Numerical solution of the time-dependent Schrdinger equation in a periodic potential

Szentmrtoni Orsolya Department of Analysis and Optimization Babe-Bolyai University Cluj-Napoca 10.04.2007

Abstract This work is based on the numerical discretization of the time-dependent Schrdinger equation, performed using the Crank- Nicholson algorithm. The particular case of a weak periodic potential is discussed in order to examine how plane wave particles propagate when their wavelength satises the Bragg condition. Keywords: nite dierence approximations, periodic potential, energy eigenstates, Bragg diraction.

Introduction

The quantum mechanics for particles in a periodic potential is a cornerstone of solid state physics. It is known from perturbative theory that the addition of a weak periodic potential modies the energy-levels 2 k2 of the particle as compared to the free particle case E = h2m . The eect is felt mostly for wavenumbers k which are close to the Bragg plane, in which case a forbidden gap in energy occurs. If a is the periodicity of the potential K = 2a is called a reciprocal lattice vector and the Bragg plane is at k = K 2 . Thus the dispersion relation for particles in a weak potential is such that the group velocity vanishes at the Bragg plane. The aim is to examine this property through the numerical algorithm.

Periodic Potential in One Dimension

Consider the simplest model of a one -dimensional conductor: H = H0 + V = p2 2m

with H 0 a gas of non interacting electrons on a line, and V periodic, that is, V (x + a) = V (x), the potential from a line of ions spaced a apart. 1

The system is taken to have N ions in a total length L, so L = Na and the boundary conditions are required to be periodic. The physics here is that without the potential the electron eigenstates are plane waves. The eect of the lattice potential is to partially reect the waves, like a diraction grating, generating components at dierent wavelengths, which is particularly important when the electron wavelength matches twice the ion spacing. 1 ikx The eigenstates of H 0 are then |k >(0) = e with eikL = 1, so k = 2n L , n is being an integer. L The unperturbed energy eigenvalues, H 0 |k >(0) = E (0) |k >(0) , are just E 0 =
2 k2

2m Suppose that the ions have two electrons each contributing to this one- dimensional conductor and assumed to move into these plane wave states. According to Fermi in the system ground state they will ll up the lowest energy states up to a maximum k -value denoted by kF (Fermi momentum). For the total of 2N electrons the allowed values of k are known from the boundary conditions kn = 2n L with n an integer. The 2N electrons will have N of each spin, each k state can take two electrons (one of each spin) so LkF / = N = L/a kF = /a . In perturbation theory one must nd the matrix elements of V (x) between eigenstates of H 0
(0)

< k |V |k >(0) =

1 L

ei(kk )x V (x)dx

which is just the Fourier component Vkk of V (x). While V is periodic with spatial period a, the only nonzero Fourier components are those having the same spatial period a Vk = 0 only if k = nK, K = 2/a, n integer. Despite many states near k = /a and k = /a that are close together in energy, for any one state k near /a the potential has only a non-zero matrix element to one other state close in energy, the state k K , that is k 2/a. Diagonalizing the full Hamiltonian in the subspace spanned by these two states |k >(0) and |k K >(0) the energy eiegenvalues are given 1 0 0 + Ek E = (Ek K ) 2 .
0 E0 Ek k K 2 ) + |Vk |2 2

0 E0 0 0 Notice that, provided |Ek kK | >> |Vk |, to leading order this gives back E = Ek , EkK the 0 E0 order depending on k . As k approaches /a, |Ek kK | becomes of order |Vk | and the energies deviate from the unperturbed values. The energy jump, or gap of 2|Vk | at k = /a means that there are no plane wave type eigenstates with energies in that range- attempting to integrate Schrodingers equation in the periodic potential fo such an energy gives exponentially decaying and growing solutions. While k increases from 0 towards /a, the plane wave initially proportional to eikx has a gradually admixture of ei(k2/a)x , until at k = /a the two will have equal weight - which means that the eigenfunction will be a standing wave. In fact there are two standing wave solutions at k = /a corresponding to the energies above and below the gap and the energy dierence between them is the origin to the energy gap. This scattering diraction that gives a left moving component to a right moving wave is known as Bragg scattering. One of standing waves piles up charges in the region between the ion cores while the other one piles up charges core points.

Finite dierences

In the late 1960s Goldberg, Schey and Schwartz had a good idea to approximate the solution of the time dependent Schrodinger equation by nite dierences, and to produce a movie using computer technology. When using nite dierences one seeks an approximate solution of the partial dierential equation at the points of a nite grid of points, and the approximation is accomplished by replacing the derivatives by appropiate dierence quotients. Explicit schemes are usually easy to implement in a program where the solution is calculated directly in terms of known quantities. Implicit schemes involve the coupling between unknowns and require solving a linear system.

3.1

The CN implicit scheme

Consider the time dependent Schrodinger equation in one dimension i with the Hamiltonian operator H (x) = 2 + V (x) (3.2). x2 = H (x) (x, t) (3.1) t

The solution is sought usually on an interval x [b, c] and is uniquely determined from the boundary and initial conditions (b, t) = (c, t) = 0 and (x, 0) = g (x). In quantum mechanics, this complex wave function | >= (x, t) modells the evolution of a particle. If the particle does not decay with time, the total probability of nding this particle somewhere in space remains exactly unity at all times < | >= | |2 dx = 1/t. In this way when solving the equation 3

numerically it is clearly an advantage to formulate a FD scheme that conserves the probability by keeping the evolution operator unitary even as it is discretized. The scheme proposed by Crank and Nicholson is a reliable and stable routine for doing this, while it preserves the norm of the wavefunction and so it is not necessary to renormalise it at each time step. This is realized by writing a formal solution of (3.1) (x, t) = eiHt (x, 0) in the so- called Cayley form, which is a long- known equation that does not violate unitarity. Multiply (3.1) by eiHt/2 and expand both sides to get i i (1 + Ht) (x, t + t) = (1 Ht) (x, t) 2 2 . Replacing the second order derivative in the Hamiltonian operator H (x) with nite dierences centered in space, one obtains the CN implicit scheme
t+t j + t+t t+t t+t + j it j 1 2j +1 t+t )= ( + Vj j 2 x2 t t t it j 1 2j + j +1 t ( + Vj j ) (3.3) 2 x2 c b N

t = j

. Dening the regular grid of dimension h = tively in time (3.3) is cast into the form in space (N the number of the meshpoints) , respec-

i k k k ( k 2j + j 1 ) iV (a + jh)j = x2 j +1

N k iHj,m m m=0

k = b + jh, k and H is a real symmetric tridiagonal matrix ( provided V (x) is real). where j The CN algorithm is of second order accuracy, i.e the discretization error decreases as 2 . It denes (0) and (nx) to be zero in order to avoid some extra calculation at the ends. When modifying the numerical scheme (3.3) to include a periodic potential (ex. V (x) = V (x + a)) the obtained result will stand for a standing wave, which is related to the theory above.

Let nx = the number of data points + 1 Let psi[x] be the complex wave function for x = 0,...,nx with psi[0] = psi[nx] = 0. Let m be the mass Let V(x) be the potential Step 1 INPUT psi[x] = input data for x = 1,...,nx-1 Set psi[0] = psi[nx] = 0 Set lambda = i/(2*m) Set u[0] = 0 Step 2 (Initialization of C-N tridiagonal matrix) For x = 1 to nx-1

Set l[x] = 1 + lambda + iV(x)/2 + lambda/2 * u[x-1] Set u[x] = -lambda/(2*l[x]) Step 3 (Iteration of C-N algorithm for nt time steps) For t from 1 to nt Set z[0] = 0 (Solution of tridiagonal system) For x from 1 to nx-1 Set z[x] = ((1-lambda) psi[x] - iV(x)/2 * psi[x] + lambda/2 * (psi[x+1] + psi[x-1] + z[x-1]))/l[x] (Back substitution) For x from nx-1 to 1 Set psi[x] = z[x] - u[x]*psi[x+1] Step 4 OUTPUT psi[1] through psi[nx-1]

References:
[1 ] A. Goldberg, H. M. Schey, J. L. Schwartz, Computer Generated Motion Pictures of One Dimensional Quantum Mechanical Transmission and Reection Phenomena, Amer. Jour. Phys. , 1967. [2 ] G. H. Golub, J. M. Ortega, Scientc computing and dierential equations, Academic Press, , 1992. [3 ] A. Jaun, Numerical methods for partial dierential equations, Ed. of the Swedish Netuniversity, 2005. [4 ] D. Harrison, Computational Physics, CiteSeer IST, PHY307F/407F, 1998. [5 ] W. H. Press, B. P.Flannery, S. A Teukolsky, W. T Vetterling, Numerical Recipes: The art of scientic computing, Cambridge University Press, 1998.