Time-dependent Schrodinger equation implementation in FEniCS

For our Bachelors thesis we're trying to solve the Schrodinger equation $$i\partial_tu = -\nabla^2u+Vu$$ in FEniCS. Given the domain $$[-5, 5]^2$$ with an initial value of $$u_0(x, y)=e^{(-2(x^2+y^2))}$$ and with $$V(x,y)=\frac{x^2}{4}+9y^2$$ at time $$T=2$$ the solution $$|u|^2$$ should look like the following image given by our supervisor. This is however the result we get: The equation is dicretized with Crank-Nicolson as follows: (where $$u=p+qi$$)

For the real part:

$$\frac{\Delta t}{2}(\nabla^2p^{n+1}-Vp^{n+1})-q^{n+1}=-q^n + \frac{\Delta t}{2}(Vp^n-\nabla^2p^n)$$

for the imaginary part:

$$\frac{\Delta t}{2}(\nabla^2q^{n+1}-Vq^{n+1})+p^{n+1}=-p^n + \frac{\Delta t}{2}(\nabla^2q^n-Vq^n)$$

In the code below, $$u$$ is $$p$$ and $$u$$ is $$q$$

We're having trouble creating the correct residual when solving for $$u$$.

In FEniCS the equation is written in weak form and when multiplying with the test function we change sign of $$dot(grad(), grad())$$ to accommodate for multiplication with $$i$$. This is were we probably are at fault because we do not know how to construct the residual. When changing the sign of the gradient for the imaginary case the solution does not converge.

T = 2.0            # final time
num_steps = 200  # number of time steps
dt = T / num_steps # time step size

mesh = RectangleMesh(Point(-5, -5), Point(5, 5), 55, 55)
V = VectorFunctionSpace(mesh, 'Lagrange', 1, dim=2)
# Define initial value
u_0 = Expression(  ( 'exp(-2*(pow(x, 2)+pow(x, 2)))', '0'),
degree=1)
u_n = interpolate(u_0, V)

def boundary(x, on_boundary):
return on_boundary

bc = DirichletBC(V,(0, 0), boundary)

u = Function(V)
v = TestFunction(V)
f = Expression(('(1/4)*pow(x, 2) + 9*pow(x, 2)', '0'), degree=2)

t=0

#Residual
f*u*v)*dx
ReHL = -u_n*v*dx + (dt/2)*(f*u_n*v + dot(grad(u_n),
f*u*v)*dx
ImHL = u_n*v*dx + dt/2*(f*u_n*v + dot(grad(u_n),

FReal = ReVL - ReHL
FIm = ImVL - ImHL

Fny = FReal + FIm

J = derivative(Fny, u)
for n in range(num_steps):
# Update current time
t += dt
# Compute solution
solve(Fny == 0, u, bc, J=J)
# Update previous solution
u_n.assign(u)

We've also tried our luck doing as the example found in this link without success: Gross-Pitaevskii in FEniCS

What are we doing wrong here?

• This seems like a question on how to use fenics, rather than a general scicomp question. You will get better answers if you ask on the fenics forums fenicsproject.discourse.group Feb 15 '20 at 9:39
• Discretizing the real and imaginary parts of a Schrodinger equation separately is something new to me... This is not how people solve time-dependent Schrodinger equation. Have you heard about time-domain finite difference? Look at here: iopscience.iop.org/article/10.1088/1751-8113/40/8/013 Feb 15 '20 at 18:17
• Unfortunately FEniCS does not support complex numbers, this is simply a workaround. Feb 15 '20 at 22:15
• @AntonScotte you don’t need any complex number if you use time-domain finite difference. Feb 15 '20 at 23:24
• Also, it seems DOLFINX supports complex numbers: igorbaratta.github.io/gsoc_blog/jekyll/update/2018/07/03/… Feb 16 '20 at 0:59