Quantum mechanics in nite-dimensional Hilbert space

A. C. de la Torrea) and D. Goyeneche

Departamento de Fsica, Universidad Nacional de Mar del Plata, Funes 3350, 7600 Mar del Plata, Argentina

Received 28 May 2002; accepted 21 August 2002 The quantum mechanical formalism for the position and momentum of a particle on a one-dimensional lattice is developed. Some mathematical features characteristic of nite-dimensional Hilbert spaces are compared with the innite-dimensional case. The construction of an unbiased basis for state determination is discussed. 2003 American Association of Physics Teachers. DOI: 10.1119/1.1514208

I. INTRODUCTION The quantum mechanical description of a physical system for which the observables have a continuous set of values requires a continuous set of states that belong to an innitedimensional Hilbert space. However, there are many physical systems for which the observables take on a discrete set of values and their corresponding states belong to a discrete and nite set, whose formal description is a nite-dimensional Hilbert space. The best known example of such an observable is angular momentum: a physical system with angular j( j 1) is described by a Hilbert space of momentum J dimension 2 j 1. A less well-known example, which we will discuss in this paper, involves the description of the position and momentum observables in a nite-dimensional Hilbert space, where the position and momentum observables do not take on a continuum of values, but instead are restricted to values on a lattice. The possibility of formalizing quantum mechanics in a nite-dimensional Hilbert space was recognized very early1,2 and has found renewed interest in many applications in, for example, quantum cryptography, quantum computers, and quantum optics. One important practical motivation for studying such systems is that a computer simulation of position and momentum will necessarily involve a nite number of sites. A highly speculative motivation is that the existence of a fundamental length scale, that is, a measure of length below which the concepts of distance and localization become meaningless, would make a discrete quantum mechanics more appropriate than a continuous one. To make this discussion more useful from the didactic point of view, the formalism of quantum mechanics in a nite-dimensional Hilbert space will be presented in a constructive way, where all the steps are logically connected. This work may therefore be a useful complement to a textbook where quantum mechanics in an innite-dimensional Hilbert space is developed. Another didactic feature of this work is that nite-dimensional quantum mechanics requires many interesting mathematical tools such as nite sums and the discrete Fourier transform, which are not usually presented at the undergraduate level. Furthermore, the important differences between nite- and innite-dimensional Hilbert spaces are emphasized, and the limit of innite dimension is considered. II. NOTATION AND DEFINITIONS We will consider a particle in a one-dimensional periodic lattice with N sites and lattice constant a. The quantum me49 Am. J. Phys. 71 1 , January 2003

chanical treatment of this system requires an N-dimensional Hilbert space H. Given any two elements of this space and , we will denote their internal product by , . We will use operators of the form A , , where the dot indicates a space holder to be occupied by the Hilbert space element on which the operator acts. The corresponding Her, . mitian conjugate is A Although we do not need to choose any particular representation for the abstract Hilbert space H, it may be convenient for didactic reasons to specialize the formalism to a three- or four-dimensional Hilbert space whose elements are column vectors of complex numbers. It is better to avoid two-dimensional spaces because they have some peculiar characteristics that do not generalize to higher dimensions. In this case , represents a complex conjugate row vector and operators are square matrices. This special representation is recommended for clarity, but it is important to emphasize to students that the formalism of quantum mechanics can be developed in an abstract Hilbert space and a particular representation is never required. The mathematical beauty of quantum mechanics is, indeed, most apparent in the abstract formulation. Any basis k in the Hilbert space will have N elements labeled by an index k with the values j, j 1, j 1 2,..., j 1,j, with j 2 ,1, 3 ,2, 5 ,..., corresponding to the di2 2 mensions N 2 j 1 2,3,4,5,... This choice of labels has some advantages and disadvantages. The main virtue of this symmetric labeling is that it corresponds to the physical property that the position and momentum can have positive and negative values. Its main shortcoming is that there are many summations and results that are usually given in textbooks with integer indices running from 0 to N 1. To reduce this shortcoming, we give in Appendix A some of these sums with a symmetric index. For the choice of symmetric indices, we must keep in mind that, for even N, j takes half-odd-integer values and we need to be careful with noninteger exponents. We will adopt a useful notation for the principal Nth root of the identity, dened by ei
2 /N

1 build a cyclic group with the important

Integer powers of property that 1

Nn (2 j

1)n

n 0, 1, 2,....

2
49

http://ojps.aip.org/ajp/

2003 American Association of Physics Teachers

III. POSITION AND MOMENTUM The position of the particle in the lattice can take any value eigenvalue ax, where the discrete number x can take j, j 1,..., j 1,j . The state of the any value in the set particle in each position is represented by a Hilbert space element x , and the set x is a basis in H. We can write the position operator X as
j

ric for N even, we would have not obtained such a simple relation as in Eq. 9 , and we would have obtained different expressions for even and odd N. In other words, we choose to dene the translation operator so that we would obtain a simple relation between the bases. With the eigenvalue given in Eq. 9 , the eigenvalue equation for T becomes T
p p p,

12

X
x j

ax

x ,

or, equivalently,
j

which clearly satises X x ax x . We can now construct a translation operator T with the property that T
x 1 x

T
p j

p ,

13

x
N 1

j
j Nj j

j .

We can now construct a Hermitian operator P as a superposition of projectors in the basis p ,

j

We will later explain the reason for dening the operator T to be periodic for odd dimensions, but antiperiodic for even dimensions. This operator is given by
j 1 j x 1 x ,

P
p j

gp

p ,

14

T
x

N 1

, ,

with its Hermitian conjugate

j 1 j x x 1 ,

where g is a real constant to be determined later. Clearly, the operator P is Hermitian and satises the eigenvalue equation P p g p p . From this eigenvalue equation, together with Eq. 12 , and the power series expansion of the exponential, we can show that T exp i 2 P . N g 15

N 1

, .

It is straightforward to check that this operator is unitary, TT T T 1, and therefore its eigenvalues are complex numbers of unit modulus and their eigenvectors form a basis see Appendix B . Now let p and p be the eigenvectors and eigenvalues of T for p j, j 1,..., j 1,j. In order to determine them, we expand p in terms of x and consider
j j j x, p x p x j x, p x.

T
x

From Eq. 4 we obtain

x 1, j p Nj p p x, j p

,
p

for x .

8a 8b

We can now give a physical interpretation to the operator P. Equation 15 , together with Eq. 4 , implies that P is the generator of translations in the position observable. We identify this observable P with the momentum, as is done in classical mechanics. If the position observable takes on values on a lattice with lattice constant a, then the momentum observable must also assume values on a lattice with lattice constant g. In Sec. IV we will see that these values must be related by ga 2 /N, that is, the momentum lattice is the reciprocal lattice of the position. In an manner identical to what was done before, we can now construct a unitary operator B that boosts the momentum states, B
p

p p 1, N 1 1

j,
j Nj j

j,

16

Up to an arbitrary phase that can be absorbed in the denition of p , the solution of the set of Eqs. 8 is
x,

and show that B

x j x x, x x j

1
p

17
x x ,

px

and
p.

,
x

9 B and and 10

18

where we have normalized p are then related by 1

x j px p j p,

The two bases

B exp i

2 X . N a

19

and 1
p j px j x.

IV. THE COMMUTATION RELATION AND THE LIMIT N\ 11 Quantum mechanics textbooks emphasize that the position and the momentum operators satisfy the commutation rela1). However, in tion X, P i we use units such that most cases it is not mentioned that this commutation relation is false in a nite-dimensional Hilbert space. It becomes
A. C. de la Torre and D. Goyeneche 50

Nx

Except for the symmetric index and a different factor, Eqs. 10 and 11 are the discrete Fourier transform. Note that if we had not dened the translation operator to be antisymmet50 Am. J. Phys., Vol. 71, No. 1, January 2003

clear that X, P i, because it can be shown that the commutation relation X, P i implies that the operators X and P are unbounded. However, in a nite-dimensional Hilbert space, all operators are bounded; therefore such a commutation relation is impossible in a nite-dimensional Hilbert space. Indeed, if the Hilbert space is nite, the basis must be nite, and a nite basis, in turn, implies that the operators must have nite-dimensional matrix representations. But for nite-dimensional matrices Tr(X P) Tr( PX) always, so Tr X, P 0, showing that X, P i leads to a contradiction in a nite-dimensional Hilbert space. An explicit calculation of the commutator in the position representation, that is, in terms of the basis x , results in
j j j s j r j

j x j

cx

dp

p,

25

where the complex coefcients c x and d p are related by the discrete Fourier transformation, dp 1 Nx
j px j

cx ,

cx

1 Np

j px j

dp .

26

X, P

ag
k j

k s r 2 k s r N
r ,

1 N . 20

Their absolute values squared c x 2 and d p 2 represent the probability distributions for position and momentum, respectively. Let (t 0 ) be the state of the system at time t 0 , which we choose to be t 0 0. In the Schrodinger picture, this state will evolve according to the time evolution unitary operator given in terms of the Hamiltonian H as U t exp iHt . 27 This description of the time evolution is equivalent to Schrodingers equation if time is represented by a continuous variable. However, in some cases it may be convenient to assume that time also takes discrete values and Eq. 27 cannot be transformed in to a Schrodinger equation. If the state is given in the position or the momentum representation, the coefcients of Eq. 25 will become functions of time, c x (t) and d p (t). Let us consider a free particle with Hamiltonian H P 2 /2m. In the momentum representation the coefcients are simply given by dp t d p 0 exp i g2p2 t 2m dp 0
p 2 t/

exp i

The sum over k can be performed, but it is advantageous to leave it as is. We can now see that in the continuous limit, where N , a0, and g0, but Nagconst, the above commutator approaches the value i, provided that agN 2 . In this limit, the sums over discrete indices k, s, and r become integrals over continuous variables , , and , according to 2 k , N 2 N
s

2 s , N , 2 N
r

2 r N
j

21a

28

,
j

. 21b

where we have introduced a time scale , dened by 2ma . g In the position representation we have
j j (p(r x) j p 2 t/ )

29

The continuous limit is then given by agN X, P i 2 expi d d 1 2 d i cr t , . 22

1 cx 0 Np

30

The sum over k, which was left unperformed, reduces to a simple form in the continuous limit. Indeed, the integral over is a well-known representation of Diracs delta function. Therefore, we have X, P i i agN 2 agN 2 d d agN 1, 2 ,

23

The second summation in Eq. 30 is a discrete Fourier transform that becomes, in the continuous limit, a Fourier integral transform of a Gaussian function with a well-known result. In the discrete case, the algebra is much more difcult and the summation cannot be evaluated in general. It took Gauss four years working with all efforts 3 to evaluate a similar summation the Gauss sum for some special values of the parameters. In any case we can see that the state is periodic, (t T) (t) with period T N if N is odd and T 4N if N is even. VI. STATE DETERMINATION AND UNBIASED BASES At an early stage in the development of quantum mechanics, Pauli4 raised the question of whether knowledge of the probability density functions for position and momentum was sufcient to determine the state of a particle. Because position and momentum are all the classically independent variables of the system, it was conjectured that this question would have an afrmative answer. However, different states with identical probability distributions were found. A review of these issues with references to the original papers can be
A. C. de la Torre and D. Goyeneche 51

where the last equality in Eq. 23 is the completeness relation. Hence, the usual commutation relation for the continuous case is recovered, provided that agN 2 . V. STATE AND TIME EVOLUTION At any instant of time, the state of the particle will be determined by a Hilbert space element . We can represent this state in the position or momentum representation, that is, in the bases x or p :
51 Am. J. Phys., Vol. 71, No. 1, January 2003

24

found in Refs. 57. If we consider the similar problem in classical statistics, that is, nd the combined distribution given the marginal distributions, we should not be surprised to nd out that the Pauli question does not have a positive answer. The marginal probability distribution functions of two random variables uniquely determine the combined distribution function only if they are uncorrelated, that is, independent random variables. Position and momentum are, however, always correlated; that is the essence of Heisenbergs uncertainty principle, and therefore we should not expect that their distributions uniquely determine the quantum state. Explicitly stated, the Pauli question is, can we nd the set of N complex numbers c x that determine the state in Eq. 25 given the knowledge of the sets c x 2 and d p 2 related by Eq. 26 ? Let us notice that the state has an arbitrary phase and is normalized; therefore, we only need to nd 2N 2 real numbers to determine the state. The numbers c x 2 and d p 2 are not independent, because the numbers of each set are probabilities and should add to 1. We therefore have 2N 2 equations at our disposal in order to nd 2N 2 unknowns. However, the equations are not linear and are insufcient to determine the state unambiguously. There is another very important feature in these 2N 2 equations. We will see that not every set of data c x 2 and d p 2 are compatible. The 2N 2 equations have solutions only if the position and momentum data satisfy a number of relations. This constraint on the data is just Heisenbergs uncertainty principle and is a consequence of the relations in Eq. 26 . These ideas are claried by an example with N 2. , and , be the position and momenLet tum bases in a two-dimensional Hilbert space. Their internal product is given by Eq. 9 . An arbitrary state, normalized and with a xed phase, is determined by 2N 2 2 numbers 2 : 0 1 and 0 ei 1
2

31

2 The independent data on position is , 2 , which directly determines and the independent data on momen2 tum is , 2 . With this last data we must deter, exp( i /4)/ 2, we obtain mine . Using that after some algebra that 2

sin

1/2
2

32

Equation 32 can have a solution only if

2

1/2
2

1.

33

If we square both sides of Eq. 33 and rearrange the terms, we nd

2

tainty principle, there is an ambiguity in the solution of Eq. 32 because if is a solution, then is also a solution. This ambiguity cannot be solved with the given data and requires more experimental information. We will next consider what observables we can measure to determine the state without ambiguity. From the previous example it is clear that we need further information besides the distribution of position and of momentum in order to determine the state of the particle. The additional information must involve an observable depending on both position and momentum, because any observable depending on only one of them will not bring new independent information. Some candidates might be X P or the correlation X P PX or any function F(X, P) symmetric or antisymmetric under the exchange X P. Perhaps the best choice of an observable that provides information on the system that is not available from the knowledge of the position and momentum distributions, is an observable whose associated basis is unbiased with respect to the position and to the momentum bases. Two bases in a Hilbert space are unbiased when they are as different as possible in the sense that any element of one basis has the same projection on all elements of the other basis. More precisely, the modulus of their internal product is a constant for all pairs. Unbiased bases are candidates for the quantum mechanical description of classically independent variables such as position and momentum; indeed, we have from Eq. 9 that x, p 1/ N, x,p. The problem of state determination leads us then to the search of a basis s unbiased to x and with respect to p . The importance of unbiased bases associated with noncommuting observables was recognized by Schwinger8 long ago, but the existence and explicit construction of maximal sets of mutually unbiased bases for any dimension is still an open problem. When the dimension N is a prime number, a set of N 1 mutually unbiased bases was presented,9,10 and the same could be achieved if N is a power of a prime number.11 To nd unbiased bases we will follow the method given in Ref. 12. We have seen that the eigenvectors p of the operator T that produces a translation or shift on the basis x build an unbiased basis to x . This result is generalized in Ref. 12, where it is shown that, if N is prime, the eigenvectors of the unitary operators T, B, TB, TB 2 ,...,TB N 1 build a set of N 1 mutually unbiased bases, where T and B are the translation operators for position and momentum dened in Eqs. 4 and 16 . In our case we want to nd a set of only three unbiased bases, and therefore we consider only the rst three operators that provide unbiased bases for any N prime or not . The rst two operators provide the bases p and x , which are related by the discrete Fourier transform and are clearly unbiased. We can easily prove using Eqs. 4 , 12 , 16 , and 17 that the operator TB is a shift operator for both bases p and x , and therefore its eigenvectors s build a basis unbiased to both of them. We have indeed
x x 1, 2 j2 j

1/2

1/2

1/2 2 .

34 x , x j, j, 35 TB
x

This relation is indeed the uncertainty principle: 2 0 or 1, or , implies 2 1/2, that is, exact localization in that is, maximal spread in momentum and, vice versa, exact momentum ( 2 0 or 1) implies maximal spread in position ( 2 1/2). Even if the data is consistent with the uncer52 Am. J. Phys., Vol. 71, No. 1, January 2003

and
A. C. de la Torre and D. Goyeneche 52

(p 1)

TB

p 1, j

p j.

j, 36

2 j2

The eigenvectors of TB in the position representation are found by expanding s in the basis x , and using Eq. 35 and the relation TB
s s s.

dimensional limit.13 The relation of S to X and P is not simple, but at least we can prove that S is antisymmetric under the exchange PX and ag. Indeed, from the Hermitian conjugate of Eq. 43 , we have exp iS gX,a P exp igX exp ia P 44

37 VII. CONCLUSION

expiS a P,gX .

In this calculation we must use with care the modular mathematics dened in Eq. 2 , resulting in the eigenvectors 1
s j 1/2 x 2 (s j 1/2)x x.

Nx

38

A similar calculation yields the eigenvectors of TB in the momentum representation, 1

s j 1/2 p 2 (s j 1/2)p p.

Np

39

Clearly s is unbiased to x and to p . The analytical calculation of discrete Fourier transforms is much more difcult than the Fourier integral transform, and therefore it is of interest that, from Eqs. 38 and 39 we can obtain the discrete Fourier transform for a family of sequences. If we equate Eqs. 38 and 39 and expand x in terms of p , we obtain 1 N
x j 1/2 x 2 bx j 1/2 p 2 bp px

We have presented the quantum mechanical formalism for the position and momentum of a particle in a onedimensional periodic lattice in a way that complements the discussion of the innite-dimensional case presented in quantum mechanics textbooks. Several mathematical subtleties related to the difference between innite- and nitedimensional Hilbert spaces and of modular mathematics were noted. We discussed the physical and mathematical relevance of unbiased bases and, as a consequence from the construction of such a basis, the discrete Fourier transform for a family of sequences is given. It is a strongly recommended that the reader reproduce all the results in terms of the asymmetric indices running from 0 to N 1. The calculations of the eigenvectors of the operator TB are useful exercises for modular mathematics.

ACKNOWLEDGMENTS b 0, 1, 2,... b 1/2, 3/2,... N even N odd . 40 This work received partial support from Consejo Nacio nal de Investigaciones Cientcas y Tecnicas CONICET , Argentina.

We rewrite this result in terms of the asymmetric indices more common in the mathematical literature as 1 N
N 1

1
n 0

1/2 n 2 bn

mn

APPENDIX A: FINITE SUMS WITH SYMMETRIC INDICES b integer. 41 All sums appearing in this appendix can be derived from the fundamental expression
N 1

1/2 m 2 bm

The unbiased basis that was found is of course not unique. It was generated by the operator TB, but there are many other operators whose eigenvectors build an unbiased basis to x n m and to or B m T n , p . Indeed, any combination, T B where n and m are not divisors of N, could do the job. We will now nd the physical meaning for the basis s . That is, we would like to nd a Hermitian operator S(X, P) with s as eigenvectors. That is, TB expiS X, P . 42 In terms of the operators X and P, and using the relation agN 2 , we have expiS X, P exp ia P exp igX . 43

zk
k 0

1 zN , 1 z

A1

valid for any complex number z. For the symmetric index, the sum is
j

zk
k j

z N/2 z z 1/2 z

N/2 1/2

for

1 j 2 ,1, 3 ,2,..., 2 N 2 j 1 2,3,4,5,..., z complex.

A2

If z takes the special values z expi (2 /N) r an arbitrary number, then

j kr k j

with r

Note that here we cannot use the BakerCampbell Hausdorff relation e P e X e P X i/2 that is valid for N , where the commutator X, P is a constant. If we could use it, then the operator S would be simply equal to gX a P. It is possible to nd the eigenvectors of the operator gX a P, but the basis so obtained is not unbiased to either x or p ; however, it becomes unbiased in the innite53 Am. J. Phys., Vol. 71, No. 1, January 2003

sin sin N

r r

A3

In our case, r will often assume integer or half-odd-integer values. For these cases we have
A. C. de la Torre and D. Goyeneche 53

j kr k j

k ,T

k ,T

1.

B1

In order to prove orthogonality, consider 1 0, 2 1

n(N 1)

for r nN,

n 0, 1, 2,...,

T T

k r

r ,T

k r, k

k r

r,

k r,

,
k

B2 . B3

for r
r/2 r,

1, 2, 3,... nN, for r 1 3 5 , , ,... . 2 2 2 A4

* r

The rst two cases correspond, with the asymmetric index, to N the well-known result k 01 kr N r,nN , and we see that this choice leads to simpler mathematics. The third case above must be handled with care in numerical evaluations because the numerator is the fourth root of exp(i 2 /N) u, with u an odd integer. Therefore it has four possible numerical results. The denominator also has two possible results. A formal derivative of Eq. A3 with respect to the parameter r leads to the result
j

If we subtract these equations from each other, we obtain 0 ( k r) r , k . Because the eigenvalues are nondegenerate, the product r , k must vanish for k r. Because T is bounded, the completeness of the eigenvectors can be proved in the usual way, and therefore k is a basis.
Electronic mail: dltorre@mdp.edu.ar H. Weyl, Theory of Groups and Quantum Mechanics Dover, New York, 1931 . 2 J. Schwinger, Quantum Kinematics and Dynamics Benjamin, New York, 1970 . 3 B. C. Berndt and R. J. Evans, The determination of Gauss sums, Bull. Am. Math. Soc. 5, 107129 1981 ; cited in J. J. Benedetto, Harmonic Analysis and Applications CRC Press, Boca Raton, 1997 . 4 W. Pauli, Die allgemeine prinzipien de wellenmechanik, Handb. Phys. 24 2 , 83272 1933 . 5 L. E. Ballentine, Quantum Mechanics. A Modern Development World Scientic, Singapore, 1998 , p. 215. 6 S. Weigert, Pauli problem for a spin of arbitrary length: A simple method to determine its wave function, Phys. Rev. A 45, 7688 7696 1992 . 7 S. Weigert, How to determine a quantum state by measurements: The Pauli problem for a particle with arbitrary potential, Phys. Rev. A 53, 2078 2083 1996 . 8 J. Schwinger, Unitary operator bases, Proc. Natl. Acad. Sci. USA 46, 570579 1960 ; reprinted in Ref. 2. 9 I. D. Ivanovic, Geometrical description of quantal state determination, J. Phys. A 14, 32413245 1981 . 10 W. K. Wooters, Quantum mechanics without probability amplitudes, Found. Phys. 16, 391 405 1985 . 11 W. K. Wooters and B. D. Fields, Optimal state-determination by mutually unbiased measurements, Ann. Phys. N.Y. 191, 363381 1989 . 12 S. Bandyopadhyay, P. O. Boykin, V. Roychowdhury, and F. Vatan, A new proof for the existence of mutually unbiased bases, quant-ph/0103162, to appear in Algorithmica Special Issue on Quantum Algorithms and Quantum Cryptography. 13 A. C. de la Torre, Relativity of representations in quantum mechanics, Am. J. Phys. 70, 298 300 2002 .
1 a

k
k j

kr

i 2

sin

r cos

r sin2

N cos r

r sin

r . A5

By taking more derivatives, again with respect to r, we can obtain other summations involving higher powers of k. APPENDIX B: BASES RELATED TO UNITARY OPERATORS In most textbooks it is proven that the nondegenerate eigenvalues of a Hermitian operator are real and their eigenvectors are orthogonal. We give here the corresponding proof for unitary operators. Let T be a unitary operator and k and k the nondegenerate eigenvalues and normalized eigenvectors. Then, we 2 1 and will prove that k r, k rk . From T k and T T 1, it follows that k k

54

Am. J. Phys., Vol. 71, No. 1, January 2003

A. C. de la Torre and D. Goyeneche

54