Post on 09-Oct-2020
Correlations after quantum quenches in the XXZspin chain: Failure of the Generalized Gibbs
Ensemble
Gábor Takács
Department of Theoretical PhysicsBudapest University of Technology and Economics
andMTA-BME “Momentum” Statistical Field Theory Research Group
B. Pozsgay, M. Mestyán, M. A. Werner, M. Kormos, G. Zaránd and G. Takács:Correlations after quantum quenches in the XXZ spin chain: Failure of the
Generalized Gibbs EnsemblePhys. Rev. Lett. 113 (2014) 117203
28th April, 2015HAS Institute for Nuclear Research, Debrecen
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Outline
1 IntroductionThermalization in isolated classical systemsThermalization in isolated quantum systemsIntegrable systems and the GGE hypothesis
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Outline
1 IntroductionThermalization in isolated classical systemsThermalization in isolated quantum systemsIntegrable systems and the GGE hypothesis
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Thermalization in isolated classical systems
Isolated classical system: N degrees of freedom, energy E, volume V
−→ microcanonical ensemble
Phase space Γ : (q1, . . . , qN , p1, . . . , pN)Dynamics: Hamiltonian H
qi =∂H∂pi
pi = −∂H∂qi
Fundamental postulate of statistical physicsThe probability density in equilibrium is
ρ(q, p) =
const. (q, p) ∈ Γ(N,E ,V )
0 otherwise
Thermalization in isolated classical systems
Isolated classical system: N degrees of freedom, energy E, volume V
−→ microcanonical ensemble
Phase space Γ : (q1, . . . , qN , p1, . . . , pN)Dynamics: Hamiltonian H
qi =∂H∂pi
pi = −∂H∂qi
Fundamental postulate of statistical physicsThe probability density in equilibrium is
ρ(q, p) =
const. (q, p) ∈ Γ(N,E ,V )
0 otherwise
Thermalization in isolated classical systems
Isolated classical system: N degrees of freedom, energy E, volume V
−→ microcanonical ensemble
Phase space Γ : (q1, . . . , qN , p1, . . . , pN)Dynamics: Hamiltonian H
qi =∂H∂pi
pi = −∂H∂qi
Fundamental postulate of statistical physicsThe probability density in equilibrium is
ρ(q, p) =
const. (q, p) ∈ Γ(N,E ,V )
0 otherwise
Thermalization in isolated classical systems
Isolated classical system: N degrees of freedom, energy E, volume V
−→ microcanonical ensemble
Phase space Γ : (q1, . . . , qN , p1, . . . , pN)Dynamics: Hamiltonian H
qi =∂H∂pi
pi = −∂H∂qi
Fundamental postulate of statistical physicsThe probability density in equilibrium is
ρ(q, p) =
const. (q, p) ∈ Γ(N,E ,V )
0 otherwise
Boltzmann ergodic hypothesis
Boltzmann (1871): ergodic hypothesisFor an observable A(q, p)
time average: A = limT→∞
ˆ T
0dtA(q(t), p(t))
mc average: 〈A〉mc =1
vol(Γ(N,E ,V ))
ˆΓ(N,E ,V )
dNqdNpA(q, p)
Ergodicity:
A = 〈A〉mc
justifies use of microcanonical ensemble.It results from dynamical chaos:
individual trajectory homogeneously spreads over Γ(N,E ,V )
Boltzmann ergodic hypothesis
Boltzmann (1871): ergodic hypothesisFor an observable A(q, p)
time average: A = limT→∞
ˆ T
0dtA(q(t), p(t))
mc average: 〈A〉mc =1
vol(Γ(N,E ,V ))
ˆΓ(N,E ,V )
dNqdNpA(q, p)
Ergodicity:
A = 〈A〉mc
justifies use of microcanonical ensemble.It results from dynamical chaos:
individual trajectory homogeneously spreads over Γ(N,E ,V )
Boltzmann ergodic hypothesis
Boltzmann (1871): ergodic hypothesisFor an observable A(q, p)
time average: A = limT→∞
ˆ T
0dtA(q(t), p(t))
mc average: 〈A〉mc =1
vol(Γ(N,E ,V ))
ˆΓ(N,E ,V )
dNqdNpA(q, p)
Ergodicity:
A = 〈A〉mc
justifies use of microcanonical ensemble.It results from dynamical chaos:
individual trajectory homogeneously spreads over Γ(N,E ,V )
Outline
1 IntroductionThermalization in isolated classical systemsThermalization in isolated quantum systemsIntegrable systems and the GGE hypothesis
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Time evolution in isolated quantum systems
Quantum systems differ fundamentally from classical ones:if thermalization occurs, it must have a fundamentally differentexplanation as time evolution is linear!
|ψ(t = 0)〉 =∑α
Cα|ψα〉 Cα : overlaps
H|ψα〉 = Eα|ψα〉
Time evolution|ψ(t)〉 =
∑α
Cαe−i~Eαt |ψα〉
Observables
〈A(t)〉 = 〈ψ(t)|A|ψ(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
Aαβ = 〈ψα|A|ψβ〉
Time evolution in isolated quantum systems
Quantum systems differ fundamentally from classical ones:if thermalization occurs, it must have a fundamentally differentexplanation as time evolution is linear!
|ψ(t = 0)〉 =∑α
Cα|ψα〉 Cα : overlaps
H|ψα〉 = Eα|ψα〉
Time evolution|ψ(t)〉 =
∑α
Cαe−i~Eαt |ψα〉
Observables
〈A(t)〉 = 〈ψ(t)|A|ψ(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
Aαβ = 〈ψα|A|ψβ〉
Time evolution in isolated quantum systems
Quantum systems differ fundamentally from classical ones:if thermalization occurs, it must have a fundamentally differentexplanation as time evolution is linear!
|ψ(t = 0)〉 =∑α
Cα|ψα〉 Cα : overlaps
H|ψα〉 = Eα|ψα〉
Time evolution|ψ(t)〉 =
∑α
Cαe−i~Eαt |ψα〉
Observables
〈A(t)〉 = 〈ψ(t)|A|ψ(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
Aαβ = 〈ψα|A|ψβ〉
Time evolution in isolated quantum systems
Quantum systems differ fundamentally from classical ones:if thermalization occurs, it must have a fundamentally differentexplanation as time evolution is linear!
|ψ(t = 0)〉 =∑α
Cα|ψα〉 Cα : overlaps
H|ψα〉 = Eα|ψα〉
Time evolution|ψ(t)〉 =
∑α
Cαe−i~Eαt |ψα〉
Observables
〈A(t)〉 = 〈ψ(t)|A|ψ(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
Aαβ = 〈ψα|A|ψβ〉
The diagonal ensembleMicrocanonical setting: start with narrow energy distribution
〈E 〉 =∑α
|Cα|2Eα
∆E =
√∑α
|Cα|2 (Eα − 〈E 〉)2 〈E 〉
Time average
〈A(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
⇓
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα
Diagonal ensemble
A = Tr ρDEA
ρDE =∑α
|Cα|2|ψα〉〈ψα|
The diagonal ensembleMicrocanonical setting: start with narrow energy distribution
〈E 〉 =∑α
|Cα|2Eα
∆E =
√∑α
|Cα|2 (Eα − 〈E 〉)2 〈E 〉
Time average
〈A(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
⇓
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα
Diagonal ensemble
A = Tr ρDEA
ρDE =∑α
|Cα|2|ψα〉〈ψα|
The diagonal ensembleMicrocanonical setting: start with narrow energy distribution
〈E 〉 =∑α
|Cα|2Eα
∆E =
√∑α
|Cα|2 (Eα − 〈E 〉)2 〈E 〉
Time average
〈A(t)〉 =∑α
C ∗αCβe− i
~ (Eα−Eβ)tAαβ
⇓
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα
Diagonal ensemble
A = Tr ρDEA
ρDE =∑α
|Cα|2|ψα〉〈ψα|
Eigenstate thermalization hypothesisStationary state: for large t
A(t) =∑α
C∗αCβe−i~ (Eα−Eβ )tAαβ → A =
∑α
|Cα|2Aαα
Microcanonical average
〈A〉mc =1
N(〈E〉 ,∆E)
∑α:Eα∈I
Aαα I = [〈E〉 −∆E , 〈E〉+ ∆E ]
Deutsch (1991): “quantum ergodicity”∑α
|Cα|2Aαα =1
N(〈E 〉 ,∆E )
∑α:Eα∈I
Aαα
Eigenstate thermalization hypothesis, ETH (Srednicki, 1994)
Aαα = 〈A〉mc (Eα) ∀α
Supposed to hold for large systems (thermodynamic limit – TDL)
Eigenstate thermalization hypothesisStationary state: for large t
A(t) =∑α
C∗αCβe−i~ (Eα−Eβ )tAαβ → A =
∑α
|Cα|2Aαα
Microcanonical average
〈A〉mc =1
N(〈E〉 ,∆E)
∑α:Eα∈I
Aαα I = [〈E〉 −∆E , 〈E〉+ ∆E ]
Deutsch (1991): “quantum ergodicity”∑α
|Cα|2Aαα =1
N(〈E 〉 ,∆E )
∑α:Eα∈I
Aαα
Eigenstate thermalization hypothesis, ETH (Srednicki, 1994)
Aαα = 〈A〉mc (Eα) ∀α
Supposed to hold for large systems (thermodynamic limit – TDL)
Eigenstate thermalization hypothesisStationary state: for large t
A(t) =∑α
C∗αCβe−i~ (Eα−Eβ )tAαβ → A =
∑α
|Cα|2Aαα
Microcanonical average
〈A〉mc =1
N(〈E〉 ,∆E)
∑α:Eα∈I
Aαα I = [〈E〉 −∆E , 〈E〉+ ∆E ]
Deutsch (1991): “quantum ergodicity”∑α
|Cα|2Aαα =1
N(〈E 〉 ,∆E )
∑α:Eα∈I
Aαα
Eigenstate thermalization hypothesis, ETH (Srednicki, 1994)
Aαα = 〈A〉mc (Eα) ∀α
Supposed to hold for large systems (thermodynamic limit – TDL)
Eigenstate thermalization hypothesisStationary state: for large t
A(t) =∑α
C∗αCβe−i~ (Eα−Eβ )tAαβ → A =
∑α
|Cα|2Aαα
Microcanonical average
〈A〉mc =1
N(〈E〉 ,∆E)
∑α:Eα∈I
Aαα I = [〈E〉 −∆E , 〈E〉+ ∆E ]
Deutsch (1991): “quantum ergodicity”∑α
|Cα|2Aαα =1
N(〈E 〉 ,∆E )
∑α:Eα∈I
Aαα
Eigenstate thermalization hypothesis, ETH (Srednicki, 1994)
Aαα = 〈A〉mc (Eα) ∀α
Supposed to hold for large systems (thermodynamic limit – TDL)
Eigenstate thermalization hypothesisStationary state: for large t
A(t) =∑α
C∗αCβe−i~ (Eα−Eβ )tAαβ → A =
∑α
|Cα|2Aαα
Microcanonical average
〈A〉mc =1
N(〈E〉 ,∆E)
∑α:Eα∈I
Aαα I = [〈E〉 −∆E , 〈E〉+ ∆E ]
Deutsch (1991): “quantum ergodicity”∑α
|Cα|2Aαα =1
N(〈E 〉 ,∆E )
∑α:Eα∈I
Aαα
Eigenstate thermalization hypothesis, ETH (Srednicki, 1994)
Aαα = 〈A〉mc (Eα) ∀α
Supposed to hold for large systems (thermodynamic limit – TDL)
Thermalization in closed quantum systems
1 ETH is time independent: every eigenstate of Hamiltonian isalready thermal. It is masked by quantum coherence, butrevealed by dephasing effect of time evolution.
2 Quantum ergodic theorem (Neumann): for a typical finitefamily of commuting observables, every initial wave functionfrom a microcanonical shell evolves so that for most times theaverages of the given observables is close to theirmicrocanonical value.Rigol & Srednicki, 2012:
Neumann’s main technical assumption is equivalentto the Eigenstate Thermalization Hypothesis
3 What observables?1 local2 few-body
Outline
1 IntroductionThermalization in isolated classical systemsThermalization in isolated quantum systemsIntegrable systems and the GGE hypothesis
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Integrable systems violate ergodicity
Fermi-Pasta-Ulam (1953; Los Alamos technical report 1955)
mxj = k(xj+1 + xj−1 − 2xj) [1 + α(xj+1 − xj−1)] j = 1, . . . ,N
Exciting a vibration mode: thermalization expected.Found instead: complicated quasi-periodic behaviour!Reason: system is integrable! (discretized version of KdV)Integrability: action-angle variables exist
[Qi ,Qj ] = 0 H ∈ Qii=1,...,N
The quantities are conserved: motion along tori inphase space −→Lissajous orbits with frequencies ω1, . . . , ωN→ may be ergodic on the tori→ ? microcanonical ensemble on the tori ?
Generalized Eigenstate Thermalization Hypothesis
Ergodicity on tori
⇒ at quantum level:
Generalized Eigenstate Thermalization Hypothesis (GETH)
ETH holds on Q-shells i.e. subspaces determined by fixing the Qi
Generalized Eigenstate Thermalization Hypothesis
Ergodicity on tori
⇒ at quantum level:
Generalized Eigenstate Thermalization Hypothesis (GETH)
ETH holds on Q-shells i.e. subspaces determined by fixing the Qi
Gibbs and generalized Gibbs ensemblesNonintegrable system: steady state is thermalDE is equivalent to a Gibbs ensemble for relevant observables
Tr ρDEA = Tr ρGEA
ρGE =1Z
e−βH Z = Tr e−βH
Determining the temperature: Tr ρGEH = 〈ψ(0)|H|ψ(0)〉 → β
Generalized Gibbs Ensemble (GGE)
natural assumption: in integrable systems quantum ergodicity holdson common eigensubspaces of Qi
ρGGE =1Z
e−∑
i βiQi Z = Tr e−∑
i βiQi
Generalized temperatures/chemical potentials
Tr ρGGEQi = 〈ψ(0)|Qi |ψ(0)〉 i = 1, . . . ,N → βii=1,...,N
GE/GGE: follow from conditional maximum entropy principle.
Gibbs and generalized Gibbs ensemblesNonintegrable system: steady state is thermalDE is equivalent to a Gibbs ensemble for relevant observables
Tr ρDEA = Tr ρGEA
ρGE =1Z
e−βH Z = Tr e−βH
Determining the temperature: Tr ρGEH = 〈ψ(0)|H|ψ(0)〉 → β
Generalized Gibbs Ensemble (GGE)
natural assumption: in integrable systems quantum ergodicity holdson common eigensubspaces of Qi
ρGGE =1Z
e−∑
i βiQi Z = Tr e−∑
i βiQi
Generalized temperatures/chemical potentials
Tr ρGGEQi = 〈ψ(0)|Qi |ψ(0)〉 i = 1, . . . ,N → βii=1,...,N
GE/GGE: follow from conditional maximum entropy principle.
Gibbs and generalized Gibbs ensemblesNonintegrable system: steady state is thermalDE is equivalent to a Gibbs ensemble for relevant observables
Tr ρDEA = Tr ρGEA
ρGE =1Z
e−βH Z = Tr e−βH
Determining the temperature: Tr ρGEH = 〈ψ(0)|H|ψ(0)〉 → β
Generalized Gibbs Ensemble (GGE)
natural assumption: in integrable systems quantum ergodicity holdson common eigensubspaces of Qi
ρGGE =1Z
e−∑
i βiQi Z = Tr e−∑
i βiQi
Generalized temperatures/chemical potentials
Tr ρGGEQi = 〈ψ(0)|Qi |ψ(0)〉 i = 1, . . . ,N → βii=1,...,N
GE/GGE: follow from conditional maximum entropy principle.
Gibbs and generalized Gibbs ensemblesNonintegrable system: steady state is thermalDE is equivalent to a Gibbs ensemble for relevant observables
Tr ρDEA = Tr ρGEA
ρGE =1Z
e−βH Z = Tr e−βH
Determining the temperature: Tr ρGEH = 〈ψ(0)|H|ψ(0)〉 → β
Generalized Gibbs Ensemble (GGE)
natural assumption: in integrable systems quantum ergodicity holdson common eigensubspaces of Qi
ρGGE =1Z
e−∑
i βiQi Z = Tr e−∑
i βiQi
Generalized temperatures/chemical potentials
Tr ρGGEQi = 〈ψ(0)|Qi |ψ(0)〉 i = 1, . . . ,N → βii=1,...,N
GE/GGE: follow from conditional maximum entropy principle.
Gibbs and generalized Gibbs ensemblesNonintegrable system: steady state is thermalDE is equivalent to a Gibbs ensemble for relevant observables
Tr ρDEA = Tr ρGEA
ρGE =1Z
e−βH Z = Tr e−βH
Determining the temperature: Tr ρGEH = 〈ψ(0)|H|ψ(0)〉 → β
Generalized Gibbs Ensemble (GGE)
natural assumption: in integrable systems quantum ergodicity holdson common eigensubspaces of Qi
ρGGE =1Z
e−∑
i βiQi Z = Tr e−∑
i βiQi
Generalized temperatures/chemical potentials
Tr ρGGEQi = 〈ψ(0)|Qi |ψ(0)〉 i = 1, . . . ,N → βii=1,...,N
GE/GGE: follow from conditional maximum entropy principle.
Closed quantum systems in reality
Can be realized with cold atoms...
Closed quantum systems in reality
Can be realized with cold atoms... ...which sometimes do not thermalize
Closed quantum systems in reality
Can be realized with cold atoms... ...which sometimes do not thermalize
T. Kinoshita, T. Wenger and D.S. Weiss:A quantum Newton’s cradle, Nature 440 (2006) 900-903.
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chainIntegrability, Bethe Ansatz and thermodynamicsQuantum quenchGGE in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chainIntegrability, Bethe Ansatz and thermodynamicsQuantum quenchGGE in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
The Heisenberg XXZ spin chainA paradigmatic integrable model: chain of 1/2 spins
. . . ↑↓↓↓↑↑↓ . . . −→ H =⊗
L
C2 ↓
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
PBC: σkL+1 ≡ σk
1
∆ > 1: gapped phase, ∆→∞: quantum Ising chain−1 < ∆ < 1: gapless phase (Luttinger liquid)
The chain is integrable and can be diagonalized by Bethe Ansatz:
|λ1, . . . , λN〉 = B(λ1) . . .B(λM)|Ω〉|Ω〉 = | . . . ↑↑↑ . . . 〉 pseudovacuum(
sin (λj + iη/2)
sin (λj − iη/2)
)L
=M∏
k 6=j
sin (λj − λk + iη)
sin (λj − λk − iη)∆ = cosh η
Qn|λ1, . . . , λN〉 =n∑
i=1
qn(λi )|λ1, . . . , λN〉
The conserved charges Qn are local → additivity!
The Heisenberg XXZ spin chainA paradigmatic integrable model: chain of 1/2 spins
. . . ↑↓↓↓↑↑↓ . . . −→ H =⊗
L
C2 ↓
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
PBC: σkL+1 ≡ σk
1
∆ > 1: gapped phase, ∆→∞: quantum Ising chain−1 < ∆ < 1: gapless phase (Luttinger liquid)
The chain is integrable and can be diagonalized by Bethe Ansatz:
|λ1, . . . , λN〉 = B(λ1) . . .B(λM)|Ω〉|Ω〉 = | . . . ↑↑↑ . . . 〉 pseudovacuum(
sin (λj + iη/2)
sin (λj − iη/2)
)L
=M∏
k 6=j
sin (λj − λk + iη)
sin (λj − λk − iη)∆ = cosh η
Qn|λ1, . . . , λN〉 =n∑
i=1
qn(λi )|λ1, . . . , λN〉
The conserved charges Qn are local → additivity!
The Heisenberg XXZ spin chainA paradigmatic integrable model: chain of 1/2 spins
. . . ↑↓↓↓↑↑↓ . . . −→ H =⊗
L
C2 ↓
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
PBC: σkL+1 ≡ σk
1
∆ > 1: gapped phase, ∆→∞: quantum Ising chain−1 < ∆ < 1: gapless phase (Luttinger liquid)
The chain is integrable and can be diagonalized by Bethe Ansatz:
|λ1, . . . , λN〉 = B(λ1) . . .B(λM)|Ω〉|Ω〉 = | . . . ↑↑↑ . . . 〉 pseudovacuum(
sin (λj + iη/2)
sin (λj − iη/2)
)L
=M∏
k 6=j
sin (λj − λk + iη)
sin (λj − λk − iη)∆ = cosh η
Qn|λ1, . . . , λN〉 =n∑
i=1
qn(λi )|λ1, . . . , λN〉
The conserved charges Qn are local → additivity!
Excitations and string hypothesis
Sz =L2−M
Ground state: unique solution with M = L/2 and all λi real
|0〉 = B(λ1) . . .B(λL/2)|Ω〉 λi ∈ R
Excitations: can be holes in the Dirac sea, or
n-string configurations
λj = λ+iη2
(n + 1− 2j)
+O(1/L)
j = 1, . . . , n
String hypothesisThermodynamics can be described by string degrees of freedom
Excitations and string hypothesis
Sz =L2−M
Ground state: unique solution with M = L/2 and all λi real
|0〉 = B(λ1) . . .B(λL/2)|Ω〉 λi ∈ R
Excitations: can be holes in the Dirac sea, or
n-string configurations
λj = λ+iη2
(n + 1− 2j)
+O(1/L)
j = 1, . . . , n
String hypothesisThermodynamics can be described by string degrees of freedom
Excitations and string hypothesis
Sz =L2−M
Ground state: unique solution with M = L/2 and all λi real
|0〉 = B(λ1) . . .B(λL/2)|Ω〉 λi ∈ R
Excitations: can be holes in the Dirac sea, or
n-string configurations
λj = λ+iη2
(n + 1− 2j)
+O(1/L)
j = 1, . . . , n
String hypothesisThermodynamics can be described by string degrees of freedom
Excitations and string hypothesis
Sz =L2−M
Ground state: unique solution with M = L/2 and all λi real
|0〉 = B(λ1) . . .B(λL/2)|Ω〉 λi ∈ R
Excitations: can be holes in the Dirac sea, or
n-string configurations
λj = λ+iη2
(n + 1− 2j)
+O(1/L)
j = 1, . . . , n
String hypothesisThermodynamics can be described by string degrees of freedom
Thermodynamic limit of Bethe Ansatz
Number of n-strings centered in an interval [λ, λ+ dλ]: Lρn(λ)dλNumber of n-string holes in an interval [λ, λ+ dλ]: Lρh
n(λ)dλ
Thermodynamic limit of BA: Bethe-Takahashi equations
ρn(λ) + ρhn(λ) = an(λ)−
∞∑m=1
(Tnm ∗ ρm)(λ)
an(λ) =1π
sinh nηcosh nη − cos 2λ
Tnm(λ) =
a|n−m|(λ) + 2a|n−m|+2(λ) + . . . m 6= n
+2an+m−2(λ) + an+m(λ)
2a2(λ) + · · ·+ 2a2n−2(λ) + a2n(λ) m = n
(f ∗ g)(λ) =
ˆ π/2
−π/2dλ′f (λ− λ′)g(λ′)
Thermodynamic limit of Bethe Ansatz
Number of n-strings centered in an interval [λ, λ+ dλ]: Lρn(λ)dλNumber of n-string holes in an interval [λ, λ+ dλ]: Lρh
n(λ)dλ
Thermodynamic limit of BA: Bethe-Takahashi equations
ρn(λ) + ρhn(λ) = an(λ)−
∞∑m=1
(Tnm ∗ ρm)(λ)
an(λ) =1π
sinh nηcosh nη − cos 2λ
Tnm(λ) =
a|n−m|(λ) + 2a|n−m|+2(λ) + . . . m 6= n
+2an+m−2(λ) + an+m(λ)
2a2(λ) + · · ·+ 2a2n−2(λ) + a2n(λ) m = n
(f ∗ g)(λ) =
ˆ π/2
−π/2dλ′f (λ− λ′)g(λ′)
Thermodynamic limit of Bethe Ansatz
Number of n-strings centered in an interval [λ, λ+ dλ]: Lρn(λ)dλNumber of n-string holes in an interval [λ, λ+ dλ]: Lρh
n(λ)dλ
Thermodynamic limit of BA: Bethe-Takahashi equations
ρn(λ) + ρhn(λ) = an(λ)−
∞∑m=1
(Tnm ∗ ρm)(λ)
an(λ) =1π
sinh nηcosh nη − cos 2λ
Tnm(λ) =
a|n−m|(λ) + 2a|n−m|+2(λ) + . . . m 6= n
+2an+m−2(λ) + an+m(λ)
2a2(λ) + · · ·+ 2a2n−2(λ) + a2n(λ) m = n
(f ∗ g)(λ) =
ˆ π/2
−π/2dλ′f (λ− λ′)g(λ′)
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chainIntegrability, Bethe Ansatz and thermodynamicsQuantum quenchGGE in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Selection of initial states: the quantum quench paradigm
Quantum quench
H(g0) −→t=0
H(g)
ground state −→ time evolution −→large t
steady state
Role of locality in our considerationsStart from ground state of local HamiltonianEvolved by a local HamiltonianRelevant observables: local operatorsExpect local charges to play a role
Example starting states for XXZ quenches:Neel: | ↑↓↑↓ . . . ↑↓〉 – ground state at ∆ =∞Dimer:(
1√2
(| ↑↓〉 − | ↓↑〉))⊗(
1√2
(| ↑↓〉 − | ↓↑〉))⊗· · ·⊗
(1√2
(| ↑↓〉 − | ↓↑〉))
ground state of Majumdar-Ghosh Hamiltonian
Selection of initial states: the quantum quench paradigm
Quantum quench
H(g0) −→t=0
H(g)
ground state −→ time evolution −→large t
steady state
Role of locality in our considerationsStart from ground state of local HamiltonianEvolved by a local HamiltonianRelevant observables: local operatorsExpect local charges to play a role
Example starting states for XXZ quenches:Neel: | ↑↓↑↓ . . . ↑↓〉 – ground state at ∆ =∞Dimer:(
1√2
(| ↑↓〉 − | ↓↑〉))⊗(
1√2
(| ↑↓〉 − | ↓↑〉))⊗· · ·⊗
(1√2
(| ↑↓〉 − | ↓↑〉))
ground state of Majumdar-Ghosh Hamiltonian
Selection of initial states: the quantum quench paradigm
Quantum quench
H(g0) −→t=0
H(g)
ground state −→ time evolution −→large t
steady state
Role of locality in our considerationsStart from ground state of local HamiltonianEvolved by a local HamiltonianRelevant observables: local operatorsExpect local charges to play a role
Example starting states for XXZ quenches:Neel: | ↑↓↑↓ . . . ↑↓〉 – ground state at ∆ =∞Dimer:(
1√2
(| ↑↓〉 − | ↓↑〉))⊗(
1√2
(| ↑↓〉 − | ↓↑〉))⊗· · ·⊗
(1√2
(| ↑↓〉 − | ↓↑〉))
ground state of Majumdar-Ghosh Hamiltonian
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chainIntegrability, Bethe Ansatz and thermodynamicsQuantum quenchGGE in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Truncated GGE
Previously: GGE only for theories equivalent to free particles.Pozsgay (2013): truncated GGE for XXZ chain– the first GGE realization in a genuinely interacting modelIdea:
use the quantum transfer method (QTM) to solve the chainkeep first few (up to 6) even chargesQ2(= H), Q4, Q6, Q8, Q10, Q12
use QTM formula to predict correlatorsObservations:
converges fast in truncation level (improved by extrapolation)truncated GGE is “the GGE” as a matter of principle
Fagotti & Essler (2013): construct full/untruncated GGEFagotti, Collura, Essler & Calabrese (2013):
compare GGE to tDMRG → some puzzles:translational invariance is not restored for dimer
Truncated GGE
Previously: GGE only for theories equivalent to free particles.Pozsgay (2013): truncated GGE for XXZ chain– the first GGE realization in a genuinely interacting modelIdea:
use the quantum transfer method (QTM) to solve the chainkeep first few (up to 6) even chargesQ2(= H), Q4, Q6, Q8, Q10, Q12
use QTM formula to predict correlatorsObservations:
converges fast in truncation level (improved by extrapolation)truncated GGE is “the GGE” as a matter of principle
Fagotti & Essler (2013): construct full/untruncated GGEFagotti, Collura, Essler & Calabrese (2013):
compare GGE to tDMRG → some puzzles:translational invariance is not restored for dimer
Truncated GGE
Previously: GGE only for theories equivalent to free particles.Pozsgay (2013): truncated GGE for XXZ chain– the first GGE realization in a genuinely interacting modelIdea:
use the quantum transfer method (QTM) to solve the chainkeep first few (up to 6) even chargesQ2(= H), Q4, Q6, Q8, Q10, Q12
use QTM formula to predict correlatorsObservations:
converges fast in truncation level (improved by extrapolation)truncated GGE is “the GGE” as a matter of principle
Fagotti & Essler (2013): construct full/untruncated GGEFagotti, Collura, Essler & Calabrese (2013):
compare GGE to tDMRG → some puzzles:translational invariance is not restored for dimer
Truncated GGE
Previously: GGE only for theories equivalent to free particles.Pozsgay (2013): truncated GGE for XXZ chain– the first GGE realization in a genuinely interacting modelIdea:
use the quantum transfer method (QTM) to solve the chainkeep first few (up to 6) even chargesQ2(= H), Q4, Q6, Q8, Q10, Q12
use QTM formula to predict correlatorsObservations:
converges fast in truncation level (improved by extrapolation)truncated GGE is “the GGE” as a matter of principle
Fagotti & Essler (2013): construct full/untruncated GGEFagotti, Collura, Essler & Calabrese (2013):
compare GGE to tDMRG → some puzzles:translational invariance is not restored for dimer
Truncated GGE
Previously: GGE only for theories equivalent to free particles.Pozsgay (2013): truncated GGE for XXZ chain– the first GGE realization in a genuinely interacting modelIdea:
use the quantum transfer method (QTM) to solve the chainkeep first few (up to 6) even chargesQ2(= H), Q4, Q6, Q8, Q10, Q12
use QTM formula to predict correlatorsObservations:
converges fast in truncation level (improved by extrapolation)truncated GGE is “the GGE” as a matter of principle
Fagotti & Essler (2013): construct full/untruncated GGEFagotti, Collura, Essler & Calabrese (2013):
compare GGE to tDMRG → some puzzles:translational invariance is not restored for dimer
Truncated GGE
Previously: GGE only for theories equivalent to free particles.Pozsgay (2013): truncated GGE for XXZ chain– the first GGE realization in a genuinely interacting modelIdea:
use the quantum transfer method (QTM) to solve the chainkeep first few (up to 6) even chargesQ2(= H), Q4, Q6, Q8, Q10, Q12
use QTM formula to predict correlatorsObservations:
converges fast in truncation level (improved by extrapolation)truncated GGE is “the GGE” as a matter of principle
Fagotti & Essler (2013): construct full/untruncated GGEFagotti, Collura, Essler & Calabrese (2013):
compare GGE to tDMRG → some puzzles:translational invariance is not restored for dimer
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolutionPrevious results: problems with dimer caseCompare to iTEBD (Werner & Zaránd)
4 The overlap TBA
5 Follow-up
6 Conclusions
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolutionPrevious results: problems with dimer caseCompare to iTEBD (Werner & Zaránd)
4 The overlap TBA
5 Follow-up
6 Conclusions
Problems with the dimer state
M. Fagotti, M. Collura, F.H.L. Essler and P. Calabrese:Relaxation after quantum quenches in the spin-1/2 HeisenbergXXZ chain, Phys. Rev. B89 (2014) 125101
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0|MG⟩ ⟶ Δ = 2
⟨σz jσ
z j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 4
⟨σz jσ
z j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 8
⟨σz jσ
z j+k⟩
t
t
k = 3
k = 1
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0|MG⟩ ⟶ Δ = 2
⟨σx jσ
x j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 4
⟨σx jσ
x j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 8
⟨σx jσ
x j+k⟩
t
t
k = 3
k = 1
Translation invariance seems broken: no relaxation at all?
Problems with the dimer state
M. Fagotti, M. Collura, F.H.L. Essler and P. Calabrese:Relaxation after quantum quenches in the spin-1/2 HeisenbergXXZ chain, Phys. Rev. B89 (2014) 125101
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0|MG⟩ ⟶ Δ = 2
⟨σz jσ
z j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 4
⟨σz jσ
z j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 8
⟨σz jσ
z j+k⟩
t
t
k = 3
k = 1
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0|MG⟩ ⟶ Δ = 2
⟨σx jσ
x j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 4
⟨σx jσ
x j+k⟩
t
t
k = 1
k = 3
0 1 2 3 4 5 6-1
-0.8
-0.6
-0.4
-0.2
0
0.2 |MG⟩ ⟶ Δ = 8
⟨σx jσ
x j+k⟩
t
t
k = 3
k = 1
Translation invariance seems broken: no relaxation at all?
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolutionPrevious results: problems with dimer caseCompare to iTEBD (Werner & Zaránd)
4 The overlap TBA
5 Follow-up
6 Conclusions
Result of iTEBD
0 1 2 3
Time0.4
0.5
0.6
0.7
0.8
0.9
1.0
<σz i
σz i+2>
Neel ∆ = 3
Result of iTEBD compared to thermal
0 1 2 3
Time0.4
0.5
0.6
0.7
0.8
0.9
1.0
<σz i
σz i+2>
Neel ∆ = 3
Result of iTEBD compared to thermal and GGE
0 1 2 3
Time0.4
0.5
0.6
0.7
0.8
0.9
1.0
<σz i
σz i+2>
Neel ∆ = 3
Result of iTEBD
0 1 2 3Time
0
0.1
0.2
0.3
<σz i
σz i+2 >
Dimer ∆ = 3
Result of iTEBD compared to GGE
0 1 2 3Time
0
0.1
0.2
0.3
<σz i
σz i+2 >
Dimer ∆ = 3
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBAQuench action methodInitial states and exact overlapsOverlap TBASpin-spin correlations
5 Follow-up
6 Conclusions
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBAQuench action methodInitial states and exact overlapsOverlap TBASpin-spin correlations
5 Follow-up
6 Conclusions
The quench action
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα Cα = 〈ψ(0)|α〉
Replace sum:∑α
−→ˆ ∞∏
n=1
Dρn(λ)eLs[ρn(λ)]
eLs[ρn(λ)]: number of Bethe states scaling to ρn(λ) in the TDL.Supposing that Cα and Aαα only depend on ρn(λ) in the TDL:
A =
ˆ ∞∏n=1
Dρn(λ)e−L(− 2LRe ln〈ψ(0)|ρn(λ)〉−s[ρn(λ)])〈ρn(λ)|A|ρn(λ)〉
In TDL: determined by saddle point of quench action functional
S[ρn(λ)] = −2LRe ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
A = 〈ρ∗n(λ)|A|ρ∗n(λ)〉
J.-S. Caux & F.H.L. Essler, 2013
The quench action
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα Cα = 〈ψ(0)|α〉
Replace sum:∑α
−→ˆ ∞∏
n=1
Dρn(λ)eLs[ρn(λ)]
eLs[ρn(λ)]: number of Bethe states scaling to ρn(λ) in the TDL.Supposing that Cα and Aαα only depend on ρn(λ) in the TDL:
A =
ˆ ∞∏n=1
Dρn(λ)e−L(− 2LRe ln〈ψ(0)|ρn(λ)〉−s[ρn(λ)])〈ρn(λ)|A|ρn(λ)〉
In TDL: determined by saddle point of quench action functional
S[ρn(λ)] = −2LRe ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
A = 〈ρ∗n(λ)|A|ρ∗n(λ)〉
J.-S. Caux & F.H.L. Essler, 2013
The quench action
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα Cα = 〈ψ(0)|α〉
Replace sum:∑α
−→ˆ ∞∏
n=1
Dρn(λ)eLs[ρn(λ)]
eLs[ρn(λ)]: number of Bethe states scaling to ρn(λ) in the TDL.Supposing that Cα and Aαα only depend on ρn(λ) in the TDL:
A =
ˆ ∞∏n=1
Dρn(λ)e−L(− 2LRe ln〈ψ(0)|ρn(λ)〉−s[ρn(λ)])〈ρn(λ)|A|ρn(λ)〉
In TDL: determined by saddle point of quench action functional
S[ρn(λ)] = −2LRe ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
A = 〈ρ∗n(λ)|A|ρ∗n(λ)〉
J.-S. Caux & F.H.L. Essler, 2013
The quench action
A = limT→∞
1T
ˆ T
0dt 〈A(t)〉 =
∑α
|Cα|2Aαα Cα = 〈ψ(0)|α〉
Replace sum:∑α
−→ˆ ∞∏
n=1
Dρn(λ)eLs[ρn(λ)]
eLs[ρn(λ)]: number of Bethe states scaling to ρn(λ) in the TDL.Supposing that Cα and Aαα only depend on ρn(λ) in the TDL:
A =
ˆ ∞∏n=1
Dρn(λ)e−L(− 2LRe ln〈ψ(0)|ρn(λ)〉−s[ρn(λ)])〈ρn(λ)|A|ρn(λ)〉
In TDL: determined by saddle point of quench action functional
S[ρn(λ)] = −2LRe ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
A = 〈ρ∗n(λ)|A|ρ∗n(λ)〉
J.-S. Caux & F.H.L. Essler, 2013
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBAQuench action methodInitial states and exact overlapsOverlap TBASpin-spin correlations
5 Follow-up
6 Conclusions
Initial states and overlapInitial states: translation invariant Neel and dimer states(T is translation by one site)
|ΨN〉 =1 + T√
2
⊗L/2
| ↑↓〉 |ΨD〉 =1 + T√
2
⊗L/2
(| ↑↓〉 − | ↓↑〉√
2
)Only overlaps with Sz = 0 states built upon “Cooper pairs”:
|λk〉C = |λ1,−λ1, λ2,−λ2, . . . , λL/4,−λL/4〉
Neel overlaps (Brockmann, De Nardis, Wouters & Caux, 2014)
〈ΨN |λk〉C√C 〈λk|λk〉C
=
L/4∏j=1
√tan(λj + iη/2) tan(λj − iη/2)
2 sin(2λj)×(Fredholm det)
Dimer overlaps (Pozsgay, 2014)
〈ΨD |λk〉 = 〈ΨN |λk〉∏k
1√2
(1− sin(λj − iη/2)
sin(λj + iη/2)
)
Initial states and overlapInitial states: translation invariant Neel and dimer states(T is translation by one site)
|ΨN〉 =1 + T√
2
⊗L/2
| ↑↓〉 |ΨD〉 =1 + T√
2
⊗L/2
(| ↑↓〉 − | ↓↑〉√
2
)Only overlaps with Sz = 0 states built upon “Cooper pairs”:
|λk〉C = |λ1,−λ1, λ2,−λ2, . . . , λL/4,−λL/4〉
Neel overlaps (Brockmann, De Nardis, Wouters & Caux, 2014)
〈ΨN |λk〉C√C 〈λk|λk〉C
=
L/4∏j=1
√tan(λj + iη/2) tan(λj − iη/2)
2 sin(2λj)×(Fredholm det)
Dimer overlaps (Pozsgay, 2014)
〈ΨD |λk〉 = 〈ΨN |λk〉∏k
1√2
(1− sin(λj − iη/2)
sin(λj + iη/2)
)
Initial states and overlapInitial states: translation invariant Neel and dimer states(T is translation by one site)
|ΨN〉 =1 + T√
2
⊗L/2
| ↑↓〉 |ΨD〉 =1 + T√
2
⊗L/2
(| ↑↓〉 − | ↓↑〉√
2
)Only overlaps with Sz = 0 states built upon “Cooper pairs”:
|λk〉C = |λ1,−λ1, λ2,−λ2, . . . , λL/4,−λL/4〉
Neel overlaps (Brockmann, De Nardis, Wouters & Caux, 2014)
〈ΨN |λk〉C√C 〈λk|λk〉C
=
L/4∏j=1
√tan(λj + iη/2) tan(λj − iη/2)
2 sin(2λj)×(Fredholm det)
Dimer overlaps (Pozsgay, 2014)
〈ΨD |λk〉 = 〈ΨN |λk〉∏k
1√2
(1− sin(λj − iη/2)
sin(λj + iη/2)
)
Initial states and overlapInitial states: translation invariant Neel and dimer states(T is translation by one site)
|ΨN〉 =1 + T√
2
⊗L/2
| ↑↓〉 |ΨD〉 =1 + T√
2
⊗L/2
(| ↑↓〉 − | ↓↑〉√
2
)Only overlaps with Sz = 0 states built upon “Cooper pairs”:
|λk〉C = |λ1,−λ1, λ2,−λ2, . . . , λL/4,−λL/4〉
Neel overlaps (Brockmann, De Nardis, Wouters & Caux, 2014)
〈ΨN |λk〉C√C 〈λk|λk〉C
=
L/4∏j=1
√tan(λj + iη/2) tan(λj − iη/2)
2 sin(2λj)×(Fredholm det)
Dimer overlaps (Pozsgay, 2014)
〈ΨD |λk〉 = 〈ΨN |λk〉∏k
1√2
(1− sin(λj − iη/2)
sin(λj + iη/2)
)
Overlaps in the thermodynamic limit
Integral form for L→∞
−2Re ln〈ψ(0)|ρn(λ)〉 =∞∑
n=1
ˆ π/2
−π/2dλρn(λ)gψn (λ)
gψn (λ) =n∑
j=1
gψ1
(λ+
iη2
(n + 1− 2j))
For our translation invariant Neel and dimer states
gN1 (λ) = − ln
(tan(λ+ iη/2) tan(λ− iη/2)
4 sin2(2λ)
)gN1 (λ) = − ln
(sinh4(η/2) cot(λ)
sin(2λ+ iη) sin(2λ− iη)
)
Overlaps in the thermodynamic limit
Integral form for L→∞
−2Re ln〈ψ(0)|ρn(λ)〉 =∞∑
n=1
ˆ π/2
−π/2dλρn(λ)gψn (λ)
gψn (λ) =n∑
j=1
gψ1
(λ+
iη2
(n + 1− 2j))
For our translation invariant Neel and dimer states
gN1 (λ) = − ln
(tan(λ+ iη/2) tan(λ− iη/2)
4 sin2(2λ)
)gN1 (λ) = − ln
(sinh4(η/2) cot(λ)
sin(2λ+ iη) sin(2λ− iη)
)
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBAQuench action methodInitial states and exact overlapsOverlap TBASpin-spin correlations
5 Follow-up
6 Conclusions
The overlap TBA equationsEntropy: 1/2 of Yang-Yang expression (only paired states!)
s[ρn(λ)] =12
∞∑n=1
ˆ π/2
−π/2dλ[ρn(λ) ln
(1 +
ρhn(λ)
ρn(λ)
)+ ρh
n(λ) ln(
1 +ρn(λ)
ρhn(λ)
)]Compute saddle point of quench action
S[ρn(λ)] = − 2L
Re ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
with the condition that the Bethe-Takahashi equations must hold
ρn(λ) + ρhn(λ) = an(λ)−
∞∑m=1
(Tnm ∗ ρm)(λ)
Result: the overlap thermodynamic Bethe Ansatz equations
log ηn(λ) = gψn (λ)− µn +∞∑
m=1
Tnm ? ln(1 + η−1m ) ηn(λ) =
ρhn(λ)
ρn(λ)
µ: chemical potential for Sz (external magnetic field).Independently obtained for the Neel case by Caux et al. before us.
The overlap TBA equationsEntropy: 1/2 of Yang-Yang expression (only paired states!)
s[ρn(λ)] =12
∞∑n=1
ˆ π/2
−π/2dλ[ρn(λ) ln
(1 +
ρhn(λ)
ρn(λ)
)+ ρh
n(λ) ln(
1 +ρn(λ)
ρhn(λ)
)]Compute saddle point of quench action
S[ρn(λ)] = − 2L
Re ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
with the condition that the Bethe-Takahashi equations must hold
ρn(λ) + ρhn(λ) = an(λ)−
∞∑m=1
(Tnm ∗ ρm)(λ)
Result: the overlap thermodynamic Bethe Ansatz equations
log ηn(λ) = gψn (λ)− µn +∞∑
m=1
Tnm ? ln(1 + η−1m ) ηn(λ) =
ρhn(λ)
ρn(λ)
µ: chemical potential for Sz (external magnetic field).Independently obtained for the Neel case by Caux et al. before us.
The overlap TBA equationsEntropy: 1/2 of Yang-Yang expression (only paired states!)
s[ρn(λ)] =12
∞∑n=1
ˆ π/2
−π/2dλ[ρn(λ) ln
(1 +
ρhn(λ)
ρn(λ)
)+ ρh
n(λ) ln(
1 +ρn(λ)
ρhn(λ)
)]Compute saddle point of quench action
S[ρn(λ)] = − 2L
Re ln〈ψ(0)|ρn(λ)〉 − s[ρn(λ)]
with the condition that the Bethe-Takahashi equations must hold
ρn(λ) + ρhn(λ) = an(λ)−
∞∑m=1
(Tnm ∗ ρm)(λ)
Result: the overlap thermodynamic Bethe Ansatz equations
log ηn(λ) = gψn (λ)− µn +∞∑
m=1
Tnm ? ln(1 + η−1m ) ηn(λ) =
ρhn(λ)
ρn(λ)
µ: chemical potential for Sz (external magnetic field).Independently obtained for the Neel case by Caux et al. before us.
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBAQuench action methodInitial states and exact overlapsOverlap TBASpin-spin correlations
5 Follow-up
6 Conclusions
Correlations on the XXZ chainWe want observables: spin-spin correlators
〈σiσi+n〉
But: it was only known how to get these in the QTM formalism!
1st idea: get from oTBA to QTM.May 2014: conference lecture by J-S Caux in New York:discrepancy between GGE and oTBA densities ρn(λ)−→ first idea killed :(
2nd idea: Hellmann-Feynman theorem
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
⇒ ∂EΨ
∂∆= L〈Ψ|(σz
1σz2 − 1)|Ψ〉
Correlations on the XXZ chainWe want observables: spin-spin correlators
〈σiσi+n〉
But: it was only known how to get these in the QTM formalism!
1st idea: get from oTBA to QTM.May 2014: conference lecture by J-S Caux in New York:discrepancy between GGE and oTBA densities ρn(λ)−→ first idea killed :(
2nd idea: Hellmann-Feynman theorem
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
⇒ ∂EΨ
∂∆= L〈Ψ|(σz
1σz2 − 1)|Ψ〉
Correlations on the XXZ chainWe want observables: spin-spin correlators
〈σiσi+n〉
But: it was only known how to get these in the QTM formalism!
1st idea: get from oTBA to QTM.May 2014: conference lecture by J-S Caux in New York:discrepancy between GGE and oTBA densities ρn(λ)−→ first idea killed :(
2nd idea: Hellmann-Feynman theorem
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
⇒ ∂EΨ
∂∆= L〈Ψ|(σz
1σz2 − 1)|Ψ〉
Correlations on the XXZ chainWe want observables: spin-spin correlators
〈σiσi+n〉
But: it was only known how to get these in the QTM formalism!
1st idea: get from oTBA to QTM.May 2014: conference lecture by J-S Caux in New York:discrepancy between GGE and oTBA densities ρn(λ)−→ first idea killed :(
2nd idea: Hellmann-Feynman theorem
HXXZ =L∑
i=1
(σx
i σxi+1 + σy
i σyi+1 + ∆σz
i σzi+1)
⇒ ∂EΨ
∂∆= L〈Ψ|(σz
1σz2 − 1)|Ψ〉
Correlations on the XXZ chain
3rd idea: a real stroke of genius (B. Pozsgay)
Hellmann-Feynman result analogous to QTM correlator formulas−→ can be extended to a conjecture for all 〈σiσi+n〉
M. Mestyán and B. Pozsgay,J. Stat. Mech. (2014) P09020.
How does it work? (without formulas)
1 Solve oTBA for the ηn functions.2 Solve Bethe-Takahashi for the ρn and ρh
n.3 Check that 〈Qi 〉 agree with initial state
(+ overlap sum rule S[ρ∗n(λ)] = 0. GGE gives −∞!)4 Solve auxiliary integral equations for some auxiliary functions5 Substitute auxiliary functions into the place of their analogues
in the QTM correlator formulas6 Compute correlator
Correlations on the XXZ chain
3rd idea: a real stroke of genius (B. Pozsgay)
Hellmann-Feynman result analogous to QTM correlator formulas−→ can be extended to a conjecture for all 〈σiσi+n〉
M. Mestyán and B. Pozsgay,J. Stat. Mech. (2014) P09020.
How does it work? (without formulas)
1 Solve oTBA for the ηn functions.2 Solve Bethe-Takahashi for the ρn and ρh
n.3 Check that 〈Qi 〉 agree with initial state
(+ overlap sum rule S[ρ∗n(λ)] = 0. GGE gives −∞!)4 Solve auxiliary integral equations for some auxiliary functions5 Substitute auxiliary functions into the place of their analogues
in the QTM correlator formulas6 Compute correlator
Correlations on the XXZ chain
3rd idea: a real stroke of genius (B. Pozsgay)
Hellmann-Feynman result analogous to QTM correlator formulas−→ can be extended to a conjecture for all 〈σiσi+n〉
M. Mestyán and B. Pozsgay,J. Stat. Mech. (2014) P09020.
How does it work? (without formulas)
1 Solve oTBA for the ηn functions.2 Solve Bethe-Takahashi for the ρn and ρh
n.3 Check that 〈Qi 〉 agree with initial state
(+ overlap sum rule S[ρ∗n(λ)] = 0. GGE gives −∞!)4 Solve auxiliary integral equations for some auxiliary functions5 Substitute auxiliary functions into the place of their analogues
in the QTM correlator formulas6 Compute correlator
Correlations on the XXZ chain
3rd idea: a real stroke of genius (B. Pozsgay)
Hellmann-Feynman result analogous to QTM correlator formulas−→ can be extended to a conjecture for all 〈σiσi+n〉
M. Mestyán and B. Pozsgay,J. Stat. Mech. (2014) P09020.
How does it work? (without formulas)
1 Solve oTBA for the ηn functions.2 Solve Bethe-Takahashi for the ρn and ρh
n.3 Check that 〈Qi 〉 agree with initial state
(+ overlap sum rule S[ρ∗n(λ)] = 0. GGE gives −∞!)4 Solve auxiliary integral equations for some auxiliary functions5 Substitute auxiliary functions into the place of their analogues
in the QTM correlator formulas6 Compute correlator
Result of iTEBD compared to GGE and oTBA
0 1 2 3Time
0
0.1
0.2
0.3
<σz i
σz i+2 >
Dimer ∆ = 3
This is very strong evidence for breakdown of GGE
1. n = 2: sublattice independent (projection is immaterial)2. Agreement with oTBA shows system has relaxed
Dimer correlators as function of ∆
1 2 3 4 5 6 7 8 9∆
-0.60
-0.58
-0.56
-0.54
<σz 1
σz 2>
iTEBDoTBAGGEtGGE
(a)
1 2 3 4 5 6 7 8 9∆
0.10
0.15
0.20
0.25
0.30
<σz 1
σz 3>
iTEBDoTBAGGEtGGE
(b)
Remarks:1. oTBA6=GGE for Neel as well (observed also by Caux et al.)... but difference is too small for iTEBD to resolve2. xx correlators show same patterns, but iTEBD is less accurate
Dimer correlators as function of ∆
1 2 3 4 5 6 7 8 9∆
-0.60
-0.58
-0.56
-0.54
<σz 1
σz 2>
iTEBDoTBAGGEtGGE
(a)
1 2 3 4 5 6 7 8 9∆
0.10
0.15
0.20
0.25
0.30
<σz 1
σz 3>
iTEBDoTBAGGEtGGE
(b)
Remarks:1. oTBA6=GGE for Neel as well (observed also by Caux et al.)... but difference is too small for iTEBD to resolve2. xx correlators show same patterns, but iTEBD is less accurate
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-upRole of bound statesBreakdown of the GETH
6 Conclusions
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-upRole of bound statesBreakdown of the GETH
6 Conclusions
Role of bound statesXXZ spectrum: n ≥ 2 strings are bound states of n = 1 magnonsin such a case the Qi do not uniquely determine the ρGGE: maximum entropy with given Qi... but that is irrelevant (S[k] = −∞)!)
A
B
k2
k3
kN
!1n
!2n
!3n
!Nn
Ii0 !GGE
k
!k
k1
! Ii0 !GGE k
G. Goldstein and N. Andrei: Phys. Rev. A 90, 043625
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-upRole of bound statesBreakdown of the GETH
6 Conclusions
Breakdown of GETHHowever: if GETH held, any state (including GGE) on the shell
Γ(ψ(0)) = |ρn(λ)〉 : 〈ρn(λ)|Qi |ρn(λ)〉 = 〈ψ(0)|Qi |ψ(0)〉
would give the same expectation value for local operators!Indeed, this is how non-integrable systems thermalize.
However: GETH is broken!B. Pozsgay: Failure of the generalized eigenstate thermalizationhypothesis in integrable models with multiple particle species,J. Stat. Mech. (2014) P09026.This also shows that GGE is broken for almost all initial states!
GGE seems to work in systems without bound statesB. Pozsgay: Quantum quenches and Generalized Gibbs Ensemble ina Bethe Ansatz solvable lattice model of interacting bosons,J. Stat. Mech., in press.
Breakdown of GETHHowever: if GETH held, any state (including GGE) on the shell
Γ(ψ(0)) = |ρn(λ)〉 : 〈ρn(λ)|Qi |ρn(λ)〉 = 〈ψ(0)|Qi |ψ(0)〉
would give the same expectation value for local operators!Indeed, this is how non-integrable systems thermalize.
However: GETH is broken!B. Pozsgay: Failure of the generalized eigenstate thermalizationhypothesis in integrable models with multiple particle species,J. Stat. Mech. (2014) P09026.This also shows that GGE is broken for almost all initial states!
GGE seems to work in systems without bound statesB. Pozsgay: Quantum quenches and Generalized Gibbs Ensemble ina Bethe Ansatz solvable lattice model of interacting bosons,J. Stat. Mech., in press.
Breakdown of GETHHowever: if GETH held, any state (including GGE) on the shell
Γ(ψ(0)) = |ρn(λ)〉 : 〈ρn(λ)|Qi |ρn(λ)〉 = 〈ψ(0)|Qi |ψ(0)〉
would give the same expectation value for local operators!Indeed, this is how non-integrable systems thermalize.
However: GETH is broken!B. Pozsgay: Failure of the generalized eigenstate thermalizationhypothesis in integrable models with multiple particle species,J. Stat. Mech. (2014) P09026.This also shows that GGE is broken for almost all initial states!
GGE seems to work in systems without bound statesB. Pozsgay: Quantum quenches and Generalized Gibbs Ensemble ina Bethe Ansatz solvable lattice model of interacting bosons,J. Stat. Mech., in press.
Breakdown of GETHHowever: if GETH held, any state (including GGE) on the shell
Γ(ψ(0)) = |ρn(λ)〉 : 〈ρn(λ)|Qi |ρn(λ)〉 = 〈ψ(0)|Qi |ψ(0)〉
would give the same expectation value for local operators!Indeed, this is how non-integrable systems thermalize.
However: GETH is broken!B. Pozsgay: Failure of the generalized eigenstate thermalizationhypothesis in integrable models with multiple particle species,J. Stat. Mech. (2014) P09026.This also shows that GGE is broken for almost all initial states!
GGE seems to work in systems without bound statesB. Pozsgay: Quantum quenches and Generalized Gibbs Ensemble ina Bethe Ansatz solvable lattice model of interacting bosons,J. Stat. Mech., in press.
Outline
1 Introduction
2 Quantum quenches in the XXZ spin chain
3 Moment of truth: comparing to real time evolution
4 The overlap TBA
5 Follow-up
6 Conclusions
Conclusions
1 The GGE description is not generally valid for genuinelyinteracting integrable systems.
2 Failure of GGE is caused by failure of thegeneralized eigenstate thermalization hypothesis.(Classical version: ergodicity on action-angle tori.)
3 Failure of GGE seems to be tied with existence of bound states.
4 The quench action correctly captures the steady state.However, it is a microscopic description!
Open issues1 Are there more local charges? Very-very likely: NO!
2 Non-local charges?Known to exist for −1 < ∆ < 1, recently also for ∆ > 1!
Must be careful (!) under what condition they can play a role:
derivation of canonical from microcanonical ensemble assumessubsystem additivity of conserved quantities (e.g. energy)!
Additivity is needed in TDL only: quasi-local charges?
3 Any other ideas?
The quest is open:the ensemble describing the steady state of integrable systems ispresently unknown.
Open issues1 Are there more local charges? Very-very likely: NO!
2 Non-local charges?Known to exist for −1 < ∆ < 1, recently also for ∆ > 1!
Must be careful (!) under what condition they can play a role:
derivation of canonical from microcanonical ensemble assumessubsystem additivity of conserved quantities (e.g. energy)!
Additivity is needed in TDL only: quasi-local charges?
3 Any other ideas?
The quest is open:the ensemble describing the steady state of integrable systems ispresently unknown.
Pozsgai Balázs: Junior Prima Díj 2014