discretize.rst

Discretizing continuous Hamiltonians

Introduction

In ":ref:`tutorial_discretization_schrodinger`" we have learnt that Kwant works with tight-binding Hamiltonians. Often, however, one will start with a continuum model and will subsequently need to discretize it to arrive at a tight-binding model. Although discretizing a Hamiltonian is usually a simple process, it is tedious and repetitive. The situation is further exacerbated when one introduces additional on-site degrees of freedom, and tracking all the necessary terms becomes a chore. The ~kwant.continuum sub-package aims to be a solution to this problem. It is a collection of tools for working with continuum models and for discretizing them into tight-binding models.

Discretizing by hand

As an example, let us consider the following continuum Schrödinger equation for a semiconducting heterostructure (using the effective mass approximation):

\left( k_x \frac{\hbar^2}{2 m(x)} k_x \right) \psi(x) = E \, \psi(x).

Replacing the momenta by their corresponding differential operators

k_\alpha = -i \partial_\alpha,

for \alpha = x, y or z, and discretizing on a regular lattice of points with spacing a, we obtain the tight-binding model

H = - \frac{1}{a^2} \sum_i A\left(x+\frac{a}{2}\right)
        \big(\ket{i}\bra{i+1} + h.c.\big)
    + \frac{1}{a^2} \sum_i
        \left( A\left(x+\frac{a}{2}\right) + A\left(x-\frac{a}{2}\right)\right)
        \ket{i} \bra{i},

with A(x) = \frac{\hbar^2}{2 m(x)}.

Using ~kwant.continuum.discretize to obtain a template

The function kwant.continuum.discretize takes a symbolic Hamiltonian and turns it into a ~kwant.builder.Builder instance with appropriate spatial symmetry that serves as a template. (We will see how to use the template to build systems with a particular shape later).

It is worth noting that discretize treats k_x and x as non-commuting operators, and so their order is preserved during the discretization process.

The builder produced by discretize may be printed to show the source code of its onsite and hopping functions (this is a special feature of builders returned by discretize):

Building a Kwant system from the template

Let us now use the output of discretize as a template to build a system and plot some of its energy eigenstate. For this example the Hamiltonian will be

H = k_x^2 + k_y^2 + V(x, y),

where V(x, y) is some arbitrary potential.

First, use discretize to obtain a builder that we will use as a template:

We now use this system with the ~kwant.builder.Builder.fill method of ~kwant.builder.Builder to construct the system we want to investigate:

After finalizing this system, we can plot one of the system's energy eigenstates:

Note in the above that we provided the function V to syst.hamiltonian_submatrix using params=dict(V=potential), rather than via args.

In addition, the function passed as V expects two input parameters x and y, the same as in the initial continuum Hamiltonian.

Models with more structure: Bernevig-Hughes-Zhang

When working with multi-band systems, like the Bernevig-Hughes-Zhang (BHZ) model [1] [2], one can provide matrix input to ~kwant.continuum.discretize using identity and kron. For example, the definition of the BHZ model can be written succinctly as:

We can then make a ribbon out of this template system:

and plot its dispersion using kwant.plotter.bands:

In the above we see the edge states of the quantum spin Hall effect, which we can visualize using kwant.plotter.map:

Limitations of discretization

It is important to remember that the discretization of a continuum model is an approximation that is only valid in the low-energy limit. For example, the quadratic continuum Hamiltonian

H_\textrm{continuous}(k_x) = \frac{\hbar^2}{2m}k_x^2

and its discretized approximation

H_\textrm{tight-binding}(k_x) = 2t \big(1 - \cos(k_x a)\big),

where t=\frac{\hbar^2}{2ma^2}, are only valid in the limit E \lt t. The grid spacing a must be chosen according to how high in energy you need your tight-binding model to be valid.

It is possible to set a through the grid_spacing parameter to ~kwant.continuum.discretize, as we will illustrate in the following example. Let us start from the continuum Hamiltonian

H(k) = k_x^2 \mathbb{1}_{2\times2} + \alpha  k_x \sigma_y.

We start by defining this model as a string and setting the value of the \alpha parameter:

Now we can use kwant.continuum.lambdify to obtain a function that computes H(k):

We can also construct a discretized approximation using kwant.continuum.discretize, in a similar manner to previous examples:

Below we can see the continuum and tight-binding dispersions for two different values of the discretization grid spacing a:

We clearly see that the smaller grid spacing is, the better we approximate the original continuous dispersion. It is also worth remembering that the Brillouin zone also scales with grid spacing: [-\frac{\pi}{a}, \frac{\pi}{a}].

Advanced topics

The input to kwant.continuum.discretize and kwant.continuum.lambdify can be not only a string, as we saw above, but also a sympy expression or a sympy matrix. This functionality will probably be mostly useful to people who are already experienced with sympy.

It is possible to use identity (for identity matrix), kron (for Kronecker product), as well as Pauli matrices sigma_0, sigma_x, sigma_y, sigma_z in the input to ~kwant.continuum.lambdify and ~kwant.continuum.discretize, in order to simplify expressions involving matrices. Matrices can also be provided explicitly using square [] brackets. For example, all following expressions are equivalent:

We can use the locals keyword parameter to substitute expressions and numerical values:

Symbolic expressions obtained in this way can be directly passed to all discretizer functions.

References

[1]	Science, 314, 1757 (2006).

[2]	Phys. Rev. B 82, 045122 (2010).