First some Physics 711. Recall that in Hamiltonian formalism
which replace the Lagrange-Euler equations
and for any
Canonical transformations preserve the symplectic structure
and generates the transformation; if
For a type-2 transformation
There is a nice way to write an infinitismal transformation as a deviation from a unit transformation
or
The canonical transformation from for which are constants can be constructed in such a manner that are the {\bf initial data}, which is equivalent to solving the Hamiltonian EOMs. In this case is also a constant, and in 711 we took it to be zero as an expedient, and the type-1 generating function is called , the Hamilton-Jacobi function,
which we solve for , and then can be inverted to get . really is the action integral, since
therefore
Despite the stark beauty of this formalism, nobody ever seems to like it much in 711.
What would this look like in quantum mechanics? The problem was considered by Dirac in 1933 and again by Feynmann in 1948. Unfortunately is not just a label (like is) in QM, it becomes an operator, which is the source of some troubles, it also becomes a state-label for the set of states associated with sharp position measurement outcomes
Consider a quantum canonical transformation (note that and probably do not commute) and the mixed matrix element ( is hermitian)
(convert to wave-mechanics notation and integrate by parts if you need to).
The problem of constructing a standard basis for operators constructed by phase-space quantization is an old one, considered by Born and Jordan in 1925, who proposed completely symmetric combinations (Weyl ordered)
and so on, and McCoy ordering
which parallels the Poincare’-Birkhoff-Witt ordering of universal enveloping algebras. We run into this question of ordering when we try to compute
so Dirac restricted attention to those functions of the form , then
Dirac then defines the -function
and shows that it is the generator of canonical transformations (type-) since then
if the derivatives and are appropriately ordered (is this even possible? What does this heuristic even mean? Probably this; compute {\bf then} replace , in the resultant)
The stage is set for the Feynmann path integral. Now let and .
Then and we can
consider transition amplitudes}or conditional probability densities such as
Why? Because with taken to be independent we know that and with constants of the motion, and
Go back to 711 where we showed that the type- generating function with with generates time translation and this is a Legendre transformation of a type- function
which we can iterate to for finite (non-infinitismal) .
An important thing to note is that the right-hand side contains no operators. As Ramond points out this formula gets us into trouble for finite because in the amplitude we can subdivide into short intervals and insert complete sets of states, resulting in a {\bf exact} quantum-mechanical formula as an alternative starting point for the iteration, because of course and therefore change as changes
but writing the right-hand side of the generating function above as a product of factors, one for each sub-interval within does not reproduce the integrations needed to make the formula exact. Feynmann observed that if the generating function formula is true for infinitismal times, we are back to our formmula gotten by semi-classical considerations
In his excellent book “Field Theory, a Modern Primer”, P. Ramond observes that as is apropos of HJ theory, we showed where this comes from explicitly above by using the relations between type- and generating functions), over this interval then can be written in terms of and , and since we must integrate over the intermediate states
and as the only contribution according to the Riemann-Lebesque theorem will come from the path with stationary action, the classical path.\\
We would say that represents the amplitude for detection at at time given a condition of detection at time resulted in a value . In other words the modulus squared of this item is a conditional probability for particle detection resulting in one value granted that it was previously detected at a specified location.
Let’s do it all (again) systematically from a quantum-mechanical starting point in the Schrodinger picture. The purpose is to obtain the amplitude as a sum over paths. We seek the conditional probability amplitude that if the particle is detected at at it will be detected at at . Since
we have
but since in general we apply these time-evolutions in order, by subdividing into many intervals
(1)
however the Baker-Campbell-Hausdorff formula allows us to say that
in which the third term on the right side contains terms generated by the commutator of the kinetic and potential energy operators. We can see that for very short times this last term can be taken to be .
Insert a complete set of momentum states into
(2)
But for the momentum state and position state
in which and are the eigenvalues of the Hermitean operators and respectively.
We arrive at
(3)
after performing the integral by completing the square. Assemble many such factors by integrating over complete sets of intermediate position states, the result is usually written as\\
In the path integral expression there are two hidden mathematical problem areas. The first is the integration measure function . In order that the path integral reproduces the correct free particle green function, this measure must be a complex number. It really should be of no surprise that if the integral is performed by dividing the path into segments of duration , that this measure depends on the number . The explicit measure (restore the mass) is
The second issue is the fact that the exponent oscillates. The common means of understanding such integrals (thanks to N. Weiner) is to give a vanishing imaginary part in order to justify the use of the Gaussian integration formula in their quadrature. In quantum field theory we refer to this as euclideanization, or a Wick rotation.
Direct evaluation for free particle
The terms within the large parenthesis can be written as
If we make an orthogonal transformation such that the matrix diagonalizes the matrix ; then since orthogonal transformations have determinant of one the integration measure is unaffected the integrations will all become decoupled
(4)
and now we can do the integrals using the Gaussian integration formulas.
Instead of position states, we may want or where is some enumerated complete basis, such as Hamiltonian eigenstates. Since any state can be expanded in such a complete basis, including position states
Consider then
(5)
absorbing the values of the wavefunctions at the now quite arbitrary endpoint of the path integration,
The oscillator
You can do exactly the same procedure for the oscillator using the results from 711
Here is the calculation: start with
we can compute the path integral by subdividing the time into parts such that
(6)
in which we computed the potential at the midpoint of each interval in order to get a symmetrical matrix representation. The exponent of the path integral will then be
in which the matrix is tridiagonal, just like it was for the free particle, with diagonal entries
and the off-diagonal entries are
and once again the vector is null with only first and last components being nonzero
Everything that we did to evaluate the free-particle path integral remains valid except for the precise form of the determinant and inverse matrix elements. We find in this case that
and so
and
using .
Furthermore
We can put the parts together and in the final expression involving the exponential of
we multiply out terms and arrange things in increasing powers of ;
We arrive at the following result (compare with the 711 action which you got by solving a generating function differential equation in problem set 11!)