spin_potential_shape.rst

More interesting systems: spin, potential, shape

Each of the following three examples highlights different ways to go beyond the very simple examples of the previous section.

Matrix structure of on-site and hopping elements

We begin by extending the simple 2DEG-Hamiltonian by a Rashba spin-orbit coupling and a Zeeman splitting due to an external magnetic field:

H = \frac{-\hbar^2}{2 m} (\partial_x^2+\partial_y^2) -
  i \alpha (\partial_x \sigma_y - \partial_y \sigma_x) +
  E_\text{Z} \sigma_z +  V(y)

Here \sigma_{x,y,z} denote the Pauli matrices.

It turns out that this well studied Rashba-Hamiltonian has some peculiar properties in (ballistic) nanowires: It was first predicted theoretically in Phys. Rev. Lett. 90, 256601 (2003) that such a system should exhibit non-monotonic conductance steps due to a spin-orbit gap. Only very recently, this non-monotonic behavior has been supposedly observed in experiment: Nature Physics 6, 336 (2010). Here we will show that a very simple extension of our previous examples will exactly show this behavior (Note though that no care was taken to choose realistic parameters).

The tight-binding model corresponding to the Rashba-Hamiltonian naturally exhibits a 2x2-matrix structure of onsite energies and hoppings. In order to use matrices in our program, we import the Tinyarray package. (NumPy would work as well, but Tinyarray is much faster for small arrays.)

For convenience, we define the Pauli-matrices first (with \sigma_0 the unit matrix):

Previously, we used numbers as the values of our matrix elements. However, ~kwant.builder.Builder also accepts matrices as values, and we can simply write:

Note that the Zeeman energy adds to the onsite term, whereas the Rashba spin-orbit term adds to the hoppings (due to the derivative operator). Furthermore, the hoppings in x and y-direction have a different matrix structure. We now cannot use lat.neighbors() to add all the hoppings at once, since we now have to distinguish x and y-direction. Because of that, we have to explicitly specify the hoppings in the form expected by ~kwant.builder.HoppingKind:

• A tuple with relative lattice indices. For example, (1, 0) means hopping from (i, j) to (i+1, j), whereas (1, 1) would mean hopping to (i+1, j+1).
• The target lattice (where to hop to)
• The source lattice (where the hopping originates)

Since we are only dealing with a single lattice here, source and target lattice are identical, but still must be specified (for an example with hopping between different (sub)lattices, see :ref:`tutorial-graphene`).

Again, it is enough to specify one direction of the hopping (i.e. when specifying (1, 0) it is not necessary to specify (-1, 0)), ~kwant.builder.Builder assures hermiticity.

The leads also allow for a matrix structure,

The remainder of the code is unchanged, and as a result we should obtain the following, clearly non-monotonic conductance steps:

Spatially dependent values through functions

Up to now, all examples had position-independent matrix-elements (and thus translational invariance along the wire, which was the origin of the conductance steps). Now, we consider the case of a position-dependent potential:

H = \frac{\hbar^2}{2 m} (\partial_x^2+\partial_y^2) + V(x, y)

The position-dependent potential enters in the onsite energies. One possibility would be to again set the onsite matrix elements of each lattice point individually (as in :ref:`tutorial_quantum_wire`). However, changing the potential then implies the need to build up the system again.

Instead, we use a python function to define the onsite energies. We define the potential profile of a quantum well as:

This function takes two arguments: the first of type ~kwant.builder.Site, from which you can get the real-space coordinates using site.pos, and the value of the potential as the second. Note that in potential we can access variables L and L_well that are defined globally.

Kwant now allows us to pass a function as a value to ~kwant.builder.Builder:

For each lattice point, the corresponding site is then passed as the first argument to the function onsite. The values of any additional parameters, which can be used to alter the Hamiltonian matrix elements at a later stage, are specified later during the call to smatrix. Note that we had to define onsite, as it is not possible to mix values and functions as in syst[...] = 4 * t + potential.

For the leads, we just use constant values as before. If we passed a function also for the leads (which is perfectly allowed), this function would need to be compatible with the translational symmetry of the lead -- this should be kept in mind.

Finally, we compute the transmission probability:

kwant.smatrix allows us to specify a dictionary, params, that contains the additional arguments required by the Hamiltonian matrix elements. In this example we are able to solve the system for different depths of the potential well by passing the potential value (remember above we defined our onsite function that takes a parameter named pot). We obtain the result:

Starting from no potential (well depth = 0), we observe the typical oscillatory transmission behavior through resonances in the quantum well.

Warning

If functions are used to set values inside a lead, then they must satisfy the same symmetry as the lead does. There is (currently) no check and wrong results will be the consequence of a misbehaving function.

Nontrivial shapes

Up to now, we only dealt with simple wire geometries. Now we turn to the case of a more complex geometry, namely transport through a quantum ring that is pierced by a magnetic flux \Phi:

For a flux line, it is possible to choose a gauge such that a charged particle acquires a phase e\Phi/h whenever it crosses the branch cut originating from the flux line (branch cut shown as red dashed line) [1]. There are more symmetric gauges, but this one is most convenient to implement numerically.

Defining such a complex structure adding individual lattice sites is possible, but cumbersome. Fortunately, there is a more convenient solution: First, define a boolean function defining the desired shape, i.e. a function that returns True whenever a point is inside the shape, and False otherwise:

Note that this function takes a real-space position as argument (not a ~kwant.builder.Site).

We can now simply add all of the lattice points inside this shape at once, using the function ~kwant.lattice.Square.shape provided by the lattice:

Here, lat.shape takes as a second parameter a (real-space) point that is inside the desired shape. The hoppings can still be added using lat.neighbors() as before.

Up to now, the system contains constant hoppings and onsite energies, and we still need to include the phase shift due to the magnetic flux. This is done by overwriting the values of hoppings in x-direction along the branch cut in the lower arm of the ring. For this we select all hoppings in x-direction that are of the form (lat(1, j), lat(0, j)) with j<0:

Here, crosses_branchcut is a boolean function that returns True for the desired hoppings. We then use again a generator (this time with an if-conditional) to set the value of all hoppings across the branch cut to fluxphase. The rationale behind using a function instead of a constant value for the hopping is again that we want to vary the flux through the ring, without constantly rebuilding the system -- instead the flux is governed by the parameter phi.

For the leads, we can also use the lat.shape-functionality:

Here, the shape must be compatible with the translational symmetry of the lead sym_lead. In particular, this means that it should extend to infinity along the translational symmetry direction (note how there is no restriction on x in lead_shape) [2].

Attaching the leads is done as before:

In fact, attaching leads seems not so simple any more for the current structure with a scattering region very much different from the lead shapes. However, the choice of unit cell together with the translational vector allows to place the lead unambiguously in real space -- the unit cell is repeated infinitely many times in the direction and opposite to the direction of the translational vector. Kwant examines the lead starting from infinity and traces it back (going opposite to the direction of the translational vector) until it intersects the scattering region. At this intersection, the lead is attached:

After the lead has been attached, the system should look like this:

The computation of the conductance goes in the same fashion as before. Finally you should get the following result:

where one can observe the conductance oscillations with the period of one flux quantum.

Footnotes

[1]	The corresponding vector potential is A_x(x,y)=\Phi \delta(x) \Theta(-y) which yields the correct magnetic field B(x,y)=\Phi \delta(x)\delta(y).

[2]	Despite the "infinite" shape, the unit cell will still be finite; the ~kwant.lattice.TranslationalSymmetry takes care of that.