Parity-time symmetry breaking physics of

dissipative Mott insulators

Vikram Tripathi2,3 , Alexey Galda3 , Himadri Barman1,2 , and

Valerii M. Vinokur3

1
Department of Physics, Zhejiang Univesrity, Hangzhou, China,
2
Department of Theoretical Physics, Tata Institute of Fundamental Research, Mumbai, India,
3
Materials Science Division, Argonne National Laboratory, Argonne, USA.
P T -symmetric quantum mechanics

Conventional quantum mechanics

• ‘Reality’ of observables: x̂† = x̂, p̂† = p̂, Ĥ = Ĥ † .
• Hermitian operators (eg. Hamiltonian) have real
eigenvalues (eg. energies).

Generalized P T -symmetric quantum mechanics

• Carl Bender (1998): P T symmetric Ĥ also has real E’s.
• Eg. Ĥ = p2 + ix3 .
P Ĥ = p2 − ix3 [p → −p, x → −x];
T Ĥ = p2 − ix3 [p → −p, i → −i];
P T Ĥ = p2 + ix3 = Ĥ.
P T -symmetric quantum mechanics

Eg. Ĥ = p̂2 + x̂2 (ix̂)ε . [ε = 0 ⇒ Harmonic oscillator.]

• For ε ≥ 0, energy eigenvalues are always +ve and real.
• Near P T -breaking (exceptional points) eigenvalues merge.
Mott transition (experiment)
Correlation driven metal-to-insulator transition (MIT)

McWhan et al. , PRL 27, 941 (’71) Limelette et al. , Science 302, 89 (’03)
• High resistivity (low conductivity) ⇒ insulator.
• Low resistivity (high conductivity) ⇒ metal.
Mott metal-insulator transition (theory)
Model: Hubbard model

Ĥ = −t ∑ c†i cj − µ ∑ c†i ci + U ∑ n̂i↑ n̂i↓ . (1)

hiji i i

d = 1 solution : Bethe ansatz [E. H. Lieb and F. Y. Wu, PRL 20, 1445
(’68)]]
• Exact.
• Finds ground state energies of M ↑’s and M 0 ↓’s : E(M, M 0 ; U).
• Finds chemical potentials µ± ≡ ±E(M ± 1, M; U) ∓ E(M, M; U).
• µ+ 6= µ− ⇒ insulator.
d > 1 solution : Dynamical mean-field theory (DMFT)
• Exact at d = ∞. [Georges et al. , RMP 68, 13 (’96)]
• Finds interacting DOS or spectral function : A(ω) = − π1 ImG(ω);
G: single particle propagator or Green’s function.
• Gap at Fermi level (ω = 0) signifies insulator.
Mott metal-insulator transition (theory)
Bethe ansatz on 1-D Hubbard model
• No phase transition, always insulator at half-filling.

R∞
Eg = U − 4 + 8 0 J1 (x)/(x(1 + exp(Ux/2))).

DMFT on Hubbard model for hypercubic lattice

• A(ω) shows metal-to-insulator transition : At U > Uc2 , opens gap at
ω = 0.

[H. Barman and N. S. Vidhyadhiraja, IJMPB 25, 2461 (’11)]

Effect of drive (electric field)

Dielectric breakdown: Landau Zener physics

• Eg. Two-level system: Ĥ = vFtσ z + ∆σ x .
[Zener, Proc. R. Soc. 145, 523 (’34)]

• Transition probability: P1→2 = e−γ ;

γ = πFth /F; Fth = ∆/vF; F ≡ eE.
πm2
• Cf. Pair production rate in QED: p = e− |eE| [Schwinger ’51].
Effect of drive (electric field)

Hubbard model in a complex gauge field ψ:

Ĥ = −t ∑ [eiΨ(t) c†iσ cjσ + h.c.] + U ∑ ni↑ ni↓ .

hijiσ j

Landau-Zener-Schwinger (LZS) generalized: Landau-Dykhne formula

1
Z χC
γ∼ Re dχ 0 [E1 (χ) − E0 (χ 0 )];
F χ

ψ(t) = Ft + iχ. In non-dissipative case: get back the usual LZS

γ ∼ ∆2 /(vF) ≡ Fth /F; v = |d∆/dt|.F .

Effect of dissipation : P T -symmetric Hamiltonian
Hubbard model with only the dissipative term in the gauge field
(imaginary)

H 0 = −t ∑ [eχ c†iσ cjσ + e−χ c†jσ ciσ ] + U ∑ n̂i↑ n̂i↓

hiji,σ i

= −t(cosh χ) ∑ [c†iσ cjσ + c†jσ ciσ ] + U ∑ n̂i↑ n̂i↓

hiji,σ i

− i(sinh χ) Ĵ. (2)

This has a generic form

Ĥ 0 = Ĥ − iλ Ĵ (3)

• For small λ , hĴi = 0⇒ real eigenvalues.

• For λ > λc , hĴi = I⇒ complex eigenvalues (P T broken !!).
1D fermionic Hubbard model
Coupled Bethe ansatz equations:
1 cos k ∞
Z
ρ(k) = − dλ θ 0 (sin k − λ )σ (λ ) ,
2π 2π −∞
1 1 ∞
Z Z
σ (λ ) = − dk θ 0 (sin k − λ )ρ(k) + dλ 0 θ 0 ((λ − λ 0 )/2)σ (λ 0 ) ,
2π C 4π −∞
Z ∞
χ(b) = b − i dλ θ (λ + i sinh b)σ (λ ) .
−∞
(4)
u ≡ U/(4t), θ (x) = −2 tanh −1 (x/u), b: parameter controlling contour C
[Fukui and Kawakami, PRB 58, 16051 (’98)].

dω J1 (ω)eω sinh(b)
 Z ∞

∆(b) = 4t u − cosh(b) + ; (5)
−∞ 2π ω(1 + 22u|ω| )

• ∆(bc ) = 0, bc = sinh −1 (u).

1
• χ 0 (b) ' C(b − bc ) ⇒ χc − χ ' C1 (b − bc )2 ⇒ ∆(χ) ' C2 (χc − χ) 2 .
1
• ⇒ ∆(F) ∼ (Fc − F) 2 ⇒ γ ∼ (Fc − F)3/2 .
2D fermionic Hubbard model
Numerics : DMFT
• Hamiltonian:

Ĥ = ∑ [[−2t(cos kx + cos ky ) − iλ sin kx − µ]c†k,σ ck,σ ] + U ∑ n̂i↑ n̂i↓ .

k,σ i

• Density of states :

U=0 U = 30t∗ .
• Finite λ leads to 2D to 1D-like crossover at U = 0: splitting of
van-Hove singularity.
• Finite λ renormalizes Mott gap, gap closes at λ > λc ' 1.1.
3D fermionic Hubbard model
Numerics : DMFT
• Hamiltonian:
Ĥ = ∑ [[−2t(cos kx + cos ky + cos kz ) − iλ sin kx − µ]c†k,σ ck,σ ] + U ∑ n̂i↑ n̂i↓ .
k,σ i
• Density of states :
0.2 0.07

0.06
λ=0 λ=0
0.15 λ=0.5 λ=0.5

Spectral function, A
λ=1.0 0.05 U=40.0 λ=1.0
λ=2.0 λ=2.0
DOS, ρ0

λ=3.0
λ=4.0 0.04
0.1
0.03

0.05 0.02

0.01
0
-20 0 20 0
Frequency, ω -50 0 50
ω

U=0 U = 40t∗ .
• Not exactly 3D-1D crossover at U = 0, two peaks arise near
boundaries instead of singularities.
• Gap renormalization happens too, gap closes at λ > λc ' 2.0
Closing of gap : universality
Critical behaviors: 2D and 3D
U

40
25 15
38

36
25
34

10 32
20
30

28 20
5

Mott gap, ∆
15

0
15
0.5 1.0 1.5 2.0

10
U = 30 1.1
0.010
0.9
10 ν=0.85 fit
A

0.005
5
0.5
0.000 5
-6 -4 -2 0 2 4 6

0
0.0 0.5 1.0 1.5
0
( - )
0.78
0 1 2 3
λc-λ
c

d=2
d = 3.
• Near the Mott transition, ∆ ∼ (λc ∀ U. − λ )ν
• 2D: ν = 0.78 ± 0.03, 3D: ν ' 0.85. (Is ν = 1 for mean-field d > dc+
limit? [Brinkman and Rice, PRB 2, 4302 (’70)])
Vortex insulator-to-metal transition
Array of Nb superconducting islands on Si/SiO2 substrate

[Poccia et al. , Science 349, 1202 (’15)]

• Increasing I ⇒ Minimum (ins) to maximum (metal) flip in
differential resistance (dV/dI).
• Critical scaling law:
I − Ic±
 
dV(f , I) dV(f , I)
− |I =Ic = F . (6)
dI dI |b|ε
• Scaling collapse ⇒ ε = 2/3 ⇒ γ ∼ (Ic − I)3/2
Vortex insulator-to-metal transition
• Landau-Ginzburg-Wilson Hamiltonian:
Z
H= d2 x [D|∇Ψ|2 + m2 |Ψ|2 + u|Ψ|4 (7)

ψ: vortex field, D: vortex stiffness, m: mass, u : interaction.

• Considering Ψ(x, y, t) = eiky y−λ t u(x) [Rubinstein, Sternberg, Ma, PRL
99, 167003 (’07)] :

Duxx + i(I/ρ)xu = −(λ − m2 − ky2 )u (8)

• ⇒ Heff = −Duxx − i(I/ρ)
• P T -symmetry limits: I = 0 ⇒ E’s real, I → ∞ ⇒ E’s imaginary.
• ξ ≡ x/a, E ≡ (λ − m2 − ky2 )/ET , ET = D/a2 ⇒

uξ ξ + i(Ia/ET ρ)u = −Eu . (9)

1
• ⇒ (E1 − E0 ) ∼ ET (1 − I/Ic ) 2
• ⇒ Same universality with 1D-fermions!.
• Landau-Dykhne ⇒ γ ∼ (Ic − I)3/2 ⇒ Supports experiment.
Summary and outlook
• Effect of electric field with dissipation on a Mott insulator can be
modeled by a P T -symmetric Hubbard model.
• The P T -symmetry broken eigenstates signals onset of a
insulator-to-metal transition (dielectric breakdown).
• 1D fermionic Mott insulator (Bethe ansatz) shows transition with
critical exponent 0.5. Vortex Mott transition in superconducting
islands reflects the same universality class.
• 2D and 3D fermionic Mott insulators (DMFT) show transitions with
critical exponents ν ' 0.78 and 0.85 (Does d > dc+ reproduces
mean-field limit ν = 1?)
• 1D LZW scheme on P T -symmetric field equation explains critical
behavior in vortex Mott transition.
• Microscopic theory could be developed for vortices (bosonic BA or
B-DMFT ?).
• Benchmarking against results for dissipation treated through a
coupled reservoir [eg. Aron, PRB 86, 085127 (’12)] .
References
• V. Tripathi, A. Galda, H. Barman, V. M. Vinokurm Parity-time
symmetry-breaking mechanism of dynamic Mott transitions in
dissipative systems, Phys. Rev. B 94, 041104(R) (2016); arXiv link:
http://arxiv.org/abs/1510.08355.
• M. Lankhorst, M. Stehno, A. Galda, V. Tripathi, H. Barman, F.
Coneri, H. Hilgenkamp, A. Brinkman, T. Baturina, A. Golubov, V.
M. Vinokur, Scaling Universality at the Dynamic Vortex Mott
Transition, Phys. Rev. B 97, 020504 (R); arXiv link:
https://arxiv.org/abs/1708.00273.
Appendix A: More about P T
symmetric QM
P T symmetric QM

Pseudo norm conservation

• hψ(t)|P |ψ(t)i = hψ(0)|P |ψ(0)i . [Miloslav Znojil,
arXiv:math-ph/0104012]

Pseudo Hermiticity
• † = η̂ Âη̂, η̂: intertwinning operator.
• η = 1 ⇒ Hermitian QM.
• η = P ⇒ P T -symmetric QM.
Appendix B: DMFT
 Classical Weiss mean-field theory

• Average magnetization at site 0:

MF MF
m= ∑ S0 eβ H / ∑ eβ H
S0 =±1 S0 =±1
β hMF −β hMF MF MF
= (e −e )/(eβ h + e−β h )
= tanh(β hMF )

• Now hMF = h + ∑ Ji0 mi = h + zJm;

i∈hi0i
z= coordination number, Jij = J for nearest neighbor interaction,
mi = m = hSi i=average magnetization per site .
• Wrote hSi i = m as well since S0 is not spin of any special site, i.e.
hSi i = hS0 i.
• Thus m = tanh β (h + zJm) ⇒ self-consistent MF equation.
Quantum version: Dynamical mean field theory (DMFT)
• No order parameter or extensive quantity, but a Green’s function
(probability amplitude of an electron moving from site i to site j
starting at time τ and ending at τ 0 ):
Gij,σ (τ − τ 0 ) ≡ −hT̂ ciσ (τ)c†jσ (τ 0 )i

• No direct mean-field Hamiltonian, but an effective or local action

for an interacting lattice model (e.g. Hubbard model):
Z β Z β
Slocal = − dτ dτ 0 ∑ c†0σ (τ)G −1 (τ − τ 0 )c0σ (τ 0 )
0 0 σ
Z β
+U dτ n0↑ (τ)n0↓ (τ)
0

• G represents Green’s function of the effective conduction bath

attached to an impurity atom. ⇒ Single impurity Anderson model
(SIAM).

Ref. Georges and Kotliar, PRB 45, 6479 (1992)

 DMFT: Introduction

• A similar picture similar to the classical Weiss mean-field theory.

• Effective single-impurity Anderson model (SIAM) (e.g. a quantum

dot attached to conducting leads)
HSIAM = Hbath + Himpurity + Hhybridization
†
= ∑ ε̃q c†qσ cqσ + [(ε0 − µ) d0σ d0σ + Un0↑ n0↓ ] + ∑(Vq c†qσ d0σ + h.c.)
qσ qσ

• Host/bath/mean-field Green’s function:

G = 1/(ω + + µ − ∆(ω)); ∆(ω) = ∑ |Vq |2 /(ω + − ε̃l )
q
 DMFT: Introduction
• The impurity represents no special site. Mapping is true for each site
in the lattice ⇒ self-consistency
Impurity sector:
• Coulomb interaction at impurity site develops a self-energy for the
host Green’s function. Impurity Green’s function obtained through
Dyson’s eq.: Gimpurity −1 = G −1 − Σimpurity .
Lattice sector:
• Further simplification: d → ∞. Only site-diagonal Green’s function
(Giiσ ) contribute and self-energy of the lattice become
k-independent.
1
Glocal = ∑ G(k, ω) = ∑
k k ω+ + µ − εd + εk − Σlocal (ω)

• Self-consistency ⇒ Gimpurity = Glocal ; Σimpurity = Σlocal

• No averaging out in the time domain, i.e. quantum dynamics intact.
• Hence the mean-field is dynamical.
 DMFT: In practice

What can we do within DMFT framework?

• Green’s function tells about the spectral density (DoS):
D(ω) = − π1 ImG(ω); can be tested through ARPES experiment.
• Transport properties using Kubo formula (e.g. conductivity):

1 ∞
Z
σ1 (ω) = dteiωt hj(q, t)j(q, 0)i
h̄ω 0
0 0
 ∗ 0 0
σ0 t∗ 2 0 nF (ω ) − nF (ω + ω ) G (ω ) − G(ω + ω )
Z
= Re dω
2π 2 ω γ(ω + ω 0 ) − γ ∗ (ω 0 )
G(ω ) − G(ω + ω 0 )
0

−
γ(ω + ω 0 ) − γ(ω 0 )
• And many more: Energy, specific heat, Hall coefficient,
susceptibility.
 DMFT: Numerical steps

1. Start with a guessed G or Σ.

2. Use and impurity solver (e.g. IPT, LMA, NRG, ED) to find Σ and G .
3. Calculate the local Green’s function for a given lattice DoS (D0 ).
D0 (ε)
Z
G(ω) = dε
ω+ − ε d − ε − Σ(ω)

4. Use Dyson’s eq. to update G or Σ:

G −1 (ω) = G−1 (ω) + Σ(ω)
• The most difficult task is to find a suitable impurity solver
 Semi-analytic impurity solver
Iterated perturbation theory (IPT)[Georges, Kotliar, Jarrell, Pruschke, Cox]

U2
Σ2 (ω) = lim ∑ G0 (iω + iνm )G0 (iωp + iνm )G0 (iωp )
iω→ω + β 2 m,p

Local moment approach (LMA) [Logan, Eastwood, Vidhyadhiraja]

U2
Z ∞
Σσ (ω) = dω 0 Gσ̄ (ω − ω 0 )Πσ σ̄ (ω 0 )
2πi −∞

• Spin symmetry broken, but restored for Fermi liquid phase by

satisfying: ∑σ σ Σ(0) = |µ|U; µ = hn̂i↑ − n̂i↓ i.
 Results: Mott metal-insulator transition

DMFT+IPT: Spectral functions and resistivity

• Mott transition with thermal hysteresis observed.

[H. Barman and N. S. Vidhyadhiraja, IJMPB 25, 2461 (2011)]
Appendix C: Bethe Ansatz
Basic formalism

1. f (x1 , · · · , xN ): Amplitude of wavefunction with electrons ↓-spin

residing in sites x1 , · · · , xM and ↑-spin residing in sites xM+1 , · · · , xN .
2. Ansatz: f (x1 , · · · , xN ) = ∑P [Q, P] exp(i ∑Nj=1 kPj xQj )
where P, Q are set of N unequal real numbers.
3. Lieb-Wu equations:
M
Lkj = 2πIj + ∑ θ (2 sin kj − 2Λβ ), j = 1, 2, . . . , N;
β =1
N M
− ∑ θ (2Λα − 2 sin kj ) = 2πJα − ∑ θ (Λα − Λβ ), α = 1, 2, . . . , M;
j=1 β =1

θ (x) = −2 tan −1 (2x/U), −π ≤ θ < π,

(10)

Ij :integers for M, Jα :integers for M 0 .