Back to 498CQM Home Calendar / Outline Previous Lecture Next Lecture |
Phys 498CQM Lecture 5Instructor: R. M. Martin |
HW2 assigned
|
where the hamiltonian operator is
Here we have allowed the potential to be an explicit function
of time.
There are two basic approaches:
Fourier Transform in time to
get equations for u(x,omega), where omega is frequency.
Explicit evolution in time. The one can Fourier transform
if desired.
We will follow the second approach as an example of
numerical methods, and the associated stability and error analysis.
Also it has become popular and used in much recent work
using "time-dependent density functional theory" or
"time-dependent Hartree-Fock". [This has been widely used
in nuclear physics (e.g. Koonin) and recent in application to
clusters (e.g. K. Yabana and G. F. Bertsch, Phys REv B54, 4484 (1996).]
Although we have not introduced
these theories yet, we can note that they all involve some
form of independent particle equations with some approximation
for V(x,t).
This is an initial value problem requiring the wavefunction to
be given at one time. Then the evolution in time is fixed.
Because of the factor or i in the equation, the
evolution of u(x,t) is manifestly unitary, i.e. the
norm of u(x,t) is conserved, which is equivalent to
conservation of particle number.
As discussed in Thijssen and Koonin, the simplest approximation is
to discretize time in steps t = n (delta t) Expanding the exponential leads to:
The error analysis in Thijssen shows this is ALWAYS unstable. A simple modification (which however makes it an implicit method requiring an inversion is to make the substitution of un by un+1 on the righ hand side. This leads to
This form is stable, BUT it is not unitary (as is obvious for the form). An explicitly unitary algorithm can be made by using the operator
which is known as the Crank-Nicholson method. The inverse means that this algorithm requires explicit inversion of an operator. [This approach is used in recent work by Argyrios Tsolodikis, U of Illinois, unpublished.] This is possible if the operator can be represented as a small matrix (e.g. quantum system with a restricted Hilbert space) and it can be done for very large bases, e.g., large grids if the operator is diagonal or nearly diagonal. This is the case for grids with the kinetic energy operator in H approximated by a finite difference formula.
Another approach is to use the "Trotter decomposition"
where H = T + V. One can use this (unitary) operator to evolve the wavefunction by time delta t. The exponential of operators is in general not simple, but in the case of diagonal operators, one has simply = exp(). How can this be done, since T and V do not commute? By a Fast Fourier Transform (FFT) one can go back and forth between real space (where V is diagonal) to reciprocal space (where T is diagonal). [This is done, e.g., in M. Suzuki, J. Phys. Soc. Jpn. 61, L3015 (1992).]
Alternatively, one can expand the exponential to higher powers and have algorithms that are sufficiently stable and unitary for desired purposes. [An analysis is given in Iitaka, et al, Comp. Phys. Commun. 90, 251 (1995) for symmetric multistep algorithms.]
Example of time dependent Schrodinger Eq. using C++ codes provided by Todd Martinez.