Two particle spin dependent interactions in HgTe/CdTe topological insulators

Marius Radu1, ∗ and Yuli Lyanda-Geller1, †

1
Department of Physics and Purdue Center for Topological Materials,
Purdue University, West Lafayette, Indiana 47907 USA
(Dated: July 12, 2016, —= draft: A=—)
HgTe forms a two-dimensional topological insulator when sandwiched between CdTe barriers for a
HgTe layer wider than the critical thickness. A considerable amount of studies have been focused on
single-particle picture in order to better understand the spin dynamics of the system. We derive two-
particle interaction Hamiltonians describing physics of this compound by using k · p band mixing
and extended Kane model on the same footing as the Coulomb interaction. We obtain a third
order contribution to the exchange interaction arising from a coupling of the spin of one charge
carrier to the relative orbital motion of the other. The model does not include terms which break
bulk inversion assymetry (BIA) or structural inversion assymetry (SIA) which makes it applicable
to any zincblende lattice. We find that the spin-dependent contribution does not conserve total
spin, leading, e.g., to Dzyaloshinskii-Moriya exchange term. Such potential in an electron-electron
collision provides a new mechanism for relaxation of spin polarization.

PACS numbers:

I. INTRODUCTION ones which have spin dependence. Controlling coupling

Interest in topological insulators (TIs) has spiked in
the last decade due to the discovery of exotic edge states
on the surface of these materials and their potential ap-
plication in developing a topological transistor. While
the existence of edge states has been predicted by Tamm
and Schckley a long time ago1,2 such states do not origi-
nate from the topological properties of the material and
therefore are strongly dependent on the bulk band struc-
ture and electronic details of the surface on an atomic
scale. This makes the states very sensitive to pertur-
bations and thus not good candidates for a topological
transistor. There is already a great deal of literature on
the experimental detection via ARPES, and sensitivity
of such surface modes in fundamental materials such as
GaAs, InSb and CuSo (cite).
In contrast to Shockley-Tamm states, topological edge
states have three crucial differences. First, due to the in-
trinsic time reversal symmetry of TI systems, edge states
cannot backscatter into each other. This essentially al-
lows each chiral state to only travel in one and only one
direction along the perimeter of the material. Second,
they are extremely robust and difficult to remove via per-
turbation or some physical material treatment, a char-
acteristic present only in topological materials exhibit-
ing edge states. And finally, it has been demonstrated,
both theoretically and experimentally that these states
demonstrate metallic characteristics, most notably, quan-
tized dissipationless spin transport3 . Such features com-
bined with the high mobility measured in most TIs (cite)
makes these materials ideal candidates for the develop-
ment of a topological transistor. The HgCdTe/HgTe het-
erostructure is a relatively simple model since construc-
tion via a 6-band k · p Kane model transforming accord-
ing to the Oh group will demonstrate all these essential
ingredients. However, an accurate physical description
should include analysis of two particle terms, especially
between spins is an essential requirement in many spin-
tronics applications and it has been the focus much re-
search.
One form of spin-spin coupling arises from the sym-
metric exchange interaction JS1 · S2 which has its origins
in the Pauli exchange principle and Coulomb potential.
This interaction is even with respect to spin permutation
and thus conserves total spin. In general however, spin-
orbit coupling in solids may result in asymmetric spin
contributions which do not preserve the total spin of the
system. Such terms can lead to spin dephasing, and in
electron-electron collisions, a spin flip process. A num-
ber such terms are already known and have been studied
extensively: i) electron and hole Dresselhaus interaction
which arises from bulk inversion asymmetry via a cou-
pling to upper conduction bands of p-symmetry (cite)
ii) Rashba coupling which arises from structural inver-
sion asymmetry such as the application of an electric
field parallel growth axis (cite). Furthermore, it has been
shown by Dzyaloshinskii-Moriya (cite) that a lack of in-
version symmetry along with spin-orbit coupling is the
microscopic mechanism which leads to an antisymmetric
exchange interaction of the form: Dij · (Si × Sj ) where
the vector Dij is usually constrained by the symmetry
of the system, and perpendicular to both the asymmetry
direction and the vector rij between the two spins.
The DM interaction is classified according to the type
of inversion symmetry breaking(A. Fert, V. Cros, and J.
Sampaio, Nat. Nanotech. 8, 152(2013).) nonetheless
both classes rely on some type of inversion symmetry
breaking, whether it be bulk or interfacial. In this study
we discuss a new contribution to the spin-spin interaction
terms of two electrons in 2D semiconductor heterostruc-
tures that is not dependent on either (BIA) or Rashba
coupling. It originates from coupling via the Coulomb
interaction of two electrons via conduction-valence band
mixing, similar in form to excitonic interactions(cite).
 2

We describe this asymmetric electron spin coupling for
2-dimensional model of HgTe/CdTe heterostructures, al-
beit the procedure is applicable to most III-V semicon-
ductors. We make detailed numerical computations for
the strength of such terms near the critical QW width of
trivial-topological transition.
The extended 6 × 6 Kane model incorporates electron
like band, Γ6 and two hole bands, Γ8 directly while con-
tributions from extended bands are accounted for pertur-
batively. In addition to the single particle Kane Hamilto-
nians the Coulomb potential is treated on equal footing.
Working in position representation where ki → ∂r∂ 1 we
have:

( ) ( )
(1) ∂ (2) ∂
H(r1 , r2 , ∂r1 , ∂r2 ) = Hkane + Hkane + Φ(|r1 − r2 |) (1)
∂r1 ∂r2

where Φ(|r1 − r2 |) term is the Coulomb interaction
and details on the contents of Hkane are given in the
Appendix. We construct an effective 2D-model by pro-
jecting the total Hamiltonian onto the zeroth-order wave-
functions which are solutions to
point where M ≈ 0 which leads to a degeneracy of the
E1 and H1 subbands and a gap of ϵL1 ≈ 10meV. For a
6-component basis we must compute 36×36 matrix ele-
ments the Fourier transform of the Coulomb potential of
the form
∫ ∞ ⟨ ⟨   ⟩ ⟩
H(∂z1 , ∂z2 )Ψ(z1 , z2 ) = E (0) Ψ(z1 , z2 ) (2) (1)  (2)   (1)  (2)
Fijkl (q) = dz1 dz2 Ψi  Ψk  e−q|z| Ψj Ψl
 ⟩  (2) ⟩ −∞
where a direct product of states |ψ⟩ = ψ (1)( ⊗ ) ψ
(4)
(1)
is chosen for basis states. Projecting Hkane ∂r∂ 1 and By symmetry arguments the number of distinguishable
( ) matrix elements reduces 9, with the majority being zero.
(2)
Hkane ∂r∂ 2 onto the direct product of single particle The effects of the L1 band are accounted for via a trans-
eigenfunctions yields the BHZ model (cite) formation analogous to the Foldy-Wouythusen in QED,
( ) which leads to an effective two particle interaction of the
(0) HD (k) 0 form:
Hbhz = ∗
0 HD (k)
A2 B2 [ (1) (1) (2) (2)
Hs(3) = −σ0 τ0 σx τx (∇2 V+65 × p2 )z
and HD (k) = ϵk) + di (k)σi , d1 + id2 = A(kx − iky ) = 2ϵ2L1
Ak− , d3 = M − B(kx2 + ky2 ) and ϵ = C − D(kx2 + ky2 ). (1)
In addition to the original basis used in BHZ we in- +σ0 τz(1) σx(2) τx(2) (∇2 V−65 × p2 )z (5)
(2) (2)
clude the lower L1 subband to form the ordered set −σx(1) τx(1) σ0 τ0 (∇1 V+65 × p1 )z
{E1+, H1+, E1−, H1−, L1+, L1−}. Terms coupling to ]
(2)
this band are +σx(1) τx(1) σ0 τz(2) (∇1 V−65 × p1 )z
 (1)

0 iA2 k− and we have defined V±56 = V 6 ± V 5 given in Eq. 29,
 (1) 
−iB2 k+ 0  (i)
HBHZ−L1 =   (3) while σj corresponding to the spin up/down space of
−iA2 k+ (1)
0  (i)
(1) the i-th particle, and τj the E1/H1 subspace of the
0 −iB3 k− ij
( ) We later show the form of V± = (αi ∓
i-th particle.
2r
A2 arises from velocity Kane matrix element between E1 αj ) ln b∥ which allows simplification of the derivatives
and L1 states, while the origin of B2 lies in interaction x1 −x2 y1 −y2
with upper projection onto cubic Dresselhaus terms B8v +
. since ∇1 Vij ij
± = −∇2 V± = (− r 2 , − r 2 ). We rewrite
∥ ∥
We construct the model close to the critical transition the off-diagonal contributionn as:

− [ ]
A2 B2 α56
Hs(3) = σx
(2) (2) (1)
τ x τ z (⃗
r × p
⃗2 ) z − σ (1) (1) (2)
x τ x τ z (⃗r × p
⃗1 ) z
2ϵ2L1 r∥2
+ [ ]
A2 B2 α56
− σx
(2) (2)
τ x (⃗
r × p
⃗ 2 ) z − σ (1) (1)
x τ x (⃗
r × p
⃗ 1 ) z (6)
2ϵ2L1 r∥2
 3

±
where the term α56 = α5 ± α6 . This coefficient is shown term there is also a diagonal spin-dependent addition to
to be significant and approximately 10% of the bare the effective Coulomb potential
Coulomb interaction. In addition to this off-diagonal

+ [ ]
(3) (A22 + B22 )α56
Hd = τ (2) (2)
z (⃗
σ · ⃗r × p
⃗ )
2 z − τz
(1) (1)
(⃗
σ · ⃗r × p
⃗ )
1 z
4ϵ2L1 r∥2
− [ ]
(A22 − B22 )α56
+ τ (1) (2)
z (⃗
σ · ⃗r × p
⃗ 2 )z − τz
(2) (1)
(⃗
σ · ⃗r × p
⃗1 ) z (7)
4ϵ2L1 r∥2
+ [ ]
(A22 − B22 )α56
+ (⃗
σ (2)
· ⃗
r × p
⃗ 2 ) z − (⃗
σ (1)
· ⃗
r × p
⃗1 ) z
4ϵ2L1 r∥2
−
(A22 + B22 )α56 [ ]
+ τz
(1) (2)
τz (⃗
σ (2)
· ⃗r × p
⃗2 )z − (⃗
σ (1)
· ⃗r × p
⃗1 ) z
4ϵ2L1 r∥2

II. METHODS The separation of the z-coordinate becomes clear since

∫
1
e−iqr cos θ dθrdr
2
We construct a new effective interaction term arising Vq = (2πe) √
r + z2
2
form the Coulomb potential between two quasiparticles ∫ [∫ 2π ]
2 r −iqr cos θ
via admixture with other bands, L1 in this case. The = (2πe) √ dr e dθ
procedure borrowed from QED where it is used to de- r2 + z 2 0
rive the Breit-Landau equation via a modified Foldy-
Wouythusen transformation, to incorporate the negative and the second integral is simply Bessel function of the
energy positrons states in the Schrodinger picture. We first kind
write the Coulomb potential as ∫
2 r
Vq = (2πe) √ dr [2πJ0 (qr)]
r + z2
2
∫
e2 J (qr)r
V (r1 , r2 ) = √ (8) = e2 (2π)
3
√0 dr (11)
2 − z2
r12 r2 + z 2
12
−q1 |z|
3 e
= e2 (2π) δ(q1 + q2 ) (12)
with r12 = r∥,1 − r∥,2 and z12 = z1 − z2 (abbreviated z). q1
The Fourier transform is
Taking the reverse Fourier Transform
∫ ∫ 2
1 d q1 d2 q2
Vq = e2 √ e−iq1 r1 e−iq2 r2 d2 r1 d2 r2 V (r1 , r2 ) = Vq eiq1 r1 eiq2 r2
2 + z2
r12 (2π)2 (2π)2
∫ 12
∫
1 1 e−q|z| iqr12
= e 2
√
2
e−iq1 r12 e−i(q1 +q2 )r2 d2 r1 d2 r2(9) = d2 q e (13)
r12 + z 2 2π q
∫ ∫
1 e−q|z|
= qdq dθeiqr12 cos θ
and upon a transformation r12 → r,r2 → r0 this is noth- 2π
∫
q
ing but
= dqJ0 (qr12 )e−q|z| (14)
∫
1
Vq = e2 √ e−iq1 r e−i(q1 +q2 )r0 d2 rd2 r0 The potential is needed in this form to treat the Coulomb
r2 + z 2 interaction between electrons on equal footing to band-
∫ ∫
1 mixing terms of the form k·p. This term is projected onto
= e2 √ e−iq1 r d2 r e−i(q1 +q2 )r0 d2 r0
r2 + z 2 the single particle wavefunctions at k∥ = 0 describing
∫ motion in the xy plane. We average ot the the z-degree
1
e−iq1 r d2 r (10)
2 2
= (2πe) δ (q1 + q2 ) √ of freedom via
r + z2
2
∫ ∫ ∞
1
e−iqr d2 r
2
= (2πe) √ ijkl
V2D (r∥1 , r∥2 ) = dqJ0 (qr12 )Fijkl (q) (15)
r + z2
2
0
 4

where the indices {i, j, k, l} run over the set decouples into four sets
{E1+, H1+, E1−, H1−, L1+, L1−} and
∫ ∞ ⟨ ⟨   ⟩ ⟩
(1)  (2)   (1)  (2) |ΨEL+ ⟩ = fEL+
1
(z) |1⟩ + fEL+
4
(z) |1⟩
kl
Fij (q) = dz1 dz2 Ψi  Ψk  e−q|z| Ψj Ψl(16)
−∞ |ΨEL− ⟩ = fEL−
2
(z) |2⟩ + fEL−
5
(z) |5⟩ (18)
⟨ ⟨  ⟨  ⟨ 
(1)  (2)  (1)  (2)  |ΨH+ ⟩ = 3
fH+ (z) |3⟩
and Ψi  Ψk  is a direct product state Ψi ⊗ Ψk 
comprising of 36 elements. Fijkl is generally a 36 × 36 |ΨH− ⟩ = fH− (z) |6⟩
6

tensor but symmetry allows great simplification since

many matrix elements of Fijkl are either identically zero.
Careful considerations leads to only 9 individual non- from which its evident that in each particle subspace the
zero integrals to compute. A general eigenstate of the matrix element must be diagonal in the upper index
Kane Hamiltonian, in the envelope function approxima-
tion could be written as: ⟨   ⟩ ∫ ∑ γ
(2)   (2)
∑
6 Ψk  e−q|z| Ψl = dz fk (z2 )e−q|z| flν (z2 )δγν
|Ψj ⟩ = fjα (z) |α⟩ (17) γν

α
where |α⟩ reffers to basis set given in Eq. A2 and j refers
to the possible solutions. In the kx = ky = 0 plane it
(19)
but not necessarily the lower index (sub-band index).
Off-diagonal matrix elements like ⟨E1+| V (z) |L1+⟩ can
in principle be non-zero. We write Fijkl

∫ ∞ ⟨ ⟨   ⟩ ⟩
(1)  (2)   (1)  (2)
Fijkl (q) = dz1 dz2 Ψi  Ψk  e−q|z| Ψj Ψl
−∞
∫∞ ∑
= dz1 dz2 fiα (z1 )fkγ (z2 )e−q|z1 −z2 | fjβ (z1 )flν (z2 )δαβ δγν (20)
−∞ αβγν
∫ ∞ ∑
= dz1 dz2 fiα (z1 )fkγ (z2 )e−q|z1 −z2 | fjα (z1 )flγ (z2 )
−∞ αγ

E1+,E1+
As an example we show explicitly the terms present in the first matrix element FE1+,E1+ where we use |E1+⟩ =
fE1+ (z) |1⟩ + fE1+ (z) |1⟩
1 4

∫ ∞
E1+,E1+
FE1+,E1+ (q) = 1
dz1 dz2 fE1+ 1
(z1 )fE1+ (z2 )e−q|z1 −z2 | fE1+
1 1
(z1 )fE1+ (z2 )
−∞
∫ ∞
+ 1
dz1 dz2 fE1+ 4
(z1 )fE1+ (z2 )e−q|z1 −z2 | fE1+
1 4
(z1 )fE1+ (z2 )
−∞
∫ ∞
+ 4
dz1 dz2 fE1+ 1
(z1 )fE1+ (z2 )e−q|z1 −z2 | fE1+
4 1
(z1 )fE1+ (z2 ) (21)
−∞
∫ ∞
+ 4
dz1 dz2 fE1+ 4
(z1 )fE1+ (z2 )e−q|z1 −z2 | fE1+
4 4
(z1 )fE1+ (z2 )
−∞


The envelope functions fij (zk ) are computed numerically  C1 e−α z if d20 < z
via finite-difference methods in Matlab from which we 1
fE1+ (z) = C1 e α z
if z < [− d20 ]

are able to extract an analytic form and use the following V1 cosh(δ z) if z ∈ − d20 , d20
ansatz for the relevant functions (see Fig. 4)

 
 −C4 e−α z if d20 < z  C3 e−β z if d20 < z
4
fE1+ (z) = C4 eα z if z < [− d20 ]
3
fH1+ (z) = C3 eβ z if z < [− d20 ] (22)
 
V4 sinh(δ z) if z ∈ − d20 , d20 V3 cos(ρ z) if z ∈ − d20 , d20
 5
3
investigated i) small q - large separation , ii) large q -
2 small separation between quasi-particles. The tendency
of limq→0 Fk must be identically 1 since for large r, the
1
structure of quasi-particles, screening is no longer rele-
vant, and the two simply must behave as point charges.
-4 -2 2 4
This is the case as we obtain
∫ ∞
-1
1
Vijkl (r ,
ef f ∥1 ∥2r ) = dqJ0 (qr12 ) = (24)
-2 0 r ∥

-3 as expected. In the limit of large q, we are left with a

term proportional to
FIG. 1: Space is split into 5 regions along the vertical
axis Z separated by the equations Z = 12 (d ± z), Z = αk (d)
Fk = (25)
1
2
(−d ± z), and four regions on the horizontal z given by q
{−∞, −d},{−d, 0},{0, d},{d, ∞} (here d = 2).
which has explicit dependence on the QW width hidden
 in the parameter αk . Numerical simulations of the mag-
 −D1 e−ρ z if d20 < z nitude of αk are performed (see Fig. 2). The following
EE EE LL
1
fL1+ (z) = D1 eρ z if z < [− d20 ]
three elements FHH , FLL , FHH have a similar form

U1 sinh(ϕ z) if z ∈ − d20 , d20 (10) (11)
Pk (q) Qk (q)
Fk = e−qd (12)
+ (12)
(26)
 −ρ z d0
Rk (q) Rk (q)
 D4 e if <z 2
4
fL1+ (z) = D4 eρ z if z < [− d20 ] but in the limit of small q behavior is the same as Eq.25.

U4 cosh(ϕ z) if z ∈ − d20 , d20 The final three matrix elements introducing off-diagonal
contributions in the direct product space of two electron
defining d0 to be the size of the QW. We perform a center- EL EL EL
wave-functions FEE , FHH , FLL vanish trivially due to
of-mass transformation, Z → z1 +z 2
2
and z → z1 −z2 . The the parity of the term e −q|z1 −z2 |
in Eq. 20. This term is
quantum well, previously a square in z1,z2 space now even with respect to interchange of particles, i.e. z1 − z2 .
becomes rhombus-like with a long axis of 2d, area given An arbitrarily small lower bound is set, b on q, due to
by the intersections of the equations Z = 12 (d ± z), Z = divergent behavior at q=0 and we perform the integral
2 (−d ± z). We treat each region separately due to the
1
in Eq. 15
absolute value quantity in the exponential. ∫ ∞
The matrix elements in Eq. 20 can be categorized ac- k αk (d)
Vef f (r∥1 , r∥2 ) = dqJ0 (qr12 )
cording to the four classes in Eq. 18. Two possibilities q
b
[ ( ) ( 2 2 )]
arise 2r∥ b2 r∥2 b r∥
= αk (d) −γ − ln + p Fq − (27)
• i=j , k=l - each can be one of three choices b 8 4
{E1, H1, L1}, irrelevant of spin up/down since
fi1 (z) = fi2 (z), fi4 (z) = fi5 (z), fi3 (z) = where( γ is the )EulerGamma constant, and
fi6 (z) for all i. This yields 6 elements b2 r∥2
EE LL HH EE EE LL p Fq 1, 1; 2, 2, 2; − 4 is the generalized hyperge-
FEE , FLL , FHH , FHH , FLL , FHH where we have
made use of the symmetry property Fiikk = Fkk ii ometric function
which amounts to exchanging the two particles, and ∞ ∏
∑ j=1p (aj )k
is possible only because of the absolute value quan- p Fq (a1 · · · ap ; b1 · · · bq ; z) = ∏ zk (28)
tity in the exponential e−q|z1 −z2 | = e−q|z2 −z1 | . k=0 j=1q (bj )k k!

• i̸=j , k=l and vice versa - which gives rise to three with (a)k is the Ponchammer symbol. For small r < a (a
EL EL EL
off-diagonal contributions FEE , FHH , FLL . Note lattice constant) limr→0 p Fq (a1 · · · ap ; b1 · · · bq ; z) = 1 −
EL LL LL LE
that FLL = FEL = FLE = FLL . b2 r∥2
32 which leaves the dominant behavior of the effective
We discuss the limiting cases now. The first three ele- potential to be logarithmic
EE
ments FEE LL
, FLL HH
, FHH have the form ( )
2r∥
(6) (7)
Vef f (r∥1 , r∥2 ) = −αk (d) ln
k
(29)
Pk (q) Qk (q) b
Fk = e−qd (8)
+ (8)
(23)
Rk (q) Rk (q) No further information about the magnitude of the length
scale of b can be extracted, but its existence is necessary,
(i) (i) (i)
where Pk , Qk , Rk are polynomials of degree i in the since a length cutoff must be introduced. However, de-
variable q. For the integral of Eq. 15 two limits are pendence on b is trivial and unphysical for our purposes
 6

e2
4 Π Ε0 potential - which allows for instantaneous communica-
tion between particles, but additional corrections arising
0.40 from QED accounting for retardation effects up to an
Α1 order of c12 :
Α2
0.35
Α3 [ ]
q1 q2 q1 q2 r˙1 · r˙2 (r12 · r˙1 )(r12 · r˙2 )
− 2 +O(c−3 )
Α4
Α5 V12 = + 3
0.30
Α6
r12 2c r12 r12
(30)
0.25 this is known as the Breit operator and for our pur-
poses this contribution is dropped since the prefactor c12
0.20
is much smaller than the energy scales we deal with here
1 1
c2 << EE1 −EL1 .
d HnmL
Because we are dealing with an effective 2D form of
6.5 7.0 7.5 8.0
the Coulomb potential, as per Eq. 15, the leading order
FIG. 2: Dependence of the αi parameters on the thickness of
term is now of the form
the HgTe quantum well. Strongest is matrix element origi- ( )
2r∥
nating from heavy-hole-like wavefunctions. Vkef f (r∥1 , r∥2 ) = −αk (d) ln (31)
b
e2
4 Π Ε0
0.5
The log form is weaker than 1r and indicates that for small
distances the particles interact via a real 2-dimensional
Coulomb interaction - electric fields do not propagate
0.4
orthogonally to the xy plane. Upon the reduction
to 2-dimensions the effective interaction between quasi-
0.3
particles in different sub-bands differs. This leads to a
Α5 -Α6 position dependent potential on the diagonal, which can-
0.2 Α5 +Α6
not be expressed using direct product of single-particle
operators such as in 3D, Φ(|r1 −r2 |)1⊗1. It instead con-
0.1 sists of the 6-matrix elements we computed and discussed
in the previous section. Equations 6 and 7 show two of
d HnmL the effective terms computed. A2 arises from P - Kane
6.5 7.0 7.5 8.0
matrix element between E1 and L1 states, while the ori-
gin of B2 lies in interaction with upper projection onto
FIG. 3: Dependence of α6 ± α5 parameters on the thickness +
of the HgTe quantum well. These two quantities enter the cubic Dresselhaus terms B8v . Note that in the absence
(3) (3)
non-trivial spin dependent Coulomb interaction. of BIA terms Hs vanishes while Hd remains non-zero
and simplifes to Eq. 32.
+
Terms arising with coefficient α56 can be shown to
since we can always add a constant term of the form be equivalent to Dzyaloshinskii-Moriya interaction pro-
ln ( bb0 ) and change b → b0 . The only difference is an addi- jected onto the quantum well. Such terms are also
tive shift, which because is a constant, is largely inconse- present in the bulk (cite badescu) since then α5 = α6
quential. The dependence of the αk (d) parameter on the however if one is considering purely electron-electron in-
thickness of the quantum well is shown in Fig. 2 - where teraction only the third term of Eq. 32 will be non-zero.
k corresponds to { the index of the 6 ordered integrals
} dis- The first term in Eq. 32 arises to the coupling of elec-
EE LL HH EE EE LL
cussed above - FEE , FLL , FHH , FHH , FLL , FHH . tron sector E1± in BHZ model to hole sector H1± via
the L1 band. Of more interest are terms with coefficient
−
α56 which vanish in the bulk. Origin of such interactions
III. DISCUSSION are the different symmetries of the electron sector, E1,
which is s-like, and hole like sector H1, which is actually
(3) (3)
Coupling terms similar to Hs and Hd are known an admixture of heavy-hole and light-hole states. Coef-
for electrons in free space as were computed by Landau- ficients α5 , α6 reflect these symmetries since they arise
Breit. Such terms arise from electron-positron band mix- from matrix elements involving the envelope functions of
ing and Coulomb interaction. However the coupling be- electron and hole like states, respectively. See Fig. 3 for
tween Dirac quasiparticles is not simply bare Coulomb dependence of α6 − α5 on QW width.
 7

+ [ ]
(3) A22 α56
Hd = 2 2 τz(2) (⃗σ (2) · ⃗r × p⃗2 )z − τz(1) (⃗σ (1) · ⃗r × p⃗1 )z
4ϵL1 r∥
− [ ]
A22 α56
+ τ (1) (2)
(⃗σ · ⃗r × p
⃗ )
2 z − τ (2) (1)
(⃗
σ · ⃗r × p
⃗ )
1 z (32)
4ϵ2L1 r∥2 z z

+ [ ]
A22 α56
+ (⃗
σ (2)
· ⃗r × p
⃗ 2 ) z − (⃗
σ (1)
· ⃗
r × p
⃗1 ) z
4ϵ2L1 r∥2
−
A22 α56 [ ]
+ τ (1) (2)
z τ z (⃗
σ (2)
· ⃗r × p
⃗2 ) z − (⃗
σ (1)
· ⃗r × p
⃗1 )z
4ϵ2L1 r∥2

APPENDIX A: KANE MODEL fourth power in k. We are left with a 6 component ba-
sis |1⟩ = |Γ6 , 1/2⟩, |2⟩ = |Γ6 , −1/2⟩, |3⟩ = |Γ8 , 3/2⟩... and
We take as the starting point the Extended 8-band the secular equation will take the form of a 6-coupled
Kane model, and since the spin-orbit coupling is strong differential set of equations for { fn (z) }
in Hg0.32 Cd0.68 Te-HgTe-Hg0.32 Cd0.68 Te QW system we  
project onto a basis at k = 0 in which this term is diago- ∑ x,y,z∑ ∑
x,y,z
 ki Dnnij
′ kj +
i
Pnn′ ki + En′ (z)δnn′
 Fn′ (z)
nal. There exists such a basis, and it corresponds to the
n′ i,j i
basis in which the total angular momentum J = L + σ
and its projection Jz along the z axis are diagonal. For = EFn (z)
the S-edge, J = 1/2 (Γ6 symmetry group). The P-edge ∑
where Fα (z) = j cj fj (z) and n and n′ runs over the
(0)
gives either J = 3/2 (Γ8 group) or J = 1/2 (Γ7 group).
The model takes into account up to all orders in k all the set above and En′ (z) are the respective band energies.
k · p and SO interactions between the bands Γc6 , Γv8 and The elements containing Pnn′ describe the coupling be-
Γv7 , while k · p with other bands (a 14 band model is the tween the n and n′ bands exactly, while Dnn
ij
′ take into ac-

starting point) are taken into account by perturbation count coupling to remote bands. The Hamiltonian takes
theory. The basis ordering, along with the phase choices the following matrix form:
are:  √ √ 
T 0 −K+ 2
K 1
K − 0
1  √3
z
√3 
u1 (r) = |Γ6 , ⟩ = S↑  
2  0 T 0 − 13 K+ 2
K K − 
 3 z 
1  −K∗ 0 
u2 (r) = |Γ6 , − ⟩ = S↓ H= √ + √0 F A R 
2  2 ∗ 1 ∗ ∗ 
 K − K A G 0 R 
 √3 
 1 ∗ √2 ∗
3 1 z 3 +
u3 (r) = |Γ8 , ⟩ = √ (X + iY ) ↑ ∗ 
2 2  3 K− K
3 z R 0 G −A 
∗ ∗ ∗
1 1 0 K− 0 R −A F
u4 (r) = |Γ8 , ⟩ = √ [(X + iY ) ↓ −2Z ↑] (A1)
2 6 (A2)
1 1
u5 (r) = |Γ8 , − ⟩ = − √ [(X − iY ) ↑ +2Z ↓] ∂
2 6 k∥ = kx2 + ky2 , k± = kx ± ky , kz = −i
∂z
3 1 ~2 [ ]
u6 (r) = |Γ8 , − ⟩ = − √ (X − iY ) ↓ T = Ec + (2F + 1) k∥ + kz (2F + 1)kz
2 2 2m
1 1 ~2 [ ]
u7 (r) = |Γ7 , ⟩ = √ [(X + iY ) ↓ +Z ↑] F = Ev − (γ1 + γ2 ) k∥ + kz (γ1 − 2γ2 )kz
2 3 2m
1
u8 (r) = |Γ7 , − ⟩
1
= √ [(X − iY ) ↑ −Z ↓] ~2 [ ]
2 3 G = Ev − (γ1 − γ2 ) k∥ + kz (γ1 + 2γ2 )kz
2m
~2 √ [( 2 ) ]
Initial estimates of the split-off energy of the Γ7 band in- R = 3 kx − ky2 γ2 − 2ikx ky γ3
dicates that it is far away from the rest of the bands abd 2m
it will not be important in the description of the QW, ~2 √
A = 3k− [γ3 kz + kz γ3 ]
is in contrast to the bulk case. Its effects are only ac- 2m
counted for perturbatively and appropriate corrections Kz = P kz
are made. The Luttinger parameters however remain 1
K± = √ P k±
unaffected since the lowest order corrections are of the 2
 8

H1+ || −0.020679eV H2+ || −0.082344eV E1− || −0.082348eV H1− / H1+ || −0.020679eV H2− / H2+ || −0.082344eV E1+ / E1− || −0.082348eV
0
0.5 0.3
0.25 0.25 0.15
−0.1
0.2
0.2 0.2
−0.2 0.1
0.1
0.15 0.15
−0.3 0
0
0.1 0.1
−0.4 −0.1 0.05

−0.2 0.05 0.05

−0.5
−0.5
0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5 −5 0 5 −5 0 5

H3+ || −0.1837eV L1− || −0.18455eV E2− || 0.26163eV H3− / H3+ || −0.1837eV L1+ / L1− || −0.18455eV E2+ / E2− || 0.26163eV
0.5 0.25 0.15
0.3 0.15
0.2
0.2 0.2
0.1 0.1
0.1 0 0.15
0
0
0.1
−0.2 0.05 0.05
−0.1
0.05
−0.2
−0.5 −0.4
0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5 −5 0 5 −5 0 5

H4+ || −0.3217eV H5+ || −0.48784eV L3− || −0.50859eV H4− / H4+ || −0.3217eV H5− / H5+ || −0.48784eV L3+ / L3− || −0.50859eV
0.5 0.5 0.25
0.15
0.2 0.2
0.2
0.15 0.1
0 0.15
0 0
0.1
0.1
−0.2 0.05
0.05 0.05

−0.5 −0.5 −0.4

0 0 0
−5 0 5 −5 0 5 −5 0 5 −5 0 5 −5 0 5 −5 0 5

FIG. 4: Envelope functions fji (z) for a centered heterostruc- FIG. 5: Probability densities of envelope functions for a cen-
tore Hg0.3 Cd0.7 Te-HgTe-Hg0.3 Cd0.7 Te with width of center tered heterostructore Hg0.3 Cd0.7 Te-HgTe-Hg0.3 Cd0.7 Te with
layer d=6.6nm. The phase choices are such that electron-like width of center layer d=6.6nm. X-asis is in units of nm.
(1,2)
envelope function (blue) is entirely real fEL± , light-hole-like
(4,5)
(green) entirely imaginary fEL± , and heavy-hole-like (red)
(3,6) depending on starting assumptions. In addition to the
entirely real fH± . X-asis is in units of nm.
Hamiltonian in Eq. A2 used by BHZ, we can incorporate
inversion symmetry-breaking terms of two kinds: BIA
terms, and hetero-interface reflection terms. The start-
Due to loss of translational symmetry in the z-direction ing model can be parametrized as follows:
kz is not a good quantum number; is to be understood
as differentiation with respect to z. ~2 ∑ 2
H6c6c = Ec + k (2F + 1) (B1)
This set is solved numerically making use of variable 2m i i
grid size finite-difference method in Matlab for k∥ = 0.
A matrix of 15,000 by 15,000 is constructed as to min- √ {
imize non-hermitian elements that are inherent to all H6c8v = 3 P T · k + iB8v
+
(Tx {ky , kz } + cp)
numerical simulations of heterostructure systems. Non- [ ]}
1 − 1 2
hermitian elements arise from the position dependent ef- + B8v (Txx − Tyy )(kz − k ) − Tzz (kx − ky )
2 2 2
2 3
fective masses which abruptly change values at the in-
(B2)
terface of two different materials. To minimize effects
of such inaccuracies a Gaussian form is assumed with a
variance large enough that the effective mass transition { [ ]
~2 1 2 2
occurs over the length span of a lattice constant. We em- H8v8v = Ev − γ1 k − 2γ2 (Jx − J )kx + cp
2 2
2m 3
ploy both the cell-centered and edge-centered method at
interface points which further allows minimization of er- −4γ3 [{Jx , Jy }{kx , ky } + cp]
}
rors by reducing the number of non-hermitian elements. 2 [ ]
Using the diagonalization procedure of sparse matrices in + √ Ck {Jx , Jy − Jz }kx + cp
2 2
(B3)
3
Matlab, we obtain a set of z-envelope functions and cor-
responding eigenvalues seen in Fig. 4, 5 for a structure and we list separately the following term proposed by
with QW width of 6.6 nm. Aleiner and Ivchenko4 , which accounts for light-hole-
heavy-hole reflection at the boundary of Hg0.3 Cd0.7 Te-
HgTe heterostructure.
APPENDIX B: EXTENDED MODELS 2 ∑
H8v8v = {Jx , Jy }a0 U ζn δ(z − zn ) (B4)
3 n
For deriving extended models, the basis set fi (z) is pro-
jected onto different bulk Hamiltonians H(kx , ky , −i ∂z ∂
). tLH ~ 2
where U = 2m 2 , zn is the coordinate of the n-th het-
0 a0
(i)
This gives rise to new effective 2D models Hij (kx , ky ) erojunction and ζn = ±1 for a heterojunction of the type
 9

AB and BA respectively. The factor ζn appears because
boundary conditions must ensure conservation of proba-
bility current for both type of heterojunctions AB or BA.
The delta-function form indicates the mixing is short-
ranged in nature, and within the range of applicability of
the effective mass method the coefficient tLH can safely
assumed to be constant and independent of energy in-
side each layer. The coefficient does in fact depend on
the composition of the material.

∗
Electronic address: radu@purdue.edu L. W. Molenkamp, X.-L. Qi, and S.-C. Zhang, Science 318,
†
Electronic address: yuli@purdue.edu 766 (2007).
1 4
I. Tamm, Phys. Z. 76, 733 (1932). I. L. Aleiner and E. L. Ivchenko, JETP Lett. 55, 692 (1992).
2
W. Shockley, Physical Review 56, 317 (1939).
3
M. König, S. Wiedmann, C. Brüne, A. Roth, H. Buhmann,