Lecture Notes on Quantum Mechanics - Part IV

Yunbo Zhang Instituter of Theoretical Physics, Shanxi University

Abstract
In this part we introduce the other formalism of quantum mechanics - the matrix mechanics developed by Heisenberg. To do this we need some basic knowledge on linear algebra and operator arithmetical rules. Three essential properties of the eigenfunctions of the Hermitian operators (reality, orthogonality, and completeness) constitute the frame of matrix mechanics.

Contents

I. Operators and its Arithmetical Rules A. Hermitian operators B. Multiplication of operators - the commutator II. General Properties of Eigenvalues and Eigenfunctions of Hermitian Operators A. Background B. Reality of eigenvalues C. Orthogonality of eigenfunctions 1. Examples of orthogonality 2. Degenerate case D. Completeness of eigenfunctions 1. Completeness of the eigenfunctions of momentum operator 2. Completeness of spherical harmonics 3. Completeness of eigenfunctions of discrete spectrum 4. Summary III. Generalized Statistical Interpretation and Uncertainty Principle A. Generalization B. The hypothesis of measurement in quantum mechanics C. Reproduce of the original statistical interpretation D. Commutative operators and common eigenfunctions E. A rigorous proof of the uncertainty relation IV. Linear Space - Mathematical Preparation A. 3D ordinary space B. Operation onto space C. Basis transformation D. Transformation of operations V. Matrix Formulation of Quantum Mechanics - Representation Theory A. The idea of representation 2

4 4 5

9 9 11 11 12 13 16 17 17 18 19 19 20 21 21 23 25 29 29 30 32 34 34 34

B. Representation of state C. Representation of operators D. Property of the matrix of a Hermitian operator E. Matrix formulation of rules of quantum mechanics VI. Unitary Transformation between two Representations A. Two sets of basis and their relations to B B. Transformation of state from representation A C. Transformation of operators VII. Diracs Notation A. States B. Rules C. |n D. Application E. Projection operators VIII. Conservation Law and Symmetry A. Constant of motion B. Several typical cases of constant of motion C. Conservation law and symmetry 1. Translation invariance and momentum conservation 2. Rotation invariance and orbital angular momentum conservation 3. Conservation laws and symmetry in general IX. Summary on Part IV

35 37 38 40 42 42 44 46 51 52 53 54 56 58 59 60 61 65 65 66 69 70

I.

OPERATORS AND ITS ARITHMETICAL RULES A. Hermitian operators

Operators are introduced in quantum mechanics to represent dynamical variables. There exists one-to-one correspondence between dynamical variables (physical quantities) F and . Here we review some basic concepts by raising some questions. operators F How is the operator obtained? Answer: Any physical quantity can be expressed in Cartesian coordinate F (x, y, z, px , py , pz ), one gets the operator by simply replacing ( the position and momentum by the corresponding operators F x, y , z , p x , p y , p z ). What is its function? Answer: Given the state of the system , the expectation value d3 r. of the physical quantity can be obtained through F = F What is a Hermitian operator? It means that F must be real. There is a general 2 = F 1 , 2 . criterion for Hermitian operator 1 , F Can we know more than the expectation value? Yes, For a special kind of states which n = fn n , we have (F )2 = 0. This is the eigenequation satises the equation F . of F Are only Hermitian operators prevail in quantum mechanics? No, in quantum theory dierent kinds of operators are used, many of them are not Hermitian (but intimately related to Hermitian operators). A typical example is A = d/dx. In summary, there is a special class of operators which are called Hermitian operators. They are of particular importance in quantum mechanics because they have the property that all of their eigenvalues are real (proved later). This is convenient for the measurement outcome of any experiment must be a real number. There are non-Hermitian operators, but they do not correspond to observable properties. All observable properties are represented by Hermitian operators (but not all Hermitian operators correspond to an observable property).

B.

Multiplication of operators - the commutator

Here we dene the algebraic manipulation of operators. The meaning of multiplication and B is that two operators act successively on a wave function of operators A B =A B A in this sense the multiplication is order-concerned! The commutation relation is the dierence of acting results. Here are some examples: The most important and fundamental one is the commutator between x and p x . For arbitrary state we have x p x = x i p x x = x i (x ) = i + x i x x

The dierence of acting results ( xp x p x x ) = i is irrelevant with the choice of state, thus x p x p x x =i We say that operators x and p x do not commute (are non-commutative) and obey the above commutation relation, or, in other words, the commutator between x and p x reads [ x, p x ] = i Take another example, x and p y . Clearly x p y = x i p y x = y (x ) = x i i y y ( xp y p y x ) = 0 or [ x, p y ] = 0 5

We say that operators x and p y commute each other (are commutative). and B , we can calculate the commutation relation between them For any two operators A B . For example, A, y = x y L yx x , L L =x ( zp x x p z ) ( zp x x p z ) x =z x p x x 2 p z z p x x +x p z x =z ( xp x p x x ) = i z y are non-commutative. Generally, we can see that the operators obey a nonx and L commutative algebra. There is another question about the Hermiticity of the product of two operators. Proposition 1 The product of two Hermitian operators are generally not Hermitian, unless they commute each other. satises Proof. The Hermitian operator A 2 = A 1 , 2 1 , A and B , however The product of A B 2 = 1 , A = i.e. B A
1 A B2 d3 r =

= A or A

1 A

2 d3 r B

A 1 B

A 1 , 2 2 d3 r = B

A = B A =B

and B can not assure the Hermiticity of their product That means, the Hermiticity of A B . The product operator A B is Hermitian only if A A =A B B B = 0. The operator L x = y i.e. A, p z z p y is composed of 4 Hermitian operators, all of them are Hermitian, and in the products y and p z , z and p y are commutative, respectively. x is Hermitian. By the same token, L 2 = L 2 + L 2 + L 2 is Hermitian, too. Thus L
x y z

Now we calculate the commutation relation between the components of the angular mo . The components of x mentum. Till now we have met several vector operators, x , p and L , 6

p are commutative evidently [ x, y ] = 0, etc. [ px , p y ] = 0, etc. x and L y do not commute, in fact The operators L x, L y = ( L yp z z p y ) ( zp x x p z ) ( zp x x p z ) ( yp z z p y ) =y p z z p x y p z x p z z p y z p x + z p y x p z z p x y p z + x p z y p z + z p x z p y x p z z p y z =y p x [ pz , z ] + x p y [ z, p z ] = i ( xp y y p x ) = i L y, L z or L x, L z , but there is no need to Of course, we could have started out with L calculate these separately - we can get them immediately by cyclic permutation of the indices ( xy , y z , z x ) z x, L y = i L L x y, L z = i L L y z, L x = i L L or , L = i L with as
xyz L

the Levi-Civita permutation symbol. It is an anti-symmetric tensor rank 3 dened

= = 1,

xxz

= 0, etc

They are the fundamental commutation relations for angular momentum; everything else follows from them. On the other hand, the square of the total angular momentum 2 = L 2 + L 2 + L 2 L x y z

x does commute with L 2 2 2 2, L x = L L x , L x + Ly , L x + Lz , L x y L y, L x + L y, L x L y + L z L z, L x + L z, L x L z =L y i L z + i L z L y + L zi L y + i L yL z =L =0 where we have used the result of Problem in the following. It follows that 2, L = 0, L Problem 2 Dene = L x iL y L Show that z, L = L L +, L = 2 L z L One can manipulate the operators in many ways, such as summation, product and more generally, given a function F (x), the power series converge if the derivatives are well dened
+

= x, y, z

F (x) =
n=0

F (n) (0) n x n!

is expressed as the function of operator A

+

) = F (A
n=0

F (n) (0) n A n!

Specially for F (x) = ex one may dene F d dx

+

= e dx =
n=0

n dn . n! dxn

x and L y in any eigenstate of L z are zero. Problem 3 Show that the expectation values of L (Zeng, p133, 13)

Problem 4 Prove the following commutator identity B, C =A B, C + B, C A A Show that [ xn , p ] = i nx n1 and more generally that [f ( x), p ] = i for any function f (x). df dx

II.

GENERAL PROPERTIES OF EIGENVALUES AND EIGENFUNCTIONS OF

HERMITIAN OPERATORS A. Background

We have learnt many eigenfunctions of Hermitian operators: For 1D momentum, the eigenfunction is p x = i d = px dx

where px is the eigenvalue and the eigenfunctions are chosen such that they can be -normalized
i 1 e px x 2 where < px < i.e. the eigenfunctions are innumerable. The 3D momentum

px (x) =

eigenfunction takes a similar form p = i and the eigenfunctions are = e (2 )3/2 1

i

= (x0 px + y0 py + z0 pz ) = p

(px x+py y +pz z )

We have also solved the eigenfunctions for the square of angular momentum and its z -component 2 Ylm = l(l + 1) 2 Ylm L z Ylm = m Ylm L 9

with the eigenvalues l = 0, 1, 2, and m = l, l + 1, l and the eigenfunctions are spherical harmonics Ylm (, ). The other example of eigenvalue problem is the eigenequation of Hamiltonian operator, the stationary Schr odinger equation n = En n H Typical exactly solvable models include the simple harmonic oscillator, square potential well, hydrogen atom etc. These examples provide us the information of micro-particle in concrete conditions and play important role in the construction of the theoretical frame of quantum mechanics. In this section we discuss the general property of eigenvalues and eigenfunctions of Hermitian operators. The spectrum of eigenvalues of Hermitian operators fall into two categories: If the spectrum is discrete (i.e., the eigenvalues are separated from one another) then the eigenfunctions constitute physically realizable states. If the spectrum continuous, (i.e., the eigenvalues ll out an entire range) then the eigenfunctions are not normalizable, and they do not represent possible wave functions (though linear combinations of them - involving necessarily a spread in eigenvalues - may be normalizable). a Some operators have a discrete spectrum only (for example, the angular momentum, the Hamiltonian for the harmonic oscillator, square potential well of innite depth), b some have only a continuous spectrum (for example, the free particle Hamiltonian, the 1D and 3D momentum operators), c and some have both a discrete part and a continuous part (for example, the Hamiltonian for a nite square well, the spectrum is discrete for E < V0 and continuous for E > V0 ; the Hamiltonian for the system of nucleus plus electron, the energy spectrum for hydrogen atom is discrete for E < 0 and continuous for the scattering of electrons E > 0) The essence of theory to be developed is the same for the three cases, but more tedious for the third case. We will presume that the eigenvalues of the system would be entirely discrete here and discuss the situation of continuous spectrum later through some examples. 10

B.

Reality of eigenvalues

Mathematically, the normalizable eigenfunctions of a Hermitian operator have the following properties: Theorem 5 Their eigenvalues are real. Proof. Suppose m = fm m F Multiplying m from the left, and integrating over entire space, we see
m = fm m d3 r m F

fm = =

m F m d3 r =

m F

m d3 r

(fm m ) m d3 r = fm

is Hermitian and m is normalized. Evidently fm where we have used the property that F are real for all Hermitian operators.

C.

Orthogonality of eigenfunctions

Theorem 6 Two eigenfunctions of the same Hermitian operator with dierent eigenvalues are orthogonal. Suppose m = fm m F n = fn n F where fm = fn . Orthogonality means
m n d3 r = 0

is a Hermitian operator, so Proof. F

m F n d3 r = m F n d3 r

11

gives fn i.e. (fm fn ) Because (fm fn ) = 0, it must be that

m n d3 r = 0 m n d3 r = fm n d3 r m

m n d3 r = 0.

1.

Examples of orthogonality

Two stationary states with dierent energy values of the Schr odinger equation are orthogonal, i.e. m = Em m H n = En n H
m n d3 r = 0

Spherical harmonics with dierent indices are orthogonal, i.e.

Ylm Yl m d = 0, l = l

Ylm Ylm d = 0, m = m

Orthogonality is for dierent eigenvalue states, while normalization is for the same state. The combination of them is called orthonormality.
(r)n (r)d3 r = mn m

(1)

For example we can write the orthogonality and normalization of spherical harmonics into
(, )Yl m (, )d = ll mm Ylm

If the spectrum of a hermitian operator is continuous, the eigenfunctions are not normalizable, and the proofs of the above two Theorems fail, because the wavefunctions are not square integrable. For example, the eigenvalues of 1D momentum are real and the eigenfunctions satisfy
i 1 e px 2

i 1 e p x dx = (p p ) 2

12

i.e.
p (x)p (x)dx = (p p )

(2)

which is strikingly reminiscent of true orthonormality equation (1) - the indices are now continuous variables, and the Kronecker delta has become a Dirac delta, but otherwise it looks just the same. Equation (2) is called Dirac orthonormality.

2.

Degenerate case

We here discuss more about the degeneracy of the energy spectrum of Schr odinger equation. If two (or more) distinct solutions to the (time-independent) Schr odinger equation have the same energy E , these states are said to be degenerate. For example, the free particle states are doubly degenerate one solution representing motion to the right, and the other motion to the left. But we have encountered no normalizable degenerate solutions, and this is not an accident. We have the following theorem: Theorem 7 In one dimension there are no degenerate bound states. Proof. Suppose there are two solutions, 1 and 2 , with the same energy E . d2 1 + V 1 = E1 2m dx2 2 2 d 2 + V 2 = E2 2m dx2 Multiply the Schr odinger equation for 1 by 2 , and the Schr odinger equation for 2 by 1 , d2 1 + V 2 1 = E2 1 2m dx2 2 d2 2 1 2 + V 2 1 = E2 1 2m dx
2 2

and subtract, we have But d dx it follows that 2

2m

d2 1 d2 2 1 dx2 dx2 = 2

=0

d1 d2 1 dx dx 2

d2 1 d2 2 1 dx2 dx2

d1 d2 1 =K dx dx 13

where K is a constant. For normalizable solutions 0 at x , this constant is in fact zero. Thus 2 so ln 1 = ln 2 + const. 1 = (constant) 2 and hence that the two solutions are not distinct. Imagine now a bead of mass m that slides frictionlessly around a circular wire ring of circumference L. This is just like a free particle, except that (x) = (x + L), since x + L is the same point as x. We try to nd the stationary states (with appropriate normalization) and the corresponding allowed energies. From the equation d2 = E 2m dx2
2

d1 d2 1 d1 1 d2 = 1 = dx dx 1 dx 2 dx

where x is measured around the circumference, we immediately know the solution is (x) = Aeikx + Beikx , The boundary condition gives Aeikx eikL + Beikx eikL = Aeikx + Beikx , and this is true for all x. In particular, for x = 0 AeikL + BeikL = A + B And for x = /2k Aei/2 eikL + Bei/2 eikL = Aei/2 + Bei/2 AeikL + Bei eikL = A + Bei or AeikL BeikL = A B Add (3) and (4) AeikL = A (4) (3) k= 2mE/ 2 .

14

Either A = 0, or else eikL = 1, in which case kL = 2n (n = 0, 1, 1, ). But if A = 0, then BeikL = B , leading to the same conclusion. So for every positive n there are two solutions: + (x) = Aei(2nx/L) (x) = Bei(2nx/L) The two independent solutions for each energy En correspond to clockwise and counterclockwise circulation (n = 0 is ok too, but in that case there is just one solution). Normalizing
L 0

1 2 + (x) dx = 1 A = B = L

Any other solution (with the same energy) is a linear combination of these 1 n = ei(2nx/L) L 2 2 2n2 2 k = En = 2m mL

, n = 0, 1, 2,

How do we account for this degeneracy, in view of the theorem that is, why does the theorem fail in this case? The theorem fails because here does not go to zero at ; x is restricted to a nite range, and we are unable to determine the constant K . Problem 8 Consider the operators
2 1 i d , Q 2 d Q d d2

where is the usual polar coordinate in two dimensions. (These two operators might rise in 1 and Q 2 Hermitian? Find a physical context if we were studying the bead-on-a-ring.) Are Q the eigenfunctions and eigenvalues. Are the spectra degenerate? It is always possible to reshue these degenerate states so that they are normalized by themselves while orthogonal to other (the rest) states. Here is an example on how to construct the orthogonal eigenfunctions from the degenerate states. , with the Example 9 Suppose that f (x) and g (x) are two eigenfunctions of an operator Q same eigenvalue q = qf Qf = qg Qg 15

It is easy to show that any linear combination of f and g h(x) = af (x) + bg (x) , with eigenvalue q (for arbitrary constants a and b) is itself an eigenfunction of Q =Q (af + bg ) = aQf + bQg = aqf + bqg = qh. Qh For example, f (x) = ex and g (x) = ex are eigenfunctions of the operator d2 /dx2 , with the same eigenvalue 1 d2 d2 x f = e = ex = f dx2 dx2 d2 d2 x g = e = (1)2 ex = g 2 2 dx dx The simplest orthogonal linear combinations are 1 x 1 e ex = (f g ) 2 2 1 x 1 cosh x = e + ex = (f + g ) 2 2 sinh x = They are clearly orthogonal, since sinh x is odd while cosh x is even.

D.

Completeness of eigenfunctions

For a 3D space, three basic vectors x0 , y0 and z0 form a orthonormal basis. Any vector can be expressed as a linear combination of the them. They are complete. If one of the basic vectors is missed, we can no more express all the vectors through the basis. They are incomplete. A periodic function f (x) of period L can be expanded into a Fourier series 2n 2n f (x) = an cos x+ bn sin x L L n=0 n=1 thus the function set cos 2n x, sin 2n x serve as a basis and are used to express periodic L L
x, is missed, they are functions. They are complete. If one of the above basis, say cos 2 L

incomplete and can not express arbitrary periodic functions at will.

16

1.

Completeness of the eigenfunctions of momentum operator

As an example, we discuss here rst the completeness of the eigenfunctions of 1D momentum operator p x = i
i 1 e px . 2 The set {p (x)} with < p < + are complete, i.e. any wave function can be expanded

d , dx

p (x) =

as a superposition of p (x)s. (x) = 1 2 (p) e

i

px

dp =

(p) p (x) dp

This is nothing but the Fourier transformation between x and p spaces. Inversely, we have (p) = so (x) = = = = dp (p) p (x) dp dx
dx p (x ) (x ) p (x) dpp (x ) p (x) (x )

1 2

(x ) e

px

dx =

p (x ) (x ) dx

dx (x x) (x )

The completeness of p (x) with p a parameter can be expressed as

dpp (x ) p (x) = (x x)

i.e.
i i 1 1 1 e px e px dp = 2 2 2

p(xx )

dp

= (x x) .

2.

Completeness of spherical harmonics

The spherical harmonics {Ylm (, )} are complete. Any function F (, ) can be expressed as a superposition of spherical harmonics F (, ) =
l,m

cl,m Ylm (, ) 17

The coecients are cl,m = Put it back we have F (, ) =

l,m Yl,m ( , ) Yl,m (, ) F ( , ) d Yl,m ( , ) F ( , ) d

( ) ( ) F ( , ) d

therefore the completeness of spherical harmonics is expressed as

Yl,m ( , ) Yl,m (, ) = ( ) ( ) l,m

3.

Completeness of eigenfunctions of discrete spectrum

Let us check the completeness of eigenfunctions for a simple harmonic oscillator n (x) = Nn e 2 with =
m
1 2 x2

Hn (x)

. {n } is complete. Any state of simple harmonic oscillator can be expressed (x) =

n

as a superposition of n s an n (x)

and an = We have (x) =

n n (x ) (x ) dx n (x) n (x ) n (x) (x ) dx n n (x ) (x ) dx

= = which gives us

(x x) (x ) dx

(x ) n (x) = (x x) n n

This can be generalized to the complete set of eigenfunctions {un (x)} of a certain operators F un (x) = fn un (x) F 18

{un (x)} is complete in that any function of the same kind can be developed upon {un (x)} (x) =
n

cn un (x)

with cn = u n (x ) (x ) dx

The completeness of {un (x)} is expressed as u n (x ) un (x) = (x x)

n

4.

Summary

The property of completeness is essential to the internal consistency of quantum mechanics, so (following Dirac) we will take it as an axiom (or, more precisely, as a restriction on the class of Hermitian operators that can represent observables/dynamical variables) Axiom 10 The eigenfunctions of an observable operator are complete: Any function can be expressed as a linear combination of them.

III.

GENERALIZED STATISTICAL INTERPRETATION AND UNCERTAINTY

PRINCIPLE

We have learnt how to calculate the probability that a particle would be found in a particular location | (x)|2 dx probability between x and x + dx (position measurement ), and how to determine the expectation value of any dynamical variables. In the section about generic problem in quantum mechanics, we have also learnt how to nd the possible outcomes of an energy measurement and their probabilities
+

n = En n , H

H =
n=1

|cn |2 En

probability of getting En |cn |2 We are now to state the generalized statistical interpretation, which subsumes all of this and enables you to gure out the possible results of any measurement and their probabilities. 19

A.

Generalization

If you measure a physical quantity F (x, p) on a particle in the state (x) which is normalized (x) (x) dx = 1, (x, i d/dx). you are certain to get one of the eigenvalues of the hermitian operator F is discrete If the spectrum of F un (x) = fn un (x) , F and {un (x)} is complete, the probability of getting the particular eigenvalue fn associated with orthonormalized eigenfunction un (x) is |cn |2 , where cn = If the spectrum is continuous, uz (x) = f (z )uz (x) F with real eigenvalues f (z ) and associated Dirac-orthonormalized, complete eigenfunctions uz (x), the probability of getting a result in the range dz is |c(z )|2 dz, where c(z ) = u z (x) (x) dx u n (x) (x) dx

Upon measurement, the wavefunction collapses to the corresponding eigenstate. In both cases, the coecients are calculated by the Fouriers trick. One can easily check the normalization of the coecients and the expectation value. Normalization of now means the total probability (summed over all possible outcomes) has got to be one |cn |2 = 1
n

and sure enough, this follows from the normalization of the eigenfunctions 1= =
m

(x) (x) dx
c m um (x) n 2

cn un (x) dx

=
n

|cn |

20

Similarly, the expectation value of F should be sum over all possible outcomes of the eigenvalues times the probability of getting that eigenvalue F =
n

|cn |2 fn

Indeed, F = =
m

(x) dx (x) F
c m um (x) F n

cn un (x) dx

=
m n

c m cn fn |cn |2 fn
n

u m (x) un (x) dx

So far at least, everything looks consistent.

B.

The hypothesis of measurement in quantum mechanics

on In summary we have the hypothesis: For a measurement of physical quantity F systems described by a wave function (x) I The measured result is a statistical one II The outcome of a single measurement could be nothing but one of the eigenvalues of , say fn F III The probability of obtaining certain eigenvalue fn is |cn |2 = u n (x) (x) dx .
2

IV Mechanism: the system, having interacted with apparatus, has changed its state from with a probability |cn |2 and gives a original (x) to one of the eigenstate un (x) of F measuring outcome fn .

C.

Reproduce of the original statistical interpretation

Can we reproduce, in this language, the original statistical interpretation for position measurements? Sure - it is real overkill, but worth checking. 21

A measurement of x on a particle in state must return one of the eigenvalues of x . But what are the eigenfunctions and eigenvalues of x ? Let gy (x) be the eigenfunction and y the eigenvalue xgy (x) = ygy (x) Here y is a xed number (for any given gy (x)), but x is a continuous variable. What function of x has the property that multiplying it by x is the same as multiplying it by the constant y ? Obviously, its got to be zero, except at the one point x = y ; it is nothing but the Dirac delta function gy (x) = A (x y ) This time the eigenvalue has to be real; the eigenfunctions are not square-integrable, but again admit Dirac orthonormality
+ gy +

(x) gy (x) dx =

|A|2 (x y ) (x y ) dx

= |A|2 (y y ) If we pick A = 1, so gy (x) = (x y ) . We see these eigenfunctions are also complete, any wavefunction (x) can be expanded as
+ +

(x) =

c (y ) gy (x) dy =

c (y ) (x y ) dy

with c (y ) = (y ) so the probability of getting a result in the range dy is |c (y )|2 dy = | (y )|2 dy not familiar? Write it as | (x)|2 dx, which is precisely the original statistical interpretation. What about momentum? We have found that the eigenfunctions of the 1D momentum operator px (x) =
i 1 e px x 2

22

which is also complete and any wavefunction (time-dependent) can be expanded as (x) = 1 2 1 = 2
+

e
+

px

(p) dp c (p) dp

px

where the coecients is given by the inverse Fourier transformation c(p) = (p) = 1 2
+

px

(x) dx

and | (p)|2 dp is the probability that a measurement of momentum would yield a result in the range dp. In momentum space, the position operator is x = i /p. More generally (x) F (x, i ) (x) dx, in position space x F (x, p) = (p) F (i , p) (p) dp, in momentum space
p

In principle you can do all calculations in momentum space just as well (though not always as easily ) as in position space. Problem 11 A particle of mass m is bound in the delta function well V (x) = (x). What is the probability that a measurement of its momentum would yield a value greater than p0 = m/ ? (Griths Example 3.4)

D.

Commutative operators and common eigenfunctions

In earlier sections we have met some situations where two or more operators may share 2 and its z component L z eigenfunctions. For example, the square of angular momentum L do admit complete sets of simultaneous eigenfunctions, the spherical harmonics Ylm (, ). In the problem of hydrogen atom, the Hamiltonian, the magnitude of the angular momentum, and the z component of angular momentum are mutually commutative operators, and one can construct simultaneous eigenfunctions of all three, labeled by their respective eigenvalues n, l, m. In general we have the following theorem:

23

 and G have common eigenfunctions set {n }, they are comTheorem 12 If operators F mutative. Proof. Suppose n = fn n F n = gn n G The set {n } is complete, thus any wavefunction can be developed on it =
n

an n

and G on gives Acting the commutator of F G G F = F G G F F

n

an n

=
n

G G F n an F an (fn gn gn fn ) n = 0
n

= therefore

G G F = F , G =0 F

Let us check the reverse proposition: and G are commutative and the eigenfunctions in the set {n } are nonTheorem 13 If F . degenerate, {n } must be eigenfunctions of G Proof. Now the premise is G G F =0 F or more precisely G nG F n = 0 F which means G n = Gf n n = fn G n F

24

 n is thus an eigenfunction of F with eigenvalue fn . On the other hand, {n } as eigenG are non-degenerate, n and G n must be the same states - the dierence functions of F between them is a constant factor which we shall call it gn n = gn n G , i.e., F and G have common eigenfuncIt turns out that {n } are also eigenfunctions of G tions. or G are degenerate, we can The case is a little complicated if the eigenfunctions of F nevertheless arrive at the same conclusion. For example, 2, L z = 0 L the combinations of spherical harmonics Ylm 1 = a1 Y11 + b1 Y10 + c1 Y1,1 2 = a2 Y11 + b2 Y10 + c2 Y1,1 3 = a3 Y11 + b3 Y10 + c3 Y1,1 2 are eigenfunctions of L 2 i = 1(1 + 1) 2 i L z . It is for i = 1, 2, 3 (three-fold degeneracy). Evidently they are not eigenfunctions of L z , i.e. Y11 can always be however always possible to reshue them to the eigenfunctions of L expressed as a combination of i . The general proof will be omitted here. ( = x, y, z ) are non commutative, do they share common eigenstates? If Problem 14 L so, what are they?

E.

A rigorous proof of the uncertainty relation

and G do not commute, F , G = 0, they Proposition 15 If two Hermitian operators F have no common eigenfunction set and their variances obey an uncertainty relation
2 F

2 G

1 2i 25

, G F

Proof. We dene =F F , F =G G G and G are also Hermitian because F and G are real. Consider now a Evidently F positive denite integration I ( ) = iG d3 r 0 F
2

where is a real constant parameter. We have I ( ) = = 2 i = 2 i Dene a= c= b=i F G

2

F F F F

+ i G

i G F G

d3 r

d3 r + F

d3 r G d3 r

G G d3 r + G

F d3 r

G G F d3 r F

d3 r = d3 r =

F G

2 = F 0 2 = G 0

G G F d3 r = i F , G F

, G is the same as F , G . Further more It is easy to show that the commutator F , G is Hermitian operator (Check it!) so its expectation value b is real. The integral i F becomes I ( ) = a 2 b + c 0. There is no real root for (the parabola doesnt intersect with the axis), so we must have b2 4ac 0 i.e.
2 2 F G 2

1 2i 26

, G F

(5)

or in the notation of standard deviation 1 2 1 F G 2 F G , G F , G F

This is the uncertainty principle in its most general form. (You might think the i makes it trivial isnt the right side of equation (5) negative? No, for the commutator carries its own factor of i, and the two cancel out.) =x For example, suppose the rst observable is position (F ), and the second is momen=p tum (G = i d/dx). We have already known [ x, p ] = i Accordingly
2 x

2 p

1 i 2i

or, since standard deviations are by their nature positive, x p /2 That proves the original Heisenberg uncertainty principle, but we now see that it is just one application of a far more general theorem: There will be an uncertainty principle for any pair of observables whose corresponding operators do not commute. We call them incompatible observables. Evidently, incompatible observables do not have shared eigenvectors at least, they cannot have a complete set of common eigenvectors. Their standard deviations can not be zero at the same time. Later we will see matrices representing incompatible observables cannot be simultaneously diagonalized (that is, they cannot both be brought to diagonal form by the same similarity transformation). On the other hand, compatible observables (whose operators do commute) share a complete set of eigenvectors, and the corresponding matrices can be simultaneously diagonalized. As an application we can estimate the ground state energy of a simple harmonic oscillator p 2 1 H= + m 2 x 2 2m 2 Apparently we have the expectation values for x , p x = 0, p =0 27

so ( p)2 = p 2 ( x)2 = x 2 and H = 1 1 ( p)2 + m 2 ( x)2 2m 2 2 1 1 2 ( x)2 2 + m 2m 4 ( 2 x)

The minimum value is determined by

2 H = 8m ( x)2

1 ( x)2
2 min 2

1 + m 2 = 0 2
2

which gives ( x) so ( x)2

min 2

4m2 2

2m

Inserting back to the expectation value of Hamiltonian we have H

min

1 1 2 2m 1 + m 2 = 2m 4 2 2m 2

which is exactly the ground state energy we have got in the 1D harmonic potential. For angular momentum we know their components are generally non-commutative. Fortunately we have 2, L = 0, L = x, y, z

2 and one of the components, say, This is why we could have common eigenfunctions of L z , i.e., the spherical harmonics Ylm (, ) are chosen to represent the angular momentum L states. Problem 16 Zeng Jinyan, page 133, 11, 12, 14.

28

IV.

LINEAR SPACE - MATHEMATICAL PREPARATION

The purpose of this chapter is to develop the alternative formalism of quantum mechanics matrix mechanics. Quantum theory is based on two constructs: wavefunctions and operators. The state of a system is represented by its wavefunction, observables are represented by operators. Mathematically, wavefunctions satisfy the dening conditions for abstract vectors, and operators act on them as linear transformation. So the natural language of quantum mechanics is linear algebra. I begin with a brief survey of linear algebra. Linear algebra abstracts and generalizes the arithmetic of ordinary vectors, as we encounter them in rst-year physics. The generalization is in two directions: (1) We allow the scalars to be complex, and (2) we do not restrict ourselves to three dimensions (indeed, we shall be working with vectors that live in spaces of innite dimension).

A.

3D ordinary space

We are already familiar with 3D space. We will rely upon it and do not repeat the sophisticated denition of linear space. 3D space has three basis vectors i, j and k and one has dened scalar products of vectors as R = ai + b j + c k i i = 1, i R = a, etc.

Here we will give the theory with slight formal modication. First, we dene the scalar product anew. By new stipulations on dot-products we can discriminate the two participants, one as x0 |, the other as |x0 i i = 1 x0 |x0 = 1 and the vector is now expressed as |R = a |x0 + b |y0 + c |z0 a = x0 |R , The square of norm of |R is R|R = (a x0 | + b y0 | + c z0 |) (a |x0 + b |y0 + c |z0 ) = a2 + b 2 + c 2 29 etc.

and similarly, the inner product of |R1 and |R2 is R1 |R2 = (a1 x0 | + b1 y0 | + c1 z0 |) (a2 |x0 + b2 |y0 + c2 |z0 ) = a1 a2 + b 1 b 2 + c 1 c 2 . The real linear space with the denition of innor product is called the Euclidean space or inner product space. Secondly, we use matrix to represent the vectors. For example, the three basis vectors are expressed as column matrices 1 0 0 |x0 = 0 , |y0 = 1 , |z0 = 0 0 0 1 and the vector |R = a |x0 + b |y0 + c |z0 a =b c

Accordingly the left vectors are represented as row matrices x0 | = (1, 0, 0) , y0 | = (0, 1, 0) , z0 | = (0, 0, 1) R| = (a, b, c) thus we have the inner products a y0 |R = (0, 1, 0) b = b c a2 R1 |R2 = (a1 , b1 , c1 ) b2 = a1 a2 + b1 b2 + c1 c2 c2
B. Operation onto space

Operation is dened as a regular, ordered and linear deformation of the system with the basis of space untouched. Here is an example of operation: We rotate the system by an

30

(a,b) (a,b)
FIG. 1: An example of operation.

angle about z axis, followed by a compression along z axis by a factor . The operation changed vector |R into |T . Suppose a a |R = b , |T = b c c From Figure (1) we know a = l cos , and a = l cos ( + ) = l cos cos l sin sin = a cos b sin b = l sin ( + ) = l cos sin + l sin cos = a sin + b cos c = c Write it in matrix form a b c cos sin 0 a = sin cos 0 b 0 0 c 31 b = l sin

In general the deformation can be expressed as a square matrix Fxx Fxy Fxz |T = F |R , F = Fyx Fyy Fyz . Fzx Fzy Fzz
C. Basis transformation

The same linear space may be expressed by two dierent orthogonal coordinates/basis sets. A change of basis is mostly described by a matrix by which the new basis vectors are given as linear combinations of the old basis vectors. Suppose we have two basis sets basis 1 : |x0 , |y0 , |z0 basis 2 : |0 , |0 , |0 A vector can be developed upon both |R
x

|R

a = a |x0 + b |y0 + c |z0 = b c = |0 + |0 + |0 =

and one must have a |x0 + b |y0 + c |z0 = |0 + |0 + |0 that is the change of basis will not aect the vector. Multiplying 0 | from the left on the above equation we have a 0 |x0 + b 0 |y0 + c 0 |z0 = similarly a 0 |x0 + b 0 |y0 + c 0 |z0 = a 0 |x0 + b 0 |y0 + c 0 |z0 = 32

Dene Sx = 0 |x0 , Sy = 0 |y0 , Sz = 0 |z0 , etc. we get

S S S x y z = Sz Sy Sz Sx Sy Sz or |R

a b c

= S |R

where S is the matrix for the basis transformation. The norm of a vector should not depend on the choice of the basis sets, which means

R|R

=x R|R

or

a (, , ) = (a, b, c) b c So T T a a a S b S b = b c c c According to (AB )T = B T AT we have

T

a b c

a a (a, b, c) S S b = (a, b, c) b c c The matrix S is xed by the transformation, while the vector (a, b, c) is arbitrary. We thus have ST S = I i.e., S is an orthogonal matrix.

33

D.

Transformation of operations

Operation F changes vector |R into |T = F |R , and this relation should be maintained under a basis transformation. In basis 1, this change is expressed by |T or more explicitly a b c In basis 2, it should read |T or
x

= Fx |R

(6)

F F F xx xy xz = Fyx Fyy Fyz Fzx Fzy Fzz = F |R

a b c

F F F = F F F F F F

Acting the matrix for the basis transformation S onto both sides of equation (6), we have S |T
x

= SFx S T S |R = |T

= F |R

we immediately know F = SFx S T or F F F F F F F F F

V.

F F F xx xy xz Fyx Fyy Fyz Fzx Fzy Fzz

S S S x y z = Sz Sy Sz Sx Sy Sz

S S S x z x Sy Sy Sy Sz Sz Sz

MATRIX FORMULATION OF QUANTUM MECHANICS - REPRESENTA-

TION THEORY A. The idea of representation

Till now, we are used to express the state of micro-system by wave functions (r), and to express the operators by some mathematical terms p x = i 34
, x

etc. Is this the unique

way to express the quantum phenomena, quantum rules and laws? In other words, whether all our knowledge of quantum mechanics could only be expressed through wave functions (r)s and operators like p x = i
? x x

The answer is no. (r)s and operators like p x = i

are only one way (style) to

express (represent) the quantum theory. They are called coordinate representation (xrepresentation) of quantum mechanics. We have others ways to represent quantum theory. We have already familiarized with the Fourier transformation of (r), i.e., (p), which can describe the state precisely just as (r) do. In the (p) frame, operators are expressed by p x = px , x = i
. px

(p) and operators like x = i

px

could also express the quan x

tum theory eectively and completely just as (r) and p x = i theory.

do (perhaps not so

conveniently). They are called momentum representation ( p-representation) of quantum There are other representations that we are going to discuss in this section.

B.

Representation of state

be an Hermitian operator, its eigenfunctions {un (x)} is a complete orthonormal Let Q set. Quantum theory can be expressed through {un (x)} set and this formulation is called Q-representation. Premise: {un (x)} set is known and forms a complete orthonormal set. Any wavefunction (x) can be expanded upon {un (x)} set (x) =
n

cn un (x)

with cn = u n (x ) (x ) dx .

State function (x) and the set of coecients {cn } are uniquely connected. That means, given the wave function (x), the coecients {cn } are completely known. On the other hand, knowing the coecients {cn }, the state function is uniquely determined. The coecients {cn } uniquely determine the state of the system just as the wavefunction (x) does. The coecients {cn } is called Q-representation of the state. Formally the Q-representation is expressed as a column matrix, i.e. the N -tuple of the

35

component of a vector {cn } , with respect to a specied orthonormal basis c1 c2 = c3 . . . the Hermitian conjugate of is denoted by = (c , c , c , ) = 1 2 3 and normalization of (x) (x) (x) dx =
m c m um (x) n

cn un (x) dx =
n

c n cn = 1

is expressed in Q-representation

c1 c2 = (c 1 , c 2 , c 3 , ) = 1. c3 . . . Similarly inner product of and another vector a 1 a2 an un (x) = , a3 . . .

=
n

generalizing the dot product in 3D, is the complex number a1 a2 dx = = (c1 , c2 , c3 , ) = a3 . . .

c n an .
n

The set of all square integrable functions constitutes a vector space, which physicists call it Hilbert space after David Hilbert, who studied linear spaces in innite dimensions. In quantum mechanics, then, Wave functions live in Hilbert Space. 36

C.

Representation of operators

In Q-representation, the states are expressed by column (row) matrix, how would an be expressed in Q-representation? operators F upon a certain wave function changes it into another The action of an operators F function (x) . (x) . (x) = F Both of them can be expanded upon the complete set {un (x)} in Q-representation (x) =
n

bn un (x) ,

(x) =
n

an un (x)

we thus have bn un (x) = F

n n

an un (x)

Multiplying u m (x) and integrating

+

u m (x)
n

bn un (x) = F
n

an un (x) dx

by means of the orthogonal properties of the eigenfunctions, we get

bm =
n

u m (x) F un (x) an dx =
n

Fmn an .

In matrix form, it is

b F F F a 2 2 21 22 23 = b3 F31 F32 F33 a3 or = F . is expressed by a square matrix The operator F F11 F12 F F 21 22 F = F31 F32 in Q-representation F13 F23 F33

b1

F11 F12 F13

a1

with the elements Fmn s dened by following integration Fmn = u m (x) F un (x) dx. 37

D.

Property of the matrix of a Hermitian operator

The matrix element Fmn and its conjugate

= Fmn

un um F

dx =

un F

um dx =

u n F um dx

satisfy
Fmn = Fnm

is a Hermitian matrix Thus the matrix F representing a Hermitian operator F F = F. Indeed F11 F 21 F = F31 F11 F 12 = F 13 F11 F12 F13 F F F F22 F23 21 22 23 = F31 F32 F33 F32 F33 F11 F12 F13 F31 F21 F F F F22 F32 21 22 23 =F = F23 F33 F31 F32 F33 F12 F13

Example 17 As an example, let us try to nd the matrix form of angular momentum y operator L y = i cos + i cot sin L 2, L z We choose the common eigenfunctions of L as (in this case we have m = 1, 0, 1) 3 sin ei 8 3 u 2 : Y1,0 = cos 4 3 u 3 : Y1, 1 = sin ei 8 u 1 : Y1,1 = for l = 1 as our basis and denote them

38

 y is accordingly a 3 3 matrix. First let us operate L y upon Y1,1 The matrix of L y u1 = L y Y 1 ,1 = i L =i 3 cos cos ei i 8 3 cot sin sin iei 8

3 cos (cos i sin ) ei 8 3 i i cos = Y1,0 = 2 4 2 we have u 1 Ly u1 d = u 2 Ly u1 d = u 3 Ly u1 d = y Y1,0 gives The result of L y u2 = L y Y1,0 = i cos L =i i = 3 sin 4 3 ( sin ) = i 4 ei + ei 2 3 sin ei 8 3 sin cos 4 Y1 ,1 Ly Y1,1 (sin dd) = 0 i Y1 ,0 Ly Y1,1 (sin dd) = 2

Y1 ,1 Ly Y1,1 (sin dd) = 0

i 3 sin ei + 2 8 2 i i = Y 1 ,1 + Y 1 , 1 . 2 2

and u 1 Ly u2 d = u 2 Ly u2 d = u 3 Ly u2 d = y is In summary the matrix form of L 0 i Ly = 2 0

i 2

i Y1 ,1 Ly Y1,0 (sin dd) = , 2 Y1 ,1 Ly Y1,0 (sin dd) = 0 i Y1 ,1 Ly Y1,0 (sin dd) = , 2

0
i 2

0 i . 2 0

39

Operator in its own-representation takes a diagonal form with diagonal elements its eigenvalues. Suppose n (x) = qn un (x) Qu then the matrix element Qmn = = qn u m (x) Qun (x) dx u m (x) un (x) dx

= qn mn which means

0 q 0 2 Q= . 0 0 q3 In the above example, we nd Lz = 0 0 0 0 0

E.

q1

0 0

2
2

0 0

2 2 , L = 0 2 0 . 2 0 0 2

Matrix formulation of rules of quantum mechanics

Here we state the rules of quantum mechanics in matrix formulation. First, in Qrepresentation =
n

cn un (x)

the expectation value of observable F F = =

m

dx F
c m um (x) F n

cn un (x) dx

=
mn

c m Fmn cn

40

This is expressed in matrix form F F F c 2 21 22 23 (c = F , 1 , c2 , c3 , ) F31 F32 F33 c3 In short the expectation value of an observable F is F = dx = F . F F11 F12 F13 c1

in the language of Secondly, we may formulate the eigenvalue equation of operator F matrix. Starting from the = F we multiply u m (x) and do the integration u m (x) F
n

cn un (x) =
n

cn un (x) dx

immediately we have cn or
n

u m F un dx = cm

Fmn cn = cm
n

In matrix form it reads

c F F F c 2 2 21 22 23 . = c3 F31 F32 F33 c3 is now expressed to nd the eigenvalues and eigenvectors Thus, eigenequation of operator F of an Hermitian matrix. An example that nds large amount of utilization is the case that F is a square matrix of nite rows and columns F11 F 21 Fk1

F11 F12 F13

c1

c1

c c F22 F2k 2 2 = , Fk2 Fkk ck ck 41

F12 F1k

c1

c1

F11 F21 Fk1

F12 Fk2

F1k F2k = 0,

F22

Fkk

which gives an algebraic equation of of k -th power, and by solving it, we can nd the eigenvalues of F to be the roots of the algebraic equation 1 , 2 , k .

VI.

UNITARY TRANSFORMATION BETWEEN TWO REPRESENTATIONS

In the section Linear Space, we introduced the basic idea of linear algebra, including basis, operation, and the transformation of them. The vectors transform as |R while the operations change according to F = SFx S T . In quantum mechanics, the eigenfunctions of Hermitian operators play the role of basis in the function space. It is naturally to consider the generalization of 3D vectors in two directions as we mentioned earlier: (1) We allow the components of the vector to be complex in order to represent the wavefunctions. This turns the transformation matrix S , which is orthogonal for 3D (real) vector space, into a unitary matrix. The transformation is therefore a unitary one, and (2) instead of working in 3D, our vectors live in spaces of innite dimension. We will show in this section a similar transformation of eigenvectors and operators arises naturally in quantum mechanics.

= S |R

A.

Two sets of basis and their relations

and B . Both of them have their own complete Suppose we have two hermitian operators A orthonormal eigenfunction sets A, B, {un (x)} { (x)} 42 u1 (x), u2 (x), un (x) 1 (x), 2 (x), (x)

 representation and B representation. The and form two independent representations, A relation between them can be established through the relation of the two basis sets {un (x)} and { (x)}. Here we use a slightly dierent method to introduce the transformation matrix (actually it is more often used in linear algebra textbook), that is, we obtain the matrix element from the transformation of basis vectors. The new basis vectors in 3D are given as linear combinations of the old ones. Likely, the eigenfunction in the set {un (x)} can be expanded upon { (x)} un (x) =

Sn (x)

(7)

where the coecients are inner product of two of the basis vectors (x)un (x)dx = (x)

Sn (x)dx = Sn

The elements of the transformation matrix are thus Sn = ( , un ) to representation B . The orthoThis matrix changes the basis vector in representation A normality of the basis vectors gives u m (x)un (x)dx =
Sm (x) Sm Sn =

Sn (x)dx =
mn

Sm Sn

= i.e.

Sn = S S

= mn

S S = I. Here S is the hermitian conjugate of S and I is unit matrix. It is too early to say the matrix S is unitary - we need prove SS = I (one may think it trivial, however the situation does occur that SS = S S ). For this purpose we calculate
Sn Sn = n n

Sn S

= SS

=
n

(x)un (x)dx

(x )u n (x )dx

(8)

43

On the other hand, the eigenfunction in the set { (x)} can be expanded upon {un (x)} (x ) =
m

cm um (x )

with cm =
u m (x ) (x )dx = (um , ) = Sm = S m

Inserting back to the right side of (8), we have SS

=
n

(x)un (x)dx
m

cm um (x )u n (x )dx cm mn
m

=
n

(x)un (x)dx (x)

n

= i.e.

cn un (x)dx =

(x) (x)dx =

SS = I. The inverse matrix of S is thus S 1 = S The matrix satisfying this is called unitary matrix, the corresponding transformation is called unitary transformation. We conclude that the transformation from one representation to another is unitary.

B.

to B Transformation of state from representation A

We now check the transformation of the state between two representations. An arbitrary representation expression as follows state function (x) has its A (x) = n an un (x) a 1 (A) a2 = a3 . . .

44

 , the same state is depicted as In representation B

(x) = b (x) b 1 (B ) b2 = b3 . . .

, we want to know the expression of in representation Now, starting from representation A . By means of equation (7) B (x) =
n

an un (x) =
n

an

Sn (x) b (x)

=
n

(Sn an ) (x) =

which gives the relation of representation of states in dierent representations b =

n

Sn an

B = SA or, in the matrix form b S S S a 2 21 22 23 2 = b3 S31 S32 S33 a3 , we know the expression of in representation A Inversely, starting from representation B an =
Sn b

b1

S11 S12 S13

a1

A = S B or a1
S31 S21 S11

a S S S b 2 2 12 22 32 = a3 S S S b 3 13 23 33 45

b1

We notice that the matrix elements are now complex (wavefunction is in nature complex) and the matrix itself is unitary (in ordinary 3D vector space it is real and orthonormal).

C.

Transformation of operators

, an operator F is expressed by a square matrix In representation A F11 F12 F13 F F F 21 22 23 FA = F31 F32 F33 with the elements Fmn = the matrix is In representation B F F F 21 22 23 FB = F F F 31 32 33 and the elements are dened as F = Inserting back the expansion (x) =
n

u m (x) F un (x) dx.

F11 F12 F13

(x) F (x) dx.

un (x) =
n

Sn un (x)

(x) =
m

Sm u m (x)

we have F =
m

Sm u m (x)F
n

Sn un (x)dx

=
mn

Sm

u m (x)F un (x)dx Sn

=
mn

Sm Fmn S

46

The operator then changes according to FB = SFA S = SFA S 1 which is reminiscent of the transformation of operation in 3D space, in that case the transformation matrix S is orthonormal. We immediately have the transformation from repre to representation A sentation B FA = S FB S = S 1 FB S Example 18 Suppose we have a miscrosystem with orbital angular momentum with eigen 2, L z as representation A value 1(1 + 1) 2 . We choose the common eigenfunctions of L u1 = Y1,1 (, ) u2 = Y1,0 (, ) u3 = Y1,1 (, ) y in this representation is a 3 3 matrix The operator L i 0 2 0 i i Ly = 0 2 2 i 0 0 2 is chosen as the common eigenfunctions of L 2, L y . We now try to The representation B y in representation B . Obviously, we should have already know nd the matrix form of L y in its own representation is a diagonal matrix with the diagonal elements its the result - L eigenvalues. y in A representation. Eigenvalues Solution 19 Step 1. Solve the eigenvalue problem of L y are determined by of L
i 0 2 i 2

0 =0 0

i 0 2 i 2

0 which gives

i2 2 i2 2 =0 2 2
3

3 The three eigenvalues are , 0, respectively. 47

=0

 The eigenvector corresponding to is obtained by solving the equations 0 1 0 a a i 1 0 1 b = b 2 0 1 0 c c which are reduced to i b=a 2 i (a c) = b 2 i =c 2 with solutions a = c, The normalization of the eigenvector gives aa + bb + cc = 1 or aa + 2aa + aa = 1 aa = thus 1 a = ei 2 We take = 0 i 1 1 a = ,b = ,c = 2 2 2 and obtain the eigenvector 1 =
1 2 i 2 1 2

b=

2ia

1 4

The equations corresponding to eigenvalue 0 are 0 1 0 a a i 1 0 1 b = 0 b 2 0 1 0 c c 48

So b=0 ac=0 The normalization gives aa + bb + cc = 1 2aa = 1 We thus have 1 a = ei 2 again we take = 0, and 1 a=c= 2 The corresponding eigenvector is
1 2

2 = 0
1 2

For the eigenvalue 0 1 0 a i 1 0 1 b = 2 0 1 0 c We have i b = a 2 i (a c) = b 2 i = c 2 It is easy to repeat the procedure above and nd the eigenvector 2 i 3 = 2 1 2 49
1

a b c

 the Step 2. Determine the new basis and transformation coecients. In representation B new basis are chosen as 1 , 2 , 3 , and they can be expanded as (x) =
1 = S11 u1 + S21 u2 + S31 u3 n

un (x)

1 1 i = u 1 + u2 u 3 2 2 2 We easily obtain the transformation matrix 1 1 1 S S S 2 2 11 12 13 2 i i S = S21 S22 S23 = 0 2 2 1 1 1 S31 S32 S33 2 2 2

which is just writing the three basis to form one square matrix with its hermitian conjugate 1 i 1 2 2 2 1 1 . S= 0 2 2 1 i 1 2 2 2 Step 3. Running the transformation through. The states change as B = SA the eigenvectors are as expected so in representation B 1 1 1 i 2 2 2 2 1 1 1 (1 )B = i2 = 0 0 2 2 1 i 1 1 0 2 2 2 2 1 i 1 1 0 2 2 2 2 1 1 (2 )B = 0 = 1 0 2 2 1 i 1 1 2 0 2 2 2 i 1 1 1 0 2 2 2 2 1 1 i (3 )B = =0 0 2 2 2 1 i 1 1 2 1 2 2 2

50

 y transforms as And the operator L y )B = S (L y )A S 1 (L 1 i 1 2 2 0 2 1 1 i = 0 2 2 2 1 i 1 0 2 2 2 1 i 1 2 2 2 2 1 1 = i2 0 2 2 1 i 1 2 2 2 2

i 2

1 2 i 2 1 2 1 2 1 2 i 0 2 1 1 2 2

0
i 2

0
i 2

0 2 0 0 i =0 0 0 0 2 0 2 0 0

We summarize the result of representation transformation in the following table. : {un (x)} Representation B : { (x)} Representation A a1 b1 a b 2 2 A = , a = (uk , ) B = , b = ( , ) a3 k b3 F11 F12 F11 F12 FA = F21 F22 , FB = F21 F22 , un ) Fmn = (um , F State transformation Operator transformation Transformation matrix A = S B FA = S FB S = S 1 FB S S11 S12 S = S21 S22 , . . .. . . . . .
Sn = (u n , )

State

Operator F

) F = ( , F B = SA FB = SFA S = SFA S 1 S11 S12 S = S21 S22 , Sn = ( , un )

Problem 20 Zeng Jinyan, textbook, page 133, 15.

VII.

DIRACS NOTATION

The theory of linear vector spaces had, of course, been known to mathematicians prior to the birth of quantum mechanics, but Diracs way of introducing vector spaces has many 51

advantages, especially from the physicists point of view. Like a vector in 3D lives out there in space, independent of anybodys choice of coordinates, the state of a system in quantum mechanics is represented by a vector, that lives out there in Hilbert space. We can express the wavefunction by means of (x) in position space, (p) in momentum space, -representation with bases {un (x)}. But its all the same state, (x), (p), and or cn in Q cn contain exactly the same information.

A.

States

Detached from concrete representations, Diracs notation for a physical state represented by a state vector in a complex vector space is called a ket and denoted by | . The vector space is a ket space. We can also introduce the notion of a bra space, a vector space dual to the ket space. We postulate that corresponding to every ket | there exists a bra, denoted -representation the ket vector | is a column matrix by |, in this dual, or bra, space. In Q a 2 | = a3 while the bra vector is a row matrix (the hermitian conjugate of the column matrix)
| = ( a 1 , a2 , a3 , )

a1

The inner product of a bra and a ket is written as a bra standing on the left and a ket standing on the right, for example, | = ( |) (| )
bra (c) ket

This product is, in general, a complex number

| = b 1 a1 + b2 a2 + b3 a3 + bn an + = n

b n an

Evidently | and | are complex conjugates of each other | = | 52

We immediately deduce that | must be a real number. Furthermore | 0, where the equality sign holds only if | is a null ket. This is sometimes known as the postulate of positive denite metric, which is essential for the probabilistic interpretation of wavefunctions. Two kets | and | are said to be orthogonal if | = 0 Given a ket which is not a null ket, we can form a normalized ket | , where | = with the property | =1 Quite generally, | is known as the norm of | , analogous to the magnitude of a vector 1 | |

in Euclidean vector space.

B.

Rules

In Diracs notation, the basic rules of quantum mechanics look much simpler. Suppose we have discrete eigenvalue spectrum {un (x)} or in Diracs notation {|n }. The orthonormality reads m|n = mn The completeness of the spectrum means that any ket | can be expanded as | =
n

cn |n

where cn = n| Inserting this back we nd | =

n

|n n|

so |n n| = 1
n

(9)

53

which is the property of completeness in Diracs notation. For continuous spectrum {uz (x)} (or {|z }) the orthonormality and completeness are expressed as z |z = (z z ) dz |z z | = 1 is written as The eigen-problem of operator F |n = fn |n F The time-dependent Schr odinger equation takes the form of i whereas the stationary equation is |n = En |n . H As an example of application of the completeness, we notice that equation (9) is of ten inserted into appropriate position to simplify the calculation. Suppose the operator F transforms ket | into | | | = F Inserting the completeness of {|n } into the above equation twice |n n| =
n n

d | | = H dt

|n n| F

Multiply the bra m| from the left and notice m|n = mn , we have m| =
n

|n n| m| F

|n is the Dirac notation of the matrix elements Fmn . Here m| F

C.

|n

The occupation-number representation of the simple harmonic oscillator is very useful in representing quantum theory. We change the notation for raising and lowering operators 54

into a and a , which are known as the creation and annihilation operator for a particle with energy unit . Let |n be the eigenvector of Hamiltonian 1 = a H a + 2 with eigenvalue En = n+ 1 2 +1 = N 2

The physical meaning of |n is thus there are n particles in the state |n and this representation is known as the occupation-number representation. The operator a ( a ) annihilate (create) one particle in the state a |n = a |n = For the ground state |0 a |0 = 0 so the rst excited state can be generated from it |1 = a |0 and we can generate any state |n from |0 1 |n = a n!
n

n |n 1 n + 1 |n + 1

|0

Simple harmonic oscillator provides us an example of representing wavefunctions and operators. The basis in this example is the eigenvectors {|n } and the matrix elements of operators are given by |n Fmn = m| F For example, the matrix elements for the annihilation operator are n |a |n = while for creation operator n |a |n = nn ,n1

n + 1n ,n+1

55

In matrix form they are 0 a= 0 0 0 1 a = 0 0 =a and the number operator N a is 0 1 0 0 0 2 0 0 0 3 0 0 0 0 0 0

0 0 0 2 0 0 0 3 0

0 1 0 0 N = 0 0 2 0 0 0 0 3
D. Application

We discuss here a crude model for (among other things) neutrino oscillations. Example 21 Imagine a system in which there are just two linearly independent states: 1 0 |1 = , and |2 = 0 1 The most general state is a normalized linear combination a | = a |1 + b |2 = , with a2 + b2 = 1. b The Hamiltonian can be expressed as a hermitian matrix; suppose it has the specic form h g H= g h 56

where g and h are real constants. If the system starts out (at t = 0) in state |1 , what is its state at time t? Solution 22 The time-dependent Schr odinger equation says i d | | = H dt

as always, we begin by solving the time-independent Schr odinger equation | = E | H . The characteristic equation that is, we look for the eigenvectors and eigenvalues of H determines the eigenvalues hE g g hE = 0 (h E )2 g 2 = 0 h E = g E = h g

Evidently the allowed energies are (h + g ) and (h g ). To determine the eigenvectors, we write h g g h h + g = (h g ) =

= (h g )

so the normalized eigenvectors are

1 | = 2 1 Next we expand the initial state as a linear combination of eigenvectors of the Hamiltonian 1 1 |(0) = = (|+ + | ) 2 0 Finally we tack on the standard time-dependence exp (iEn t/ ) 1 |(t) = ei(h+g)t/ |+ + ei(hg)t/ | 2 1 1 1 = ei(h+g)t/ + ei(hg)t/ 2 1 1 1 iht/ eigt/ + eigt/ = e 2 eigt/ eigt/ cos ( gt/ ) = eiht/ i sin (gt/ ) 57

In the model of neutrino oscillation |1 represents the electron neutrino, and |2 the muon neutrino; if the Hamiltonian has a non-vanishing o-diagonal term g , then in the course of time the electron neutrino will turn into a muon neutrino, and back again. In 2002 Nobel Prize in Physics was awarded with one half jointly to: Raymond Davis Jr, Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, USA, and Masatoshi Koshiba, International Center for Elementary Particle Physics, University of Tokyo, Japan, for pioneering contributions to astrophysics, in particular for the detection of cosmic neutrinos. In order to increase sensitivity to cosmic neutrinos, Koshiba constructed a larger detector, Super Kamiokande, which came into operation in 1996. This experiment has recently observed eects of neutrinos produced within the atmosphere, indicating a completely new phenomenon, neutrino oscillations, in which one kind of neutrino can change to another type. This implies that neutrinos have a non-zero mass, which is of great signicance for the Standard Model of elementary particles and also for the role that neutrinos play in the universe.

E.

Projection operators

The license to treat bras as separate entities in their own right allows for some powerful and pretty notation. For example, if | is a normalized vector, the operator | | P picks out the component of any other vector that lies along | | = | | P we call it the projection operator onto the one-dimensional subspace spanned by | . It is 2 = P easy to show that projection operators are idempotent: P 2 | = P P | P | | = | =P | P | = | | | = | | = P

To say two operators are equal means that they have the same eect on all vectors. Since 2 | = P | for any vector | , P 2 = P . P . If | is an eigenvector of One may also determine the eigenvalues and eigenvectors of P with eigenvalue , then P | = | P 58

and it follows that 2 | = P | = 2 | P 2 = P , and | = 0, so 2 = , and hence the eigenvalues of P are 0 and 1. Any But P , with eigenvalue 1; any vector orthogonal (complex) multiple of | is an eigenvector of P , with eigenvalue 0. to | is an eigenvector of P be an operator with a complete set of orthonormal eigenvectors Q |n = qn |n . We Let Q can be written in terms of its spectral decomposition can show that Q = Q
n

qn |n n|

An operator is characterized by its action on all possible vectors, so what we must show is that for any vector | = Q
n

qn |n n| |

Indeed we expand | as | = | = Q
n

n cn

|n with cn = n| , n| qn |n =
n n

n |n = Qc

qn |n n| |

so = Q
n

qn |n n| .

Problem 23 Griths, page 123, 3.22, 3.23, 3.26, 3.35 (example), 3.37.

VIII.

CONSERVATION LAW AND SYMMETRY

In classical mechanics, these exist many conservation laws of dynamical variables, such as energy conservation, momentum conservation, etc. These conservation laws are closely connected with symmetries of the system. In quantum mechanics, the corresponding conservation law still hold and the relation between conservations and symmetries remains the same.

59

A.

Constant of motion

Let us rst check how the expectation value of operator changes with time. Consider the mean value of a certain dynamical variable F F = d3 r F

on a state (r, t) which satises the time dependent Schr odinger equation i The time derivative is calculated as d d F = dt dt = = = d3 r F t

=H t

d3 r + F d r +
3

F t

d3 r +

d3 r

F t

1 H i d3 r

d3 r + F

1 H d3 r i

1 F + F,H t i

satisfy the operator relation If operator F 1 F + F,H = 0 t i is called a constant of motion. Generally, the operator doesnt Then d F /dt = 0 and F /t = 0, so the criterion for a dynamical variable to be a constant of motion rely on time, F is , H =0 F , H = 0, the expectation value of F on any state does not depend on time. That is, if F is indeFurthermore, we can show the probability distribution of nding eigenvalues of F and H (this is possible pendent of time. We can choose the common eigenfunction set of F because they commute each other) {un, } n, = En un, Hu un, = f un, . F 60

as our complete and orthonormal basis, with

3 u n, um, d r = mn

Thus any state can be expanded upon them (r) =

n,

cn, un, cn, un, e

n,
i

(r, t) =

En t

on state (r, t) is and the expectation value of F F = =

n,

(r, t)d3 r (r, t)F

c n, un, e
i

En t m,

f cm, um, e cn, f

2

Em t 3

dr

=
n,

c c f = n, n,
n,

The probability of a measurement giving result f , that is cn, , is therefore independent of time for given n.

B.

Several typical cases of constant of motion

(a) Momentum conservation: If the space is homogeneous (free particle), or the particle feels no potential, the momentum is conserved = 1 p H 2 2m obviously = 0. p , H

(b) Angular momentum conservation: If the space is isotropic (central force eld), e.g. the particle feels the same potential in any direction = 1 p 2 + V (r) H 2m 2 2 1 2 L = ( r ) + + V (r) 2m r2 r r 2mr2 61

with 2 = L
2

1 1 2 (sin ) + sin sin2 2

The angular momentum operator and its components contain only and , and has nothing to do with r. Thus L 2 = 0 H, and L x = H, L y = H, L z = 0 H, i.e. the angular momentum is conserved, including its three components. doesnt depend on t, then evidently (c) Energy conservation: If the Hamiltonian itself H H =0 H, which is the conservation of energy in quantum mechanics. (d) Parity conservation: Dene an operation of space inversion r r (r, t) = (r, t) P is known as the parity operator. We have P 2 (r, t) = P (r, t) = (r, t) P 2 is 1 and the eigenvalues for P is 1. So i.e. the eigenvalues of P 1 = 1 P 2 = 2 P The eigenfunction 1 (2 ) is known as the even (odd) parity state of the system. If the Hamiltonian is unchanged under space inversion (r) = H (r) H then H (r, t) = H (r)P (r, t) P (r)P (r, t) =H 62

i.e. H =H P P and H share common eigenfunction Thus the parity is a constant of motion and P set, which means that the eigenfunctions of the system have denite parity and do not change with time. This is the conservation of parity in quantum mechanics. Constant of motion is an important point in quantum mechanics. It should not be confused with the concept of stationary state. We emphasize here the distinction between . It is the result of them. Explicitly, the stationary state is a particular eigenstate of H separation of variables from the Schr odinger equation i which gives the stationary equation (r) = E (r) H and the time-dependent factor i and (r, t) = (r)e
i

(r, t) (r, t) =H t

d f (t) = Ef (t) dt
Et

(constant of motion or not) on a staThe expectation value of any dynamical variable F tionary state F = = does not change with time. ( (r)e
i

Et

( (r)e i Et )d3 r )F

(r)d3 r (r)F

On the other hand, the constant of motion is a that commutes with H particular observable represented by an operator F , H = 0. F The expectation value of a constant of motion on any state (stationary or not) F = (r, t)d3 r (r, t)F 63

does not change with time. Here is an additional notes on expectation value of conserved quantities. The conserved quantities do not necessarily take denite values, unless the system is in one of the eigenstates of that operator. For example, in the case of free particles, the constant of motion or the conserved quantity is the momentum p, whose eigenstates are plane waves eipx/ . The state of a free particle is generally a wave packet, i.e. the superposition of plane waves. The momentum does not take denite value - there is a probability distribution. Another example is the particle in central force eld. The constant of motion or conserved quantity is the angular momentum L 2 Ylm (, ) = l(l + 1) 2 Ylm (, ) L z Ylm (, ) = m Ylm (, ) L The system may be in a state like = c1 Y11 (, )+ c2 Y20 (, ), for example or more generally =
lm

clm Ylm (, )

2 and L z at all. which is not an eigenstate of L Generally the initial condition determines whether the system would stay in the eigenstate of the conserved quantity at time t. If the system is prepared initially (t = 0) in k , one of , it will stay there forever. On the contrary, or more the eigenstate of conserved quantity F ) practically, if the system is initially in a combination state (not an eigenstate of F (r, 0) =
k

ak (0)k

at time t it is (r, t) =
k

ak (t)k

and the probability of nding the particle in one of the eigenstates does not change with time d |ak (t)|2 = 0 or |ak (t)|2 = |ak (0)|2 dt are known as good quantum numbers. The quantum numbers of conserved quantity F

64

C.

Conservation law and symmetry

Noethers theorem is an amazing result which lets physicists get conserved quantities from symmetries of the laws of nature. Time translation symmetry gives conservation of energy; space translation symmetry gives conservation of momentum; rotation symmetry gives conservation of angular momentum, and so on. This result, proved in 1915 by Emmy Noether shortly after she rst arrived in G ottingen, was praised by Einstein as a piece of penetrating mathematical thinking. Its now a standard workhorse in theoretical physics. If someone claims Noethers theorem says every symmetry gives a conserved quantity, they are telling a half-truth. The theorem only applies to certain classes of theories. In its original version it applies to theories described by a Lagrangian, and the Lagrangian formalism does most of the work in proving the theorem. There is also a version which applies to theories described by a Hamiltonian. Luckily, almost all the theories studied in physics are described by both a Lagrangian and a Hamiltonian.

1.

Translation invariance and momentum conservation

A translation operation is dened as a displacement of the system bodily along x-direction by an innitesimal segment a so that x x = x + a. Suppose (x) is the original wave (a) is the translation operator, then function, (x) is the translated one, D (a) (x) (x) = D Evidently (x ) = (x) i.e. (a) (x + a) = (x) D Replacing x with x a we have
2 (a) (x) = (x a) = (x) a d + 1 (a)2 d + D dx 2! dx2 n d 1 d = (a)n n (x) = ea dx (x) n! dx n=0

65

The translation operator along x by a is thus expressed as

d i x (a) = ea dx D = e h ap

where the innitesimal operator is nothing but the momentum operator p x = ih d dx

1 p2 2m

= Suppose the space is homogeneous, the Hamiltonian H (a) tion D (x) = H D (x) H D that is

is invariant by a transla-

H =H D D, H =0 p =0 D x , H which is exactly the condition for conservation of momentum. We conclude that homogeneity means conservation of momentum.
i ap x (a) = e h It is easy to show translation operator D is unitary. Indeed, the Hermitian (a) is conjugate of D i ap x (a)1 = D (a) (a) = e h =D D

so (a)D (a) = 1 D (a) is thus a unitary transformation. Operators are transformed Translation realized by D as =D (a)F D (a) . F Generalization to 3D is straightforward.

2.

Rotation invariance and orbital angular momentum conservation

Rotate the system bodily around z -axis by an angle , the wavefunction (r, , ) changes as () (r, , ) (r, , ) = R Similarly (r, , ) = (r, , ) () (r, , + ) = (r, , ) R 66

Replacing with we have () (r, , ) = (r, , ) R = (r, , ) + ()

1 2 (r, , ) + ()2 2 (r, , ) + 2!

=
n=0

1 ( )n (r, , ) n!

= e (r, , ) The rotation operator around z -axis by is thus expressed as

i () = e R = e Lz

where the innitesimal operator is the z -component of the angular momentum operator z = ih L Now we consider an innitesimal rotation around an arbitrary axis n by R(n, ) (r) = (r = (r)

r) y (r) y z (r) + z

(r) x

=
n=0

1 ( n!

r )n (r)

As Figure 2 shows,

r = (n r)

R(n, ) (r) =
n=0

1 ()n ((n r) )n (r) n!

By means of the relation (A B) C = A (B C) we know (n r) = n (r ) so

R(n, ) (r) =
n=0

1 [()n (r )]n (r) n! 1 i n (r p) n!

) (nL n

=
n=0

(r)

= e

(r)

67

r 0

r+ r

FIG. 2: A rotation around an arbitrary axis n by .

The rotation operator around an arbitrary axis n by is thus expressed as R(n, ) = e More explicitly n x0 , R(x0 , ) = e n y0 , R(y0 , ) = e n z0 , R(z0 , ) = e
i i

) (nL

x L y L

z L

Suppose the Hamiltonian is isotropic or the potential has spherical symmetry (central force eld) = 1 p 2 + V (r) H 2m

68

we thus have = R(y0 , ), H = R(z0 , ), H =0 R(x0 , ), H x, H = L y, H = L z, H =0 L x, L y, L z are conservative, so is operator L 2 = L 2 + L 2 + L 2 . Isotropy thus means i.e. L x y z the conservation of momentum. In cases the potential V is invariant under rotation around certain direction, the component of angular momentum on that direction is conservative.
1 1 2 2 2 For example, in the case of a cylindric symmetry potential V = 2 m (x2 + y 2 ) + 2 mz z , z is conservative. L

Rotation operator is also a unitary operator and transformation accomplished by such an operator is an unitary transformation. Indeed R(n, ) = e
i

) (nL

i R (n, ) = e (nL) = R(n, )

and wavefunctions and operators are transformed according to the following rule (r) (r) = R(n, ) (r) = e
i

) (nL

(r)

) ) F = R(n, )F R (n, ) = e i (nL e i (nL F F

3.

Conservation laws and symmetry in general

, which transforms into = Q , More generally, we consider a linear transformation Q the Schr odinger equation in invariant under this transformation, i.e. i We have i 1 gives Applying Q i then 1 H Q =H H Q =Q H Q, H =0 Q 69 1 H Q = H =Q t Q Q = H t and i = H =H t t

 forms a symmetric group and the conservation of probaThe symmetric transformation Q is unitary, i.e. Q Q = Q Q = I. bility requires that Q For symmetry transformations that dier innitesimally from the identity transformation I , we can write = I + iF Q means where is an innitesimal parameter. The unitary of Q = I iF Q Q I + iF

F + O(2 ) = I + i F =I so =F F is Hermitian operator, known as the Hermitian generator of Q . Then i.e. F , H = 0 = d F = 0 F dt corresponding to symmetry transformation In general, there exists a constant of motion F . Q H = 0 = Q, The following table summarizes the main result in this section. Symmetry Homogenous Space/Translational Invariance Isotropic Space/Rotational Invariance Space Inversion Invariance Time Translation Invariance
IX. SUMMARY ON PART IV

Conserved Quantity Momentum Angular Momentum Parity Energy

Two cornerstones in quantum mechanics are wavefunctions and operators. In wave mechanics, wavefunctions are mathematical functions representing wave nature of the microparticle, while operators are mathematical derivatives acting on the wavefunctions. There exists a special class of operators in quantum theory - Hermitian operators. They usually obey a non-commutative algebra and their eigenvalues are real, the eigenfunctions are orthonormal and complete. 70

We can nevertheless describe the theory by means of another totally dierent language - matrix. In matrix mechanics, wavefunctions are column or row matrices, whereas operators are square matrices. The eigenvalue problem is reduced to nd the eigenvalues and eigenfunctions of a Hermitian matrix. The completeness of the eigenvectors of a Hermitian operator enables us to represent the quantum theory using the eigenvectors as basis just like we use three basic vectors to express any vectors in 3D space. In this sense, wavefunctions live in Hilbert space, which is complex and innite dimensions. Moreover, we can jump between two dierent bases - the wavefunctions and operators are connected by a unitary matrix. Dirac invented much simpler notations to express quantum theory. In this new picture, states are distinguished as bras and kets. An especially useful representation is the occupation number representation of harmonic oscillator. on a wave function (x) the result could be For a measurement of physical quantity F , say fn . And probability of obtaining fn is |cn |2 . If nothing but one of the eigenvalues of F two operators are commutative, one can nd the common eigenfunctions set and measure them simultaneously. On the other hand, if they fail to commute, there always exists a limit for their variances - the uncertainty relation. These conservation of physical quantities are closely connected with the symmetries of corresponding system. According to Noethers theorem, there exists a constant of motion F . to symmetry transformation Q

71

 F ,G
0 0

f : , F n n n n m n mn

f n real
* n

Common eigenfunction set

Uncertainty relation

n

( x) n ( x ') x x '

on ( x) Measurement of F Result f n , Probability cn

2

Hermitian Operators
F , , F
1 2 1 2

(r )

( x, y, z , i , i , i ) F x y z

Wavefunctions
Hilbert Space: complex & infinite dimension

Operators

a1 a2

F11 F21
F11 F21

F12 F22

Representation Transformation

Hermitian Matrices

B S A
SF S F B A
Diracs Notation

F12

F22 0

and
n ,

n
n

n 1

F11 F21

F12 c1 c1 F22 c2 c2

Symmetry
Unitary Transformation Q

Conservation Law
Hermitian Constant of Motion F

FIG. 3: Summary on Part IV.

72