686 W. Zheng et al. / Appl. Math. Comput.

153 (2004) 685695

There are three electrons out of the nucleus of the lithium atom; and each of
them has three freedoms (without considering its spin). So the system is de-
scribed by a nine-dimensional Schr odinger equation (for xed-nucleus prob-
lem). It is very dicult to solve it by variational methods, nite dierence
methods and other methods directly. Usual approaches in quantum mechanics
are mean-eld methods [1]. For them, every electron is considered indepen-
dently to be in a central electric eld formed by the nucleus and other electrons.
The original problem is transferred into a system of nonlinear partial dier-
ential equations of low-dimension, then solve it iteratively. So the physical
model itself is simplied approximately.
The main works of the nite element method (FEM) applied to atomic and
molecular problems appeared in 1970s, and in one- or two-dimensional cases
[24]. In 1975, Askar calculated the energies of hydrogen atom in the ground
state and the rst excited state [2]. Then Nordholm and Bacsky applied FEM to
more general bound state problems [4]. Fridman and Rosenfeld analyzed two
model problems of two-dimensional Schr odinger equations with FEM in 1977
[3]. All these works showed that FEM was simple and ecient for one- and two-
dimensional atomic and molecular problems. The rst work of FEM applied to
three-dimensional case was due to Levin and Shertzer [5]. They calculated the
helium in the ground state in 1985. The six-dimensional Schr odinger equation
can be transferred into three-dimensional systems of equations rigorously [6].
Most of the works about three-dimensional FEM applied to three-body
problems appeared hereafter. They are Braun et al. [7], Scrinzi [8], Ackermann
[9], Zheng and Ying [10], and so on. They obtained very precise results.
Shi and Ding [11] solved the HartreeFock equation of the ground lithium
numerically by one-dimensional piecewise linear nite element method. It is a
system of one-dimensional nonlinear partial dierential equations(whose
Coulomb integral and exchange integral are double). The relative error be-
tween their result and the HartreeFock limit value is )4.1  10
4 . After
considering correlative correct, the relative error between their approximate
eigenvalue and the experimental value is )4.9  10
4 . That is to say, the results
by the HartreeFock FEM can only be compared with the limit value of the
HartreeFock method.
To the best of our knowledge, we have not seen any other nite element
result for the lithium atom. In this paper, we transfer the nine-dimensional
Schr odinger equation into a six-dimensional partial dierential eigenvalue
equation rigorously for the ground state of the lithium atom. From the vari-
ational formulation of the Schr odinger equation, we have also derived an six-
dimensional weak form of the eigenvalue problem which is equivalent to the
original variational formula for the ground lithium.
We construct a kind of six-dimensional nite element methods to solve the
transferred energy equation. With very coarse mesh, the relative error of our
approximate eigenvalue and the nonrelativistic limit value is order of 10
5 in
 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695 687

atomic unit. Up to now, the nite element methods are only applied to partial
dierential equations of dimension no more than three. This work is the rst
try to construct highly dimensional FEM; and it is also valuable in applying
FEM to quantum mechanics.
The paper is arranged as follows. The nine-dimensional Schr odinger equa-
tion is transferred into a six-dimensional partial dierential equation in Section
2. A six-dimensional variational equation is given in Section 3. In Section 4, a
kind of six-dimensional nite element methods are described in detail to ap-
proximate the obtained variational equation. Numerical results are given in
Section 5.
We use atomic unit in this paper, i.e. Bohr Radius a0 for length, Hartree for
energy. We consider the stationary nucleus, nonrelativistic and spin-indepen-
dent case.

2. Six-dimensional dierential equation of the ground lithium

There are three electrons out of the nucleus of the lithium atom. In the
Cartesian coordinates, its Schr
odinger equation is:
H w Ew; 2:1
where the Hamiltonian operator H is dened as
1 3 3 3 1 1 1
H
D1 D2 D3


; 2:2
2 r1 r2 r3 r12 r13 r23
o2 o2 o2
Dj 2 2 2 ; 1 6 j 6 3
oxj oyj ozj
are Laplacian operators. Let
ri xi ; yi ; zi ri sin hi cos /i ; sin hi sin /i ; cos hi
~ 2:3
be coordinates of the ith electron in the space-xed system of coordinates
o
xyz and spherical coordinates. rjk j~
rj
~
rk j is the distance between the
jth electron and the kth electron.

rj ~
~ rk rj2 rk2
j~ rk j2
rj
~
cos hjk ; 2:4
rj rk 2rj rk
where hjk is the inter-electronic angle between radial vectors ~ rj and ~rk , rj j~
rj j.
We know from quantum mechanics [12,13], that the ground state of the
lithium atom is the stablest and has the lowest energy. It is the most symmetric
state under rotations about the vector ~ r1 (or equivalently about ~ r2 or ~
r3 ); and it
depends only on six variables r1 ; r2 ; r3 ; h12 ; h13 ; h23 . So we can suppose that the
wave function w in (2.1) depends only on them for the ground state, that is,
w wr1 ; r2 ; r3 ; h12 ; h13 ; h23 .
688 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695

In spherical coordinates, the Hamiltonian of the system can be written as





1 1 o 2 o 1 o 2 o 1 o 2 o
H
r r r
2 r12 or1 1 or1 r22 or2 2 or2 r32 or3 3 or3

L2 L2 L2 3 3 3 1 1 1

21
22
23


; 2:5
r1 r2 r3 r1 r2 r3 r12 r13 r23
where L2i is the square of the angular momentum operator of the ith electron
dened as
L2i L2ix L2iy L2iz

2
2
2
o o o o o o

y1
z1
z1
x1
x1
y1 :
oz1 oy1 ox1 oz1 oy1 ox1
2:6
By (2.3) and (2.4), we have
8
> or1 x1 or2 or3 oh23
>
> ; 0;
>
> ox r1 ox1
ox1 ox1
>
< oh1 
12 1 x1 cos h12 x2

; 2:7
>
> ox1 sin h12
r12 r1 r2 
>
> oh13 1 x1 cos h13 x3
>
>
:
:
ox1 sin h13 r12 r1 r3
We can also obtain partial derivatives or=ox, r r1 ; r2 ; r3 ; h12 ; h23 ; h13 and
x xi ; yi ; zi , i 1; 2; 3. But they are left out for briefness. The cross product of
any two vectors is dened as
8
< Q1 ~
> r2 ~r3 q11 ; q12 ; q13 y2 z3
y3 z2 ; z2 x3
z3 x2 ; x2 y3
x3 y2 ;
Q2 ~ r3 ~r1 q21 ; q22 ; q23 y3 z1
y1 z3 ; z3 x1
z1 x3 ; x3 y1
x1 y3 ;
>
:
Q3 ~ r1 ~r2 q31 ; q32 ; q33 y1 z2
y2 z1 ; z1 x2
z2 x1 ; x1 y2
x2 y1 :
2:8
By the law of chain, it follows


o o iq33 ow iq23 ow
L1z w i x1
y1 w
: 2:9
oy1 ox1 r1 r2 sin h12 oh12 r1 r3 sin h13 oh13
Since
( !
2
q33 ow i q33 cos h12 ow
L1z
x1 x2
y1 y2
r1 r2 sin h12 oh12 r1 r2 sin h12 r1 r2 sin2 h12 oh12
)
2
q33 o2 w q33 q23 o2 w

; 2:10
r1 r2 sin h12 oh212 r1 r3 sin h13 oh12 oh13
 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695 689
( !
2
q23 ow i q23 cos h13 ow
L1z x1 x3 y1 y3

r1 r3 sin h13 oh13 r1 r3 sin h13 r1 r3 sin2 h13 oh13
)
q23 2 o2 w q23 q33 o2 w

; 2:11
r1 r3 sin h13 oh213 r1 r2 sin h12 oh12 oh13

So thanks to (2.9)(2.11), the following is true:



2 iq33 ow iq23 ow
L1z w L1z L1z w L1z
L1z
r1 r2 sin h12 oh12 r1 r3 sin h13 oh13
(
 )
3 2 2
1 q3 cos h12 ow q33 o2 w

x1 x2 y1 y2
2 2
r1 r2 sin h12 r1 r2 sin h12 oh12 r1 r2 sin h12 oh12
(
2  2
)
1 q33 cos h13 ow q23 o2 w

x1 x3 y1 y3
2 2
r1 r3 sin h13 r1 r3 sin h13 oh13 r1 r3 sin h13 oh13
2q33 q23 o2 w
: 2:12
r12 r2 r3 sin h12 sin h13 oh12 oh13

Similarly, we have
(
2  2
)
1 q3 cos h12 ow q32 o2 w
L21y w
x1 x 2 z1 z2
2 2
r1 r2 sin h12 r1 r2 sin h12 oh12 r1 r2 sin h12 oh212
(
2  2
)
1 q2 cos h13 ow q22 o2 w

x 1 x 3 z1 z3
2 2
r1 r3 sin h13 r1 r3 sin h13 oh13 r1 r3 sin h13 oh213
2q32 q22 o2 w
; 2:13
r12 r2 r3 sin h12 sin h13 oh12 oh13
( ! )
2 2
2 1 q31 cos h12 @w q31 @2w
L1x w
y1 y2 z1 z2
2 2
r1 r2 sin h12 r1 r2 sin h12 @h12 r1 r2 sin h12 @h12
( ! )
2 2
1 q22 cos h13 ow q21 o2 w

y1 y3 z1 z3
2
2
r1 r3 sin h13 r1 r3 sin h13 oh13 r1 r3 sin h13 oh13
2q31 q21 o2 w
: 2:14
r r r sin h12 sin h13 oh12 oh13
2
1 2 3

Since

Q3  Q2 ~
r1 ~
r2  ~
r3 ~
r1 ~
r1 ~
r3 ~
r2 ~
r1
~
r1 ~
r1 ~
r3 ~
r3
r1 r3 cos h13  r1 r2 cos h12
r12 r2 r3 cos h23
r12 r2 r3 cos h12 cos h13
cos h23 ; 2:15
690 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695

Combining (2.6) with (2.12)(2.15) gives

L21 w L21x L21y L21z w



1 o o 1 o o

sin h12 w
sin h13 w
sin h12 oh12 oh12 sin h13 oh13 oh13
2cos h12 cos h13
cos h23 o2 w
: 2:16
sin h12 sin h13 oh12 oh13

Similarly, we have

L22 w L22x L22y L22z w



1 o o 1 o o

sin h12 w
sin h23 w
sin h12 oh12 oh12 sin h23 oh23 oh23
2cos h12 cos h23
cos h13 o2 w
: 2:17
sin h12 sin h13 oh12 oh23
L23 w L23x L23y L23z w



1 o o 1 o o

sin h13 w
sin h23 w
sin h13 oh13 oh13 sin h23 oh23 oh23
2cos h13 cos h23
cos h12 o2 w
: 2:18
sin h13 sin h23 oh13 oh23

Combining (2.5) with (2.16)(2.18), we can transfer the Hamiltonian into the
following form:




1 1 o 2 o 1 o 2 o 1 o 2 o
H
r r r
2 r12 or1 1 or1 r22 or2 2 or2 r2 or3 3 or3


3

1 1 1 o o 1 1 1
2
2
sin h 12 2
2
r1 r2 sin h12 oh12 oh12 r1 r3 sin h13




o o 1 1 1 o o
 sin h13 sin h 23
oh13 oh13 r22 r32 sin h23 oh23 oh23
2cos h12 cos h13
cos h23 o2 2cos h12 cos h23
cos h13



r12 sin h12 sin h13 oh12 oh13 r22 sin h12 sin h13

o2 2cos h13 cos h23
cos h12 o2


oh12 oh23 r32 sin h13 sin h23 oh13 oh23
3 3 3 1 1 1



: 2:19
r1 r2 r3 r12 r13 r23
The partial dierential equation (2.19) is six-dimensional and can be solved
by nite dierence methods, variational methods or nite element methods.
 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695 691

3. Six-dimensional variational equation

In order to obtain nite element approximations to the energy of the ground

lithium, we look for the six-dimensional weak formula of the Schr odinger
equation. The expected formula will be derived from the original equation (2.1)
instead of (2.19).
Let R > 0 and dene domains in R6 and R9 as follows:
X 0; R  0; R  0; R  0; p  0; p  0; p; 3:1
 q 
b 9 2 2 2
X x1 ; y1 ; . . . ; z3 2 R j xi yi zi 6 R; i 1; 2; 3 : 3:2

We consider the approximate problem of (2.1) in the bounded domain X b


H w Ew; in X; b
3:3
w 0; on o X;b
b is the boundary of X.
where o X b The equivalent variational formula of (3.3) is
b and w 6 0, such that
nd E; w 2 R1  H01 X
aw; u Ew; u; b
8u 2 H01 X; 3:4
where
Z X 3

1 ow ou ow ou ow ou
aw; u dx
2 X^
i1
oxi oxi oyi oyi ozi ozi
Z

1 1 1 3 3 3



wu dx; 3:5
X^ r12 r13 r23 r1 r2 r3
Z
w; u wu dx; 3:6
^
X
where dx dx1 dy1    dz3 is the volume element of X. b Thanks to [14], the
spectrum of H in (3.3) is discrete and bounded below and approximates the
discrete spectrum of HTin (2.1) as R ! 1. If w is an eigenfunction of (3.3) or
b
(3.4), then w 2 H 2 X b
H01 X.
Denote X 0; R  0; R  0; R 
1; 1 
1; 1 
1; 1 without con-
fusion. Let
l12 cos h12 ; l13 cos h13 ; l23 cos h23 : 3:7
Like in Section 2, we also suppose that the wave function of the ground lithium
depending only on six variables,
w wr1 ; r2 ; r3 ; l12 ; l23 ; l13 ; u ur1 ; r2 ; r3 ; l12 ; l23 ; l13 3:8
Substituting w; u into (3.5) and (3.6), in view of (2.3), (2.4) and (2.7), we can
transfer (3.4) into the following form: Find E; w 2 R1  V and w 6 0, such that
ar w; u Ew; ur ; 8u 2 V ; 3:9
692 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695

where
Z
w; ur wur12 r22 r32 dr1 dr2 dr3 dl12 dl23 dl13 : 3:10
X
Z "

1 ow ou ow ou ow ou 2 2 2
ar w; u r rr
2 X or1 or1 or2 or2 or3 or3 1 2 3
ow ou 2 ow ou 2
r r22 r32 1
l212 r r32 r22 1
l213
ol12 ol12 1 ol13 ol13 1
ow ou 2
r r32 r12 1
l223
ol23 ol23 2


ow ou ow ou
r2 r2 l
l12 l13
ol12 ol13 ol13 ol12 2 3 23


ow ou ow ou
r2 r2 l
l12 l23
ol12 ol23 ol23 ol12 1 3 13


ow ou ow ou
r2 r2 l
l13 l23
ol13 ol23 ol23 ol13 1 2 12
2r12 r22 r32 2r12 r22 r32
p p
r12 r22
2r1 r2 l12 r12 r32
2r1 r3 l13
! #
2r12 r22 r32 2 2 2 2 2 2
p
6r1 r2 r3
6r1 r2 r3
6r1 r2 r3 wu
r22 r32
2r2 r3 l23
 dr1 dr2 dr3 dl12 dl23 dl13 : 3:11
We dene the function space V as
V fwjwjoX 0; kwkV < 1g; 3:12
where oX is the boundary of X dened as
oX : fr1 ; r2 ; r3 ; l12 ; l23 ; l13 2 Xjr1 R or r2 R or r3 Rg;
Z 
2
2
2


2
2 ow ow ow 1 1 ow
kwkV : 1
l212 2 2

X or1 or2or3 r1 r2 ol12


2

2
1 1 ow 1 1 ow
1
l213 2 2 1
l223 2 2
r1 r3 ol13 r2 r3 ol23
2l23
l12 l13 ow ow 2l13
l12 l23 ow ow

r12 ol12 ol13 r22 ol12 ol23

2l12
l13 l23 ow ow

r32 ol13 ol23
 r12 r22 r32 dr1 dr2 dr3 dl12 dl13 dl23 w; wr : 3:13
 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695 693

4. Finite element approximations

Let fTh g be the family of subdivisions of the domain X  R6 (dened as in

Section 3). Each K in Th is a rectangle in R6 . Use notations similar to those in
[15], the nite element triple is dened as:
K r1;0 ; r1;0 h1   r2;0 ; r2;0 h2   r3;0 ; r3;0 h3   l12;0 ; l12;0 h4 
 l13;0 ; l13;0 h5   l23;0 ; l23;0 h6 six-dimensional rectangle;
RK faijklmn r1;0 ih1 =2; r2;0 jh2 =2; r3;0 kh3 =2; l12;0 lh4 =2;
l13;0 mh5 =2; l23;0 nh6 =2ji; j; k; l; m; n 0; 1; 2g;
( )
X2 
i j k l m n  1
PK p aijklmn r1 r2 r3 l12 l13 l23 aijklmn 2 R ; 4:1
i;j;k;l;m;n0

where r1;0 ; r2;0 ; r3;0 ; l12;0 ; l13;0 ; l23;0 are coordinates of some vertex of K.
Choose a basis of PK as
r1
r1;i1 r1
r1;i
1 r2
r2;j1 r2
r2;j
1 r3
r3;k1
pijklmn
r1;i
r1;i1 r1;i
r1;i
1 r2;j
r2;j1 r2;j
r2;j
1 r3;k
r3;k1
r3
r3;k
1 l12
l12;l1 l12
l12;l
1 l13
l13;m1

r3;k
r3;k
1 l12;l
l12;l1 l12;l
l12;l
1 l13;m
l13;m1
l13
l13;m
1 l23
l23;n1 l23
l23;n
1
 ;
l13;m
l13;m
1 l23;n
l23;n1 l23;n
l23;n
1
4:2
where all subscripts and superscripts recur in turn according to 0, 1, 2. Denote
r1;i r1;0 ih1 =2 and dene other notations similarly. Obviously, the following
relations are true:
pijklmn aIJKLMN dijklmn;IJKLMN

1; i; j; k; l; m; n I; J ; K; L; M; N ;
4:3
0; i; j; k; l; m; n 6 I; J ; K; L; M; N :
Dene the nite element space as
Vh fw 2 C 0 XjwjoX 0; wjK 2 PK ; 8K 2 Th g; 4:4
where oX is the boundary of X. Clearly, Vh  H 1 X  V . So the conforming
nite element approximation of (3.9) is: Find Eh ; wh 2 R1  Vh and wh 6 0,
such that
ar wh ; uh Eh wh ; uh r ; 8uh 2 Vh : 4:5
Since the dimension of Vh is nite, (4.5) is equivalent to the following gener-
alized algebraic eigenvalue problem:
AX Eh MX ; 4:6
694 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695

where A is the global stiness matrix and M is global mass matrix. Their orders
are equal to the dimension of Vh . Furthermore, both of them are large and
banded matrices. In real computations, we only need to calculate all element
stiness matrices AK and element mass matrices M K and assemble them to A
and M [16]. The number of freedoms in each inner element equals to 729.

Remark. The convergence and error estimates of the nite element scheme will
be our future work.

5. Numerical results

We use inverse iteration method [16] to solve the generalized eigenvalue

problems (4.6), and use the standard Gaussian integral formula to obtain all
matrix elements. We carry out our computation on PC: Intel PIII750 with 1G
SDRAM. Since the dimension of the domain is six, the number of freedoms
grows very quickly as the mesh is rened. We can not get very small meshes,
nor can we choose very big R for the domain X because of the limit of our
computer. Although the number of grid points for each variable is very small,
the relative error of our result is order of 10
5 . So our result can be improved
as capabilities of computers are improved.
We place same grid points along three radial directions and also along three
angular directions. The subdivisions and computational results are listed in the
Table 1 (NGPs means number of Gaussian points).
Nonrelativistic limit value 11:
7:47807 a:u:
Subdivisions (grid points):

1. Radial directions (unit a0 ): 0.0, 1.0, 3.0,

Angular directions: )1.0, 1.0.
2. Radial directions (unit a0 ): 0.0, 0.8, 2.0, 4.0,
Angular directions: )1.0, 1.0.
3. Radial directions (unit a0 ): 0.0, 0.4, 1.0, 2.5, 5.0,
Angular directions: )1.0, 1.0.

Table 1
Approximate energies of the ground lithium by FEM (a.u.)
Subdivisions of Approximate Cuto radiuses Numbers of NGPs in each
the domain energies Ra0 freedoms direction
1 )7.450617 3 1620 3
2 )7.47699138 4 5319 5
3 )7.47794341 5 13 149 6
 W. Zheng et al. / Appl. Math. Comput. 153 (2004) 685695 695

Acknowledgements

Project supported by the China State Major Key project for Basic Re-
searches and the Science Fund of the Ministry of Education of China.

References

[1] Ch. Froese-Fischer, The HartreeFock Method for Atoms, Wiley-Intersciece, New York,
1977.
[2] A. Askar, Finite element method for bound state calculations in quantum mechanics, J. Chem.
Phys. 62 (1975) 732.
[3] M. Fridman, Y. Rosenfeld, A. Rabinovitch, R. Thieberger, Finite element method for solving
the two-dimensional Schr odinger equation, J. Comput. Phys. 26 (1978) 269.
[4] S. Nordholm, G. Bacsky, Generalized nite element method applied to bound state
calculation, Chem. Phys. Lett. 42 (1976) 259263.
[5] F.S. Levin, J. Shertzer, Finite-element solution of the Schr
odinger equation for the helium
ground state, Phys. Rev. A 32 (1985) 3285.
[6] W. Zheng, Application of the HylleraasBreit transformation to the Schr
odinger equations of
heliumlike atoms, Chinese Journal of Atomic and Molecular Physics 19 (3) (2002).
[7] M. Braun, W. Schweizer, H. Herold, Finite-element calculations for the S states of helium,
Phys. Rev. A 48 (1993) 1916.
[8] A. Scrinzi, A 3-dimensional nite elements procedure for quantum mechanical applications,
Comput. Phys. Commun. 86 (1995) 67.
[9] J. Ackermann, Finite-element expectation values for correlated two-electron wave functions,
Phys. Rev. A 52 (1995) 1968.
[10] W. Zheng, L. Ying, Finite element calculation for helium atom, Rearch Report, School of
Mathematical Sciences and institute of Mathematics, Peking University, 69, 2001.
[11] A. Shi, P. Ding, Solution of the ground state of lithium by HartreeFock nite-element
method, Chinese J. Computat. Phys. 8 (2) (1991).
[12] R.I.M. Alastair, Quantum mechanics, third ed., Institute of Physic Pub, 1992.
[13] C. Cohen-Tannoudji, B. Diu, F. Laloe, Mecanique quantique (English), Translated from the
French by Susan Reid Hemley, Nicole Ostrowsky, Dan Ostrowsky, New York, Wiley, 1977.
[14] W. Zheng, L.Ying, Finite Element Approximation to the Discrete Spectrum of the Schrodinger
Operator with Coulomb Potential, Rearch Report, School of Mathematical Sciences and
institute of Mathematics, Peking University, 5, 2002.
[15] P.G. Ciarlet, The Finite Element Method For Elliptic Problems, North-Holland, Amsterdam,
New York, Oxford, 1978.
[16] K. Bathe, E. Wilson, Numerical Methods in Finite Element Analysis, Prentice-Hall, New
Jersey, 1973.