You are on page 1of 4

Bounds on the ground state energy: Application of the variational principle

Bibhas Bhattacharyyaa
Department of Physics, Jadavpur University, Kolkata 700 032, India

Received 19 March 2008; accepted 19 September 2008 The variational principle is revisited in the context of nding the upper and lower bounds to the ground state energy. It is shown how the variational principle can be employed to nd the lower bound by partitioning the Hamiltonian into several parts. We demonstrate how the variational principle can be used to nd the exact ground states in some special cases. We consider the harmonic oscillator and the Ising Hamiltonian on a bipartite lattice which are familiar yet instructive examples for students. 2009 American Association of Physics Teachers. DOI: 10.1119/1.2998200 I. INTRODUCTION The variational principle is a very powerful tool in quantum mechanical calculations and is part of the standard curricula on quantum mechanics at the undergraduate and graduate level. In many textbooks14 the proof of the variational principle is given along with its application to nding approximate ground states for problems that are not exactly solvable. Many beautiful applications of the variational calculations can be found in the quantum chemistry literature.5 Most of these calculations exploit the fact that the variational principle yields a good upper bound to the exact ground state energy of a given Hamiltonian by choosing a suitable trial wave function for the ground state. Some papers have explored the applications of the variational principle to study the properties of the ground state wave function6 and to understand some basic results of Rayleigh-Schrdinger perturbation theory.7 In this paper we demonstrate how the variational principle can also be used to nd a lower bound. Such an application of the variational principle is discussed in Refs. 8 and 9. We employ here a different and simpler technique based on partitioning the Hamiltonian into several positive semi-denite parts. This approach is useful in many-body problems. We further show that we can nd an exact solution when the lower bound merges with the upper bound. We explore familiar examples such as the harmonic oscillator and the Ising Hamiltonian, for which exact solutions are known. Some nontrivial applications of this technique can be found in recent work on many-body physics.1012 ai*a j ai 2 .
i

=
i,j

Also, we have H =
i,j

ai*a j

=
i

a i 2E i

E0
i

ai 2 , 3

where H H

= Ei

; E0 = min Ei . Hence, 4

E0 .

In the coordinate representation Eq. 4 becomes *H d E0 , 5

* d

where d is the innitesimal volume element, and the integral extends over all coordinate space. To nd the upper bound consider to be a function of one or more variational parameters and nd the minimum of the expectation E = H . 6

II. THE VARIATIONAL PRINCIPLE We review the statement and the proof of the variational principle for the sake of completeness. The variational principle states that3,4 given a trial state , the expectation value of the energy in that state will be an upper bound to the ground state energy E0 of the Hamiltonian H, that is, H E0 . 1

In this way we can nd the best possible upper bound Emin . In principle, we can always try to improve this with adjustable parameters . bound by a better choice of The present method of nding the lower bound requires the Hamiltonian to be partitioned into a set of subHamiltonians. We write H = H1 + H2 + + Hn . 7

Proof: Because the orthonormal eigenstates i of H form a complete set, we can expand an arbitrary state as a linear combination of the i ; that is, = iai i , where the ai are complex quantities. Thus, for an orthonormal basis i ,
44 Am. J. Phys. 77 1 , January 2009 http://aapt.org/ajp

We assume that such a partition is possible. For example, we can partition the Hamiltonian into kinetic and potential energy parts. A more involved partitioning may be required for generating a better lower bound see the examples in Secs. III and IV . Let the exact ground state energy of Hi be i i = 1 , 2 , . . . , n and the corresponding ground state be i . Thus
2009 American Association of Physics Teachers 44

Hi

.
0

8 of H corresponds to the ground Hi


0 0

A=

1/4

14

The actual ground state state energy E0. Thus E0 =


0

H
0 0

The corresponding energy expectation of the Hamiltonian in Eq. 13 is E = = 2 * x H


1/2

=
i

x dx e
x2

Note that the ground state i of Hi is in general different from the ground state 0 of H. According to the variational principle Eq. 1 0 Hi
0 0 0 i.

d2 1 + m 2m dx2 2
2

2 2

x e

x2

dx 15a

10 = 11

Hence, E0
i i.

2m

m 8

15b

Thus, the sum of the ground state energies of the subHamiltonians constitutes a lower bound of E0. We can improve on this bound by suitably partitioning the Hamiltonian H. If Emin = i i, then Emin = E0 = i i, and the exact solution is achieved. This case may appear to be rather elusive but we will see an example in the following. III. GROUND STATE OF THE HARMONIC OSCILLATOR We now illustrate the method for the one-dimensional harmonic oscillator described by the Hamiltonian, 2 p + 1m H= 2m 2
22

E yields an upper bound to the ground state energy E0 of the Hamiltonian given in Eq. 13 . We minimize this bound and obtain with respect to the variational parameter = m / 2 and the upper bound,4 E
min =

1 2

E0 .

16

To nd the lower bound we start from Eq. 12 and intro duce the ladder operators a :1416 a+ = ip + m x 2 m ip + m x 2 m . , 17a

a =

17b

The Hamiltonian in Eq. 12 in terms of a is given by 1 H = H 1 + H 2 = a +a + 2 . 18

x ,

12

where p and x represent the momentum and position operators, respectively, m is the mass, and is the frequency of the oscillator. Here we show how the variational theorem yields the well-known exact solution1315 by nding the best possible upper and lower bounds to the ground state energy of the Hamiltonian in Eq. 12 . We work in the coordinate representation so that Eq. 12 becomes H= d2 1 + m 2m dx2 2
2 2 2

It is evident that a+ is the Hermitian conjugate of a and vice = a a 0. The versa. Therefore, the eigen-values of H1 + in Eq. 18 is a constant. Therefore, using Eq. 11 , term H2 we obtain a lower bound for E0: E0 1 2 . 19

x .
2

13

We choose a trial wave function x = Ae x , where A is a normalization constant and is a variational parameter. Although the choice of a variational wave function is never unique and may be improved by trial and error, physical insight is often helpful. In the present case the increase of the potential energy term with x is unbounded, and therefore we expect a rapid decrease of the wave function as x increases. The parity of the Hamiltonian in Eq. 13 requires that the wave function be symmetric. We also know that the ground state wave function is likely to have a maximum about x = 0. These requirements are satised by the choice of 2 a Gaussian x = Ae x . This choice is not unique see examples in Ref. 4 , and its usefulness can be justied only by explicit calculations. The normalization condition * x x dx = 1 yields4
45 Am. J. Phys., Vol. 77, No. 1, January 2009

The same lower bound can be obtained using a different approach.8,9 The bounds given in Eqs. 16 and 19 yield the exact ground state energy E0 = / 2 and hence the exact ground state wave function 0 x . As we have mentioned, our choice of the partitioning is not unique. We could have made the trivial partitioning H1 2 = x2m 2 / 2. Because H and H are both = p / 2m and H2 1 2 positive semi-denite operators, we can conclude that E0 0. Hence we obtain a lower bound to E0 which is worse than that we have obtained in Eq. 19 .

IV. THE ISING MODEL We next consider the the Ising model17,18 on a bipartite hypercubic lattice for example, a linear chain or a square lattice . The Hamiltonian is given by
Bibhas Bhattacharyya 45

HIsing = J
i,j

S iS j ,

20

where J is the exchange interaction between spins at sites i and j, and Si measures the projection in units of / 2 onto 1 the axis of quantization the z-axis of a spin 2 moment at site i. In Eq. 22 i , j denotes pairs of nearest neighbors i and j. The eigenvalues of Si can take only the values 1. We represent the up spin down spin state by i i , which belongs to the eigenvalue +1 1 of S . Thus,
i

To nd the lower bound we rewrite the Hamiltonian in Eq. 20 in a convenient way. As mentioned in Sec. II, this partitioning is not unique and we will demonstrate two such representations of the Hamiltonian given in Eq. 20 : HIsing = H1 + H2 = J 2 J Si S j 2 2 S2 + S2 i j
i,j

case I ,

26a

i,j

HIsing = H1 + H2 = J 2 J Si + S j 2 + 2 S2 + S2 i j
i,j

Si = i i i i , and Si can be represented by the Pauli matrix, 1 0 0 1 .

21

case II . 26b

i,j

22

The interaction between the spins is responsible for ordering; J 0 corresponds to a ferromagnet Si = + 1 or 1 for all i to minimize the energy in the ground state. For J 0 the system favors antiferromagnetic ordering Si = 1 with S j = 1 for j a nearest neighbor to a given i . Although we can intuitively determine the ground state of the model, we formally establish it by an exact calculation. To nd the upper bound we consider trial states dened on a hypercubic lattice of N sites,
F

We have made use of the fact that Si and S j commute with each other. For J 0 in case I there is a positive semi-denite operator in the rst term on the right-hand side. The only eigenvalue of S2 is +1 and, therefore, the second term yields i a constant NJZ / 2 after performing the sum over i , j in Eq. 26a . Thus, from Eq. 26a , we obtain a lower bound to E0, 1 Elow = NJZ 2 J 0 . 27

=
i A

j B

j, j,

23a

Similarly, for J 0 in case II there is a positive semi-denite operator in the rst term on the right-hand side and the rest of the terms yields a value +NJZ / 2. Thus, from Eq. 26b we obtain another lower bound to E0, 1 Elow = NJZ 2 J 0 . 28

AF

=
i A

23b

j B

where the products imply direct products of states dened on all the sites. A and B denote the two sublattices each with N / 2 sites into which the bipartite hypercubic lattice can be divided. The nearest neighbors of a given site i A are Z the coordination number sites belonging to the sublattice B and vice versa. The states F and AF respectively denote ferromagnetic and antiferromagnetic states. It is easy to verify that the states given by Eq. 23 are eigenstates of the Hamiltonian HIsing in Eq. 20 : HIsing HIsing
F

1 = NJZ 2 1 = NJZ 2

24a

AF

AF

24b

It is yet to be proved that either of them constitutes the ground state. We use these two states as the trial states in Eq. up up 1 to generate two upper bounds EF and EAF to the exact ground state energy E0, 1 up EF = NJZ, 2 1 up EAF = NJZ. 2
up EF is a better upper bound for J J 0.

We compare Eqs. 25a and 27 and nd the exact ground low up state energy EF = EJ 0 = E0 = NJZ / 2 when J 0. Therefore, we have a ferromagnetic ground state given by F as in Eq. low up 23a . Similarly, for J 0 we have EAF = EJ 0 = E0 = NJZ / 2. Correspondingly, we have an antiferromagnetic ground state AF . Note that if we replace all the s by s and vice versa in Eqs. 23a and 23b we arrive at the same results, which means that the ground state is doubly degenerate due to the spin-reversal symmetry of the Hamiltonian in Eq. 20 . Several model systems of interacting electrons on a lattice have been treated using this method. The Hubbard model19 contains a kinetic energy term together with short range interaction terms. Several variants of this Hamiltonian have been studied, but exact solutions to these models are rare. In some special cases it may be possible to guess the nature of the ground state; however it remains to be established by an exact calculation. One way to proceed is provided by the method we have discussed in this paper. The success of the technique lies in a suitable partitioning of the Hamiltonian for nding the lower bound. There are many such interesting applications in the recent literature.1012
a

25a

25b
up 0, and EAF works for

Electronic mail: bibhas.bhattacharyya@saha.ac.in L. I. Schiff, Quantum Mechanics, 3rd ed. McGraw-Hill, New York, 1968 , Chap. 8. 2 J. L. Powell and B. Crasemann, Quantum Mechanics Addison-Wesley, Reading, MA, 1961 , Chap. 11. 3 F. Schwabl, Quantum Mechanics Springer-Verlag, Berlin, Heidelberg, 1992 , Chap. 11. 4 D. J. Grifths, Introduction to Quantum Mechanics, 2nd ed. Prentice
1

46

Am. J. Phys., Vol. 77, No. 1, January 2009

Bibhas Bhattacharyya

46

Hall, Englewood Cliffs, NJ, 2004 , Chap. 7. R. McWeeny, Coulsons Valence, 3rd ed. Oxford U.P., Oxford, 1979 , Chap. 3. 6 J. Mur-Petit, A. Polls, and F. Mazzanti, The variational principle and simple properties of the ground-state wave function, Am. J. Phys. 70 8 , 808810 2002 ; S. K. Foong, D. Kiang, and Y. Nogami, Comment on The variational principle and simple properties of the ground-state wave function, Am. J. Phys. 71 7 , 731732 2003 ; J. Mur-Petit, A. Polls, and F. Mazzanti, Reply to Comment on The variational principle and simple properties of the ground-state wave function, Am. J. Phys. 71 7 , 732 2003 . 7 V. Hushwater, Application of the variational principle to perturbation theory, Am. J. Phys. 62 4 , 379380 1994 . 8 L. Pauling and E. B. Wilson, Introduction to Quantum Mechanics, McGraw-Hill, New York, 1935 , Chap. 7. 9 J. Lee, The upper and lower bounds of the ground state energy using the variational method, Am. J. Phys. 55 11 , 10391040 1986 . 10 R. Strack and D. Vollhardt, Hubbard model with nearest-neighbor and bond-charge interaction: Exact ground-state solution in a wide range of
5

parameters, Phys. Rev. Lett. 70 17 , 26372640 1993 ; R. Strack and D. Vollhardt, Rigorous criteria for ferromagnetism in itinerant electron systems, Phys. Rev. Lett. 72 21 , 34253428 1994 . 11 J. de Boer and A. Schadschneider, Exact ground states of generalized Hubbard models, Phys. Rev. Lett. 75 23 , 42984301 1995 . 12 B. Bhattacharyya and S. Sil, The Hubbard model with bond-charge interaction on a triangular lattice: A renormalization group study, J. Phys.: Condens. Matter 11 17 , 35133523 1999 . 13 See Ref. 1, Chap. 4. 14 See Ref. 4, Chap. 2. 15 See Ref. 3, Chap. 3. 16 W. A. Harrison, Applied Quantum Mechanics World Scientic, Singapore, 2001 , Chap. 16. 17 F. Reif, Fundamentals of Statistical and Thermal Physics McGraw-Hill, New York, 1965 , Chap. 10. 18 C. J. Thompson, Classical Equilibrium Statistical Mechanics Clarendon, Oxford, 1988 , Chap. 6. 19 O. Madelung, Introduction to Solid-State Theory Springer-Verlag, Berlin, 1978 , Chap. 8.

Kohlrausch Slide Wire. The Kohlrausch Wire is used as a variable resistance when making resistance measurements using bridge techniques. The instrument in the picture appears in the 1911 Leeds and Northrup catalogue No. 44 Resistance Boxes and Wheatstone Bridges at $60. The bridge wire is mounted upon a marble cylinder 15 cm in diameter. There are ten turns of wire, giving a total length of 470 cm... The position of the contact is read by means of the vertical glass scale... It is easily possible to estimate to thousandths of one complete resolution, one tenthousandths of the total motion of the contact point. The total resistance is about 7 ohms. It is in the Greenslade Collection. Photograph and Notes by Thomas B. Greenslade, Jr., Kenyon College

47

Am. J. Phys., Vol. 77, No. 1, January 2009

Bibhas Bhattacharyya

47

You might also like