetry, invariance and conservation. As an example,
for a one-dimensional (1D) element with three translational DOF we may start from
11 12 13
def
(21.4)
Me = m e 12 22 23 ,
13 23 33
in which m e is the total element mass, and the sum of all row sums is one. EW is often applied to
entries of a deviator matrix that measures the change from a baseline matrix such as MC . For
example, see the three-node bar template (23.2).
Because of its generality, EP can be expected to lead to optimal customized instances. But it is
restricted to simple (usually 1D) elements because the number of parameters grows quadratically
in the matrix size, whereas for the foregoing two schemes either it grows linearly, or stays constant.
21.3.4. Multilevel Parametrization
A hierarchical combination of parametrization schemes can be used to advantage if the kinetic energy
can be naturally decomposed from physical considerations. For example, the Timoshenko beam
element covered in 24.2 uses a two-matrix-split template combined by a weighted form similar to
(21.2) as top level. (The energy split is between translational and rotational inertia, respectively.)
The two components are constructed by spectral and entry-weighted parametrization, respectively.
Such combinations fall under the scope of multilevel (ML) parametrization.
21.3.5. Selective Mass Scaling
Selective Mass Scaling, or SMS, is a method proposed recently (references given in H.4) in which
the mass matrix is modified by a scaled version of the stiffness matrix. Thus M becomes
MK = M +
K
K.
r2e f
(21.5)
Here K 0 is a dimensionless scaling factor whereas r2e f is a reference frequency used to

homogenize physical dimensions. The modification (21.5) may be done at the element or system
level. The objective is to filter down high frequencies in explicit DTI for applications such as
contact-impact; e.g., vehicle crash simulation. Filtering aims to reduce spurious noise as well as
increasing the stable timestep. It thus follows under customization scenarios RHFP and MSTS of
Table 21.1. The basic idea can be explained as follows. Let i and vi denote the natural frequencies
218
21.3
and associated orthonormalized eigenvectors, respectively, whereas i and v i are their counterparts
for the modified eigenproblem (M K + i2 K) v i = 0. By inspection the eigenvectors are preserved:
v i = vi . Taking the Rayleigh quotient shows that the modified frequencies are
i2 =
i2
,
Ri
in which
Ri = 1 + K
i2
.
r2e f
(21.6)
Choosing K > 0 cuts down each frequency by Ri > 0. For low frequencies the modification
is negligible if K and r2e f are appropriately selected so that Ri 1. For nonphysical high
frequencies (mesh modes) the reduction can be significant In fact note that if i2 >> K /r2e f , i2
2
cannot exceed the fixed bound max
= r2e f / K . The downside is that low frequency accuracy
may suffer significantly, as illustrated later.
Although SMS may be presented as a variant of the MW parametrization technique of 21.3.1, it
deserved to be considered on its own for the reasons stated in H.4.
21.3.6. Singular Mass Matrices
A thread linked to SMS but independently developed is that of singular mass matrices. This
has been primarily advocated for multibody dynamics, as well as dynamical systems leading to
differential-difference EOM that occur in active control with time lags. References are provided in
H.5. The objective is roughly similar to SMS: reduce high frequency noise pollution triggered by
rapid transients and/or time lags. But now this is done by raising the optical branch (or branches)
so as to widen the acoustoptical gap pictured described in 23.1.1 and illustrated in Figure 23.2.
Noisy frequencies that fall in the gap decay exponentially.
There are several ways to produce such matrices. Under the template framework, the use of spectral
parametrization (SP) is particularly convenient, as observed in 21.3.2. Other approaches include
reduced numerical integration or injection of a convenient null space using mass matrix projection.
21.3.7. Constant Optical Branch Variant
Instead of rising the optical branch (or branches) by making Me singular, one may try to make the
OB frequency independent. Templates that accomplish that feat are tagged as having a Constant
Optical Branch, or COB for short. They form subsets collectively identified as the COB variant.
The group velocity pertaining to a COB vanishes, so associated waveforms with that particular
frequency do not propagate. COB templates were discovered during the course of this work, and
are briefly studied in 23.1.13 for the three-node bar element.
21.3.8. Mass-Stiffness Template Pairs
The concept of template was first developed for element stiffness matrices, as a natural generalization
of its decomposition into basic and higher order parts. A brief historical account is provided in H.7.
Normally the stiffness template is optimized by imposing superconvergence conditions dealing with
higher order patch tests while element aspect ratios are kept arbitrary. That optimal instance, if
found, is kept fixed while a mass matrix template is subsequently investigated.
Maximum customization for dynamics can be expected if both stiffness and mass matrix templates
can be simultanously adjusted. This is known as a mass-stiffness (MS) template. These may
219
be of interest when improving dynamic behavior is paramount. Presently there is relatively little
experience with this more ambitious approach. A note of caution: highly optimized MS templates
may be abnormally sensitive to geometric or material perturbations away from a regular mesh.
21.3.9. Frequency Dependent Templates
One final generalization should be mentioned: allowing free parameters to be function of the
frequency. If this is done for the mass matrix, we speak of a frequency dependent mass (FDM)
template. If this is done for both the mass and stiffness matrices, we call the combination a frequency
dependent mass-stiffness (FDMS) template. Both cases are illustrated in 22.1.1122.1.13 for the
two-node bar element.
Although this ultimate complication is largely a curiosity, it might be occasionally useful in problems
that profit from transformation to the frequency domain. For example: a linear dynamic system
driven by a harmonic excitation of slowly varying frequency, if only the long term (steady-state)
response is considered. Such systems may arise in parametric stability and active control.
2110
22
Mass Templates
for Bar2 Elements
221
Chapter 22: MASS TEMPLATES FOR BAR2 ELEMENTS
TABLE OF CONTENTS
Page
22.1
The Two-Node Bar Element

. . . . . . . . . .
22.1.1
Bar2 Entry Weighted Template . . . . . . .
22.1.2
Bar2 One Parameter Mass Template . . . . .
22.1.3
Bar2 Alternative Parametrization
. . . . . .
22.1.4
Bar2 Angular Momentum Conservation . . . .
22.1.5
Bar2 Fourier Analysis . . . . . . . . . .
22.1.6
Bar2 Dispersion Diagrams . . . . . . . .
22.1.7
Best By Low Frequency Fitting . . . . . .
22.1.8
Folding Frequency
. . . . . . . . . .
22.1.9
Bar2 Test: Vibrations of a Fixed-Free Bar Member
22.1.10 Other Customization Options . . . . . . .
22.1.11 Bar2 Frequency Dependent Mass
. . . . . .
22.1.12 Bar2 Frequency Dependent Mass-Stiffness Pair .
22.1.13 Bar2 Frequency Dependent Mass Instances . . .
222
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
.
.
.
.
.
.
223
223
223
224
224
224
228
229
229
229
2211
2211
2212
2213
22.1
THE TWO-NODE BAR ELEMENT
22.1. The Two-Node Bar Element

The template approach is best grasped through an example that involves the simplest nontrivial
structural finite element: a two-node prismatic bar of mass density , area A and length , that can
only move along the longitudinal axis x. See Figure 22.1(a). This element is often acronymed Bar2
for brevitys sake. The well known consistent and diagonally-lumped mass matrix forms are

me 2 1
me 1 0
e
e
,
ML =
.
(22.1)
MC =
6 1 2
2 0 1
in which m e = A is the total element mass. These are derived in 16.4.
22.1.1. Bar2 Entry Weighted Template
The most general mass matrix form for Bar2 is the entry-weighted template
e

e
M11 M12
11 12
11 12
e
e
= A
=m
.
M =
e
e
M21
M22
21 22
21 22
(22.2)
The first form is merely a list of entries. Next the element mass m e = A is factored out. The
emerging parameters 11 through 22 are numbers, which illustrates a general rule: template free
parameters should be dimensionless. This simplifies analysis and implementation. To cut down on
parameters one looks at configuration constraints. The most obvious ones are:
Matrix symmetry: Me = (Me )T . For the expression (22.2) this requires 21 = 12 .
Physical symmetry: For a prismatic bar, Me in (22.2) must exhibit antidiagonal symmetry: 22 =
11 .
Conservation of total translational mass: same as conservation of linear momentum or of kinetic
energy. Apply the uniform velocity field u = v to the bar. The associated nodal velocity vector is
u e = ve = v [ 1 1 ]T . The kinetic energy is T e = 12 (ve )T Me ve = 12 m e v 2 (11 + 12 + 2 + 22 ).
This must equal 12 m e v 2 , whence 11 + 12 + 21 + 22 = 2(11 + 12 ) = 1.
Nonnegativity: Me should not be indefinite. [This is not an absolute must, and it is actually relaxed
in some elements discussed later.] Whether checked by computing eigenvalues or principal minors,
this constraint is nonlinear and of inequality type. Consequently it is not often applied ab initio,
unless the element is quite simple, as in this case, or can be stated through simple expressions.
22.1.2. Bar2 One Parameter Mass Template
On applying the symmetry and conservation rules three parameters of (22.2) are eliminated. The
remaining one, called , is taken for convenience to be 11 = 22 = (2 + )/6 and 12 = 21 =
(1 )/6. This rearrangement gives

2+ 1
e
1
= (1 )MCe + MeL .
(22.3)
M = 6 A
1 2+
Expression (22.3) shows that the general Bar2 mass template can be recast as a linear combination
of the CMM and DLMM instances listed in (22.1). Summarizing, we end up with a one-parameter,
matrix-weighted (MW) template that befits the LCD form (21.2). If = 0 and = 1, (22.3)
223
(a)
1
(e)
Total mass A
(b)
2 x
e
= L
massless connector
Figure 22.1. The two-node prismatic bar element: (a) element
configuration; (b) direct mass lumping to end nodes.
reduces to MCe and MeL , respectively. This illustrates another requirement: the CMM and DLMM
forms must be instances of the mass template.
Finally we can apply the nonnegativity constraint. For the two principal minors of Me to be
nonnegative, 2 + 0 and (2 + )2 (1 )2 = 3 + 6 0. Both are satisfied if 12 .
Unlike the others, this constraint is of inequality type, and only limits the range of .
The remaining task is to select the parameter. This is done by introducing an optimality criterion
that fits the problem at hand. This is where customization comes in. Even for this simple case the
answer is not unique. Thus the statement the best mass matrix for Bar2 is so-and-so has to be
qualified. Two specific optimization criteria are considered in 22.1.4 and 22.1.5.
22.1.3. Bar2 Alternative Parametrization
An alternative template expression that is useful in some investigations, such as those undertaken in
Appendix V, is obtained by reparametrizing via = 1 + 2, the inverse of which is = ( 1)/2.
The resulting form is

3+ 3
e
1
.
(22.4)
M = 12 A
3 3+
This is called the form of the general Bar2 mass template. Observing that its determinant is
A , Me is seen to be singular if = 0, and nonnegative if 0.
22.1.4. Bar2 Angular Momentum Conservation
This criterion can only be applied in multiple dimensions, since angular rotations do not exist in
1D. Accordingly we allow the bar to move in the {x, y} plane by expanding its nodal DOF to
ue = [ u x1 u y1 u x2 u y2 ]T , whence (22.3) becomes a 4 4 matrix
2+
0
1
0
2+
0
1
(22.5)
Me = 16 A 0
1
0
2+
0
0
1
0
2+
Apply a uniform angular velocity about the midpoint. The associated node velocity vector at
= 0 is u e = 12 [ 0 1 0 1 ]T . The discrete and continuum energies are
/2

2
e e
e
3
e
1 e T
1
1
A x d x = 24
A3 .
(22.6)
T = 2 (u ) M u = 24 A (1 + 2), T =
/2
Matching
= T gives = 0. So according to this criterion the optimal mass matrix is the
consistent one (CMM). Note that if = 1, Te = 3T e , whence the DLMM overestimates the
element rotational (rotary) inertia by a factor or 3.
Te
224
22.1
CONTINUUM BAR
infinite continuum
bar with , E and A
constant along x
(a)
c0
u0(x,t)
x
Plane axial wave
with phase velocity
c0 and wavelength 0
(axial displacement drawn

transversally to bar for
visualization conveniency)
(b)
Over lattice,
phase velocity
and wavelength
change to c and
, respectively
FEM-DISCRETIZED BAR
FEM discretization
as infinite lattice
(c)
(d)
c
uj (t)
xj
xj d
xj +d
j
xj
j1
j+1
2-element,
3-node patch
extracted at
generic node j
Figure 22.2. Propagation of a harmonic plane wave over an infinite, prismatic, elastic bar: (a) propagation
over a continuum bar; (b) FEM discretization as infinite regular lattice; (c) propagation of plane wave over
Bar2-discretized lattice; (d) extraction of a typical two-element patch. For visualization convenience, the
wave-profile axial displacement u(x, t) is plotted normal to the bar.
22.1.5. Bar2 Fourier Analysis

For longitudinal motions, a more useful customization criterion is to improve accuracy in the long
wavelength, low-frequency limit; this is labeled LFCF in Table 21.1. This is carried out by a well
known tool: Fourier analysis. Physical interpretation: probe the fidelity with which planes waves
are propagated over a FEM-discretized regular lattice, when compared to the propagation over a
continuum bar. The essentials are illustrated in Figure 22.2. The top half depicts the continuum bar
whereas the bottom half shows stages of the Fourier analysis of its FEM-discretized counterpart.
Symbols used for the analysis of plane wave propagation are collected in Table 22.1 for the readers
convenience. [The same notation is reused in later Sections.] Corresponding nomenclature for
the FEM-discretized two-node bar lattice is collected in Table 22.2. The continuum-versus-lattice
notational rule is: corresponding quantities use the same symbol but the zero subscript is suppressed
in the lattice. For example, the continuum wavelength 0 becomes the lattice wavelength .
Plane wave propagation over a regular spring-mass lattice is governed by the semidiscrete linear
equation of motion (EOM):
M u + K u = 0,
(22.7)
225
Table 22.1
Quantity
, E, A
() , ()
u 0 = E u 0
u 0 (x, t)
B0
0
k0
0
0
f0
T0
0
c0
Nomenclature for Harmonic Plane Wave Propagation over Continuum Bar

Meaning (physical dimension in brackets)
Mass density, elastic modulus, and cross section area of bar
Abbreviations for derivatives with respect to space x and time t, respectively
2
2
2

Bar wave equation. Frequency domain
forms:
0 u = 0.

0 u = c0 u and u + k
Plane wave function u 0 = B0 exp i(k0 x 0 t) [length], in which i = 1
Wave amplitude [length]
Wavelength [length]
Wavenumber k0 = 2/0 [1/length]
Dimensionless wavenumber 0 = k0 0
Circular (a.k.a. angular) frequency 0 = k0 c0 = 2 f 0 = 2c0 /0 [radians/time]
Cyclic frequency f 0 = 0 /(2) [cycles/time: Hz if time in seconds]
Period T0 = 1/ f 0 = 2/0 = 0 /c0 [time]
Dimensionless circular frequency 0 = 0 T0
= 0 0 /c0
Group wave velocity c0 = 0 /k0 = 0 /T0 = E/ [length/time]. Often
abbreviated to wavespeed. Physically, c0 is the longitudinal speed of sound.
Unsubscripted counterpart symbols, such as k or c, pertain to a discrete FEM lattice; cf. Table 22.2
Table 22.2
Quantity
u(x, t)
u
Mu + Ku = 0
B
Ne
f
T

c
c
Nomenclature for Harmonic Plane Wave Propagation over Bar2 Lattice

Meaning (physical dimension in brackets)
Plane wave function (22.8) [length]
Node displacement vector, constructed by evaluating u(x, t) at nodes [length]
Semidiscrete lattice wave equation (22.7). K and M are infinite Toeplitz matrices
Wave amplitude [length]
Bar element length [length]
Wavelength = 2/k = 2/ [length]
Wavenumber k = 2/ = / [1/length]
Dimensionless wavenumber = k = 2 /
Number of elements per wavelength: /: same as signal sampling rate
Circular (a.k.a. angular) frequency = c0 / [radians/time]
Cyclic frequency f = /(2) [cycles/time: Hz if time in seconds]
Period T = 1/ f = 2/ = /c [time]
Dimensionless circular frequency = /c0
Group wave velocity over lattice: c = /k = c0 (/) [length/time]
Wavespeed ratio c/c0 = / from discrete to continuum
Quantities unchanged from continuum to lattice, such as E,are not repeated in this Table. Note
that the definition of uses the continuum wavespeed c0 = E/; not the discrete wavespeed c.
in which M and K are infinite, tridiagonal Toeplitz matrices. This EOM can be solved by Fourier
methods. Figure 22.2(b) displays two characteristic lengths: and . The element length-towavelength ratio is called = /. The floor function of its inverse: Ne = / is the
number of elements per wavelengths. Those ratios characterize the fineness of the discretization,
as illustrated in Figure 22.2(b).
Within constraints noted later the lattice can propagate real, travelling, harmonic plane waves of
wavelength and grpup velocity c, as depicted in Figure 22.2(b,c). The wavenumber is k = 2/
226
22.1
(a) Wavelength = , dimensionless wavenumber = 0.

Sampling rate = elements per wavelength N e =
c>0
(b) Wavelength =8 , dimensionless wavenumber = /4.

Sampling rate = elements per wavelength N e = 8
c>0
(c) Wavelength =2 , dimensionless wavenumber = ,

Sampling rate = elements per wavelength Ne = 2
c=0
(folding wavenumber)
(d) Wavelength = , dimensionless wavenumber = 2,

Sampling rate = elements per wavelength Ne = 1
c<0
Figure 22.3. Selected plane waves of various wavelengths, illustrating the physical meaning of
the dimensionless wavenumber (DWN) = k = 2 /. The number of elements per wavelength
is Ne = / = 2/, in which . denotes the floor function. (This is equivalent to the
spatial sampling rate of filter technology.) The case = 2 pictured in (c) pertains to the folding
or Nyquist frequency, at which = , Ne = 2, and the group velocity c vanishes.
and the circular frequency = 2/T = 2 c/ = k c. The range of wavelengths that the lattice
may transport is illustrated in Figure 22.3.
To study plane wave solutions it is sufficient to extract a two-element patch, a process depicted in
Figure 22.2(d). A harmonic plane wave of amplitude B is described by the function

(22.8)
u(x, t) = B exp [ j (kx t)] = B exp j x c0 t ], j = 1.
Here the dimensionless wavenumber and dimensionless circular frequency
were introduced as
= k = 2/ = 2 and = /c0 , respectively, in which c0 = E/ is the elastic bar
group velocity, which for the continuum is the same as the phase velocity. (In physical acoustics
c0 is the sound speed of the material.) Using the well-known Bar2 static stiffness matrix and the
mass template (22.3) gives the patch equations
A
6
2+
1
0
1
4 + 2
1
0
1
2+
u j1
u j
u j+1
EA
+
1
1
0
1
2
1
0
1
1
u j1
uj
u j+1

= 0.
(22.9)
From this one takes the middle (node j) equation, which repeats in the infinite lattice:
A
[1
6
4 + 2
u j1
1 ] u j
u j+1
EA
+
[ 1
227
u j1
1 ] u j
u j+1

= 0.
(22.10)
(b)
6
5
4
CMM: = 0
2
Continuum Bar
1.5
Continuum Bar
BLFM: = 1/2
Wavespeed ratio c = c/c0
Dimensionless frequency = /c0
(a)
1
0.5
BLFM: = 1/2
DLMM: = 1
0.5
2
1
DLMM: = 1
0
5
1
4
2
3
CMM: = 0
1.5
5
1
4
2
3
Figure 22.4. Results from Fourier analysis of Bar2 infinite regular lattice for three choices of , plus
continuum: (a) dimensionless dispersion diagram (DDD); (b) dimensionless group velocity diagram (DGVD).
Evaluate the wave motion (22.8) at x = x j1 = x j , x = x j and x = x j+1 = x j + while

keeping t continuous. Substitution into (22.10) gives the wave propagation condition

c0 t
c0 t
A c02
2
2
6 (2 + ) 6 (1 ) cos cos
i sin
B = 0. (22.11)
3

If this is to be zero for any t and B, the expression in brackets, called the characteristic equation,
must vanish. Solving gives the dimensionless frequency versus wavenumber relation
2 =
Its inverse is
6(1 cos )
1 2 4
= 2 +
+ C6 6 + . . .
2 + + (1 ) cos
12
(22.12)

6 (2 + )2
1 2 3 9 20 + 202 5
= arccos
+
+ ...
=

6 + (1 )2
24
1920
(22.13)
Transforming (22.12) to physical wavenumber k = / and circular frequency = c0 / gives

2

6c0
1 cos(k)
12 2 2
2
2 2
4 4
(22.14)
= c0 k 1 +
k + C6 k + . . .
=
2 2 + + (1) cos(k)
12
in which C6 = (1 10 + 102 )/360.
22.1.6. Bar2 Dispersion Diagrams
An equation that links frequency and wavenumber: = () as in (22.12), or = (k), as in
(22.14), is a dispersion relation. A plot of the dispersion relation with k and along horizontal
and vertical axes, respectively, is called a dispersion diagram. When this is done in terms of
dimensionless wavenumber and dimensionless frequency , the plot is called a dimensionless
dispersion diagram, or DDD. Such diagrams exhibit a 2 period: () = ( + 2 n) for integer
228
22.1
n. Thus it is enough to plot () over either [, ] or [0, 2 ], a range called a Brillouin zone.
All DDD in this paper use the [0, 2] range choice.
Why is = 0 at = 2? The wavelenth = pictured in Figure 22.3(d) has the same value at
all nodes for each time t. This nodal sampling cannot be distinguished from the case = (that
is, = 0) shown in Figure 22.3(a). They must share the same frequency, which is zero; associated
plane waves propagate with the same speed but in opposite directions. Similar arguments can
be made to justify the dispersion curve symmetry about wavenumber = , as well as the 2
periodicity.
22.1.7. Best By Low Frequency Fitting
An oscillatory dynamical system is nondispersive if is linear in k, in which case c = /k is
constant and the wavespeed (the group velocity) is the same
for all frequencies. The physical
dispersion relation for the continuum bar is c0 = 0 /k0 = E/. Hence all waves propagate with
the same speed in this model. Group and phase velocities coalesce.
The FEM-discretized lattice group velocity is c = /k = c0 (/), which differs from c0
except at = = 0. The Bar2 discrete model is dispersive for any fixed , since from (22.14) we
get

1
1 2 2 1 20 + 202 4
6(1 cos )
c

=
=1+
+
+ ...
=
c =
c0
2 + + (1 ) cos
24
1920
(22.15)
Plainly the best fit to the continuum for small wavenumbers = k<<1 is obtained by taking
= 1/2, which makes the second term of the series (22.12) or (22.15) vanish. So for LFCF
customization the best mass matrix is the average of the lumped and consistent ones:

A 5 1
e
e
e
e
1
1
.
(22.16)
M B L F M = M
1 = 2 MC + 2 M L =
=
12 1 5
2
This instance is labeled BLFM, for best low-frequency match. Figure 22.4(a) plots the dimensionless
dispersion relation (22.12) for the CMM ( = 0), DLMM ( = 1) and BLFM ( = 12 ) mass
matrices, along with the continuum-bar relation 0 = 0 . The superior small- fit provided by the
BLFM is evident.
22.1.8. Folding Frequency
The maximum lattice frequency occurs at the folding wavenumber = k = or = 2, which
is waveform (c) in Figure 22.3. The sampling rate Ne is then 2 values per element. This is called
the folding or Nyquist frequency, and is denoted as
12
.
(22.17)
a2 f =
1 + 2
(The a in the subscript
stands
for acoustic branch; this notation is explained in 23.1.1). This

varies from a f = 12 = 2 3 for the CMM through a f = 2 for the DLMM. Frequencies higher
than a f cannot be propagated over the lattice. As shown in Figure 22.4(b), the lattice wavespeed
vanishes at the folding wavenumber = , and is negative over the range (, 2]. Waveforms in
that rage move with negative speed: c < 0. As discussed in 22.1.6, the waveform with = , or
= 2 , cannot be distinguished from a rigid motion such as that pictured in Figure 22.3(a), and
the lattice frequency falls to zero.
229
=E=A =1 throughout
;;
=E=A =1 throughout
;;
(a)
(b)
L=/2
L=/2
(a)
LFCF Bar2 Template
Instances, Fix-Free Mode 1,
Exact Frequency 1 = 1
BLFM
4
DLMM
2
0
CMM
1
Number of elements Ne
16
(b)
LFCF Bar2 Template
6
4
BLFM
DLMM
CMM
16
Correct digits in computed frequency
Figure 22.5. Fixed-free homogeneous prismatic elastic bar member used in vibration test for Bar2 and Bar3 template
instances. Both pictured discretizations display 4 elements. (a): member modeled with Bar2 elements; results reported
in Table 22.3 and Figure 22.6. (b): member modeled with Bar3 elements; results reported in Figure 23.6.
(c)
LFCF Bar2 Template

BLFM
2
DLMM
0
CMM
16
Figure 22.6. Performance of selected Bar2 template instances in predicting the first three natural frequencies
i , i = 1, 2, 3 of the fixed-free prismatic homogeneous bar shown in Figure 22.5(a). This is a graphical,
log-log representation of the results of Table 22.3. Horizontal axis shows number of elements while vertical
axis displays correct digits of computed frequency. See text for details of what is shown along each axis.
22.1.9. Bar2 Test: Vibrations of a Fixed-Free Bar Member

Natural frequency predictions of three Bar2 template instances are compared for predicting natural
frequencies of longitudinal vibrations of the fixed-free elastic bar member pictured in Figure 22.5.
The member is prismatic, with constant E = 1, A = 1, and = 1. The total member length is taken
as L = /2 for convenience. With those numerical properties the continuum eigenfrequencies are

(2i 1) E
= 2i 1, i = 1, 2, 3, . . .
(22.18)
0i =
2L
The member is divided into Ne identical elements, with Ne = 1, 2, . . . 16. Figure 22.5(a) pictures
the case Ne = 4. Three template instances are compared: CMM ( = 0), DLMM ( = 1)
and BLFM ( = 1/2). Numerical results obtained for the first three frequencies are collected in
Table 23.1. The O( 4 ) convergence of BLFM is obvious. For example, 4 elements give 2 correct
to 4 digits while both CMM and DLMM, which converge as O( 2 ), give only 2. As expected,
CMM overestimates the continuum frequencies while DLMM underestimates them.
The results of Table 22.3 are graphically reformatted in Figure 22.6, as accuracy versus elements
log-log plots. The horizontal axis shows number of elements Ne in log2 scale. The vertical axis
displays correct digits of computed frequency, computed as
d = log10 |i | ,
in which
2210
i = i 0i .
(22.19).
22.1
Table 3. Bar2 Instance Results for Vibrations of a Fixed-Free Bar Member

Instance
Ne
CMM
1
2
4
8
16
1.102658
1.025859
1.006437
1.001607
1.000402
3.583726
3.174947
3.043539
3.010855
5.767394
5.202396
5.050339
DLMM
1
2
4
8
16
0.900316
0.974495
0.993587
0.998394
0.999598
2.352640
2.829496
2.956815
2.989169
4.234640
4.801608
4.949951
BLFM
1
2
4
8
16
0.986247
0.999188
0.999950
0.999997
1.000000
2.781352
2.987344
2.999237
2.999953
4.827222
4.989971
4.999389
* frequency not provided by discrete FEM model
Here i is the frequency error of computed values with respect to continuum frequencies 0i =
2 i 1, given by (22.18). The plots clearly show at a glance that, for the same Ne , BLFM roughly
doubles the number of correct digits provided by the other two instances. It also illustrates that
CMM and DLMM give the same error magnitude (within plot accuracy) although of different signs.
Thus log-log plots such as those in Figure 22.6 are unable to show whether the convergence is from
above or below, because of the taking of absolute values in (22.19). That visualization deficiency
should be kept in mind should error signs be important.
22.1.10. Other Customization Options

The last three customization options listed in Table 21.1 are not relevant to this element. RHFP is
unnecessary because the dispersion diagram does not have an optical branch. MSTS is pointless
because the DLMM in (22.1) is unique. Finally, RDAW does not apply to 1D elements.
22.1.11. Bar2 Frequency Dependent Mass
As noted in 21.3.9, it is occasionally useful to make the mass and/or stiffness matrix frequency
dependent. The goal is to exactly match the continuum dispersion relation = for all frequencies,
or at least a finite range that includes = = 0. Such an exact fit allows for coarser discretizations.
The cost paid is that matrix entries become trigonometric functions of frequency. Both the EOM
and associated eigenproblems become trascendental.
Unless the frequency is specified beforehand (for example, in pure harmonic excitation) an iterative
process is unavoidable. Therefore exactness gains might be illusory: the dog chases its own tail.
2211
Early publications that follow this approach are cited in H.6. For reasons indicated there, those
formulations are not necessarily instances of the general template derived in 22.1.12.
The simplest way to introduce frequency dependency is to allow the mass template parameter
in (22.3) to be frequency dependent, while the stiffness matrix is held fixed. To find the expression
of , set in the characteristic equation extracted from (22.11):

(22.20)
6 (2 + )2 6 (1 )2 cos = 0,
in which has been renamed . Solving for it gives
= 1 +
6
3
1 2
4
6
...
2
1 cos
2
40
1008 28800
(22.21)
Since = for the continuum,

= 1 +
6
3
1 2
4
6
...
2
1 cos
2 40 1008 28800
(22.22)
As 0 or 0 both (22.21) and (22.22) approach 0/0. The indeterminacy is removed by the
Taylor expansions given above, which show that the limit is 12 , as may be expected. As
or grows, decreases so the template gradually favors the CMM more. Two interesting values
should be noted. If = 3.38742306673364, = 0, which makes the CMM frequency exact;
this occurs at the intersection of the continuum and CMM dispersion curves in Figure 22.4(a). If
= lim = 4.05751567622863, = 12 , which makes Me singular. If > lim , Me becomes
indefinite. It follows that the match (22.21) or (22.22) is practically limited to the DWN range
0 < 4.
22.1.12. Bar2 Frequency Dependent Mass-Stiffness Pair
The most general FDMS template for Bar2 has 8 free parameters. These are chosen as deviations
from the optimal frequency-independent matrices:

5 1
11 12
1 1
11 12
e
e
+
, K = CK
+
,
M = CM
21 22
21
22
1 5
1
1
(22.23)
in which C M = A/12 and C K = E A/. All parameters may be frequency dependent. For
brevity that dependency will not be explicitly shown unless necessary. If all ij vanish, Me reduces
to (22.16), which is BLFM optimal. If all ij vanish, Ke reduces to the well known stiffness
of a 2-node prismatic bar. Thus in the zero-frequency (static) limit all parameters must vanish,
which provides useful checks. To cut down on parameters, we impose diagonal and antidiagonal
= 12
, and 22
= 11
. In addition
symmetry conditions a priori: 21 = 12 , 22 = 11 , 21
= 21
= 11
avoids singularities in the static limit, as noted later. Thus (22.23) reduces
setting 12
to

5 1
11 12
1 1
11 11
e
e
+
, K = CK
+
.
M = CM
12 11
11
11
1 5
1
1
(22.24)
2212
22.1
Table 4. General FDMS Template For Bar2

Free parameters Linkage equation (top line); Taylor series at = = 0 (bottom line)
11 , 12
2
2
11
= ( (5 +11 ) + (12
+ 2 (1 312 )) cos 12)/(12(1 cos ))

11 0 = [ 11 0 12 0 ] + O( )
11
, 12
11 = (12 + 1211
5 2 (12 + 1211
+ 2 (1 12 )) cos )/ 2

2
3
11 0 = [ 11 0 12 0 ] + O( )
11
, 11
12 = (12 + 1211
+ 2
2 (5 + 11 )) sec )/ 2
(122 + 1211

12 0 = [ 11 0 12 0 ] + O( 3 )
Template parameters are generally functions of or ; e.g., 11 = 11 () = 11 (), etc.

Parameter arguments are usually omitted to reduce clutter unless necessary.
In the bottom-line series, , and denote parameter values at = = 0.
11 0
11 0
12 0
which has 3 free parameters: 11 , 12 and 11

. These matrices are nonnegative if
4 + 11 + 12 0,
6 + 11 12 0,
1 + 11
0.
(22.25)
Imposing the plane wave motion (22.8) on a two-element patch, extracting the middle node equation
and dropping extraneous factors yields the complex characteristic equation

12(1+11 ) (5+11 ) 2 12 + 1211 + (112 ) 2 cos exp( j) = 0.
(22.26)
Since the complex exponential never vanishes, it may be dropped and (22.26) reduces to the real
equation

) (5+11 ) 2 12 + 1211
+ (112 ) 2 cos = 0.
12 (1+11
(22.27)
To match the continuum, is replaced by , whence
) (5+11 ) 2 (12 + 1211

+ (112 ) 2 ) cos = 0.
f cm = 12 (1+11
(22.28)
This establishes a linear constraint among the 3 parameters. Consider these as functions of :
= 11
(), 11 = 11 (), and 12 = 12 (). Expanding in Taylor series about = 0 yields
11

f cm = 6

11
0
11 0
12 0

11
11
12

+
3 + . . . = 0, (22.29)
+ 6

0
0
0

This shows that

= = 0 the continuum equation is identically satisfied. If
in the static limit

2
,
however,
a
term
in
appears in (22.29); this is the reason for presetting 12

= 11
.
12 |0 = 11
0
Further developments depend on which parameter pair is kept. Table 22.4 lists three possibilities:
, 11 ), (11
, 12 ), and (11 , 12 ).
(11
2213
22.1.13. Bar2 Frequency Dependent Mass Instances
= 0 in (22.24). Taking
Some relatively simple FDM instances can be obtained by setting 11
12 = 11 and solving for the latter gives
11 = 12 = 1 +
2
4
6
12
6
=
...
2
1 cos
20 540 14400
(22.30)
The resulting Me is indefinite if > 4.05752. This is the equivalent of the FDM instance considered
in 22.1.11. The difference between (22.30) and (22.22) lies in the choice of baseline matrix for
= 0 yields
null free parameters. On the other hand, setting 12 = 0 along with 11
11 =
12 + 5 2 + (12 + 2 ) cos
4
11 6
=
...,
2
40 14400
(22.31)
This correction is smaller than (22.30) if < /2. The resulting Me is indefinite if > 4.46192.
2214
23
Mass Templates
for Bar3 Elements
231
TABLE OF CONTENTS
Page
23.1
The Three-Node Bar Element . . . . . . . . . .

23.1.1
Dispersion Diagram Terminology
. . . . . .
23.1.2
Bar3 General Mass-Stiffness Template . . . .
23.1.3
Bar3 Alternative Mass Template
. . . . . .
23.1.4
Bar3 Patch Equations . . . . . . . . . .
23.1.5
Bar3 Fourier Analysis . . . . . . . . . .
23.1.6
Bar3 Standard Template Instances
. . . . .
23.1.7
Bar3 Low-Frequency Fitting . . . . . . . .
23.1.8
Bar3 Lumped Mass Template Variant
. . . .
23.1.9
Bar3 MSTS Customization
. . . . . . . .
23.1.10 Reducing High Frequency Pollution . . . . .
23.1.11 Bar3 Spectral Mass Variant
. . . . . . . .
23.1.12 Bar3 Selective Mass Scaling Variant . . . . .
23.1.13 Bar3 Constant Optical Branch Variant . . . . .
23.1.14 Bar3 Test: Vibrations of a Fixed-Free Bar Member
232
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
.
.
.
.
.
.
.
.
233
233
233
234
236
236
237
238
2310
2311
2312
2313
2314
2315
2319
23.1
THE THREE-NODE BAR ELEMENT
23.1. The Three-Node Bar Element

The three-node bar element configuration is shown in Figure 23.1(a). The element is prismatic with
length = L e , uniform cross section area A and mass density . Midnode 3 is at the center. The
element DOFs are arranged as ue = [ u 1 u 2 u 3 ]T . The element name is often abbreviated to Bar3
in the sequel. We will consider only frequency independent templates here.
Despite its simplicity, the Bar3 template is sufficiently feature rich so it may be used to illustrate
most of the customization scenarios listed in Table 21.1. Two reasons: it has multiple dispersion
branches, and the stiffness has a free parameter. But additional terminology on dispersion diagrams
has to be introduced first. Readers familiar with the topic should skip to 23.1.2.
23.1.1. Dispersion Diagram Terminology
The characteristic equation of the Bar3 element, derived in 23.1.5 below, gives two positive real
frequencies for each plane wave wavelength. The dimensionless forms are identified by a and o ,
ordered so a o . The functions a () and o (), in which is the dimensionless wavenumber,
are called acoustic and optical branches, respectively, of the DDD. This terminology originated
in crystal physics, in which both branches have physical meaning in modeling molecular level
oscillations. [In crystallography, acoustic waves are lower frequency waves caused by sonic-like
disturbances, in which adjacent molecules move in the same direction. Optical waves are higher
frequency oscillations caused by interaction with light or electromagnetics, in which adjacent
molecules move in opposite directions. Textbook references are provided in ?.]
In FEM discretization work, only the acoustic branch has physical meaning because for small
(that is, long wavelengths) it approaches the continuum bar relation = , as plainly illustrated
by the a2 series in 22.1.5. On the other hand, the optical branch is spurious. It is caused by the
discretization and pertains to higher frequency lattice oscillations, also known as mesh modes.
Figure 23.2 displays nomenclature used for a two-branch dispersion diagram, such as that exhibited
by the Bar3 element. As noted, the acoustic branch is the long-wavelength counterpart of the
continuum model, for which = ; thus a |0 = 0. On the other hand, the optical branch has a
nonzero frequency oc = o |0 called the cutoff frequency (COF) that cannot vanish, although it
may go to infinity under certain conditions; such as a singular mass matrix. Also of interest are the
values of a and o at the folding frequency f = ; these are denoted a f and o f , respectively.
The lowest and highest values of o are called max
and min
o
o , respectively, whereas the largest
max
a is called a . For the plots drawn in Figure 23.2 (note disclaimer on the right):
amax = a f ,
min
= o f ,
o
max
= oc .
o
(23.1)
Often, but not always, min

and amax occur at = , the folding (Nyquist) wavenumber at which
o
> amax , the frequency range min
> > amax is
group velocities vanish. In any case, if min
o
o
called the acoustoptical frequency gap. Frequencies within the gap are said to pertain to portion I
of the stopping band or stopband, a term derived from filter technology. Frequencies > max
o
pertain to portion II of the stopping band.
A frequency that falls within a stopping band cannot propagate as plane wave over the FEM lattice,
since there the characteristic equation has complex roots with negative real parts. This causes
exponential attenuation so any periodic disturbance with that frequency will die out.
233
(a)
(b)
,E,A = const
1
(e)
3
= Le
= 1
j2
Midnode 3
@ center
2 x
j1
j+1
j+2
Two-element
patch
xj
=1
coordinate
x
j
xj
xj +
Figure 23.1. Three-node prismatic bar element: (a) configuration; (b)

extraction of two-element patch from a regular lattice.
23.1.2. Bar3 General Mass-Stiffness Template

We begin by introducing a general template for the mass-stiffness (MS) pair. The mass template is
given four parameters:

4 + 1 1 + 3 2 + 4
1 3 4
e
e
m
m
(23.2)
Me =
1 + 3 4 + 1
2 + 4 = MCe M M +
3 1 4 .
30
30
2+
2+
16 +
4
is the consistent mass matrix (CMM), obtained for 1 = 2 =

in which m = A . Here
3 = 4 = 0, which is derived in 17.1. The template (23.2) incorporates matrix, geometric
and fabrication symmetries ab initio. It includes all DLMM by setting 3 = 1 and 4 = 2.
Because of its practical importance, however, that lumped mass subset is studied in 23.1.8 using
a two-parameter template variant.
e
MCe M M
For (23.2) to be nonnegative definite (NND), three inequality constraints have to be satisfied. Those
are more elegantly expressed in terms of the alternative -form derived in 23.1.3.
Conservation of total element mass A (invariance of linear momentum) imposes the following
homogeneous constraint:
(23.3)
21 + 2 + 23 + 44 = 0.
This constraint is not always preimposed as it may complicate intermediate expressions, but it
is eventually applied at some point. Conservation of angular momentum in 2D or 3D requires
1 = 3 , as verified by the CMM. This is ignored, however, as it hinders customization.
As regards the stiffness matrix, the following one-parameter template is used

1
1 2
1 1 0
e
4
k
Ke = Keb + Keh = k e 1 1 0 +
1
1 2 .
3
2 2 4
0
0 0
(23.4)
in which k e = E A/. Here Keb and Keh denote the basic and higher-order stiffness matrices,
respectively. This decomposition was introduced by Bergan and coworkers in the 1980s for the
development of the Free Formulation; references are provided in H.7. The higher order stiffness
is scaled by the free parameter 0. Setting = 1 produces the well known stiffness of the
quadratic (isoparametric) displacement model, whereas = 0 reduces Ke to the Bar2 stiffness
considered in 22.1.2.
234
23.1
=0
=2
Brillouin zone
Dimensionless frequency
(branches have period 2)
stopping band II
oc
oc
optical branch
o= o()
cutoff
frequency
of
max
folding optical
frequency
of
acoustoptical gap: same as stopping band I
af
folding acoustic
frequency
af
acoustic branch
a= a() folding
0
wavenumber
For the diagram as drawn
a = af
omax = oc
min
o = of
These do not generally hold

in more complex elements,
multiple space dimensions
or irregular lattices
Figure 23.2. Nomenclature for a two-branch dispersion diagram typical of 1D structural elements such as
Bar3. The stopping band is the union of I and II. (Symmetry and monotonicity about the folding wavenumber
f = is typical of prismatic, simple 1D elements in regular lattices; else those features are typically lost. In
addition, multiple optical branches will appear for characteristic equations with more than two roots for each .)
23.1.3. Bar3 Alternative Mass Template

An alternative configuration of the general Bar3 mass template is obtained by changing the four i
free parameters to three: 1 , 2 , and 3 , through the replacement rule
1 = 1 + 2 4, 2 = 14 + 41 413 , 3 = 1 2 + 1, 4 = 13 21 2. (23.5)

in which 13 = 30 1 3 . The mass conservation condition (23.3) is identically satisfied by
(23.5), which is why the number of free parameters can be cut by one. Conversely, if the i are
given and do satisfy (23.3), the i can be computed from
1 =
3 =
1
(3 + 1 + 3 ),
2

1
41 (40 + 2
480
2 = 12 (5 + 1 3 ),

23 ) + 40 (8 + 2 + 43 ) 421 (2 23 )2 .
(23.6)
(If 1 = 2 = 3 = 0, this gives 1 = 3/2, 2 = 5/2 and 3 = 2/3 for the CMM.) On inserting
(23.5) into (23.2) the so-called -form of the Bar3 mass template emerges:
1 + 2
1 2
21 + 13
e
m
(23.7)
Me =
1 2
1 + 2
21 + 13
30
21 + 13 21 + 13 30 + 41 413
An attractive feature of (23.7) is that mass matrix admissibility can be readily correlated to parameter
values. Specifically, Me is positive definite (PD) if and only if 1 , 2 and 3 are positive. This can
be proven from the following properties:
1 (Me ) =
m e 2
,
15
det(Me ) =
m e 2 3
,
225
det(Me 22 ) =
m e 1 2
,
225
(23.8)
The first equality gives one eigenvalue of Me (the other two have more complicated expressions),
whence PD mandates 2 > 0. The last equalities give the determinants of Me and of its 2 2
235
upper principal minor, respectively. Accordingly, PD requires also 3 > 0 and 1 > 0. For Me to
be nonnegative definite (NND) simply change > to . Those conditions can be harked back to the
i of Me using (23.6), but the expressions are noticeably messier. The second equality of (23.8)
gives another nice feature: Me becomes singular if and only if 2 = 0, or 3 = 0, or both.
The main advantage of Me over Me is the linear dependence of entries on the i . This simplifies
patch analysis as well as reparametrization for several template variants studied later.
23.1.4. Bar3 Patch Equations
To assess wave propagation and dispersion performance of the MS template defined by (23.2) and
(23.4), we carry out the Fourier analysis of the infinite bar lattice shown in Figure 17.1(b). Extract a
typical two node patch as illustrated. The patch has five nodes: three endpoints and two midpoints,
which are assigned global numbers j2, j1, . . . j+2. The unforced semidiscrete dynamical
equations of the patch are M p u P + K p u P = 0, in which
4 + 1
2 + 4
16
+ 2
2
+
4
me
Mp =
1 + 3 2 + 4
30
0
0
0
0
3 + 4 8
16
E A 8
Kp =
3 + 4 8
3
0
0
0
0
u P = [ u j2 u j1 u j u j+1
1 + 3
0
0
2 + 4
0
0
2(4 + 1 ) 2 + 4 1 + 3 ,
2 + 4
16 + 2 2 + 4
1 + 3
2 + 4
4 + 1
3 + 4
0
0
8
0
0
6 + 8 8 3 + 4 ,
8
16
8
3 + 4 8 3 + 4
u j+2 ]T .
(23.9)
Note that the element mass conservation constraint (23.3) is not preimposed as it would complicate
intermediate expressions. It is enforced later. Keep the second and third equations, namely those
for nodes j 1 and j. This selection picks the equations for a typical corner and midpoint node.
Accordingly, the patch equations are
p u P = 0.
p u P + K
M
(23.10)
p result on deleting rows 1,4,5 of M p and K p , respectively.

p and K
The 2 5 matrices M
23.1.5. Bar3 Fourier Analysis
We study the propagation of harmonic plane waves of wavelength , wavenumber k = 2/, and
circular frequency over the lattice of Figure 17.1(b). For convenience they are separated into
corner and midpoint waves:

u c (x, t) = Bc exp j (kx t) , u m (x, t) = Bm exp j (kx t) , j = 1.
(23.11)
Wave u c (x, t) propagates only over corner nodes and vanishes at midpoints, whereas u m (x, t)
propagates only over midpoints and vanishes at corner nodes. Both have the same wavenumber
and frequency but different amplitudes and phases. [The wave pair (23.11) can be combined to
236
23.1
form a single waveform that propagates over all nodes. The combination has two components that
propagate at the same speed but in opposite directions. This is useful when studying boundary
conditions or transitions in finite lattices, but unecessary for a periodic infinite lattice.]
As in 22.1.5, we will work with the dimensionless frequency = /c0 with c0 = E/, and
the dimensionless wavenumber = k. Inserting (23.11) into (23.10), passing to dimensionless
variables, removing scale factors, and requiring that solutions exist for any t yields the characteristic
equation

1
A (160(16+2 )2 )
A1 A3 (80+(2+4 )2 )
Bc
2 3
= 0.
Bm
A1 (80+(2+4 )2 ) 30+40(4+1 )2 +A1 (30+40+(13 )2 )
(23.12)
in which A1 = cos , A2 = cos(/2) and A3 = cos(/2) j sin(/2). For nontrivial solutions the
determinant of the characteristic matrix must vanish, which provides a quadratic equation in 2 .
For each wavenumber , solving the equation gives two squared frequencies. Their expressions,
found by Mathematica, are
P Q
P+ Q
2
2
, o = 5
,
(23.13)
a = 5
R
R
in which coefficients P, Q, and R are given by
P = c1 + 3 c2 + c9 cos ,
R = c5 4 4 24 + c8 cos ,
Q = 192 (cos 1) (c5 c3 + c8 cos ) + (c1 + 3 c2 + c9 cos )2 ,

c1 = 4 (40 + 4 1 + 2 + 4 4 ), c2 = 16 + 2 , c3 = 4 (4 + 4 ),
c4 = 4 (5 + 3 + 4 ), c5 = 60 + 4 2 + 1 c2 , c6 = 16 3 c3 ,
c7 = 4 (2 + c4 ), c8 = c6 + 2 (3 1) 20, c9 = c7 3 c2 .
(23.14)
Subscripts a and o stand for acoustic and optical branches, respectively, a terminology explained
in 23.1.1. If Me is positive definite and the conservation condition (23.3) holds, the branch
frequencies (23.13) have small (low frequency, long wavelength) Taylor series of the generic
form
a2 = 2 +
C4 4
C6 6
C8 8
+
+
+ ...,
4!
6!
8!
2o = D0 +
D2 2
D4 4
+
+ ...,
2!
4!
(23.15)
Coefficients Cn and Dn were obtained through the Mathematica built-in Series function up to
n = 10 and are displayed for some interesting instances below.
23.1.6. Bar3 Standard Template Instances
We start by considering two instances available in the FEM literature since the mid 1960s. The
CMM instance MCe is obtained for 1 = 2 = 3 = 4 = 0. Using = 1 for Ke we get
P = 208 + 32 cos , Q = 128 (237 + 224 cos 11 cos(2)), and R = 20 (3 cos ). The
squared frequencies have the small- expansions
a2 = 2 +
6
11 8
+ ...,
720 151200
2o = 60 20 2 +
237
19 4
+ ....
3
(23.16)

(b)
2
1.5
6
Optical
5
4
Continuum
Continuum
0.5
0
0.5
Acoustical
Acoustical
1
Optical
1.5
1
0
Velocity ratio c =c/c0
(a)
8
(d)
7
Continuum
6
5
Optical
1.5
Continuum
1
0.5
0
0.5
Acoustical
Acoustical
Optical
1.5
1
0
(c)
8
Figure 23.3. DDD and DGVD plots for well known Bar3 template instances treated in 23.1.6. (a,b):
diagrams for CMM instance; (c,d): diagrams for SLMM (Simpson DLMM) instance. Acoustic and optical
branches shown in red and blue, respectively. Continuum case = and c = 1 shown in black.
The SLMM (Simpson lumped diagonal mass matrix) instance derived in (17.3) of 17.1 results
if 1 = 1, 2 = 4, 3 = 1, and 4 = 2. Using = 1 in (23.4) gives P = 220 + 20 cos ,
Q = 200 (147 + 140 + cos(2)), and R = 100. The squared frequencies have the small-
expansions
a2 = 2
8
6
+ ...,
1440 48384
2o = 24 2 2 +
4
+ ....
12
(23.17)
For small , SLMM fits the continuum better than CMM. Dispersion diagrams for the foregoing
instances are plotted in Figures 23.3(a,c). Corresponding group velocity diagrams are shown in
Figures 23.3(b,d). As in the case of the two-node bar, the consistent mass overestimates the
continuum frequency = for 0 , whereas the lumped mass underestimates it.
23.1.7. Bar3 Low-Frequency Fitting
Inspection of the coefficient of 6 in (23.16) and (23.17) suggests combining one third of MCe with
two thirds of MeL to cancel it. Setting
1 = 2/3,
2 = 8/3,
3 = 2/3,
in (23.4) gives
MCe L
A
= 13 MCe + 23 MeL =
90
238
4 = 4/3,
14
1
2
1
14
2
= 1,

2
2 .
56
(23.18)
(23.19)
23.1
For this instance, labeled BLCD, P = 24 (9 + cos ), Q = 32 (927 + 884 cos 11 cos 2), and
R = 20 (13 cos )/3. It has the small expansions
a2 = 2
8
+ ...,
37800
2o = 30
15 2
11 4
+
+ ....
4
32
(23.20)
Dispersion and group velocity diagrams are shown in Figure ?(a,b). Despite the O( 8 ) accuracy
achieved in the acoustic branch of BLCD, it is shown next that this instance is not optimal.
Considering next the general MS template (23.2)(23.4), let us find the MS pair for which the
acoustic branch a best matches the continuum = forsmall . Given the expansion of a2 in
(23.15), the goal is to make as many coefficients beyond 2 vanish as possible, and to minimize
the magnitude of the first surviving one. The analysis was actually performed using the -form
(23.7) of the general mass template. Four free parameters are available: 1 , 2 , 3 , and . Only a
procedural summary and final results are given.
It is possible to make C4 = C6 = 0 without difficulty, which permits elimination of 1 and 2 . But
all solutions of C8 (3 , ) = 0 are imaginary, so the term in 8 cannot be cancelled. Extremization
to 3 and gives only one constraint: 160 2 1203 + 932 = 0, from which
of C8 with respect
3 = 4 (5 15)/3. Both signs give the same C8 . Taking = 1 for convenience, the sign in
3 , and working back we get

3
85
4
4
(23.21)
2
2 375, 2 = 5 12 15, 3 =
5 15 , = 1.
1 =
6
5
3
The sign for 1 gives better conditioned mass matrices and still the same C8 , so we pick that one.
The numeric values to 16 places are 1 = 2.7835604012611213, 2 = 3.0635083268962915,
and 3 = 1.50268887172344. The resulting minimum of |C8 | is C8best = 64/3 28 3/5 =

0.355373. This is about 3 times smaller than the C8 = 8!/37800 = 16/15 from (23.19).
Converting to the i parameters via (23.5) yields
9112 a1 7 a2
328 a2
6112 a1 a2
46+6 a1 +4 a2
, 2 =
, 3 =
, 4 =
, (23.22)
6
3
6
3
in which a1 = 31/4 53/4 = 4 375 and a2 =

15. Numerical values to 16 places are
1 = 1.8470687281574132, 2 = 0.3387110767802213, 3 = 0.7200520743648302, and
4 = 1.3682381704561768. The resulting mass matrix, labeled BLFM (for best low frequency
match), given to 16 places, is

0.1949022909385804 0.0093315975211724 0.0210587276514608
MeB L F M = m e 0.0093315975211724 0.1949022909385804 0.0210587276514608 .
0.0210587276514608
0.0210587276514608 0.5446237025593408
(23.23)
e
=
A .
Its eigenvalues (to 6 places) are positive: 0.547077, 0.204234, and 0.183117 times m
Hence
For this instance P = 8 (35 2 15
this mass matrix is admissible
and well scaled.
(5
2 15)
cos ), Q = 1920 (15 + 4 15 + (15 4 15) cos ) cos(/2)2 , and R = 20 (53 10 15 +
(7 2 15) cos )/3. It has the small- expansions
5(29
8
1127
291
30
15
15)) 2
a2 = 2 +
8 + . . . 2o =
. . . (23.24)
3024(5 15)3
5 15
32(5 15)3
1 =
239

(b)
2
7
Continuum
6
5
4
(a)
8
Optical
Continuum
1
0.5
0
0.5
Acoustical
Acoustical
Optical
1.5
1
0
1.5
7
Continuum
Optical
3
Acoustical
1.5
Continuum
1
0.5
0
Acoustical
Optical
(e)
(f)
7
6
1.5
Continuum
Optical
4
3
1.5
Continuum
1
0.5
0
0.5
Acoustical
Acoustical
Optical
1.5
1
0
0.5
2
0
(d)
6
4
(c)
8
Figure 23.4. DDD and DGVD plots for RHFP instances derived in 23.1.7 and 23.1.8. (a,b): diagrams for the
BLCD instance (23.19). (c,d): diagrams for the BLFM instance (23.23). (d,e): diagrams for the BLFD instance
(23.29). Acoustic and optical branches shown in red and blue, respectively. Continuum case = and c = c/c0 = 1
shown in black.
Dispersion and group velocity diagrams are shown in Figure 23.4(c,d). Note that
values at thefolding
(Nyquist) wavenumber = are identical: a2 () = 2o () = 12(10 15)/(23 4 15) =
9.792694126734647, which is amazingly close to the continuum value of 2 = 9.86960440108935.
(In fact, the AB for < and the continuum are indistinguishable at plot resolution.) There is no
acoustoptical gap; instead we observe a bifurcation point.
23.1.8. Bar3 Lumped Mass Template Variant
Although diagonally lumped mass matrices (DLMM) plainly form a subset of the general template
(23.2), their practical importance justifies the use of a more compact two-parameter form. This is
done by taking
1 = L1 + 1, 2 = L2 + 4, 3 = 1, 4 = 2.
(23.25)
2310
23.1
Replacing into (23.2) produces the lumped mass template variant

0
= MeS L M M
0 .
L2
(23.26)
The baseline mass matrix is the DLMM (17.3) produced by Simpsons 3-point integration rule, and
now labeled SLMM. The stiffness matrix template is still (23.4). Parameters L2 and can be
eliminated in favor of L1 through
A
MeL =
30
5 + L1
0
0
0
5 + L1
0
0
0
20 + L2
2 L1 + L2 = 0,
A
+
30
L1
0
0
0
L1
0
(10 L1 )2
.
20 (5 L1 )
(23.27)
The first constraint expresses element mass conservation while the second one enforces C4 = 0
and makes the AB agree with the continuum through 4 in the expansion (23.15). (This agreement
is considered essential as otherwise there would be no advantage in using this element instead of
Bar2.) As only one parameter remains, customization is straightforward. The admissible range in
L1 for PD mass is 5 < L1 < 10, but if L1 > 5, < 0 and Ke becomes indefinite.
On applying (23.27), the first NZ term in the ABTS beyond 2 is C6 6 /6!, in which C6 = (5
3 L1 + 2L1 )/( L1 10). Trying to attain O( 6 ) accuracy by setting C6 = 0 is futile since the L1
roots are complex conjugate.
Solving C6/ L1 = 0 gives two real solutions: L1 = 5(2 3).
Of these only the one with 3 keeps Me admissible. Replacing into (23.27) gives the signature
L1 = 5(2
3),
L2 = 2 L1 = 10(2
3),
3
.
4( 3 1)
(23.28)
Numerical values to 16 places are L1 = 1.339745962155614, L2 = 2.679491924311228,

and = 1.024519052838329. The 6 term in the AB is about 36% smaller than that of SLMM:
6 /2246 versus 6 /1440. The template instance, labeled BLFD, is
MeB L F D
KeB L F D

0
A 5 3 5 0
=
0 ,
0
5 35
30
0
0
20 10 3

1
1
4E A
E A 1 1 0
=
1 1 0 +
1
1

(
3
1)
0
0 0
2 2

2
2 .
4
(23.29)
The Taylor series of the dispersion branches are

a2
10 3 10 6
...,
=
720
2
2o
3
2 . . .
=

2
( 3 1)
4 36
12
(23.30)
The DDD and DGVD are shown in Figure 23.4(e,f). As in the case of theBLFM, pictured
in (c,d)
2
2
of that figure, the branches intersect at = , where a f = o f = (62 3)/(2 3) 9.4641.
2311
23.1.9. Bar3 MSTS Customization

Since DLMM are often used in explicit DTI, it is of some interest to find whether the stable time
stepsize can be maximized while still satisfying O( 4 ) accuracy. This goal pertains to the MSTS
customization of Table 21.1. Let max be the maximum of a and o over the Brillouin zone
[0, 2 ]. To maximize the stable time step, one minimizes max over free parameters, while
trying to keep both mass and stiffness admissible. This procedure can be streamlined by assuming
a DDD configured as in Figure 23.2, whence only frequency values at = 0 and = need to be
considered. Since a |=0 = 0, the search only involves oc , a f and o f , or (for convenience)
their squares. For the DLMM template variant (23.26) under the accuracy constraints (23.27) the
process boils down to solving the max-min problem in one variable:

2
60
60(10 L1 ) 4(10 L1 )
2
2
,
,
.
(23.31)
min max oc , a f , o f = min max
L1
L1
L1 5
5 + L1
25 2L1
A simple plot shows that the cutoff frequency dominates for admissible L1 (5, 5), so it is
sufficient to minimize with respect to 2oc . This again leads to the solution (23.25). Consequently
the BLFD instance also maximizes the explicit DTI time step. The reward, however, is marginal
with respect to SLMM: only about a 3.5% gain.
To get a more significant improvement, it is necessary to keep free, and accept that O( 4 ) accuracy
is lost. It may be verified that the largest possible stable timestep is produced by the signature
L1 = 5/2,
L2 = 2 L1 = 5,
= 3/8.
(23.32)
which apportions nodal masses as 1:1:2, while substantially modifying the stiffness matrix. Setting
(23.32) gives an instance with a constant optical branch (COB) 2o = 8. Its stable stepsize is 1.673
times that of BLFD. But its LF performance is exactly the same as that of the lumped-mass Bar2,
which does not have an OB. So it is largely a curiosity.
23.1.10. Reducing High Frequency Pollution
The presence of the optical branch (OB) does not affect vibration calculations in structural dynamics.
One simply ignores those eigenfrequencies as nonphysical. However, the OB may become a
nuisance in direct time integration (DTI) for problems that involve discontinuities, such as pulse
propagation, or contact-impact, because it may feed spurious noise. To alleviate this problem three
approaches may be tried at the template level:
1.
Singular Mass Matrix. If Me is made singular with an appropriate null eigenvector, the optical
branch is raised. In fact it becomes infinite at = 0. The net effect is that the acoustoptical gap
is increased at low wavenumbers. This heps to filter out frequencies that fall in the gap, since
they will decay exponentially. One drawback of singularity is that explicit DTI is excluded,
even if Me is diagonal.
2.
Selective Mass Scaling. A scaled stiffness matrix is added to the mass. As discussed in
21.3.5, eigenvectors are unchanged but higher natural frequencies are effectively reduced.
The effect is similar to that of adding stiffness proportional damping, but without altering
vibration modes. It may be done at the element or assembly (master) level. In the study of
23.1.12 it is done at the element level.
2312
23.1
3.
Constant Optical Branch. A constant optical branch (COB) is one independent of wavenumber.
It stays at a constant frequency o = oc over the entire Brillouin zone, and has zero group
velocity since oc / = 0. To be effective in cutting noise pollution, oc a f , in which
a f is the folding acoustic frequency. If that holds, the stopping band above a f is effectively
maximized. (Even if a mesh frequency hits oc exactly, it will not propagate since its group
velocity vanishes.) A COB template is one that possesses that property (for each OB should
there be more than one).
The three foregoing approaches are studied below for the Bar3 element.
23.1.11. Bar3 Spectral Mass Variant
Making Me singular is not sufficient; it is important to have the correct null eigenvector. To achieve
that it is convenient to use the Spectral Parametrization (SP) outlined in 21.3.2. Select three
generalized coordinates: g0 , g1 and g2 as amplitudes of three physically transparent eigenmotions:
g0
Amplitude of rigid body motion: v0 = [ 1
g1
Amplitude of acoustic bar motion: : v1 = [ 1
g2
Amplitude of optical bar motion: : v2 = [ 1
2 ]T .
1 ]T .
1
1
0 ]T .

3, 2 and
Those
three
vectors
are
mutually
orthogonal.
To
make
them
orthonormal,
divide
by
6, respectively. Stacking the orthonormalized vectors as columns, the linkage between physical
and generalized coordinates can be expressed as

1/3 1/ 2 1/6
g0
ue = [ u 1 u 2 u 3 ] = 1/3 1/ 2
(23.33)
1/ 6 = g1 = HT g.
0
2/ 6
g2
1/ 3
The inverse relation is g = H ue because H is orthogonal by construction, and thus H1 = HT .
As mass matrix in generalized coordinates we stipulate the 3 3 diagonal matrix D of entries
m e S0 /3, m e S1 /45, and m e S2 /15, in which m e = A , and the scaling factors were chosen
for convenience in cleaning up downstream expressions. Element mass conservation will require
S0 = 1, so the first entry is simply m e /3. Transforming to physical coordinates yields the spectral
mass template variant

10 + S1 + S2 10 S1 + S2 10 2 S2
15 0
0
e
e
m
m
MeS =
HT 0 S1 0
H=
10 S1 + S2 10 + S1 + S2 10 2 S2 .
45
90
10 2 S2
10 2 S2
10 + 4 S2
0 0 3 S2
(23.34)
The variant (23.34) is a subset of the general template (23.2) that results by taking
1 = 13 ( S1 + S2 2), 2 = 13 (4 S2 38), 3 = 13 (13 S1 + S2 ), 4 = 13 (42 S2 ).
(23.35)
e
e
e
By construction, the eigenvalues of (23.34) are m /3, m S1 /45 and m S2 /15, whence the nonnegativity condition is fulfilled if S1 and S2 are nonnegative. To make MeS singular, set S2 = 0,
which produces

10 + S1 10 S1 10
e
m
(23.36)
MeS = HT D H =
10 S1 10 + S1 10 .
90
10
10
10
2313
Solving C4 = 0 and C6 = 0 yields two solutions for and S1 , of which we pick that with larger
S1 (to get a better Me eigenvalue). This gives
5 + 10
3(5 + 10)
= 12.24341649025257, =
= 0.6801898050140316,
S1 =
2
12
(23.37)
in addition to S2 = 0. Inserting into (23.39) gives the instance labeled BSSM for Best Singular
Spectral Mass. The mass matrix, with numerical values given to 6 places, is

M11 M12 M13
0.247149 0.024927 0.111111
e
m
MeB SS M =
0.247149 0.111111 . (23.38)
M12 M22 M23 m e 0.024927
180 M
M
M
0.111111
0.111111 0.111111
13
23
33
in which M11 = M22 = 35+ 10, M12 = 53 10, and M13 = M23 = M33 = 20. The associated
stiffness matrix, with numerical values given to 6 places, is

K 11 K 12 K 13
1.906920 0.093080 1.813839
e
k
KeB SS M =
1.906920 1.813839 . (23.39)
K 12 K 22 K 23 k e 0.093080
9 K
K 23 K 33
1.813839 1.813839
3.627679
13
in which k e = E A/,
K 11 = K 22 = 14 + 10, K 12 = 4 + 10, K 13 = K 23 = 10 2 10
and K 33 = 20 + 4 10. Dispersion and group velocity diagrams are shown in Figure 23.5(a,b).
The Taylor series of the dispersion branches are
a2 = 2
8
...,
6048
2o = 240 2 +
5 4
+ ...
126
(23.40)
The O( 6 ) AB accuracy of this element is comparable to that of BLCD and BLFM, but its OB gets
out of the way. Is this the template instance for all seasons? Only future experimentation in direct
time integration (DTI) will tell.
23.1.12. Bar3 Selective Mass Scaling Variant
In the Selective Mass Scaling (SMS) approach outlined in 21.3.5, the mass matrix is modified by
adding a scaled version of the stiffness matrix:
MeK = Meu + c K Ke .
(23.41)
Here Me is an unmodified mass matrix, and c K a scaling coefficient with appropriate physical
dimensions. Both Me and Ke may generally be template forms. Since Me and Ke have different
physical dimensions, it is convenient to change the raw expression (23.41) to
MeK = Meu + K s e Ke ,
(23.42)
in which s e is a scaling coefficient with dimension of mass-over-stiffness (equivalently, 1/s e has

dimensions of squared physical frequency) while K is a dimensionless free parameter. For the
Bar3 element we take s e = ( A )/(E A/) = 2 /E. This can be maneuvered to the following
equivalent form, which is convenient for implementation:
.
MeK = Meu + K m e K
e
2314
(23.43)
23.1
e is a dimensionless stiffness matrix

Here m e = A is (as usual) the element mass, whereas K
u
obtained by setting E = 1, A = 1 and = 1. To reduce the overall number of parameters, we pick
Meu to be the diagonally-lumped template subset (23.26); this agrees with the common use of SMS
e . Hence
in explicit DTI. The general stiffness template (23.4) with unit E, A and is used for K

MeK = m e
1
30
5 + L1
0
0
0
5 + L1
0
0
0
20 + L2
4 K
+
3
1
1
2
1
1
2
2
2
4

.
(23.44)
Mass conservation is enforced if L2 = 2 L1 . Inserting this in (23.44) we have three free

parameters: L1 , K and . This MeK with L2 = 2 L1 is a particular case of the general mass
template if
1 = 1 + L1 + 10 (4 + 3) K , 2 = 4 2 L1 + 160 K ,
(23.45)
3 = 1 + 10 (4 3) K , 4 = 2 80 K .
Unlike previous variants, now appears in the mass template. The linkage (23.45) becomes linear
if is preset, for example to 1, and nonlinear otherwise.
Further experimentation with the SMS template variant (23.44) was confined to L1 = L2 = 0,
which takes SLMM as original mass matrix. That leaves out two free parameters: K and .
Suppose K is chosen. Then O( 4 ) AB accuracy can be maintained by taking
=
1
.
1 12 K
(23.46)
If K > 1/12, < 0 and Ke becomes indefinite. But setting 0 K 1/12 hardly change the
higher frequencies. For that one needs a much larger K ; say K = O(1). If so, adjusting Ke as
per (23.46) is precluded: the cure is worst than the disease. One may as well set = 1. The high
frequencies are cut down, but LF accuracy is seriously lost.
This tradeoff is vividly displayed in the vibration benchmarks reported in 23.1.14. Three instances
labeled SMS1, SMS2 and SMS3, are tested there. Their signatures are { K = 1/24, = 2},
{ K = 1/2, = 1}, and { K = 2, = 1}, respectively. Dispersion and group velocity diagrams
for SMS2 are shown in Figure 23.5(c,d). The poor LF fit is obvious.
23.1.13. Bar3 Constant Optical Branch Variant
The investigation of the general Bar3 template (23.2)(23.4) for COB instances was done under two
preset conditions: C4 = 0, which enforces order O( 4 ) accuracy in the acoustic branch (AB), and
= 1 in the stiffness template (23.4). Several one-parameter families satisfying these conditions
were found. The two that produced simpler mass matrices were retained, reparametrized, and
labeled COBA and COBB. Associated mass matrices are subscripted OA and OB, respectively.
The COBA family is defined by
MCe O B A
me
=
12
6 A
2 A
2 + 2 A
2 A
6 A
2 + 2 A
2315
2 + 2 A
2 + 2 A
4 4 A

.
(23.47)

(b)
2
7
6
(a)
8
Continuum
Optical
5
4
3
Continuum
1
0.5
0
0.5
Acoustical
Acoustical
1
Optical
1.5
1
0
1.5
1.5
Continuum
5
4
3
1
0
Optical
Continuum
0.5
0
Optical
Acoustical
(e)
(f)
2
7
6
1.5
Acoustical
Continuum
5
4
3
Optical
1
1
Continuum
1
0.5
0
Optical
Acoustical
1.5
Acoustical
0
1.5
0.5
2
0
0.5
(d)
2
(c)
8
Figure 23.5. DDD and DGVD plots for three RHFP instances derived in 23.1.1123.1.13. (a,b) Diagrams for the
BSSM instance (23.19). (c,d) Diagrams for the SMS2 instance (23.23). (e,f) Diagrams for the COB0 instance; first of
(23.53). Acoustic and optical branches shown in red and blue, respectively. Continuum case = and c = c/c0 = 1
shown in black.
in which m e = A . The determinant is (1 A )/18. MC O B A is PD if A < 1. Parameter A is

linked to those of the general template (23.2) by
1 = 11
5 A
,
2
2 = 2(3 + 5 A ),
3 = 6
5 A
,
2
4 = 7 + 5 A .
(23.48)
The COBB family is defined by

MCe O B B
me
=
432
96 36 B B2
24 12 B + B2
48 + 24 B
24 12 B + B2
96 36 B B2
48 + 24 B
48 + 24 B
48 + 24 B
384

.
(23.49)
The determinant is (36 12 B B2 )2 /34992. MC O B B is positive definite (PD) if 6( 2 + 1) <

2316
23.1
Table 1. Bar3 Instances Compared In Fixed-Free Bar Vibrations Tests
Variant
ref. eqn.
Full or
diag Me
Signature
First NCT
in ABTS
Cutoff & folding freq

oc
a f
o f
CMM
(23.2)
=1, 1 =2 =3 =4 =0
+ 6 /720
7.746 3.162 3.464
SLMM
(23.26)
=1, L1 = L2 =0
6 /1440
4.899 2.828 3.464
BLCD
(23.2)
See (23.18)
8 /37800
5.477 2.928 3.464
BLFM
(23.2)
See (23.22)
8 /113458
5.159 3.129 3.129
BLFD
(23.26)
See (23.28)
6 /2246
4.732 3.076 3.076
BSSM
(23.34)
See (23.37)
8 /6048
SMS1
(23.44)
=2, K =1/24
6 /640
4.000 2.828 3.098
SMS2
(23.44)
=1, K =1/2
4 /2
1.359 1.265 1.309
SMS3
(23.44)
=1, K = 2
2 4
0.700 0.686 0.692
COB0
(23.47)
=1, A = 5/3
6 /240
2.449 2.449 2.449
Instance
name
2.711 5.714
NCT: non-continuum term, ABTS: AB Taylor series of 2 wrt , centered at = 0

a
: at = 0; : at = ; : at = .
oc
o
af
a
of
o
B < 6( 2 1). Parameter B is linked to those of the general template (23.2) by

1 =
8 5 B
5 2
B,
3
2
72
2 =
32
,
3
3 =
8 5 B
5 2
+ B,
3
6
72
4 =
16 + 5 B
.
3
(23.50)
These two families are taken to collectively define the Bar3 template variant identified as COB.
They coalesce only for A = 5/3 and B = 6, which produces an instance discussed below.
An interesting result is that the acoustic branch is identical for all COB instances:
a2 =
12(1 cos )
6
8
= 2
...
5 + cos
240 6048
(23.51)
whereas the constant OB value is family and parameter dependent:

2oc A =
16
,
1 A
2ocB =
432
.
36 12 B B2
(23.52)
It follows that the only role played by A and B is to adjust the OB height along the vertical
DDD axis. As noted in 23.1.10, it should equal or exceed the folding acoustic frequency a2 f = 6,
which is the same for all COB instances on account of (23.51). This requires A 5/3 and
B 6. As A 1 and B 12 the OB moves to and the mass matrices assume different
2317
BLCD
BLFD
BLFM
6
4
CMM
SLMM
2
0
1
16
BLFM
6
4
SLMM
SMS1
COB0
SMS2
SMS3
0
1
16
BSSM
16
LFCF Bar3 Template

BLFM
10
CMM
SLMM
0
2
(f)
SMS1
BSSM
COB0
SMS2
SMS3
1
16
RHFP Bar3 Template

BLFD
BLCD
(e)
RHFP Bar3 Template

10
(d)
8
CMM
10
BLFD
BLCD
10
(c)
LFCF Bar3 Template

16
10
(b)
LFCF Bar3 Template

(a)
10
RHFP Bar3 Template

8
6
4
SMS1
BSSM
SMS2
0
2
COB0
SMS3
16
Figure 23.6. Performance of ten Bar3 template instances in predicting the first three natural
frequencies i , i = 1, 2, 3 of the fixed-free prismatic homogeneous bar shown in Figure 22.5(b).
See text in 22.1.9 for a detailed description of the log-log plots.

limits. For COBA, MCe O B A A 1 is the optimal Bar2 matrix (22.16), which is PD. On the other

hand the limit MCe O B B B 12 falls in the indefinite range. Three noteworthy instances of the mass
matrices produced by these two families are

23
11 16
3
1 1
e
e
m
m
MCe O B0 =
11
23 16 , MCe O B1 =
1
3 1 ,
36 16 16
6 1 1
32
2
(23.53)

4
1
2
e
m
MCe O B2 =
1
4 2 .
18 2 2 16
MCe O B0 is the unique mass matrix for which 2oc = a2 = 6; that is, the COB passes through the
folding (Nyquist) frequency. It emerges by setting either A = 5/3 in (23.47) or B = 6 in
(23.49). MCe O B1 which gives 2oc A = 16, is the simplest mass matrix that produces a COB. It is
obtained by setting A = 0 in (23.47). Finally MC O B2 , which yields 2ocB = 12, was the first COB
instance discovered, as noted in H.7. It is obtained by setting B = 0 in (23.49).
Dispersion and group velocity diagrams for COB0 are shown in Figure 23.5(e,f). The DDD for
COB1 and COB2 would possess an identical AB branch but the flat OB would appear higher,
whereas the DGVD would be identical. Those two diagrams are omitted to save space.
2318
23.1
23.1.14. Bar3 Test: Vibrations of a Fixed-Free Bar Member

The natural frequency benchmark test presented in 22.1.9 for three Bar2 discretizations is repeated
for the ten Bar3 template instances listed in Table 23.1. The fixed-free bar member is pictured in
Figure 22.5(b). It is prismatic, with constant E = 1, A = 1 = 1. The total member length
is L = /2. With those numerical properties, the continuum eigenfrequencies 0i are given by
(22.18). The member is divided into Ne identical elements, with Ne = 1, 2, . . . 16.
To reduce cluttering the instances in Table 23.1 are divided into two groups of five each. Results
are presented in number of correct digits versus number of elements for the first three frequencies,
exactly as described for the Bar2 test in 22.1.9. Group 1 include CMM and SLMM as well as
instances constructed with optimal LFF customization in mind: BLCD, BLFM and BLFD. Results
are displayed in Figure 23.6(a,b,c). Group 2 includes instances derived with RHFP in mind: BSSM,
SMSx (x = 1, 2, 3) and COB0. Results are displayed in Figure 23.6(d,e,f).
BLFM is the clear winner in the first group, with BLCD close behind, while the others, with
only O( 4 ) AB accuracy, lag appreciably. In the second group, BSSM is the clear winner, with
performance comparable to BLFM and BLCD of the first group. SMS1 and COB0 are way behind,
while SMS2 and SMS3 are highly inaccurate. (As observed in 23.1.12, SMS1 would hardly effect
any HF reduction, so its reasonable LF accuracy is misleading.)
2319
24
Mass Templates
for Plane Beam
Elements
241
Chapter 24: MASS TEMPLATES FOR PLANE BEAM ELEMENTS
TABLE OF CONTENTS
Page
24.1
24.2
The Bernoulli-Euler Plane Beam Element

. . . . . . . . . .
24.1.1
The BE Beam Mass Template
. . . . . . . . . . .
24.1.2
BE Beam Template Fourier Analysis . . . . . . . . . .
The Timoshenko Plane Beam Element
. . . . . . . . . . .
24.2.1
Ti-Beam Continuum Elastodynamic Analysis . . . . . . .
24.2.2
Ti-Beam Element
. . . . . . . . . . . . . . .
24.2.3
The Ti-Beam Mass Template
. . . . . . . . . . . .
24.2.4
Ti-Beam Full Mass Parametrization
. . . . . . . . .
24.2.5
Ti-Beam Block-Diagonal Mass Parametrization . . . . . .
24.2.6
Ti-Beam Fourier Analysis . . . . . . . . . . . . .
24.2.7
Ti-Beam Selected Template Instances
. . . . . . . . .
24.2.8
Ti-Beam Vibration Analysis Example . . . . . . . . .
242
243
243
243
246
246
249
249
2410
2411
2412
2413
2414
24.1
THE BERNOULLI-EULER PLANE BEAM ELEMENT
24.1. The Bernoulli-Euler Plane Beam Element

This Section and the next one study templates for two-node plane beam elements constructed from
the Bernoulli-Euler (BE) and Timoshenko models, respectively. To keep the material relatively
compact, two restrictions are observed:
Only mass matrix templates are developed.
The only customization is low frequency continuum fit (LFCF)
To enforce the first one, the optimal stiffness matrix for statics (optimal means that it satisfies the
homogeneous static equilibrium equations over the element) is chosen and kept fixed. Simultaneous
adjustment of the mass and stiffness templates to form MS pairs is relegated to future research.
Prior experience in this regard, cited in H.7, suggest that the improvement is marginal.
24.1.1. The BE Beam Mass Template
The Bernoulli-Euler (BE) beam model is a special case of the Timoshenko model treated in 24.2.
Nevertheless it is useful to build its mass template separately, since results provide a valuable cross
check with the more complicated Timoshenko beam. The well known CMM of this element is
derived in 18.1, to which the reader is referred for notation; the derivation assumes a prismatic
two-node element with four nodal DOF with the standard cubic shape functions. This matrix is
augmented to produce the following entry-weighted template:
13
Me
=m
35
+ 11
11
( 210
+ 12 )
1
+ 22 )2
( 105
9
70
13
( 420
13
35
+ 13
+ 23 )
+ 11
13
( 420
+ 14 )
1
( 140
+ 24 )2
11
( 210 + 12 )
(24.1)
1
+ 22 )2
( 105
symm
in which m e = A . The parameters in (24.1) are i j , in which i j identifies the mass matrix
entry. The template (24.1) accounts for matrix symmetry and some physical symmetries. Three
more conditions can be imposed right away:
14 = 23 ,
13 = 11 ,
212 = 11 + 222 + 223 224 .
(24.2)
The first comes from prismatic fabrication, and the others from conservation of total translational
mass and angular momentum, respectively. Four free parameters remain: {11 , 22 , 23 , 24 }.
For the stiffness matrix we take the well known one for a plane prismatic homogeneous BE beam
element
12 6 12 6
EI
42 6 22
(24.3)
Ke = 3
12 6

symm
42
in which I = Izz is the second moment of inertial of the cross section with the respect to z, which
is chosen to go along the neutral axis. If the FEM model contains only prismatic beams, this Ke
is nodally exact, and consequently statically optimal. (It can also be derived from the equilibrium
equations.) This stiffness is kept fixed throughout the Fourier analysis.
243
24.1.2. BE Beam Template Fourier Analysis

The Fourier analysis procedure should be by now familiar to the reader. An infinite lattice of
identical beam elements of length is set up. This will look like Figure 22.2(bd), except that the
member is now a plane beam. Plane waves of wavenumber k and frequency propagating over
the lattice are represented by

(24.4)
v(x, t) = Bv exp j (kx t , (x, t) = B exp j (kx t , j = 1.
At a typical lattice node j there are two DOF: v j and j . Two patch equations are extracted, and
converted to dimensioneless form on defining = k and = c0 /, in which c0 = E I /( A4 )
is a reference phase velocity. The condition for wave propagation gives the characteristic matrix
equation

Cvv Cv
(24.5)
= Cvv C Cv C v = 0,
det
C v C
in which

Cvv = 8402(13+3511 ) 2 (840+(97011 ) 2 ) cos /35,

C v = Cv = j 2520 + (13+42023 ) 2 sin /210,

C = 1680 4(1+10522 ) 2 + (840 + 3(1+14024 ) 2 ) cos /210.
(24.6)
The condition (24.5) gives a quadratic equation in 2 that provides two dispersion solutions:
acoustic branch (AB) a2 () and optical branch (OB) 2o (). The AB represent genuine flexural
modes, whereas the OB is a spurious byproduct of the FEM discretization. The small- (low
freqnecy, long wavelength) expansions of these roots are
a2 = 4 + C6 6 + C8 8 + C10 10 + C12 12 + . . . ,
2o = D0 + D2 2 + D4 4 + . . . ,
(24.7)
2
2
in which C6 = 11 222 423 + 224 , C8 = 1/720 + 11 + 422 + 223 /3 + 1622 23 +
16223 + 11 (1/12 + 422 + 823 424 ) 24 822 24 1623 24 + 4224 , etc.; and D0 =
2520/(1 + 42022 42024 ), etc. Mathematica calculated these series up to C14 and D4 .
The continuum dispersion curve is 2 = 4 , which automatically matches a2 as 0. Thus
four free parameters offer the opportunity to match coefficients of four powers: { 6 , 8 , 10 , 12 }.
But it will be seen that the last match is unfeasible if Me is to stay nonnegative. We settle for a
scheme that agrees up to 10 . Setting C6 = C8 = C10 = 0 while keeping 22 free yields two sets
of solutions, of which the most useful one is

11 = 422 67/540 (4/27) 38/35 10822 ,

(24.8)
23 = 43/1080 222 + 95/14 67522 /54,

24 = 19/1080 22 + 19/70 2722 /27.
The positivity behavior of Me as 22 is varied is shown in Figure 24.1(a). M(e) is indefinite for
22 < min
22 = (27 4 35)/5040 = 0.0006618414419844316. At the other extreme the solutions
of (24.8) become complex if 22 > max
22 = 19/1890 = 0.010052910052910053.
244
24.1
THE BERNOULLI-EULER PLANE BEAM ELEMENT
d 4 = det(Me)
0
5
d1
max
22 =
d2
min
22
0.01005291 10
0.02
b22 =0.00281659
=0.0283026
C12 106
a =0.00412698
22
= 0.00066184
15
d3
0.03
z22
0.01
22
0.03
0.01
0.02
0.01
22
0.01
Figure 24.1. Behavior of Me as function of 22 with other parameters given by (24.8): (a) determinants dk of
max
12 in ABTS
principal minors of order k of Me , showing legal positivity range {min
22 , 22 }; (b) coefficient C 12 of
series.
Figure
24.1(b)
plots C12 (22 ) = (111545 3008 + z15120(525 + 4)22 )/685843200, with
= 70 19 189022 . This has one real root 22 = 0.02830257472322391, but that
max
gives an indefinite
mass matrix. For 22 in the legal range [min
22 , 22 ], C 12 is minimized for
b22 = (25 105 171)/30240 = 0.0028165928951385567, which substituted gives the optimal
mass matrix:
a11
1788 a13 732
me
a22 2 732 a24 2
MeL F F O pt =
a33
1788
30240
symm
a44 2
(24.9)
0.389589 0.059127 0.110410 0.024206

0.0123402 0.024206 0.0055482
= me
.
0.389589 0.059127
0.0123402
in which a11 = a
2724 + 60 105, a22 = a44 = 117 + 25 105 and
33 = 12396 60 105, a13 =
a24 = 219 + 5 105. For this set, C12 = (25 105 441)/91445760 = 2.021 106 . Another
interesting value is 22 = 13/3150 = 0.004126984126984127, which substituted in (24.8) yields
rational values for the other parameters: 11 = 13 = 23/2100, 12 = 14 = 23 =
23/4200, 24 = 23/4200 and 24 = 17/12600. Substitution into (24.1) gives
4818 729 1482 321

me
1722 321 732
MeB L F M =
4818 729
12600
symm
1722
(24.10)
0.382381 0.057857 0.117619 0.025476

0.0136512 0.025476 0.0057942
= me
.
0.382381 0.057857
symm
0.0136512
For this matrix, C12 = 41/18144000 = 2.26 106 . Its magnitude is only about 10% higher
than for the truly LFF optimal (24.9). Since its entries are simpler, (24.10) is adopted as BLFM
245
matrix for the BE element, and used as a baseline for the Timoshenko beam element investigated
in 24.2.
24.2. The Timoshenko Plane Beam Element
This last example is far more elaborate than the previous ones. The goal is to construct a mass
template for the prismatic, plane beam Timoshenko model, a name often abbreviated to Ti-beam.
It includes the BE model as special case; consequently results can be crosschecked with those of
24.1. One interesting feature of this model is that the continuum dispersion diagram has two
branches, both of which are physical.
The acoustical-like branch, which has zero frequency at zero wavenumber, corresponds to lowerfrequency bending oscillations in which the beam displaces transversely. The optical-like branch,
which exhibits a nonzero cutoff frequency, corresponds to higher frequency shear oscillations.
Because of these interpretations, they are called the flexural frequency branch (FFB) and the shear
frequency branch (SFB), respectively. They are identified by subscripts f and s, respectively, for
the continuum model, while a and o are reused for the FEM discretization. Both branches are
dispersive, meaning that group velocity depends on wavenumber. Those velocities tend to finite
values, except in the BE limit.
The upshoot of these complications is that LFCF customization, which was so clear-cut with Bar3,
becomes ambiguous: do we want to fit the FFB or the SSB? For thin beams, (as well as in the
BE limit, in which case the SSB moves to ) the FFB is dominant. But as the beam becomes
progressively thicker (as measured by a slenderness coefficient introduced in 24.2.1) the situation
is less clear: for an extremely thick beam the shear oscillations may well dominate. (Of course in
that case the Timoshenko model is questionable.)
The continuum model is first studied in some detail, since frequency expansion formulas applicable
to template customization by characteristic root fitting are not available in the literature.
24.2.1. Ti-Beam Continuum Elastodynamic Analysis
Consider a structural beam member modeled as a shear-flexible Timoshenko plane beam (Ti-beam),
as illustrated in Figure 24.2. This figure provides the notation used below. Section properties
{, E, A, As , I, I R } are constant along x. The beam is transversally loaded by line load q(x, t)
(not shown in figure), with dimension of force per length. The primary kinematic variables are the
transverse deflection v(x, t) and the total cross-section rotation (x, t) = v (x, t) + (x, t), where
= V /(G As ) is the mean shear rotation. The kinetic and potential energies in terms of those
variables are
L
L

2
2
2
2
1
1
1
E
I
(v
)
+
G
A
(
v
qv
d x.
A v + I R d x, [v, ] =
T [v, ] = 2
s
2
2
0
(24.11)
where superposed dots denote time derivatives. The equations of motion (EOM) follow on forming
the Euler equations from the Lagrangian L = T :
L
= 0 G As ( v )+ Av = q,
v
L
= 0 E I +G As (v ) I R = 0. (24.12)
246
24.2
Section-averaged
shear rotation
Deformed
cross
y, v section
Normal to
deformed
longitudinal
axis
v(x)
THE TIMOSHENKO PLANE BEAM ELEMENT
slope
v' = dv/dx
A positive transverse shear

force V = GAs produces
a CCW rotation (+) of
the beam cross section
, E, G, A, As , I and IR
constant along beam
M
x
x, u
+V
Positive bending
moment and transverse
shear conventions
Figure 24.2. Plane beam member modeled as Ti-beam, illustrating notation followed in 24.2.1. Transverse load
q(x) not shown to reduce clutter. Infinitesimal deflections and deformations grossly exaggerated for visibility.
An expedient way to eliminate is to rewrite the coupled equations (24.12) in transform space:

As 2 G As p 2
v
q
G As p
=
,
(24.13)
E I p 2 G As I R s 2
0
G As p
q}
in which { p, s, v,
,
denote transforms of {d/d x, d/dt, v, , q}, respectively (Fourier in x and
Laplace in t). Eliminating and returning to the physical domain yields

AE I
2 AI R ....
IR
E I

q +
q.
(24.14)
v +
v =q
E I v + Av I R +
G As
G As
G As
G As
(This derivation does not preset I I R , as usually
done in textbooks.)
For the unforced case q = 0,

(24.14) has plane wave solutions v = B exp i (k0 x 0 t) . The propagation condition yields a
characteristic equation relating k0 and 0 . To render it dimensionless, introduce a reference phase
velocity c02 = E I /( AL 4 ) so that k0 = 0 /c0 = 2/0 , a dimensionless frequency = 0 L/c0
and a dimensionless wavenumber = k0 L.
As dimensionless measures of relative bending-to-shear rigidities and rotary inertia take
0 = 12E I /(G As L 2 ),
r R2 = I R /A,
0 = r R /L .
(24.15)
The resulting dimensionless characteristic equation is

1
4 2 ( 12
0 + 02 ) 2 2 +
1

12 0
02 4 = 0.
(24.16)
This is quadratic in 2 . Its solution yields two kinds of squared-frequencies, which will be denoted
by 2f and 2s because they are associated with flexural and shear modes, respectively. Their
expressions are listed below along with their small- (long wavelength) Taylor series:
P Q
2
1
1
f = 6
= 4 ( 12
0 + 02 ) 6 + ( 144
20 + 14 0 02 + 04 ) 8
0 02
(24.17)
1
( 1728
30 +
1
2 2
24 0 0
+ 12 0 04 + 06 ) 10 + . . . = A4 4 + A6 6 + A8 8 + . . .
247
(a)
40
Shear branches of
Timoshenko model
250
200 Cutoff
=0
frequencies
150
= 1/2
100
(b)
Shear branches of
Timoshenko model
35
Bernoulli-Euler
model
=0
Dimensionless speed /
300
30
25
= 1/2
20
= 1/2
15
Bernoulli-Euler
model
10
Flexural branches of
Timoshenko model
50
0
=0
5
15
10
20
25
=0
= 1/2
Flexural branches of
Timoshenko model
25
5
15
10
20
Figure 24.3. Spectral behavior of continuum Ti-beam model for a narrow b h rectangular cross section.
(a): dispersion curves () for = h/ = 1/4 and two Poissons ratios; Timoshenko flexural and shear
branches in red and blue, respectively; Bernoulli-Euler curve = 2 in black. (b) Wavespeed /.

P+ Q
12
12
1
1
=6
=
+
+ 2 2 4 + ( 12
0 + 02 ) 6 + . . . = B0 + B2 2 + . . .
2
2

0 0
0 0
0
0
(24.18)
1
0 ) and Q = P 2 13 4 0 02 . The dispersion relation 2f () defines
in which P = 1 + 2 (02 + 12
the flexural frequency branch (FFB) whereas 2s () defines the shear frequency branch (SFB). If
0 0 and 0 0, which reduces the Ti- model to the BE one, (24.16) collapses to 2 = 4 or
(in principal value) = 2 . This surviving branch pertains to flexural motions whereas the shear
branch disappears; more precisely, 2s () .
2s
It is easily shown that the radicand Q in the exact expressions is strictly positive for any {0 >
0, 0 > 0, 0}. Thus for any such triple, 2f and 2s are real, finite and distinct with 2f () <
2s (). Further { 2f , 2s } increase indefinitely as . Following the dispersion-diagram
nomenclature introduced in Figure 23.2, the value s at = 0 is called the cutoff frequency.
To see what branches look like, consider a beam of narrow rectangular cross section of width b and
height h, fabricated of isotropic material with Poissons ratio . Accordingly E/G = 2(1 + )
and As /A 5/6. [Actually a more refined As /A ratio would be 10(1 + )/(12 + 11), but that
makes little difference in the results.] We have A = bh, I = I R = bh 3 /12, r R2 = I R /A = h 2 /12,
1 2
h /L 2 and 0 = 12E I /(G As L 2 ) = 12(1 + )h 2 /(5L 2 ). Since 0 /12 =
02 = r R2 /L 2 = 12
2
12(1 + ) 0 /5, the first-order effect of shear on 2f , as measured by the 6 term in (24.17), is 2.4
to 3.6 times that from rotary inertia, depending on . Replacing into (24.17) and (24.18) yields

2
2
2
2
60 + 2 (17 + 12)2 240 4 (1 + )4
(17
+
12)
60
+
f
=
2(1 + )4
2s
4
(24.19)
1
1
60 (17 + 12)2 6 + 3600 (349 + 468 + 144 2 )4 8 + . . .
= 60 + (17 + 12)2 2 (1 + )4 4 + . . .
(1 + )4
in which = h/L. Dispersion curves () for = h/L = / and = {0, 12 } are plotted
248
24.2
Deformed
cross
section
1
v'1 =[dv/dx]1
y, v
1
v'2 =[dv/dx]2
v1
v2
v(x)
x, u
Figure 24.4. Two-node element for Timoshenko plane beam, illustrating kinematics.
in Figure 24.3(a). Phase velocities / are shown in Figure 24.3(b). The figure also shows the
flexural branch of the BE model. The phase velocities of the Timoshenko model tend to finite
values in the shortwave, high-frequency limit , which is physically correct. The BE model
is physically wrong in that limit because it predicts an infinite propagation speed.
24.2.2. Ti-Beam Element
The shear-flexible plane beam member of Figure 24.2 is discretized by two-node elements. An
individual element of this type is shown in Figure (24.4), which illustrates its kinematics. The
element has four nodal freedoms arranged as
ue = [ v1 1 v2 2 ]T
(24.20)
Here 1 = v1 + 1 and 2 = v2 + 2 are the total cross section rotations evaluated at the end nodes.
The dimensionless properties (24.15) that characterize relative shear rigidity and rotary inertia are
redefined using the element length:
= 12E I /(G As 2 ),
r R2 = I R /A,
= r R /.
(24.21)
If the beam member is divided into Ne elements of equal length, = L/Ne whence = 0 Ne2
and = 0 Ne . Thus even if 0 and 0 are small with respect to one, they can grow without
bound as the mesh is refined. For example if 0 = 1/4 and 02 = 1/100, which are typical values
for a moderately thick beam, and we take Ne = 32, then 250 and 2 10. Those are no
longer small numbers, a fact that will impact performance as Ne increases. The stiffness matrix to
be paired with the mass template is taken to be that of the equilibrium element:
12
6
12
6
EI
6 2 (4 + ) 6 2 (2 )
(24.22)
Ke = 3
.
6
12
6
(1 + ) 12
6 2 (2 ) 6 2 (4 + )
This is known to be optimal in static analysis for a prismatic beam member. It will be kept fixed in
the ensuing derivations. It reduces to the stiffness matrix (24.3) of the BE model if = 0.
24.2.3. The Ti-Beam Mass Template
FEM derivations usually split the 4 4 mass matrix of this element into Me = Mev + Me , where
Mev and Me come from the translational inertia and rotary inertia terms, respectively, of the kinetic
249
energy functional T [v, ] of (24.11). The most general mass template would result from applying
a entry-weighted parametrization of those two matrices. This would require a set of 20 parameters
(10 in each matrix), reducible to 9 through 11 on account of invariance and conservation conditions.
Attacking the problem this way, however, leads to unwieldy algebraic equations even with the help of
a computer algebra system, while concealing the underlying physics. A divide and conquer approach
works better. This is briefly outlined next and covered in more detail in the next subsections.
(I) Express Me as the one-parameter matrix-weighted form Me = (1 0 ) MeF + 0 MeD . Here
MeF is full and includes the CMM as instance, whereas MeD is 2 2 block diagonal and includes
the DLMM as instance. This is plainly a generalization of the LC linear combination (21.2).
(II) Decompose the foregoing mass components as MeF = MeF T + M F R and MeD = MeDT + MeD R ,
where T and R subscripts identify their source in the kinetic energy functional: T if coming from
the translational inertia term 12 A v 2 and R from the rotary inertia term 12 I R 2 .
(III) Both components of MeF are expressed as parametrized spectral forms, whereas those of MeD
are expressed as entry-weighted. The main reasons for choosing spectral forms for the full matrix
are reduction of parameters and physical transparency. No such concerns apply to MeD .
The analysis follows a bottom up sequence, in order (III)-(II)-(I). This has the advantage that if
a satisfactory custom mass matrix for a target application emerges during (III), stages (II) and (I)
need not be carried out, and that matrix directly used by setting the remaining parameters to zero.
24.2.4. Ti-Beam Full Mass Parametrization
As noted above, one starts with full-matrix spectral forms. Let denote the natural iso-P coordinate
that varies from 1 at node 1 to +1 at node 2. Two element transverse displacement expansions in
generalized coordinates are introduced:
vT ( ) = L 1 ( ) cT 1 + L 2 ( ) cT 2 + L 3 ( ) cT 3 + L 4 ( ) cT 4 = LT cT ,
L 4 ( ) c R4 = L R c R ,
v R ( ) = L 1 ( ) c R1 + L 2 ( ) c R2 + L 3 ( ) c R3 +
L 1 ( ) = 1, L 2 ( ) = , L 3 ( ) = 12 (3 2 1), L 4 ( ) = 12 (5 3 3 ),

L 4 ( ) = 1 5 3 (5 + 10) = L 4 ( ) (1 + 5).
(24.23)
The vT and v R expansions are used for the translational and rotational parts of the kinetic energy,
respectively. The interpolation function set {L i } used for vT is formed by the first four Legendre
L 4 to
polynomials over = [1, 1]. The set used for v R is the same except that L 4 is adjusted to
produce a diagonal rotational mass matrix. All amplitudes cT i and c Ri have dimension of length.
Unlike the usual Hermite cubic shape functions, the polynomials in (24.23) have a direct physical
interpretation. L 1 : translational rigid mode; L 2 : rotational rigid mode; L 3 : pure-bending mode
L 4 : bending-with-shear mode antisymmetric about = 0.
symmetric about = 0; L 4 and
With the abbreviation (.) d(.)/d x = (2/)d(.)/d , the associated cross section rotations are
compactly expressed as
T = vT + T = LT cT + T ,
R = v R + R = LR c R + R ,
(24.24)
in which the mean shear distortions are constant over the element:
T =
2
10
vT =
cT 4 ,
12

R =
2410
2
10
vR =
c R4 .
12

(24.25)
24.2
The kinetic energy of the element in generalized coordinates is

T =
e
1
2
A v T2

+ I R R2 d x =
4

A v T2 + I R R2 d = 12 c TT DT c T + 12 c TR D R c R .
(24.26)
Both generalized mass matrices turn out to be diagonal as expected:

DT = m e diag [ 1
1
3
1
5
1
7
],
D R = 4m e 2 diag [ 0 1 3 5 ] ,
in which as usual m e = A . To convert DT and D R to physical coordinates (24.20), vT , v R , T and

R are evaluated at the nodes by setting = 1. This establishes the transformations ue = GT cT
1
and ue = G R c R . Inverting: cT = HT ue and c R = H R ue with HT = G1
T and H R = G R . A
symbolic calculation yields for HT :
30(1 + )
5(1 + ) 30(1 + ) 5(1 + )
1
3
36 + 30
3
36 30
HT =
(24.27)
.
0
5(1 + )
0
5(1 + )
60(1 + )
6
3
6
3
Matrix H R differs only in the second row:
30(1 + ) 5(1 + )
1
15
30
HR =
0
5(1 + )
60(1 + )
6
3
30(1 + )
30
0
6
5(1 + )
15
.
5(1 + )
3
(24.28)
The difference comes from adjusting L 4 to

L 4 in (24.23). To map this into a spectral template,
inject six free parameters in the generalized masses while moving 4 2 inside D R :
DT = m e diag [ 1
1
1
1
3 T1 5 T2 7 T3
],
D R = m e diag [ 0 R1 3 R2 5 R3 ] .
(24.29)
The transformation matrices (24.27) and (24.28) can be reused without change to produce MeF =
HTT DT HT + HTR D R H R . If T 1 = T 2 = T 3 = 1 and R1 = R2 = R3 = 4 2 one
obtains the well known consistent mass matrix (CMM) as a valuable check. The configuration
(24.29) already accounts for linear momentum conservation, which is why the upper diagonal
entries are not parametrized. Enforcing also angular momentum conservation requires T 1 = 1
and R1 = 4 2 , whence the template is reduced to four parameters:
1 0
0
0
0
0
0
0
2
0 13
0
0
0
0
e T 0 4
+
m
H
H
MeF = m e HTT
H R . (24.30)
T
R
0 0 1 T 2
0
0
0
0
3 R2
5
1
0
0
0
5 R3
0 0
0
7 T3
Since both HT and H R are nonsingular, choosing all parameters in (24.30) to be nonnegative
guarantees that MeF is nonnegative. This property eliminates lengthy a posteriori checks.
Setting T 2 = T 3 = R2 = R3 = 0 and = 0 yields the correct mass matrix for a rigid beam,
including rotary inertia. This simple result highlights the physical transparency of spectral forms.
2411
24.2.5. Ti-Beam Block-Diagonal Mass Parametrization

Template (24.30) has a flaw: it does not include the DLMM. To remedy the omission, a block
diagonal form, with four free parameters: T 1 , T 2 , R1 , and R2 is separately constructed:
1
T 1
0
0
0
0
0 R1
2
2
0
0
0
0
T 2 2
e R1 R2
MeD = m e T 1
(24.31)
+m
.
1

0
0
0
0
0
T1
R1
2
0
0 T 1 T 2 2
0
0 R1 R2 2
Four parameters can be merged into two by adding the foregoing matrices:
1
1
0
0
2
0
0
2 2
MeD = m e 1
.
1
1
0
0
2
0
0
1 2 2
(24.32)
in which 1 = T 1 + R1 and 2 = T 2 + R2 . Sometimes it is convenient to use the split form

(24.31), for example in lattices with varying beam properties or lengths, a topic not considered there.
Otherwise (24.32) suffices. If 1 = 0, MeD is diagonal. However for computational purposes a block
diagonal form is just as good and provides additional customization power. Terms in the (1,1) and
(3,3) positions must be as shown to satisfy linear momentum conservation. If angular momentum
conservation is imposed a priori it is necessary to set 2 = 12 2 , and only one parameter: 1 ,
remains.
The general template is obtained as a linear combination of MeF and MeD :
Me = (1 0 )MeF + 0 MeD
(24.33)
In summary, there is a total of 7 parameters to play with: 4 in MeF , 2 in MeD , plus 0 . This is
less that the 9-to-11 count that would result from a full entry-weighted parametrization, so not all
possible mass matrices for this element are included by (24.33).
24.2.6. Ti-Beam Fourier Analysis
An infinite lattice of identical Ti-beam elements of length is set up in th usual manner. As in
24.1, plane waves of wavenumber k and frequency propagating over the lattice are represented
by

(24.34)
v(x, t) = Bv exp i(kx t , (x, t) = B exp i(kx t .
At each typical lattice node j there are two freedoms: v j and j . Two patch equations are extracted,
and converted to dimensionless form on defining = k and = c/, in which c = E I /( A4 )
is a reference phase velocity. (These should not be confused with c0 .) The condition for wave
propagation gives the characteristic matrix equation

Cvv Cv
det
(24.35)
= Cvv C Cv C v = 0,
C v C
in which the coefficients are complicated functions computed by Mathematica and omitted for
brevity. Solving the equation provides two equations: a2 and 2o , where a and o denote acoustic
2412
24.2
and optical branch, respectively. These are expanded in powers of for matching to the continuum.
For the full mass matrix one obtains
a2 = 4 + C6 6 + C8 8 + C10 10 + . . . ,
2o = D0 + D2 2 + . . .
(24.36)
Coefficients of terms up to 12 were computed by Mathematica. Those relevant for parameter

selection are
C6 = /12 2 ,

C8 = 2 15 R2 T 2 + 5(1 + ) + 60(1 + 3) 2 + 720 4 /720,

C10 = 44 + 35T 2 3T 3 282 + 525 R2 (1 + ) 105 R3 (1 + )+
1575 R2 (1 + ) (3T 3 35T 2 (4 + 3) + 35(17 + 5(3 + )))+
(2940 + 12600 R2 (1 + ) + 420(2T 2 (1 + ) 5(7 + 6(2 + )))) 2

25200(2 + (7 + 6)) 4 302400(1 + ) 6 / 302400(1 + ) ,

D0 = 25200(1 + )/ 7 + 105 R2 + 3T 3 + 21002 2 ,

D2 = 2100(1 + )(56 35T 2 + 3T 3 63 + 3T 3 + 105 R3 (1 + )
525 R2 (1 + )2 35T 2 (2 + )) + 2100(1 + )(3360 + 63002 +

2
21003 ) 2 + 529200002 (1 + ) 4 / 7 + 105 R2 + 3T 3 + 21002 2 .
(24.37)
For the block-diagonal template (24.32):
a2 = 4 + F6 6 + F8 8 + F10 10 + . . . ,
2o = G 0 + G 2 2 + . . .
(24.38)
28802 5 + 3602 1 5 + 52

F6 = 242 , F8 =
720
6
242 + 2
, G2 =
.
G0 =
2 (1 + )
22 (1 + )
(24.39)
in which
The expansions for the 7-parameter template (24.33) are considerably more involved than the above
ones, and are omitted for brevity.
24.2.7. Ti-Beam Selected Template Instances
Seven useful instances of the foregoing templates are identified and described in Table 24.1. Table 24.2 lists the template signatures that produce those instances. These tables include two well
known mass matrices (CMM and DLMM) re-expressed in the template context, and five new ones.
The latter were primarily obtained by matching series such as (24.37) and (24.38) to the continuum
ones (24.17) and (24.18), up to a certain number of terms as indicated in Table 24.2.
For the spectral template it is possible to match the flexure branch up to O( 10 ). Trying to match
O( 12 ) leads to complex solutions. For the diagonal template the choice is more restrictive. It is
only possible to match flexure up to O( 6 ), which leads to the instance called FLMM. Trying to
go further gives imaginary solutions. For the 7-parameter template (24.33) it is again possible to
2413
Table 1. Ti-Beam Selected Template Instances

Instance
name
Description
Comments
CMM
Consistent mass matrix derived in

18.2. Matches FFB up to O( 6 ).
A popular choice. Fairly inaccurate, however,

as beam gets thicker. Grossly overestimates
intermediate frequencies.
FBMS
FFB matched to O( 10 ) with spectral

(Legendre) template (24.30).
Converges faster than CMM. Performance degrades as beam gets thicker, however, and
becomes inferior to CDLA.
SBM0
SFB matched to O( 0 ) while flexure

fitted to O( 10 )
Custom application: to roughly match SBF and

cutoff frequency as mesh is refined. Warning:
indefinite for certain ranges of and : use
with caution.
SBM2
SBF matched to O( 2 ) while flexure

fitted to O( 8 )
Custom application: to finely match SFB and

cutoff frequency as mesh is refined. Warning:
indefinite for wide ranges of and : use with
extreme caution.
FLMM
Diagonally lumped mass matrix with

rotational mass picked to match FFB
to O( 6 ).
Obvious choice for explicit dynamics. Accuracy degrades significantly, however, as beam
gets thicker. Underestimates frequencies. Becomes singular in the BE limit.
CDLA
Average of CMM and FLMM. Matches

FFB to O( 8 ).
Robust all-around choice. Less accurate than

FBMS and FBMG for thin beams, but becomes
top performer as beam gets thicker. Easily
constructed if CMM and FLMM available in
code.
FBMG
FFB matched to O( 10 ) with 7parameter template (24.33).
Known to be the globally optimal positivedefinite choice for matching flexure in the BE
limit. Accuracy, however, is only marginally
better than FBMS. As in the case of the latter,
performance degrades as beam gets thicker.
match up to O( 10 ) but no further. The instance that exhibits least truncation error while retaining
positivity is FBMG. This is globally optimal for the BE limit = = 0, but the results are only
slightly better for the reasons discussed below. Matching both flexure and shear branches leads to
instances SBM0 and SBM2, which have the disadvantages noted in Table 24.1.
The exact dispersion curves of these instances are shown in Figure 24.5 for = 48/125 and
2 = 1/75, which pertains to a thick beam. On examining Figure ?(c) it is obvious that trying to
match the shear branch is quite difficult; the fit only works well over a tiny range near = 0.
24.2.8. Ti-Beam Vibration Analysis Example
The vibration analysis performance of the seven Ti-beam template instances listed in Tables 24.1
2414
24.2
Table 2. Signatures of Selected Ti-Beam Template Instances
Instance
name
Templ.
form
CMM
(24.30)
FBMS
(24.30)
26/3 4 2 +/3
SMB0
(24.30)
SMB2
(24.30)
FLMM
(24.32)
CDLA
(24.33)
4 2
FBMG
(24.33)
c3
c4
c5
T 2 T 3
Template signature
R2
R3
1
Fit to continuum freqs.

2f (flexural) 2s (shear)
4 2
up to 6
none
c1
up to 10
none
7/3 4 2 +/3 20 2
up to 10
up to 0
7/3
up to 8
up to 2
up to 6
none
4 2
c2
20 2
0
1 2

2
4 2
1 2

2
1/2
up to 8
none
c6
1/12
1 2

2
c7
up to 10
none
c1 = 253 + 120 2 + 2 (45 300 2 ) + 3(7 20 2 + 1200 4 ) / 15(1 + ) ,
c2 = 19 + 102 (90 2 1) 30(1 26 2 + 120 4 ) / 75(1 + )2 ,
c3 = (9 + 105)/10, c4 = (61 105 483)/18, c5 = ( 105 1)/30,
c6 = 48 + 7272 + 8403 + 22128 2 + 19848 2 100802 2 113040 4
+120960 4 + 5 105 (48 + 872 + 403 ) 24(6 + 21 + 202 ) 2

+720(3 + 8) 4 / 60(21 + 105)(1 + ) , c7 = (3 5 5/21)/8.
and 24.2 is evaluated on a simply supported (SS) prismatic plane beam. The beam has length L
and is divided into Ne identical elements. The cross section is rectangular with width b and height
h. The material is isotropic with Poissons ratio = 0. Three different height-to-span ratios h/L
that characterize a thin, moderately thick and thick beam, respectively, are considered. Results for
the three configurations are collected in Figures 24.6, 24.7 and 24.8, respectively, for the first three
vibration frequencies. All calculations are rendered dimensionless using appropriate scaling.
The accuracy of the computed frequencies is depicted using log-log plots of dimensionless natural
frequency error versus Ne . The error is displayed as d = log10 (| comp exact |, which gives at
a glance the number of correct digits d, versus log2 Ne for Ne = 1 through 32. If the LF error is
approximately controlled by a truncation term of the form m , the log-log plot should be roughly
a straight line of slope m, inasmuch as = k = k L/Ne . (Note that the accuracy curves for
CMM and FLMM are virtually on top of each other for Ne 4, although errors have opposite
signs; that is why their average CDLA does much better.)
The results for the Bernoulli-Euler (BE) model, shown in Figure 24.6, agree perfectly with the
truncation error in the 2f branch as listed in Table 24.1. For example, the top performers FBMG
and FBMS gain digits twice as fast as CMM, DLMM and SBM2, since the formers match 2f
to O( 10 ) whereas the latter do that only to O( 6 ). Instances CDLA and SMB0, which agree
2415

60
cutoff
50
60
s continuum
40
o CMM
30
f continuum
a CMM
10
a 2DLMM3
5
4
cutoff
a SBM0
5
1
4
2
3
0
60
50
cutoff
40
s continuum
50
s continuum
f continuum
a SBM2
o SBM0
10
60
40
o CDLA
30
f continuum
o FBMG
20
o FBMS
0
a FBMS
CDLA
5
4
2 a
3
f continuum
30
20
10
o SBM2
30
20
s continuum
40
o DLMM
20
cutoff
50
a FBMG
10
Figure 24.5. DDD of selected Ti-beam mass template instances for = 48/125 and 2 = 1/75.
through O( 8 ), come in between. The highly complicated FBMG is only slightly better than the
much simpler FBMS. The case for their high accuracy should be emphasized. For example, four
FBMS elements give 1 to six figures: 9.86960281 . . . versus 2 = 9.86960440 . . ., whereas
CMM gives less than three: 9.87216716 . . .. The accuracy ceiling of about 11 digits for FBMS
and FBMG observable for Ne > 16 is due to the eigensolver working in double precision ( 16
digits). Rerunning with higher (quad) floating point precision, the plots continues marching up as
straight lines before leveling off at approximately 25 digits.
On passing to the Timoshenko model, the well ordered BE world of Figure 24.6 unravels. The
culprits are and . These figure prominently in the branch series and grow without bound as Ne
increases, as discussed in 24.2.2. Figure 24.7 collects results for a moderately thick beam with
h/L = 1/8, which corresponds to 0 = 3/80 and 02 = 1/768. The BE top performers, FBMS
and FBMG, gradually slow down and are caught by CDLA by Ne = 32. All other instances trail,
with the standard ones: CMM and FLMM, becoming the worst performers. Note that for Ne = 32,
CMM and FLMM provide only 1 digit of accuracy in 3 , although there are 32/1.5 21 elements
per wavelength.
Figure 24.8 collects results for a thick beam with h/L = 2/5, corresponding to 0 = 24/625 and
02 = 1/75. The trends of Figure 24.7 are exacerbated, with FBMS and FBMG running out of
steam by Ne = 4 and CDLA clearly emerging as best for Ne 8. Again CMM and FLMM trail
badly.
The reason for the performance degradation of FBMS and FBMG as the Ti-beam gets thicker is
unclear as of this writing. Eigensolver accuracy is not responsible since rerunning the cases of
Figures 24.7 and 24.8 in quad precision did not change the plots. A numerical study of the 2f
truncation error shows that FBMS and FBMG fit the continuum branch better than CDLA even for
2416
24.2
12
10
FBMS
FBMG
CDLA
8
SBM0
6
SBM2
2
CMM
FLMM
1
10
FBMG
CDLA
SBM0
4
SBM2
FLMM
CMM
32
16
12
Digits of accuracy in third vibration frequency
10
Digits of accuracy in second vibration frequency
12
Digits of accuracy in first vibration frequency
16
FBMG
6
SBM0
SBM2
FLMM
CMM
32
Number of elements along span
FBMS
CDLA
32
16
Figure 24.6. Accuracy of first 3 natural vibration frequencies of SS prismatic beam using mass matrices of
instances listed in Tables 24.1 and 24.2. Bernoulli-Euler model with 0 = 0 = 0. Exact (12-decimal) frequencies
1 = 2 = 9.869604401089, 2 = 4 2 = 39.478417604357 and 3 = 9 2 = 88.826439609804. Cutoff
frequency +.
10
10
FBMG
FBMS
CDLA
SBM0
4
SBM2
FLMM
CMM
1
12
16
10
FBMS
FBMG
CDLA
SBM2
CMM
32
12
12
SBM0
FLMM
16
CDLA
FBMS
FBMG
SBM2
SBM0
FLMM
CMM
32
32
16
Figure 24.7. Accuracy of first 3 natural vibration frequencies of SS prismatic beam using mass matrices of instances
listed in Tables 24.1 and 24.2. Timoshenko model with 0 = 3/80 = 0.0375 and 02 = 1/768 = 0.00130, pertaining
to a rectangular x-section with h/L = 1/8 and = 0. Exact (12-decimal) frequencies 1 = 9.662562122511,
2 = 36.507937703548 and 3 = 75.894968024537. Cutoff frequency cut = 12/(0 02 ) = 495.741868314549.
10
10
CDLA
6
FBMS
FBMG
SBM0
SBM2
FLMM
CMM
1
12
16
32
10
CDLA
4
FBMS
FBMG
SBM0
SBM2
FLMM
CMM
1
12
12
16
32
CDLA
FBMS
FBMG
SBM0
SBM2
FLMM
CMM
16
32
Figure 24.8. Accuracy of first 3 natural vibration frequencies of SS prismatic beam using mass matrices of instances
listed in Tables 24.1 and 24.2. Timoshenko model with 0 = 24/625 = 0.384 and 02 = 1/75 = 0.0133, pertaining
to a rectangular x-section with h/L = 2/5 and = 0. Exact (12-decimal) frequencies 1 = 8.287891683498,
2 = 24.837128591729 and 3 = 43.182948411234. Cutoff frequency cut = 12/(0 02 ) = 48.412291827593.
2417
very thick beams. Possible contamination of vibration mode shapes with the shear branch was not
investigated.
2418
25
Mass Templates
for Plane Stress
Elements
251
Chapter 25: MASS TEMPLATES FOR PLANE STRESS ELEMENTS
TABLE OF CONTENTS
Page
25.1
25.2
25.3
25.4
Longitudinal Wave Propagation in a Continuum Thin Plate . . . .

Trig3 LCD Template . . . . . . . . . . . . . . . . . .
Trig3 Template Fourier Analysis . . . . . . . . . . . . . .
Can Plane Waves Actually Propagate Over a 2D FEM Lattice?
. .
252
253
253
254
256
25.2
TRIG3 LCD TEMPLATE
This chapter presents an example of the construction and LFF custiomization of a mass template
for the simplest two-dimensional element, namely the three-node plane stress (membrane) linear
triangle, identified as Trig3 in the sequel. The example is intended to illustrate three additional
features that must be considered in multiple dimensions; in particular mesh directionality, multiple
plane wave types, and (for an isotropic material) Poissons ratio. It is the only multidimensional
template presented in this paper, since the subject has barely lifted off the ground. The results
are taken from the a recent thesis [320] in which intermdiate calculation details (omitted here for
brevity) may be found. The thesis also works out a more complicated version of this element type
that includes corner drilling freedoms. That development is too complex to be briefly covered here.
25.1. Longitudinal Wave Propagation in a Continuum Thin Plate
Wave propagation in continuum models of elastic solids is a classical subject of elastodynamics
that is well covered in several textbooks and monographs; e.g., [833,835,315,836,837]. Here
we consider longitudinal waves propagating in an isotropic, homogeneous, elastic flat thin plate.
Thickness effects are ignored. The continuum model has elastic modulus E, Poissons ratio ,
shear modulus G = E/(2(1 + )), mass density , and uniform thickness h. Two plane wave types
are possible in that model:
Pressure waves or P-waves, in which material points harmonically oscillate in the direction of
the wave propagation. Also known as longitudinal, dilatational and push waves. A P-wave
with E = E/(1 2 ).
propagates at speed c P0 , in which c2P0 = E/,
Shear waves or S-waves, in which material points harmonically oscillate in the direction normal to
wave propagation. Also known as transverse, rotational and shake waves. An S-wave propagates
at speed c S0 , in which c2S0 = G/. The dispersion equations are
2P0 = c2P0 k 2 ,
2S0 = c2S0 k 2 ,
(25.1)
in which k is the wavenumber while P0 and S0 denote the characteristic frequencies for the P- and
S-wave, respectively. In dimensionless variables = k a, 2P0 = 2P a 2 /c2P0 , and 2S0 = 2S a 2 /c2S0 ,
these reduce to
2P0 = 2 ,
2S0 = 2 .
(25.2)
Note that both plane wave types are nondispersive (i.e., speed is independent of frequency) and can
propagate along any in-plane direction, as befits isotropy. That is not the case in a FEM lattice of
Trig3 elements, as studied next.
25.2. Trig3 LCD Template
The Trig3 mass template is assumed to be of the LCD form (21.2): it linearly combines the CMM
and DLMM given in (19.2) and (19.3), respectively, through a weight parameter:
2 0 1 0 1 0
0 2 0 1 0 1
e
e
m
(1
)
1 0 2 0 1 0 m
I6 (25.3)
+
Me = (1 )MCe M M + MeDL L M =
0 1 0 2 0 1
12
3
1 0 1 0 2 0
0 1 0 1 0 2
253
y
ym
(a)
Normal to wave
propagation direction
Mesh
symmetry
axes
(b)
ym
Wave propagation
direction
xm
xm
4
Square cells of
side dimension a
(c)
3
1
6
7
six-triangle,
seven-node
patch
Figure 25.1. Mesh for Fourier analysis of Trig3 LCD mass template: (a) infinite regular lattice of
square cells; (b) rotated lattice to account for wave propagation directionality, forming angle with
respect to wave propagation direction (x-axis); (c) six-triangle, seven-node patch.
This is paired with the well known Trig3 stiffness Ke , which is kept fixed. To study the performance
of (25.3) an infinite 2D regular lattice such as pictured in Figure 25.1(a) was chosen. The triangles
form square cells of side a. To account for the effect of wave directionality in the FEM lattice,
the direction of propagation of the P- and S-waves is assumed to be always along the x axis. The
lattice is rotated by an angle with respect to x, as illustrated in Figure 25.1(b). Thus the mesh
symmetry axes {xm , ym } for an angle with the propagation axis {x, y}. The advantage of this
choice, as opposed to rotating the propagation direction, is that the wave functions are kept simple.
The kinematic expressions for the P- and S-waves are
u x = Bx exp(k x P t), u y = 0,
S-wave: ux = 0, u y = B y exp(k x S t).
P-wave:
(25.4)
in which Bx and B y are nonzero amplitudes. Note that the wavenumber k is kept the same in both
expressions. All calculations summarized below were carried out symbolically with Mathematica.
25.3. Trig3 Template Fourier Analysis
From the 2D infinite lattice pictured in Figure 25.1(b), extract a (repeating) six-triangle, seven-node
patch highlighted in (c) of that Figure. Assemble the 14 14 patch mass matrix M p from (25.3),
p and the
and the 14 14 patch stiffness matrix K p . From these extract the 2 14 mass matrix M
p for the patch center node. The dynamic force residual equations at that
2 14 stiffness matrix K
node are
p ) uB ,
p 2 M
(25.5)
r = (K
in which r = [ r x r y ]T is the residual vector measuring force equilibrium in the x and y directions,
and u p is a 14-vector containing wave displacements of the seven patch nodes. Two u P : u p P and
254
averaged optimal
25.3
TRIG3 TEMPLATE FOURIER ANALYSIS
1.0
0.8
best for P-wave
0.6
0.4
best for S-wave
0.2
0.0
0.0
0.1
0.2
0.3
Poisson's ratio
0.4
0.5
Figure 25.2. Best -averaged values of free parameter as function of

Poissons ratio , for two plane wave types (P and S).
u pS are constructed by evaluating the P- and S- plane waves, respectively at the patch nodes, and
inserted in (25.5). For the P-wave, u y = 0 and setting r x = 0 we solve for 2 , renamed 2P .
For the S-wave, u x = 0 and setting r y = 0 we solve for 2 , renamed 2S . Thse are then Taylor
series expanded at k = 0 in powers of k, up to k 4 . Passing to dimensionless variables: = k a,
2P = 2P a 2 /c2P0 , and 2S = 2S a 2 /c2S0 , the result for the P-wave is
2P = 2
2C1 + 2(1 3) cos(4) + C2 sin(2) 3(1 + ) sin(6) 4

+ ...
384
(25.6)
in which C1 = 32 + 3 17 and C2 = 32 + 9 23. For the S-wave:

2S = 2
2D1 + 2(5 + ) cos(4) + D2 sin(2) + 3(1 + ) sin(6) 4

+ ...
192(1 )
(25.7)
in which D1 = 7 +16(1)13 and D2 = 7 +16(1)25. Comparing to the continuum

expressions (25.2), which may be rewritten we observe a O( 4 ) error than depends on and . To
eliminate the effect of we integrate that error term over [/2, pi/2] and divide by to get
an average error. Setting that to zero and solving for , we get two averaged-optimal values for
the free parameter, one for each wave type, denoted by P and S :
45 + 40 3 (45 24 3)
45 16 3 + (45 + 32 3)
P () =
, S () =
, (25.8)
48
24(1 )
These are plotted as functions of Poissons ratio in Figure 25.2. As can be observed, the dependence
on is failrly mild. For the S wave it is approximately 0.5 whereas for the S-wave it varies
from 0.72 to 1.00. Which one to pick? Assuming the given is uniform, select either P or S
according to the wave type expected to dominate the solution. This may require some interaction.
If S-waves dominate = 1/2 is recommended for any Poissons ratio.
255
25.4. Can Plane Waves Actually Propagate Over a 2D FEM Lattice?

The answer to the title question is: only for selected direction angles. To check that for the P-wave,
pick the frequency 2P obtained by solving r x = 0, and insert that solution into r y . It will be
observed that for [90 , 90 ], r y = 0 is only satisfied exactly at = 90 , = 45 , and
= 0 . For those angles the assumption of plane wave propagation is exact. Othwerwise r y = 0
means that the plane wave is distorted by the FEM discretization even if the lattice is regular. A
similar result is obtained for the S-wave on replacing 2S into r x . In practice the distortion is not
of great concern since the goal of this analysis is to find reasonable values for the free parameter,
rather than finding a solution to the dynamic problem. Those are two distict objectives.
256

Matrix Finite Element Methods in Dynamics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Matrix Finite Element Methods in Dynamics

Uploaded by

Copyright:

Available Formats

D

Appendix D: MASS MATRIX DIAGONALIZATION

Repeat for all coordinate directions.

Appendix D: MASS MATRIX DIAGONALIZATION

One-Dimensional Lobatto Integration Rules

Common names: Trapezoidal rule and Simpsons rule for p =

One-Dimensional Newton-Cotes Integration Rules

Same as 2-point Lobatto; see Table D.1

D.2. Lobatto Mass Lumping

LOBATTO MASS LUMPING

Appendix D: MASS MATRIX DIAGONALIZATION

CONGRUENTIAL MASS TRANSFORMATION

If H is nonsingular, the inverse mapping is M S = GT MT G, in which G = H1 . We will say

Appendix H: A SHORT HISTORY OF MASS MATRICES

CONSISTENT MASS MATRICES APPEAR

The three-century astronomical heritage of Newtonian Mechanics. For unperturbed orbit

Appendix H: A SHORT HISTORY OF MASS MATRICES

FREQUENCY DEPENDENT MATRICES

H.4. Selective Mass Scaling

These are counteracted, however, by two disadvantages:

H.6. Frequency Dependent Matrices

Appendix H: A SHORT HISTORY OF MASS MATRICES

stiffness by partial inversion. The procedure is described in detail in [572]. It is restricted to 1D

CONCLUSIONS AND FUTURE WORK

Mesh directionality effects that require angular averaging

Additional dependence on elastic material properties

The free parameter explosion as matrices get larger

Appendix H: A SHORT HISTORY OF MASS MATRICES

Material property dependence. Optimal free parameters become dependent on additional

Appendix I: IMPLEMENTATION OF BAR3 TEMPLATE IN MATHEMATICA

Appendix I: IMPLEMENTATION OF BAR3 TEMPLATE IN MATHEMATICA

Table I.1. Bar3 Template Signature List Specification

# of Signature format, identified as

{ "GEN",{ },{ 1 , 2 , 3 , 4 } } 21 +2 +23 +44 = 0

Constant Optical (23.47,23.49)

The arguments are:

Signature of the target form.

ELEMENT LEVEL MODULES

The function returns

Example 1. Let gmpars={ 1 ,2 ,3 ,4 } and tsign={ "LUM",{ },{ L1 , L2 } }. The module

Appendix I: IMPLEMENTATION OF BAR3 TEMPLATE IN MATHEMATICA

Figure I.3. Bar3 element mass and stiffness template modules.

The arguments are:

Template signature. See Table I.1 for configuration details.

Logical flag. If True, process in floating point. If False, process symbolically.

As function values the modules return

3 3 element mass matrix

3 3 element stiffness matrix

ASSEMBLY LEVEL MODULES

Figure I.4. Bar3 master mass and stiffness assembler modules.

Figure I.5. Bar3 mass and stiffness patch extraction modules.

I.3. Assembly Level Modules

Appendix I: IMPLEMENTATION OF BAR3 TEMPLATE IN MATHEMATICA

The arguments are:

Number of elements in lattice. The number of freedoms is numdof=2*numele+1

Element length. Total member length will be Le*numele

Template definition argument. Two possibilities:

Logical flag; see I.2.1

As function values the modules return

Master mass matrix of order ndofndof

Master stiffness matrix of order ndofndof

I.3.2. Lattice Patch Modules

Patch mass equations as a coefficient matrix of order 2 5; see (?)(?)

Patch stiffness equations as a coefficient matrix of order 2 5; see (?)(?)

Dimensionless space coordinate = x/ [1, 1] over patch.

Dimensionless characteristic frequency /c0

Dimensionless time = tc0 /