Vacancy-induced low-energy states in

undoped graphene
Bhatt-Lee physics in the Kitaev model & crossover
behaviour in Chiral Orthogonal universality class. . .

Kedar Damle, IACS Dec. 2016

Tata Institute, Bombay
Motivation

I Random impurity potentials localize electron wavefunctions in

d=1
Weak-localization in d = 2, mobility edge in d = 3 . . .
(Gang of Four 1979)
I What about hopping disorder?
I Right way to ask this question: Are symmetries preserved by
disordered Hamiltonian different?
Bipartite random hopping problem

I Random (real) nearest neighbour hopping on bipartite lattices

H has bipartite or “chiral” symmetry: Every state at energy  has
partner at energy − (wavefunction changes sign on one
sublattice)
I Bipartite symmetry broken by random potentials or next-nearest
neighbour hops
The question

I  = 0 is special in such problems

Natural question: Does anything interesting happen?
Some answers in d = 1

I Density of states diverges very strongly as  → 0

ρ() ∼ ||−1 log3 (Ω/||)
1

Dyson (’53), Theodorou & Cohen (’76), Eggarter & Riedinger

(’78)
Simplest example of strong-disorder renormalization group fixed
point
Motrunich, KD, Huse (’00,’01)
I Also: Diverging length scale as  → 0 limit—Loosely identified
with localization length (not-quite . . . )
Some answers in d = 2

I Gade & Wegner (’91-93): Density of states diverges somewhat

less strongly as  → 0
ρ() ∼ ||−1 exp(−b log1/x (Ω/||)) with x = 2
Think this is arcane?
I Then try this:
Real answer has x = 3/2, due to a strong-disorder effect.
Motrunich, KD, Huse (’02); Motrunich, Ph.D thesis (’01).
confirmed field-theoretically by Mudry, Ryu, Furusaki (’03)
Vacancy-disorder

I Another kind of disorder: Missing sites in tight-binding model

Natural if substitutional impurities correspond to missing orbital
I Question: Does vacancy-disorder change the asymptotic
low-energy behaviour of ρ()?
Notice: No change in symmetries of microscopic H
Our focus today

I Vacancies in tight-binding model for graphene

I Vacancies modeled by deleteting sites
No interactions, no warping, no substrate charges, single-band
model . . .
Simplest possible abstraction of complicated system
Switch topics: Bhatt-Lee physics in Si:P

e−
P+
e−
P+
e−
+
P

I Low density of P dopants in Si → Half-filled “Hubbard model” on

random lattice
Electrical insulator
I At low energies: Physics of S = 1/2 local moments
Low energy spin physics

S2 S4
S1

S5
S3
S6
~ · ~Sj with broad distribution of Jij
P
I
i,j Jij Si
I Singlet pairs with broad distribution of binding energies
I Tχ(T): Pairs with binding energy < T
χ(T) ∼ N(T) 1
T ∼ T α with α set by concentration of P
(Bhatt & Lee)
Asymptotically exact?

I In d = 1, picture asymptotically exact for the random-exchange

antiferromagnetic chain
Γ−2
χ(T) = TT as T → 0.
(ΓT ≡ log(J/T) [J: overall scale of antiferromagnetic exchange].)
(Dasgupta & Ma, D. S. Fisher)
I For d > 1, status unclear (strong-disorder RG inconclusive)
(Motrunich & Huse)
Bhatt-Lee physics of diluted SU(2) symmetric
Majorana spin liquid

I Tractable example of a disordered SU(2) symmetric Majorana

spin liquid in d = 2
T)
with χ(T) = 4T C
+ N(Γ
4T as T → 0
I N(ΓT ) consistent with Bhatt-Lee physics
N(ΓT ) ∼ Γ−y
T for Tcr  T  J
1/3 2/3
N(ΓT ) ∼ ΓT exp(−cΓT ) for T  Tcr
Asymptotically exact realization of Bhatt-Lee physics

I Following Bhatt-Lee—
C → Density of free-moments
N(ΓT ) → Density of singlet-pairs with binding energies smaller
than T
Raises (interesting?) question: Alternate Strong-disorder RG
approach to go beyond tractable limit?
Connection to chiral orthogonal universality class and
Graphene

I χ(T) ∝ κ(T) for particle-hole-symmetric canonical free-fermions

with vacancy disorder.
N(Γ) → integrated DOS for single-particle energies
0 < || < J × 10−Γ (i.e. excluding zero modes)
I Vacancy-induced crossover in DOS in chiral orthogonal
universalitty class
Another example of same crossover: Undoped graphene with
vacancy disorder
Setting: Honeycomb model of Yao & Lee

X X
H=J τ~rλ τ~rλ0 ~S~r · ~S~r0 − ~B · ~S~r . (1)
h~r~r0 i λ ~r

I ~τ : “Orbital degrees of freedom that remain dynamical at low

energy
I ~S = ~σ2 : spin-half moments
I Original motivation: Low-energy effective Hamiltonian for a
frustrated S = 1/2 model on the decorated honeycomb lattice
with multi-spin interactions
Each ~S: Low-energy projection of total spin of three spins.
τ z = ±1: Two different low energy doublets that make up low
energy sector
Majorana representation

I σ~rz = −ic~xr c~yr

τ~rz = −ib~xr b~yr
and cyclic permutations
I c~λr and b~λr are Majorana (real) fermion operators.
Single-site Hilbert space doubled by this representation
Constraint on fermion states

I D~r ≡ −ic~xr c~yr c~zr b~xr b~yr b~zr = +1 at each site ~r

Curious fact: D = −1 sector also provides faithful representation
of ~σ and ~τ .
→
No “unphysical” states. Instead: Two copies of physical states at
each site
I In D = +1 sector: σ~rα τ~rβ = ic~α β
r b~r
Similar reduction in D = −1 sector
Reduction leads to exact solution

I On bond with orientation λ (λ = x, y, z) hrr0 iλ, get term:

uhrr0 iλ (i~cr · ~cr0 )
with uhrr0 iλ = −ibλr bλr0
I Three copies of Kitaev’s non-interacting Majorana model, all
coupled to same static Z2 gauge field
Majorana fermion Hamiltonian

J X X X
H= uh~r~r0 iλ (ic~α α
r c~r0 + h.c.) + B ic~xr c~yr (2)
2 α=x,y,z 0
h~r~r iλ ~r

where ~B = Bẑ.
I Convenient: Canonical fermions f~r = (c~xr − ic~yr )/2
I S~rz = ic~xr c~yr = f~r† f~r − 1/2
z 2
Want to compute: mz ≡
P
r hS~r i/2L as function of B and obtain
I
z
χ(T) = dm
dB at B = 0
Calculating susceptibility

I Hamiltonian H for f fermions:

Tight-binding model with static Z2 gauge-fields u determining
signs of each hopping matrix element t = u|J|
I χ(T) related to f fermion compressibility κ(T) at
particle-hole-symmetric chemical potential µ ≡ B = 0.
I cz Majorana plays no role in susceptibility calculation
I χ(T) = 1 dρtot. () /T e/T
R
T (e +1)2
where ρtot () is full DOS of H
Projection issues?

I In usual Kitaev model: Projection gives subleading corrections in

thermodynamic limit
Subtle for impurity susceptibility etc
(Pedrocchi-Chesi-Loss, Zschocke-Vojta)
(building on: Willans-Chalker-Moessner,
Baskaran-Mandal-Shankar, Yao-Zhang-Kivelson)
I What happens here?
Again: Only subleading corrections in general.
I For specific boundary conditions: Coefficient of subleading
corrections zero
Flux-binding

I Lieb-Loss heuristics:
Each vacancy binds static π-flux in ground-state sector.
Gap to other flux sectors
(Kitaev, Willans-Chalker-Moessner)
I At low temperature, χ dominated by this flux-sector
Flux-attachment

I Send flux-strings off to one open edge

-1
-0.5 0 0.5 1 1.5 2 2.5 3
Connection to vacancy-impurities in Graphene

I Without flux-attachment, H is tight-binding model for graphene

with compensated vacancies
I Study numerically with and without flux attachment
Dilution

I Remove honeycomb lattice sites at random

I Global “compensation”: Equal number of vacancies on A and B
sublattices
Protocol: Randomly/alternately pick sublattice, then pick random
site to remove, ensure global compensation
I Exclusion constraints on vacancies—prevent disconnecting
lattice into clusters, prevent dangling bonds.
I Choices eliminate “graph zeroes”
Choice of geometry

I Semi-open L × L unit cells (2L2 sites in undiluted sample) and

armchair edges
I Vacancies excluded from interrupting armchair edge
I L chosen even and antiperiodic boundary conditions or odd and
periodic boundary conditions
I Choices eliminate boundary-induced graph zeroes.
I Any zero modes will now be “nontrivial”
Computational details

I Form H 2 , the square of the tight-binding Hamiltonian H (with

hopping amplitude t = 1 between nearest-neighbours), and work
with the (1 − nv )L2 × (1 − nv )L2 block (TAB )† TAB where TAB is the
matrix of hopping amplitudes from undeleted B sublattice sites to
undeleted A sublattice sites
I Fully multiprecision implementation of the ALGOL routines in
Wilkinson’s handbook to count eigenvalues of (TAB )† TAB below
10−2Λ .
I Results checked at moderate L and moderate Λ against
LAPACK routines.
Formulary

I ρtot () = ρ() + wδ()

R 10−Γ
I N(Γ) = 2 ρ()d
0
(Γ = log10 (1/||))
I Modified Gade-Wegner form:
1/x
ρ(E) ∼ || 1
e−b| ln |
equivalently:
1
1
N(Γ) = aΓ1− x e−bΓ x
(x = 3/2, two free parameters a and b)
I Dyson form:
1
ρ() ∼
||[log[1/||]]1+y
equivalently:
N(Γ) = cΓ−y (two free parameters c and y)
Computational details—II

I For each sample, computations first done in a coarse-grid of

Λ = Γi , then Ntot (Γ) “filled in” iteratively when needed. Final grid
spacing ∆(Γ) = 0.5.
I Lowest-nonzero gap 10−Γg in a given sample obtained with
accuracy of ∆(Γg ) = 0.5.
I wL (number of zero modes per unit volume) given by Ntot (Γ, L)
after “last” downward step.
I “Last”: Grid extends to Γmax as high as 100—stability checked by
varying precision
I Study NL (Γ) = Ntot (Γ, L) − wL and wL for ∼ 3000 samples
Alternate (less painful) protocol

I Keep track of differences Ntot (Γi+1 ) − Ntot (Γi )

I Obtain ||ρ() directly
I Poorer statistical properties, but finesses zero mode question
Zero modes

nv= 0.05
1
nv= 0.0625
nv= 0.075
0.99 nv= 0.1

0.98
fL

0.97

0.96

0.95
0 0.002 0.004 0.006 0.008
1/L
Zero modes
0.008
0.004
nv= 0.05
nv= 0.0625
nv= 0.075
w 0.002
0.006 nv= 0.1

0
0.04 0.06 0.08 0.1
nv
wL

0.004

0.002

0
0 0.002 0.004 0.006 0.008
1/L
Rigorous lower-bound on zero modes

+1
−1 −1
+1 −1 −1
−1 +1 +2 +1
+1 −1 −1
−1 −1
+1

h 4-triangle
i and R6 motifs
(i) (i) (i) 2
wL ≥ max(N∆4A , N∆4B )/L
Implies w ≥ n∆4 (concentration of 4-triangles)
Robust to hopping disorder

I R6 mode robust to bond disorder (but not 4-triangles).

I More general R-type zero modes possible, also robust to bond
disorder
I Dominate over 4-triangles except for asymptotically small nv (out
of reach)
N(Γ) for nv = 0.05
nv= 0.05
−2
10
L=220
L=200
L=180
N1D, y= 1.56(1)
−3 NGW
10

Γc
NL(Γ)

−4
10

−5
10 2 4 8 12 16 20 24
Γ
Taking the thermodynamic limit
nv= 0.05

0.00235
NL(Γ= 4)

0.0023

0.00225

0.0022
0 0.001 0.002 0.003 0.004 0.005 0.006
1/L
nv= 0.05
0.00025

0.00024

0.00023
NL(Γ= 16)

0.00022

0.00021

0.0002

0.00019

0.00018
0 0.001 0.002 0.003 0.004 0.005 0.006
1/L
Thermodynamic limit: nv = 0.05
nv= 0.05
-2
10

Thermodynamic Limit
N1D , y=1.55(1)
NGW
-3
10
N(Γ)

Γc

-4
10

-5
10 2 4 8 12 16 20 24
Γ
Γ∗gap
nv= 0.05

L=180
0.03 L=200
L=220

0.02
P(Γg)

0.01

0 10 20 30 40 50 60 70
Γg
N(Γ) for nv = 0.0625
nv= 0.0625
−2
10
L=200
L=180
L=160
N1D( Γ), y=1.76(1)
−3 NGW
10 Γc
-2
10
NL(Γ)

-3
10
N(Γ)

−4 Γc
10
-4
10 Thermodynamic limit
N1D, y=1.82(1)
NGW
-5
10 2 4 Γ 6 8 10 12 141618
−5
10 2 4 6 8 10 12 14 16 18
Γ
N(Γ) for nv = 0.075
nv= 0.075
−2
10
L=180
L=160
L=140
N1D, y= 1.91(1)
−3 NGW
10 Γc
-2
10
NL(Γ)

-3
10
Γc
N(Γ)

−4
10
-4
10 Thermodynamic limit
N1D, y= 1.96(1)
NGW
-5
10 2 4 6 8 10 12
Γ
−5
10
2 4 6 8 10 12
Γ
N(Γ) for nv = 0.1
nv= 0.1
−2
10
L=180
L=160
L=140
N1D(Γ), y= 2.38(5)
Γc
−3 NGW
10
-2
10
NL(Γ)

-3 Γc
10
N(Γ)

−4
10
-4
10 Thermodynamic limit
N1D(Γ), y= 2.43(5)
NGW
-5
10 2 4 6 8
Γ
−5
10
2 4 6 8
Γ
Γ∗gap
nV= 0.075
0.08

L=140
L=160
L=180
0.06
P(Γg)

0.04

0.02

0
0 10 20 30 40
Γg
Γ∗gap
Γg
0 10 20 30 40 50
0.05

0.04 nv= 0.0625

L= 160
0.03 L= 180
P(Γg)

L= 200
0.02

0.01

0
0.12
0.1 nv= 0.1
L= 140
0.08 L= 160
P(Γg)

0.06 L=180
0.04
0.02
0
0 5 10 15 20 25
Γg
Physics:

I w depends on nv and correlations between positions of

vacancies
I Is crossover Γc and intermediate asymptotic exponent y
“quasi-universal”?
I Operational definition of “Quasi-Universality”:
Vary nv , vary correlations.
If resulting w is same, crossover Γc and y same...
Toy-model: Dilution by 4-triangles (nv = 0.0049)
40 4-triangles on each sublattice
−2
10
L=180
N1D, y=1.69(2)
NGW

−3
10
NL(Γ)

Γc
−4
10

−5
10
2 4 6 8 10 12 14 16 18 20
Γ
Check crossover systematics
3
15
Thermodynamic
Thermodynamic
Largest size Largest size
10
Γc

2.5
5

0
0 0.001 0.002 0.003 0.004
y

w
2

1.5

0 0.001 0.002 0.003 0.004

w
Filled symbol: 4-triangle diluted sample
Another aspect of crossover systematics
50 Thermodynamic limit data
Largest size data

40

30
lc

20

10

0
0 10 20 30 40 50
lw
p
lc ≡ 1/ N(Γc )
√
lw ≡ 1/ w
Caution: Not claiming lc = lw , only lc ∼ lw .
Kitaev: Zero modes

1 nv= 0.05
nv= 0.0625
nv= 0.075
0.98 nv= 0.1
fL

0.96

0.94

0.92
0 0.002 0.004 0.006 0.008
1/L
Kitaev: Zero modes
0.008
0.004
nv= 0.05
nv= 0.0625
nv= 0.075
C
0.002
0.006 nv= 0.1

0
0.05 0.07 0.09
nv
CL

0.004

0.002

0
0 0.002 0.004 0.006 0.008
1/L
Kitaev: N(Γ)
nv= 0.05
−2
10
L=200
L=180
L=160
N1D( Γ), y=1.30(0)
−3 NGW( Γ)
10
-2
10 Γc
NL(Γ)

-3
10
N(Γ)

−4 Γc
10
-4
10 Thermodynamic limit
N1D, y=1.31(1)
NGW
-5
10 2 4 6 Γ 8 10 12 16 20 24
−5
10
2 4 6 8 10 12 16 20 24
Γ
Kitaev: N(Γ)
nv= 0.075
−2
10
L=180
L=160
L=140
N1D, y= 1.59(1)
−3 NGW
10 Γc
-2
10
NL(Γ)

-3
10 Γc
N(Γ)

−4
10
-4
10 Thermodynamic limit
N1D, y= 1.65(2)
NGW
-5
10 2 4 Γ 6 8 10 12 14
−5
10
2 4 6 8 10 12 14
Γ
Kitaev: Crossover systematics

50 Thermodynamic limit data

Largest size data

40

30
lc

20

10

0
0 10 20 30 40 50
lw
Comments on other work

Evers group: 0 < y < 1

(Hafner et. al. 2014)
Mirlin group prediction: y = 0.5
(Ostrovsky et. al. 2014)
Willans-Chalker-Moessner (in gapped phase of Kitaev): y = 0.7
???...
Acknowledgements

I Collaborators:
Sambuddha Sanyal (ICTS-TIFR) and Olexei Motrunich (Caltech)
I Computational resources at TIFR and ICTS-TIFR