Theory of Wigner Distributions in Quantum Mechanics

March 8-14, 2015

Contents
1 Introduction: Kinematics of Cartesian Quantum Systems 1

2 The Weyl map and the Moyal Bracket 2

3 The Wigner distribution and its properties 5

4 Extension to many degrees of freedom 7

5 Wigner Distributions in finite dimensions- The spin 1/2 case 9

1 Introduction: Kinematics of Cartesian Quantum Systems

The basic ideas of this subject go back to Weyl and Wigner, and slightly later Moyal. Our discussion
will be largely at the kinematic level. We will use Diracs notation extensively.
To begin with, we discuss Cartesian systems in Quantum Mechanics, or Cartesian Quantum
Mechanics. This means systems whose basic dynamical variables are Cartesian positions and
conjugate Cartesian momenta. So spin is excluded. Later we will look at spin and other finite
dimensional quantum systems.
For one degree of freedom (dof), we have one canonical pair of hermitian operators obeying the
Heisenberg canonical commutation relation (CCR).

q = q; p = p; [q, p] = i~ (1.1)

In addition, they are assumed to be irreducible:

[q or p, A] = 0 A = aI, multiple of identity (1.2)

Informally this also means that every operator A pertaining to the system is expressible as a function
of q and p. The meaning and spirit of this will be clarified as we proceed.
Such operators q and p are unbounded, they can not be applied to all state vectors or wave
functions. They have definite domains on which alone they can act. To avoid such domain problems,
these CCR can be represented in the unitary Weyl form:

eiaq eibp = ei~ab eibp eiaq (1.3)

where a and b are real. Equivalently for all real q, p, q 0 , p0 :

i 0 0 i i 0 0 i 0 0
e ~ (p qq p) e ~ (pqqp) = e 2~ (p qq p) e ~ [(p +p)q(q +q)p] (1.4)

Here all these operators are unitary; i.e. they have no domain problems. The most familiar repre-
sentations of q and p are in terms of Schrodinger wave functions in position space, or wave functions

1
IISER Mohali N. Mukunda March 2015

in momentum space; in a Hilbert space:

( Z 2 )
 
H = (q)  dq | (q) <

( Z 2 )
 
= (p)  dp | (p) < ,

Z
1
(p) = dqeipq/~ (q);
2~
d(q)
(q)(q) = q(q), (p)(q) = i~ ;
dq
d(p)
(p)(p) = p(p), (q)(p) = i~ (1.5)
dp
Each of q and p can be applied only to wave functions in its domain. Formally eqs. (1) can be
verified.
We use Dirac notation and idealized eigen functions of q and p. Each of them has all real numbers
as possible eigenvalues:

q|qi = q|qi; hq 0 |qi = (q 0 q), < q 0 , q < ,

Z
dq|qihq| = 1 on H;

p|pi = p|pi; hp0 |pi = (p0 p), < p0 , p < ,
Z
dp|pihp| = 1 on H;

1
hq|pi = eiqp/~ (1.6)
2~
The wave functions (q), (p) in eqs.(1.5) are the following overlaps or inner products.

|i H : (q) = hq|i, (p) = hp|i (1.7)

Since both q and p have all real numbers as eigenvalues, it is meaningful to displace or shift them
by any c-numbers. For any real q and p, clearly q q and p p obey eqs. (1.1), so they must be
unitarily related to q and p respectively. This is achieved by so called displacement operators which
are just what appear in eqs. (1.4). So we define:

D(q, p) = ei(pqqp) /~, < q, p < ,

D(q, p) D(q, p) = 1 (1.8)

Their actions on operators and ket vectors are obtained quite easily:
i 0 0
D(q 0 , p0 )D(q, p) = e 2~ (p qq p) D(q 0 + q, p0 + p); (a)
1
D(q, p)(q or p)D(q, p) = q q or p p; (b)
i
(q 0 + 2q )p
D(q, p)|q 0 i = e ~ |q 0 + qi,
i 0 p
D(q, p)|p0 i = e ~
(p + 2 )q
|p0 + pi (c) (1.9)

2 The Weyl map and the Moyal Bracket

For a classical system with one Cartesian dof, dynamical variables are (real) functions f (q, p); i.e.,
functions over the two dimensional phase space plane. For the corresponding quantum system,
dynamical variables are (hermitian)operators F acting on the Hilbert space H in eqs. (1.5). As q

2
IISER Mohali N. Mukunda March 2015

and p are an irreducible pair, every F is in principle some function of q and p. The Weyl map is a
rule or a convention to set up a one-to-one correspondence between classical (real) functions f (q, p)
on the one hand, and (hermitian) operators F on the other.
To begin, consider finite degree polynomials in q and p. For linear expressions, we of course set

q q, p p, q p q p (2.1)

For higher degree expressions, we use the idea of symmetrised polynomials to allow for the fact
that q and p do not commute. The Weyl rule can be stated in several equivalent ways. For any non
negative integers m and n, we define the correspondence:
(m + n)! m
q m pn = coefficient of ( )n in (q p)m+n
m!n!
(m + n)! m
m pn ) =
(q\ coefficient of ( )n in ( q p)m+n (2.2)
m!n!
As examples, we have:

(q
d m ) = (q)m ;

1
\
(q m p) = (q m p + q m1 pq + q m2 pq 2 + ..... + pq m ); ......;
(m + 1)
(p
d n ) = (p)n (2.3)

This is formal as all these are unbounded operators. Nevertheless we can see that this rule does
map classical real f (q, p) to hermitian quantum F .
There are other ways to set up rules to go from classical f (q, p) to quantum F used in other
contexts, but for the purpose of the Wigner distribution, the Weyl rule is the appropriate one.
Incidentally the same rule eq. (2.2) can be expressed as follows: for any real and ,

ei(q p) ei(q p) D(~, ~) (2.4)

Based on this, we can use the methods of Fourier analysis to express the Weyl rule in yet another
way. Suppose a classical f (q, p) has the Fourier integral representation:
Z Z
1
f (q, p) = d d f(, )ei(q p) (2.5)
2
Then by eq. (2.4) and linearity, we have:
Z Z
1
f (q, p) F = d d f(, )ei(q p)
2
Z Z
1 p q
= 2
dqdpf( , )D(q, p) (2.6)
2~ ~ ~
Now we can see that, using many nice properties of the D(q, p), this can be inverted. Using first
eq. (1.9 c) and then eq. (1.9 a):
Z
T r[D(q, p)] = dq 0 hq 0 |D(q, p)|q 0 i

Z
i 0 q
= dq 0 hq 0 |q 0 + qie ~ (q + 2 )p

= 2~(q)(p); (a)

T r[D(q 0 , p0 ) D(q, p)] = T r[D(q 0 , p0 )D(q, p)]

i 0 0
= e 2~ (q pp q) T r[D(q q 0 , p p0 )]
= 2~(q q 0 )(p p0 ) (b) (2.7)

3
IISER Mohali N. Mukunda March 2015

This means that in the space of operators on H and with respect to the Hilbert Schmidt inner
product, the D(q, p) form an orthonormal basis. Incidentally, we can say eq. (2.6) is the quantum
operator form of the Fourier integral representation (2.5). Combining eqs. (2.6) and (2.7), we get:

1 p0 q 0
 
0 0
T r[D(q , p ) F ] = f , (2.8)
~ ~ ~

So via eq. (2.5), we arrive at the inverse to the Weyl map (2.6):
Z Z
1 0 0
f (q, p) = dq 0 dp0 ei(p qq p)/~ T r[D(q 0 , p0 ) F ] (2.9)
2~

Let now g(q, p) be another classical function with Fourier transform g(, ), mapped by the Weyl
rule to the operator G. Since by Fourier inversion we know how to pass from f to f and g to g and
eq (2.9) shows how to pass F f , G g, we can put it all together and can obtain the relation
Z Z
1
T r[G F ] = dqdpf (q, p)g(q, p) (2.10)
2~

Thus, square integrable functions over the classical phase plane are taken by the Weyl rule to
operators with finite Hilbert-Schmidt norm.
A somewhat surprising instance of the f (q, p) F correspondence will be found useful later on.
The choice
1
f (q, p) = (q)(p), i.e, f(, ) = (2.11)
2
leads to Z Z
1
F = dqdpD(q, p) (2.12)
(2~)2
What is this operator? If we apply it to |q 0 > and use eq(1.9c) we get
Z Z
1 0 q
F |q 0 > = 2
dqdpei(q + 2 )p/~ |q 0 + q >
(2~)
Z
1 q 1
= dq(q 0 + )|q 0 + q >= | q 0 >,
2~ 2 ~
1
i.e. F = P , P = Parity operator (2.13)
~
Thus
P
f (q, p) = (q)(p) F = (2.14)
~
where of course
P (q or p)P 1 = q or p, P 2 = I (2.15)

The Moyal Product and Bracket

Since the Weyl rule gives a one-to-one correspondence between classical phase space functions
and quantum operators, it follows that the associative law of non-commutative multiplication of
operators can be expressed in terms of the classical functions themselves. Introduce the notation:

W eyl
f (q, p) F = (f (q, p))w (2.16)

Then we want to find the structure of the * product in

(f (q, p))w (g(q, p))w = (f (q, p) g(q, p))w (2.17)

4
IISER Mohali N. Mukunda March 2015

For exponential functions, after relabelling variables, eq(1.4) says:

 00 qq 00 p)/~ 0 0
  00 q 0 q 00 p0 )/2~ 00 qq 00 p)/~ 0 0

ei(p )w (ei(p qq p)/~ = ei(p ei(p ei(p qq p)/~
w w
 
i(p00 qq 00 p)/~ i(p0 qq 0 p)/~
= e e ,
w
00 qq 00 p)/~ 0 0 i 00 q 0 q 00 p0 ) i (p00 qq 00 p) i (p0 qq 0 p)
ei(p ei(p qq p)/~ = (p
e 2~  e~ e~

i ~ ~
i~ i~ i(p00 q2 q 00 p2 )/~ i(p0 q1 q 0 p1 )/~
= e 2~ i q2 p1 p2 i q1
e e (2.18)

By linearity we can generalize to any f (q, p) and g(q, p):

 
i~ 
p
f (q, p) g(q, p) = e 2 q2 p1 2 q1 f (q2 , p2 )g(q1 , p1 ) (2.19)

q1 =q2 =q,p1 =p2 =p

This is the Moyal product or multiplication rule for classical phase space functions, an exact ren-
dering of multiplication of quantum operators. For the commutator, we have:

[(f (q, p))w , (g(q, p))w ] = (f (q, p) g(q, p) g(q, p) f (q, p))w ,

f (q, p) g(q, p) g(q, p) f (q, p)

"   #
i~
= exp ( ) (f (q2 , p2 )g(q1 , p1 ) g(q2 , p2 )f (q1 , p1 ))
2 q2 p1 p2 q1
q1 =q2 =q, p1 =p2 =p
 
~ 
= 2i sin ( ) f (q2 , p2 )g(q1 , p1 )

2 q2 p1 p2 q1 q1 =q2 =q, p1 =p2 =p

= i~ {f (q, p), g(q, p)} + O(~3 ) (2.20)

The leading term is essentially the classical PB {f, g}, discovered by Dirac in September 1925!

3 The Wigner distribution and its properties

The Weyl map f (q, p) F = (f (q, p)w ) is such that the physical dimensions of f and F are the
same. This is clear from eqs. (2.5, 2.6). Now let be the density matrix of a (pure or mixed) state
of the quantum system, obeying
= 0, T r() = 1 (3.1)
In the pure case,
= | >< |, < | >= 1 (3.2)
In the state the expectation value of any observable F is given by:

< F > = T r(F ) (3.3)

The idea of the Wigner distribution is to express this in the classical looking form familiar from
classical statistical mechanics. For any F = (f (q, p))w , we want a function W (q, p) to represent
so that Z Z
T r(F ) = dqdpW (q, p)f (q, p) (3.4)

The general result (2.10) tells us how to define W (q, p): setting G = there, which makes g(q, p)
real, W (q, p) must be such that
1
(W (q, p))w = (3.5)
2~

5
IISER Mohali N. Mukunda March 2015

Thus the rule to pass from to W (q, p) has this additional factor compared to the Weyl map rule.
From eq (2.9), and taking account of this extra factor, we find:
Z Z
1 0 0
W (q, p) = 2
dq 0 dp0 ei(p qq p)/~ T r(D(q 0 , p0 ) )
(2~)
Z Z
1 0 0
= 2
dq 0 dp0 ei(q pp q)/~ T r(D(q 0 , p0 )
(2~)
Z Z Z
1 0 0 0 0
= dq dp dq 00 ei(q pp q)/~ < q 00 |D(q 0 , p0 )|q 00 >
(2~)2
Z Z Z 0
1 0 0 00 i(q 0 pp0 q)/~+i(q 00 + q2 )p0 /~
= dq dp dq e < q 00 ||q 00 + q 0 >
(2~)2
q0
Z Z
1 0
= dq 0 dq 00 eipq /~ (q 00 + q) < q 00 ||q 00 + q 0 >
2~ 2
Z
1 0
= dq 0 eipq /~ < q q 0 /2||q + q 0 /2 > (3.6)
2~

For a pure state = | >< | we get:

Z
1 0
W (q, p) = dq 0 eipq /~ (q q 0 /2)(q + q 0 /2) (3.7)
2~

The basic properties of the Wigner distribution follow from those of :

Z Z

= = W (q, p) real; T r = 1 = dqdpW (q, p) = 1 (3.8)

However, 0 does not mean W (q, p) 0. In general, taking F = 0 in eq (3.4) we find :

f (q, p) = 2~W0 (q, p),

Z Z
0
T r( ) = 2~ dqdpW0 (q, p)W (q, p) (3.9)

Since the left hand side can certainly vanish, for example mutually orthogonal pure states, in prin-
ciple W (q, p) can become negative in some parts of the phase plane.
On the other hand we find that point wise the value of W (q, p) is bounded in magnitude. From
1
eqs (2.14, 3.4), taking F = ~ we see that

1
T r(P ) = ~W (0, 0), |W (0, 0)| (3.10)
~
This is true all over the phase plane since from eqs (1.9c, 3.6):

0 = D(q0 , p0 )1 D(q0 , p0 ) W 0 (q, p) = W (q + q0 , p + p0 ) (3.11)

So we can extend eq.(3.10):

1
|W (q, p)| (3.12)
~
We can also combine eqs (3.10, 3.11) to see that

1
W (q, p) = T r(D(q, p)P D(q, p)1 ) (3.13)
~
So the Wigner distribution at each (q, p) is an expectation value. This formula was first found by
me in 1978. Later Wootters (1987) gave the name phase point operators for the expression in eq
(3.13), for the finite dimensional case.

6
IISER Mohali N. Mukunda March 2015

As examples, we calculate the Wigner distributions for the ground state and first excited state
of the simple harmonic oscillator. For simplicity, we set m = = ~ = 1, so the Hamiltonian is:
1 2
q + p2


H = (3.14)
2
The normalized wave functions are:

1 q 2 /2 2 2 /2
0 (q) = e , 1 (q) = qeq (3.15)
1/4 1/4
Then using the standard integrals
Z
2
dqeq (1, q 2 ) = (1, 1/2) (3.16)

from eq. (3.7), we find

1 2 2



q 2 +p2 q 2 +p2
W0 (q, p) = e q + p2 1/2 e


, W1 (q, p) = (3.17)

Since
Z Z
dqdpW0 (q, p)W1 (q, p) = 0, (3.18)

W1 (q, p) does indeed have to become negative sometimes!

There are however limits to how negative W (q, p) can become. If we go back to eq. (3.4) and
consider functions of q alone, f (q),then clearly F = f (q). And for these we do have quantum
mechanically determined non-negative probability distributions. Similarly also for f (p), when F =
f (p). For a general state and then in the pure case = |ih|, we recover the marginals:
Z
dp W (q, p) = hq||qi or |(q)|2 ;
Z

dq W (q, p) = hp||pi or |(p)|2 (3.19)

So this is how the Wigner distribution makes contact with genuine probabilities and why it is called
a quasi-probability distribution.

4 Extension to many degrees of freedom

All the formulae in Sections II and III generalize easily, we present only some of them and develop
useful notations as well.
For n Cartesian canonical pairs, we have 2n basic hermitian operators qj , pj , j = 1, 2, ......, n
obeying the CCRs
[qj , pk ] = i~jk , [qj , qk ] = [pj , pk ] = 0 (4.1)
A compact way of expressing them is to define a 2n component column vector with operator entries:

1


2


= 3 = a , a = 1, 2, ..., 2n; a = a ;



.

.
2n
j = qj , j+n = pj (4.2)

7
IISER Mohali N. Mukunda March 2015

Then the CCRs (4.1) can be written as

 
0 Inn
[a , b ] = i~ab , = (4.3)
Inn 0

This 2n 2n matrix is very important and will play a role in the discussion of the symplectic
groups Sp(2n, R). Its basic properties, apart from reality and anti-symmetry, are

= 1 = , 2 = I2n2n (4.4)

We can use both the split q p notation and the notation as convenient.
The Weyl form of the CCRs (4.1, 4.3) is directly expressed in terms of the displacement operators
which are unitary.
T /~
D(q, p) = D() = ei(p.qq.p)/~ = ei ,
T 2n n n
= (q1 ...qn , p1 ...pn ) R , q R , p R (4.5)

Then the CCRs appear as the composition law for the D():
i 0 T
D( 0 )D() = e 2~
D( 0 + ) (4.6)

The standard representations in terms of wave functions are:

Z
H = {(q) C| dn q|(q)|2 < }
n
ZR
= {(p) C| dn p|(p)|2 < },
n
Z R
1
(p) = dn qeip.q/~ (q) ;
(2~)n/2 Rn

(qj )(q) = qj (q), (pj )(q) = i~ (q);
qj

(qj )(p) = i~ (p), (pj )(p) = pj (p) (4.7)
pj

The main properties of the D(), in addition to (4.6), are:

D()a D()1 = a a ;
i 0 q
D(q, p)|q 0 i = e ~ p.(q + 2 ) |q 0 + qi,
i p
q.(p0 + 2 )
D(q, p)|p0 i = e ~ |p0 + pi (4.8)

Of course we supplement (4.7) with

(q) = hq|i, (p) = hp|i;

1
hq|pi = eiq.p/~ (4.9)
(2~)n/2

The Weyl map or rule to go from classical (real) phase space functions to Hermitian quantum
operators on H can be given at several levels:

(.q .p)N (.q .p)N , N = 1, 2, ....;

ei.(.q .p) ei(.q .p) (4.10)

8
IISER Mohali N. Mukunda March 2015

For general classical functions on the phase space R2n , we can use the Fourier route generalizing
(2.5, 2.6):
Z Z
1
f (q, p) = dn dn f(, )ei.(.q .p)
(2)n
Z Z
1
F = dn dn f(, )ei(.q .p ) (4.11)
(2)n
Then generalizing eqns (2.7 , 2.10), we have
T r(D( 0 ) D()) = (2~)n (2n) ( 0 );
Z
1
T r(G F ) = d2n g() f () (4.12)
(2~)n R2n
The basic equations for the Wigner distribution are straight forward generalisations of Sections II,
III:
Z
1 0 1 1
W () = W (q, p) = n
dn q 0 eip.q /~ hq q 0 ||q + q 0 i
(2~) Rn 2 2
Z
1 0 1 1
= dn q 0 eip.q /~ (q q 0 )(q + q 0 ) in pure case; (a);
(2~)n Rn 2 2
Z
T r(F ) = d2n W ()f (), (b)
R2n
Z
T r(0 ) = (2~)n d2n W 0 ()W (); (c)
R 2n
Z
dn pW (q, p) = hq||qi or |(q)|2 ,
n
ZR
dn qW (q, p) = hp||pi or |(p)|2 ; (d)
Rn
= D(0 )1 D(0 ) W 0 () = W ( + 0 )
0
(e) (4.13)

5 Wigner Distributions in finite dimensions- The spin 1/2 case

The preceding Sections depend crucially on the structure and consequences of the CCRs (1.1,
4.1, 4.3) for Cartesian systems. For finite dimensional quantum systems, when dimH < , new
features are to be expected though the Cartesian case can be a guide. We now describe a treatment
by Feynman for two dimensional quantum systems - particle with spin 21 , two-level atom, photon
polarization states etc.
As motivation recall the following features of the one - dimensional Cartesian case:
(i) W (q, p) is a real function of two independent real continuous c- number arguments q and p;
we regard them as possible eigen values of the non commuting operators q and p respectively.
(ii) The (marginal) integrals of W (q, p) over p (respectively over q) give the quantum mechanical
probability distributions for position q (respectively momentum p) in the concerned quantum state.
(iii) We have the relation (3.4) re-expressing the operator form of a general quantum mechanical
expectation value in completely c- number terms.
These feature are carried over to the spin 12 case as follows. The three basic operators (apart
from the unit operator) are the hermitian 2 2 Pauli matrices x , y , z , or j , j = 1, 2, 3 obeying
the familiar commutation relations.
[j , k ] = 2ijkl l . (5.1)
We change the Cartesian scheme in this way:
q z , q  = 1 = eigen values of z ;
p x , p 0 = 1 = eigen values of x ;
phase space point (q, p) four 0 points0 (, 0 ) (5.2)

9
IISER Mohali N. Mukunda March 2015

The most general state is a 2 2 density matrix obeying

= 0, T r = 1 (5.3)

As is well known, it can be expressed in the form

1
= (I + n.), n = real 3 dimensional vector, |n| 1;
2
|n| = 1 pure state, |n| < 1 mixed state. (5.4)

Given , the probabilities for z =  = 1, and for x = 0 = 1, are given by standard quantum
mechanics as
1 1
Prob.(|z = ) = T r(. (1 + z )) = (1 + nz );
2 2
1 1
Prob.(|x = 0 ) = T r(. (1 + 0 x )) = (1 + 0 nx ) (5.5)
2 2
We now need two things to put together the Weyl-Wigner picture: each state should be represented
by a corresponding c- number Wigner distribution W (, 0 ) ; and each of the operators j should be
represented by corresponding c- number functions fj (, 0 ) in such a way that the analogue of eqn
(3.4) is obeyed. Here are the Feynman prescriptions which achieve all this:
1
W (, 0 ) = T r{(1 + z + 0 x + 0 y )}; (a)
4
x , y , z fx (, 0 ), fy (, 0 ), fz (, 0 ) = 0 , 0 ,  (b) (5.6)

Then indeed we find:

X
T r(j ) = W (, 0 )fj (, 0 );
,0 =1
X
W (, 0 ) = Prob.(|z = ),
0
X
W (, 0 ) = Prob.(|x = 0 );

X
W (, 0 ) = 1 (5.7)
,0

Then general expectation values, marginals and overall normalisation all work out satisfactorily.
Using eqn (5.4) in (5.6) we find:
1
W (, 0 ) = (1 + 0 n1 + 0 n2 + n3 ) (5.8)
4
This is just a collection of four real numbers, normalized and with non negative marginals. How
now is the Wigner quality recognized or expressed? If any four real numbers W (1, ) are given,
do they make up a physically acceptable Wigner distribution corresponding to a quantum state?
The necessary and sufficient condition for this is that they be expressible in the form of the right
hand side of eq. (5.8) for some n with |n| 1.
In the same year, 1987, as Feynmans work, Wootters also published an important paper on
finite dimensional Wigner distributions, in which the phrase phase point operators appeared. Since
2005-2006, Chaturvedi, Simon, myself and some others initiated an approach based on early ideas
of Dirac. A fair amount of work is done, but it is still work in progress. An important point is
that the problems in odd and in even dimensions are very different, the former being considerably
simpler.

10