Introduction

In this document, Q1D Simple Dirac Fermion TI (which I shall write as Q1D SDF TI for short) refers to systems where the x dimension is assumed to be of innite / semi-innite length, and the y dimension is of mesoscopic length scale so the quantum connement eect giving rise to discrete y modes is present with the Hamiltonian H = v f p + M + U I (1) where M is an externally imposed magnetic eld, and U a spin-independent externally imposed magnetic eld.

Eigenstates
We rst derive the energy eigenstates of the Q1D system. Let the width of the system be L, so that the spatial position of the TI is from y = 0 to y = L. In order to fulll the boundary conditions in the y direction, L we write the eigenmode l (x, y) satisfying 0 dy l (x, y)H l (x, y) = E l as a superposition of the transverse modes n = l (x, y) = eikm x
n 2 L

sin(n/Ly) : (2)

n (y)l n

The l s are (2 by 1) spinor which is to be determined. Note that in writing n l in the above form we have absorbed the weightings of the various transverse modes into their respective l so that l are not normalized to unity. n n From Hl (x, y) = El l (x, y) , we have dy eikl x m (y)Hl (x, y) = El dy eikm x m (y)l (x, y). (3)

Expanding out the matricres of H of Eq. 1 in full, the left hand side of the previous equation reads dy
n

m vf

0 (kx + y )

(kx y ) 0

Mz + U M

M+ Mz + U

n l n

where M = Mx My . Noting that dy m (y)n (y) = mn and introducing mn = dy m (y)(n n (y)), Eq. 3 reduces to (Mz + U El )mn mn (vf kx + M ) + mn mn (vf kx + M+ ) mn (Mz + U El )mn l = 0 n

which we can split into vf k l

n

0 mn

mn 0

l = n
n

(El Mz U )mn mn mn M

mn mn M+ (El + Mz U )mn

l . n

(4) In theory, the transverse mode (n) summation of Eq. 2 should go to innity. This is not, however, not possible in practice and we cut o the summation at a number Nm of transverse modes. Eq. 4 can then be written as a generalized eigenvalue equation vf kl A = B(El ) where A and B are 2Nm 2Nm matrices, a 2Nm 1 column vector. kl and can then be solved for numerically. Notice that in our approach we supply El as the input, and seek those values of kl s which give rise to energy eigenstates with energy El . Since the matrices involved are 2Nm 2Nm in size, there will be 2Nm eigenvalues and eigenmodes. In the absence of an externally applied magnetic eld, Nm of these kl are real and positive, or imaginary with positive imaginary parts corresponding to right propagating / decaying states. The remaining Nm kl s are real and negative, or imaginary with negative imaginary parts. 0.0.1 Dispersion relationship

Connement along the y direction opens up an energy gap due to the quantization of the energy modes. The application of an external magnetic eld M results in further splitting of the energy states. Somewhat unexpectedly, the shape of the dispersion relation is independent of the direction of the external magnetic eld.

Wavefunction matching across interfaces of different widths

We consider an interface between 2 TIs of dierent widths. We write the wavefunction on the left of the interface
Nm

I =
n

aI (eikn x n
m

I I ) m n,m

where the an are the weights (to be determined) of the nth eigenstate given by I I n = eikn x m I I . Similarly, we write the wavefunction on the right of m n,m the interface as
Nm

II =
i

aII (eikn i

II

x j

II II ). m i,j

For ease of discussion, we take the section on the left to be located between 0 to LI in the y direction, and the section on the right to be the narrower one located from yII to yII + LII so that II = L2 sin(n/LII (y yII )) < LI . n II We need the wavefunctions at both sections to match at the interface at x = x0 , , i.e. that I (x0 , y) = II (x0 , y). For an arbitrary ath transverse mode II , we have a
yII yII

dy II (y)I (x0 , y) = a
yI yI

dy II (y)II (x0 , y) a

aI eikn x0 ( n
n

( (
j

dy II I )I ) = a m n,m ( dyII II )II ). a j i,j

yII yI

aII e i
i yII yI

II iki x0

Noting that dyII (y)II (y) and writing a j the above reduces to aI (eikn x0 n
n m
I

dy II (y)I (y) = Cam , m a

II

Cam I ) = n,m
i

aII (eiki n

x0

which actually consists of 2 seperate equations since each spinor actually has 2 components. There are Nm dierent II s, giving rise to a total of 2Nm a equations which solves for the correct number of unknowns taking the Nm right-propagating/decaying states on the left of the interface, there are Nm left-propagating/decaying states on the left of the interface corresponding to reected incident waves at the interface, and Nm right propagating/decaying states on the right of the interface corresponding to the incident waves transmitted across the interface. Footnote : Notice that in the above we are integrating the wavefunctions on both sides of the interface with a transverse mode of the shorter side across the width of the shorter side. By right the integration ought to be done with a transverse mode of the longer side, but in practice I found that this does not quite work matching the Fourier weights of the wavefunctions on both sides of the interface with a nite number of modes does not gurantee that the wavefunctions themselves match. Unfortunately, attempting to match the wavefunctions across segments of dierent widths gives rise to ill-behaved results.

Magnetoresistance in a 3 segment Q1D system

Actually Im not very sure exactly how to dene the transmission for this sort of many-mode system, but here I calculate the quantity rk =
idrain right propagating modes

|bi |2 ki |a2 |kj j

jsource right propagating modes

where ai and bj are the coecients of the modes in the source and drains respectively, and right propagating modes refer to those modes with real, positive ks. The basic setup for the system under consideration is shown below. As an example, we calculate the magnetoresistance for the magnetizations of the leads and the central region both lying in the xy plane and having the same magnetide as a function of the energy E and the magnetization direction of the middle region. The transmission is high when the magnetization in the middle region is parallel to those of the two leads. Somewhat surprisingly, the transmission is also high when the lead and central magnetizations are exactly anti-parallel to each other. The tranamission decreases for intermediate relative orientations of the lead and central magnetizations.

BHZ Hamiltonian
Here I attempt to build an NEGF formulation of the work in PRB83, 081402(R) which uses the 4-band BHZ Hamiltonian H = = H(k) 0 0 H (k)
k

+ Mk Ak+ 0 0

Ak k Mk 0 0

0 0 k + Mk Ak

0 0 Ak+ k Mk

2 2 2 2 where k = C +V (x, y)D(kx +ky ), Mk = M B(kx +ky ), k = kx iky , and A, B, C, D, and M are parametrs describing the band structure of a HgTe/CdTe QW. Unpacking and recombining the various terms, the upper subblock looks like

H(k) =

e+ + V m + k 2 Ak+

Ak e + V m k 2

2 2 where e = (C M ), m = (D B) and k 2 = kx + ky . The lower subblock looks like this as well except that Ak is replaced by Ak . I think the NEGF formulation works, although this it is rather slow and memory intensive. Here I reproduce someething which looks a bit similar to Fig. 3 of the paper, although in this case the width of the strip is 300 nm while that in the paper is 200 nm. So far I havent seen anyone do NEGF formulation of the BHZ Hamiltonian. Probably one reason is that this is very computationally intensive - it took me about 15 minutes of computational time, and about 3.5GB of RAM, on my Core 2 Q9550 computer overclocked to 3 GHz to generate the graphs above. Compared to the other method of wavefunction matching, the NEGF is probably

superior in that its easier to set up arbitary geometries (like holes in the middle) as well as add multiple leads to the edges. Incidentally, the picture above is interesting because it shows that in the regions where the TI is narrow (near the middle x direction of the picture) on either side of the central hole, substantial particle (I dont know which corresponds to electrons/holes) densities can be found on both transverse edges of the narrow regions but in the wider TI regions (those away from the central hole) substantial particle density can only be found on the top edge. Apparently, one needs substantial transverse width before non-trivial TI behaviour can be exhibited.