of 5
Supplementary Material
Dynamical anyon generation in Kitaev honeycomb non-Abelian spin liquids
Yue Liu,
1
Kevin Slagle,
1, 2
Kenneth S. Burch,
3
and Jason Alicea
1, 2
1
Department of Physics and Institute for Quantum Information and Matter,
California Institute of Technology, Pasadena, CA 91125, USA
2
Walter Burke Institute for Theoretical Physics, California Institute of Technology, Pasadena, CA 91125, USA
3
Department of Physics, Boston College, Chestnut Hill, Massachusetts 02467, USA
(Dated: June 23, 2022)
Appendix A: Effective lattice model
Spin Liquid
γ
R
γ
L
γ
1
γ
2
γ
L
γ
1
γ
L
γ
2
γ
R
γ
1
γ
R
γ
2
L
b
L
h
/
2
L
h
/
2
FIG. 1. Mapping from the dumbbell setup to a helical Majorana chain. The dashed lines refer to edges of the dumbbell. We
imagine flattening the holes along the vertical arrows and obtain a helical chain whose two edge modes are represented by solid
lines.
In this appendix, we map the Majorana edge modes of the dumbbell to an effective spin model that can be simulated
more readily. The original system hosts chiral Majorana fermions
γ
1
,
2
around two holes and two counter-propagating
Majorana fermions
γ
L,R
along the edges of the bridge region. We imagine flattening the two holes along the vertical
arrows in Fig. 1. In the resulting deformed geometry, we identify
γ
1
on the upper half of the flattened hole with a
left-mover
γ
L
; on the lower half of the flattened hole we similarly identify
γ
1
with a right-mover
γ
R
(at the end of the
left hole, the fields obey boundary conditions
γ
L
=
γ
R
). An analogous identification maps
γ
2
to left- and right-movers.
As Fig. 1 summarizes, at this point we have mapped the dumbbell onto a helical Majorana chain with fields
γ
L/R
propagating over an effective length
L
eff
=
L
h
/
2 +
L
b
+
L
h
/
2
.
The Hamiltonian of the dumbbell is
H
=
L
eff
0
dx
[
ivγ
R
x
γ
R
+
ivγ
L
x
γ
L
]
L
h
/
2+
L
b
L
h
/
2
dx
[
κ
(
γ
R
x
γ
R
) (
γ
L
x
γ
L
)]
.
(A1)
We can recover the same low-energy Hamiltonian from the lattice model
H
=
iJ
L
eff
1
j
=1
γ
j
γ
j
+1
+
λ
L
h
2
+
L
b
2
j
=
L
h
2
+3
γ
j
2
γ
j
1
γ
j
+1
γ
j
+2
iδJ
(
γ
L
h
2
+1
γ
L
h
2
+2
+
γ
L
h
2
+
L
b
1
γ
L
h
2
+
L
b
)
,
(A2)
with
v
J
,
κ
λ
,
δJ
λ
, and
{
γ
i
j
}
= 2
δ
i,j
. One can establish this correspondence by expanding
γ
j
γ
L
+ (
1)
j
γ
R
, which yields
j
γ
j
+1
∼ −
R
x
γ
R
+
L
x
γ
L
. The
κ
term describing interaction in the bridge region
can be similarly recovered by expanding four-fermion terms to leading order:
γ
j
2
γ
j
1
γ
j
+1
γ
j
+2
(
γ
R
x
γ
R
) (
γ
L
x
γ
L
)
.
Finally, the
δJ
term counteracts the explicitly generated mass at endpoints of the bridge (due to the loss of translation
symmetry at the endpoints).
We can map the Kitaev chain model to an Ising spin chain using the usual Jordan-Wigner transformation:
γ
2
i
1
=
Z
i
i
1
j
=1
X
j
, γ
2
i
=
iX
i
γ
2
i
1
,
(A3)
2
where
X
j
and
Z
j
are Pauli operators. The resulting spin Hamiltonian is a critical transverse-field Ising model
supplemented by three-spin interactions in the bridge and boundary transverse fields at the bridge endpoints:
H
0
=
J
L
eff
2
1
j
=1
Z
j
Z
j
+1
J
L
eff
2
j
=1
X
j
,
H
int
=
λ
L
h
4
+
L
b
2
2
j
=
L
h
4
+1
(
X
j
Z
j
+1
Z
j
+2
+
Z
j
Z
j
+1
X
j
+2
)
,
δH
=
δJ
(
X
L
h
4
+1
+
X
L
h
4
+
L
b
2
)
.
(A4)
Above we separated the Hamiltonian into pieces involving
J
,
λ
, and
δJ
, as used in the main text.
The above mapping establishes that the lattice model
H
0
and continuum Hamiltonian
H
manifestly contain the
same low-energy degrees of freedom. One may ask, however, whether the continuum edge Hamiltonian adequately
captures all of the low-energy physics of the dumbell in the spin liquid setup. Since the emergence of a chiral Ising edge
conformal field theory (CFT) is a universal property of the non-Abelian spin liquid (the edge mode can also be derived
from the Kitaev honeycomb model [1–3]), the question is equivalent to asking whether
H
misses additional low-energy
degrees of freedom that need to be included. One possibility is the dumbbell supports a
non-chiral
gapless edge mode
riding along with the chiral Ising CFT; such an outcome is technically compatible with the bulk topological order but
would require fine-tuning and so can be safely dismissed. A more plausible possibility is that ‘accidental’ low-energy
bulk fermions or visons reside at discrete locations near the edge (e.g., due to disorder, atomically sharp corners,
etc. [1, 4–9]) and hybridize with the chiral edge degrees of freedom that we include. The chiral edge—consisting of
gapless, extended states—would however ‘eat up’ these states, i.e., they would be absorbed by the edge and become
delocalized, at most changing the overall topological sector of the dumbbell. If we assume that the overall topological
charge around the dumbbell is trivial then our scheme continues to hold without modification. Finally, the boundaries
of our dumbbell should be viewed as ‘blurred’ on the scale of the bulk correlation length, which we assume here is
large compared to the atomic spacing. The dumbbell will thus not suffer from sharp corner effects in the sense that
the edge mode wavefunctions will be smeared out over many lattice spacings, spending little time near any atomically
sharp corners (which need not exist in any case in our setup due to the manner in which we envision manipulating
the spin liquid using magnetic tunnel junctions).
Instead of studying the effective lattice model, one could alternatively contemplate exploiting exact solvability of the
full 2D Kitaev honeycomb model to simulate the dynamics of the dumbbell setup more directly. We are interested,
however, in attacking the problem in an experimentally relevant regime that is potentially far from the exactly
solvable point. In this generic situation, gauge fluxes are no longer static, thus complicating an explicit solution to
the 2D problem, particularly for simulating nontrivial time dependence. Our anyon-generation scheme circumvents
this challenge by relying only on universal edge physics, which depends on a small number of parameters (e.g., the
edge velocity, the hole/bridge sizes, and the interaction between counter-propagating edge modes). The 1D effective
lattice model derived above directly incorporates these parameters while ‘filtering out’ unimportant bulk degrees of
freedom. Correspondingly, the effective lattice model enables efficient simulation of time dependence as carried out in
the main text. That said, in future work it would be interesting to develop efficient methods for studying dynamics
of the non-Abelian phase in the full 2D Kitaev honeycomb model—particularly in the presence of experimentally
relevant perturbations that spoil exact solvability.
Appendix B: Effective-model level structure and simulation details
Figure 2 plots the 10 smallest energy eigenvalues of the effective Hamiltonian [Eq. (A4)] versus
λ
, obtained from
DMRG with
(
L
h
,L
b
) = (320
,
120)
and
(320
,
56)
; note that the ground state energy has been fixed to zero for each
λ
.
The many-body energies at
λ
= 0
can be extracted from the theory of a single chiral Majorana fermion propagating
over length
L
= 2
L
h
+ 2
L
b
, as arises for the trivial-bridge spin liquid setup that the Hamiltonian emulates in this
limit. That is, the many-body energies follow from summing up single-fermion energies of the form
E
ψ
=
2
πv
L
(
k
+ 1
/
2)
with integer
k
0
. Dotted grey lines correspond to levels with odd fermion number, while solid black lines have even
fermion number. The former represent states that are not physically accessible in our spin-liquid problem (where we
assume trivial total topological charge for the dumbbell).
3
0
0.1
0.2
0.3
0.4
0.5
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
I
×
I
ψ
×
ψ
σ
×
σ
λ
/J
TCI
I
×
I
ψ
×
ψ
σ
×
σ
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0
0.1
0.2
0.3
0.4
0.5
λ
/J
Energy
/J
Energy
/J
(
L
h
,L
b
) = (320
,
120)
(
L
h
,L
b
) = (320
,
56)
FIG. 2. Lowest 10 levels (with the ground state energy subtracted) obtained by DMRG for
(
L
h
,L
b
) = (320
,
120)
and
(320
,
56)
.
Black curves indicate levels that have trivial total topological charge and thus represent physically accessible states in our spin-
liquid problem. Grey dotted curves corresponds to levels with an odd number of fermions—which we assume are inaccessible
in the spin liquid context.
The levels in Fig. 2 at first vary weakly with
λ
, but undergo more dramatic evolution near the TCI point (
λ
= 0
.
428
J
,
corresponding to the vertical dashed line). At
λ
= 0
.
5
—where our protocol terminates—the lowest levels that are
accessible in the spin liquid problem correspond to the following anyon sectors:
I
×
I
(ground state),
σ
×
σ
(
1
st
excitation),
ψ
×
ψ
(
2
nd
excitation), etc. See the labels on the right side of Fig. 2. Notice that the first excited state is
actually two-fold degenerate, with one of the degenerate levels representing an inaccessible, odd-fermion-parity state.
This pair of levels corresponds to the two possible fusion channels for Ising anyons (i.e.,
σ
×
σ
=
I
or
ψ
). In the spin
representation, one can identify the states in the
σ
×
σ
sector by the properties
Z
j
Z
j
+1
〉≈
1
and
X
j
〉≈
0
for sites
emulating the bridge region; the
I
×
I
and
ψ
×
ψ
states instead have
Z
j
Z
j
+1
〉≈
0
and
X
j
〉≈
1
in the bridge sites.
Our time-dependent protocol simulations initialize the system into the
λ
= 0
ground state. Figure 2 indicates that
the ground state does not exhibit any level crossings en route to
λ
= 0
.
5
; consequently, diabatic evolution is required
to access
σ
×
σ
(and other excited states). Throughout the simulation, the truncation error is set to be
10
9
and the
time step is set to 0.005 (in
J
= 1
units). We checked that the error due to truncation and the finite time step is
small by doubling the truncation error and the time step in a system with
L
h
= 320
,L
b
= 56
, and
τ
= 1200
, which
only affected the amplitudes for the
I
×
I
and
σ
×
σ
states by
<
2%
.
Evolution from the ground state to the
σ
×
σ
=
ψ
fusion sector is forbidden even in the spin representation (due to
Ising symmetry), and even for diabatic evolution—modulo TEBD simulation errors from truncation and finite time
steps. When computing the probability of entering the
σ
×
σ
state at the end of the protocol, we use DMRG to
compute the pair of degenerate
σ
×
σ
eigenstates and then sum the probabilities for both. The probability of the
σ
×
σ
=
ψ
sector due to TEBD error is negligible (e.g.,
4
×
10
8
for
L
h
= 160
,
L
b
= 80
and
τ
= 1200
).
In Fig. 3 we show an example of when the probability is transferred from
I
×
I
state to
σ
×
σ
state. We plot the
probabilities as a function of
λ
(
t
)
for a
(
L
h
,L
b
) = (80
,
80)
dumbbell with
τ
= 400
. When
λ < λ
TCI
, the transition rate
is small. The major probability transition appears just after the TCI point where kinks of spontaneously generated
mass in the bridge carry Majorana zero modes and facilitate the generation of
σ
×
σ
state. When the bridge is deep
in the gapped phase (
λ > λ
TCI
), such kink excitations become gapped and difficult to generate, and the transition
rate correspondingly decreases to nearly zero again.
4
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
0.2
0.4
0.6
0.8
1
P
σ
×
σ
|

σ
×
σ
(
λ
)
|
H
(
λ
)
|
I
×
I
(
λ
)
!
|
TCI
λ
/J
(
L
h
,L
b
) = (80
,
80)
FIG. 3. Transition probability from the
I
×
I
to
σ
×
σ
state (defined as a function of
λ
via adiabatic continuity) for a system
with
(
L
h
,L
b
) = (80
,
80)
and
τ
= 400
. In the legend
H
(
λ
)
dH
(
λ
)
/dλ
denotes the derivative of the interaction term, and
|〈
σ
×
σ
(
λ
)
|
H
(
λ
)
|
I
×
I
(
λ
)
〉|
measures the transition rate in the language of time-dependent perturbation theory.
σ
σ
J
κ
T
1
f
(
n
σ
)
T
7
/
4
T
δ
T
T
+
δ
T
FIG. 4. Dumbbell setup incorporated in a thermal anyon interferometer.
Appendix C: Interferometric detection of Ising anyons in the dumbbell setup
Unitary time evolution in our protocol generically yields a superposition of
I
×
I
and
σ
×
σ
sectors for the holes in
our dumbbell. (For simplicity, here we neglect the
ψ
×
ψ
sector, which is suppressed in the desired operating regime.)
A projective measurement of the anyon charge on one of the holes is therefore required. The protocol repeats until
projective measurement returns the desired
σ
×
σ
sector.
Our dumbbell setup can be naturally incorporated into previously proposed interferometer designs to enable anyon
detection. In this appendix, we focus on the example of thermal interferometry (see Refs. 10 and 11), which can
be implemented in the device shown in Fig. 4. The five hatched regions are assumed to be dynamically tunable
between spin liquid and trivial phases using magnetic tunnel junctions that locally control the Zeeman field in the
Kitaev material. The hatched bridge connecting the holes in the dumbbell was discussed in detail in the main
text. Four additional hatched regions serve to bend the outermost spin-liquid edge states inward during the readout
stage—creating constrictions that allow tunneling of Ising anyons between the upper and lower edges. A temperature
difference
2
δT
between cold and hot reservoirs at the left and right ends, respectively, generates a heat current
J
; the
corresponding thermal conductivity is
κ
=
J/
(2
δT
)
.
If the bulk thermal conductance is sufficiently small, the heat exchange is dominated by the chiral edge mode—
yielding quantized thermal Hall conductivity proportional to the central charge in the absence of constrictions. As
shown in Refs. 10 and 11, when the two constrictions are created, edge Ising anyons tunneling through the constrictions
introduce a correction to the heat conductance [see Eq. (39) in Ref. 11] that, crucially, is sensitive to the anyon charge
5
located in the enclosed hole in the dumbbell:
κ/κ
=
T
7
/
4
{
t
2
L,σ
+
t
2
R,σ
+ [1 + (
1)
n
σ
]
t
L,σ
t
R,σ
}
.
(C1)
Here,
T
is the mean temperature;
t
L
(
R
)
is the tunneling amplitude of the edge Ising anyons at the left (right)
constriction; and
n
σ
= 0
,
1
is the number of Ising anyons generated in the enclosed hole. In particular, the
t
L,σ
t
R,σ
contribution reflects an interference term—which
vanishes
when an Ising anyon resides in the enclosed hole. Physically,
the absence of interference in this case arises because an edge Ising anyon encircling a bulk Ising anyon changes the
fusion channel for the Ising anyon pair in the dumbbell, producing a state that is orthogonal to that arising when
edge Ising anyons do not encircle the hole. Therefore one can detect an Ising anyon in the hole by measuring the
thermal conductance of the above setup, providing a means of projective measurement required for our protocol to
deliver well-defined anyon charge in the dumbbell.
[1] M. Thakurathi, K. Sengupta, and D. Sen, Phys. Rev. B
89
, 235434 (2014), arXiv:1310.4701 [cond-mat.mes-hall].
[2] T. Mizoguchi and T. Koma, Phys. Rev. B
99
, 184418 (2019).
[3] T. Mizoguchi, T. Koma, and Y. Yoshida, Phys. Rev. B
101
, 014442 (2020).
[4] A. J. Willans, J. T. Chalker, and R. Moessner, Phys. Rev. Lett.
104
, 237203 (2010).
[5] A. J. Willans, J. T. Chalker, and R. Moessner, Phys. Rev. B
84
, 115146 (2011).
[6] F. Zschocke and M. Vojta, Phys. Rev. B
92
, 014403 (2015).
[7] M. Vojta, A. K. Mitchell, and F. Zschocke, Phys. Rev. Lett.
117
, 037202 (2016).
[8] W.-H. Kao, J. Knolle, G. B. Halász, R. Moessner, and N. B. Perkins, Phys. Rev. X
11
, 011034 (2021).
[9] S.-H. Jang, Y. Kato, and Y. Motome, Phys. Rev. B
104
, 085142 (2021).
[10] Z. Wei, V. F. Mitrović, and D. E. Feldman, Phys. Rev. Lett.
127
, 167204 (2021).
[11] K. Klocke, J. E. Moore, J. Alicea, and G. B. Halász, Phys. Rev. X
12
, 011034 (2022).