Fabian Essler (Oxford)

Transcription

1 Quantum Quenches in Integrable Models Fabian Essler (Oxford) Collaborators: J.S. Caux (Amsterdam), M. Fagotti (Oxford) earlier work: P. Calabrese (Pisa), S. Evangelisti (Oxford), D. Schuricht (Aachen) M. Fagotti and F.H.L. Essler, JSTAT P07012 (2013) J.S. Caux and F.H.L. Essler, PRL 110, (2013) M. Fagotti and F.H.L. Essler, PRB 87, (2013) F.H.L. Essler, S. Evangelisti & M. Fagotti, PRL 109, (2012) Kyoto, August 2013

2 Outline 1. Introduction to Quantum Quenches and some general notions. 2. A new formalism for analyzing quenches in integrable models. 3. Generalized Gibbs Ensembles and the late-time behaviour after quenches in integrable models GGE and dynamical correlations. Time evolution of reduced density matrices (TFIM). GGE expectation values of local observables for quenches in the spin-1/2 XXZ chain.

3 A. Introduction and some general notions

4 Quantum Quenches in isolated many-particle systems A. Consider an isolated quantum system in the thermodynamic limit; Hamiltonian H(h) (short-ranged), h e.g. bulk magnetic field B. Prepare the system in the ground state of H(h 0 ) C. At time t=0 change the Hamiltonian to H(h) D. (Unitary) time evolution (t) = exp(-ih(h)t) E. Goal: study time evolution of local (in space) observables (t) (x) (t), (t) O 1 (x) 2 (y) (t), (t) O 1 (x,t 1 ) 2 (y,t 2 ) (t)

5 Local Relaxation Given that we are considering an isolated system, does the system relax in some way? It can never relax as a whole.

6 Local Relaxation Given that we are considering an isolated system, does the system relax in some way? It can never relax as a whole. Initial state after the quench is a pure state (t) = exp(-ih(h)t) = n exp(-ie n t) <n n>. Can always choose observables O that never relax, e.g. O=O = (t) O (t) = A cos([e 1 -E 2 ]t+ )

7 Local Relaxation Given that we are considering an isolated system, does the system relax in some way? It can never relax as a whole. It can relax locally (in space). B A Entire System: A B Take A infinite, B finite Ask questions only about B: Expectation values of local ops: (t) O B (x) (t) Physical Picture: A acts like a bath for B.

8 Local Relaxation Given that we are considering an isolated system, does the system relax in some way? It can never relax as a whole. It can relax locally (in space). A No time-averaging involved!!! B Physical Picture: A acts like a bath for B.

9 Subsystems and Reduced Density Matrices = initial (pure) state of the entire system A B (A infinite) Density matrix: (t)= (t) (t) A Reduced density matrix: B (t)=tr A (t) B B contains all local correlation functions in B: j =0,x,y,z for B=[1,...,ll] in a spin-1/2 quantum spin chain

10 Definition of the Stationary State If lim t B (t)= B ( ) exists for any finite subsystem B: system approaches a stationary state; (t) B (x) (t) become time-independent for all local operators.

11 How about quenches in quantum integrable models?

12 How about quenches in quantum integrable models? Have local integrals of motion [I m, I n ]=[I m, H(h)]=0.

13 Example: transverse-field Ising chain Grady 82 Prosen 98 define operators S j,j+` = j zj+1... z j+` 1 j+` I n involve spins on n+2 neighbouring sites

14 How about quenches in quantum integrable models? Have local integrals of motion [I m, I n ]=[I m, H(h)]=0. (t) I m (t) independent of time Expectation: Time evolution of local operators/stationary state should be special.

15 How about quenches in quantum integrable models? Have local integrals of motion [I m, I n ]=[I m, H(h)]=0. (t) I m (t) independent of time Expectation: Time evolution of local operators/stationary state should be special. Stationary state: M. Rigol, V. Dunjko, V. Yurosvki & M. Olshanii, PRL98, (2007) Generalized Gibbs Ensemble

16 RDM formulation of the Generalized Gibbs Ensemble Let I m be local integrals of motion [I m, I n ]=[I m, H(h)]=0 Define GGE density matrix by: m fixed by Reduced density matrix of B: gg =exp(- m I m )/Z gg tr[ gg I m ]= (0) I m (0) gg,b =tr A gg The system is described by a GGE if for any finite subsystem B B ( )= gg,b cf Barthel & Schollwöck 09

17 B. A new formalism for quenches in integrable models

18 B. A new formalism for quenches in integrable models In integrable models we can construct a basis of eigenstates We then want to calculate This is difficult. To study thermodynamic limit we must sum infinitely many terms & in general must deal with late time ( infrared ) divergencies. cf Calabrese, Essler & Fagotti 11, 12

19 Idea: follow Yang&Yang approach to thermodynamics 1. i! i, X (...)! Z D[ ]e S (...). 2. evaluate path integral by saddle point approximation

20 Caux&Essler 13 > = simultaneous eigenstate of all local conservation laws I m = m such that (0) I m (0) = m > can be constructed e.g. by a generalized TBA Mossel& Caux 12 once the expansion of (0) in eigenstates of H(h) is known.

21 Caux&Essler 13 Much more efficient way of calculating dynamics! e.g. Late-time dynamics dominated by small excitations over Have reproduced some known results for Ising using this formalism (rather non-trivial checks). Applications to sine-gordon and Lieb-Liniger models under way. talk by J.-S. Caux Bertini, Essler, Schuricht

22 Description of the stationary state GGE: lim t tr A (t)>< (t) =tr A [ gg ] Our description: lim t tr A (t)>< (t) =tr A >< So expectation values of local operators in the stationary state are given by using a single simultaneous eigenstate of all conservation laws. analogous result for Gibbs ensemble: Goldstein, Lebowitz, Tumulka, Zanghi 06 Similar to generalized microcanonical ensemble Cassidy, Clark & Rigol 11

23 So locally (in space) looks the same as the GGE (and the diagonal ensemble ). Globally they all differ.

24 C. Some new results on GGEs & integrable models

25 1. The GGE applies to dynamical correlation functions. Essler, Evangelisti & Fagotti 12 More generally: if lim h (t) O 1...O n (t)i =Tr[ stat O 1...O n ] t!1 then lim h (t) O 1(t 1 )...O n (t n ) (t)i =Tr[ stat O 1 (t 1 )...O n (t n )] t!1 for fixed t 1,t 2,...,t n ultimately follows from Lieb&Robinson 72 Bravyi, Hastings& Verstraete 06 Explicit results for in Ising.

26 2. Some conservation laws are more important than others. Transverse Field Ising Chain Fagotti & Essler 13

27 Transverse Field Ising Chain Hamiltonian: Phase Diagram: T Quantum Critical Point order parameter: ( always) h 0 T>0: order melts

28 Transverse Field Ising Chain Hamiltonian: quenches: h 0 h 0 1 h 0 Barouch, McCoy & Dresden 70 Igloi & Rieger 00, 11 Rossini et al 09, 10 Calabrese, Essler & Fagotti 11, 12 Schuricht & Essler 12 Essler, Evangelisti & Fagotti 12 Foini, Cugliandolo & Gambassi 12 Heyl, Polkovnikov & Kehrein 12 Viehmann et al 13

29 How fast is the approach to the t limit? - How close is B (t) to gg,b? Define a distance: h 0 >1 (Z 2 unbroken): Can reduce this to expression in terms of 2N 2N matrix ( B is 2 N 2 N matrix). h 0 <1 (Z 2 broken): B is not Gaussian use cluster decomposition + causality ( Calabrese-Cardy horizon ) Calabrese & Cardy 05

30 D (GGE) = D( `(t), gg,`) Distance between quench and generalized Gibbs reduced density matrices for sub-system sizes ` = 10,...,150 J D (GGE) t -3/2 at late times. Holds for any quench h 0 h

31 Difference between GGE and thermalization (Gibbs)? Distance to a Gibbs ensemble at the appropriate temperature 1 D (Gibbs) 0.1 Distance between quench and Gibbs reduced density matrices for sub-system sizes J t h 0 =1.2 h=3 D (Gibbs) const at late times.

32 Do we really need all conservation laws? Define a truncated GGE by keeping only the y most local conservation laws: D (y) GGE (Gibbs) l=10 J t h 0 =1.2 h=3 Distance D (y) = D( `(t), (y) between quench and truncated GGE reduced density matrices for subsystem size =10. tgge,`) Keeping more conservation laws gives a better description

33 Do we really need all conservation laws? 5 50 D (y) ` = 50 1x10-7 ` =5 h 0 =1.2! h = J y= # of conservation y laws h 0 =1.2 h=3 kept Good description as soon as y!

34 Which conservation laws are most important? Leave out the q th conservation law: 0.1 1x10-4 D d(+q) 1x10-7 1x q 5 50 h 0 =1.2 h=3 q= index of removed conservation law 5 50 h 0 =1.2! h =3 The more local the conservation law, the more important it is!!!

35 Which conservation laws are most important? Conservation laws Subsystem size : must keep all I m with m< +n 0 (h,h 0 )

36 3. GGE and quenches in the spin-1/2 Heisenberg chain.

37 3. GGE and quenches in the spin-1/2 Heisenberg chain. X H (1) = J X j S x j S x j+1 + S y j Sy j+1 + Sz j S z j+1 1 4, > 1 Higher conservation laws are known; generated by transfer matrix sin H (k) k log (i + ) =0. [H (k),h (n) ]=0. (i + ) = Tr[L L ( )L L 1 ( )...L 1 ( )], L j ( ) = 1+ z 2 z j + sinh( 2 ) sinh(i + 2 ) 1 z j z 2 + sinh(i ) sinh(i + 2 ) ( + j + + j ),

38 GGE density matrix GGE = 1 Z GGE exp X l=1 lh (l)!. Can be viewed as thermal density matrix of integrable Hamiltonian H = X l=1 l 1 H (l) Can use (Quantum Transfer Matrix) formalism developed for finite temperature correlators! Boos, Göhmann, Klümper et al Boos, Miwa, Jimbo, Smirnov, Takeyama 06-09

39 Two tasks remain: 1. Construct Quantum Transfer Matrix formalism for generalized Hamiltonian. straightforward generalization of H = 2X l=1 l H (l) Klümper and Sakai 02 X1 2. Determine the m must solve lim L!1 h 0 H (l) 0i L = lim L!1 Tr GGE H (l). L This is hard: cf Poszgay 13

40 Idea: evaluate generating function Ω Ψ0 (λ) = 1 L Ψ 0 τ (i + λ)τ 1 (i + λ) Ψ 0 = i k=1 ( γ ) k λ k 1 sin γ (k 1)! Ψ 0 H (k) Ψ 0 L, using QISM Ω Ψ0 (λ) M j (λ) = 1+τ z σ z j 2 L x x=λ 1 Sp Ψ 0 V L (x, λ)...v 1 (x, λ) Ψ 0, L x x=λ + sinh( γ λ 2 ) 1 τ zσz j sinh( iγ + γ λ 2 ) 2 + sinh( iγ ) sinh( iγ + γ λ 2 )(τ + σ + j + τ σ j ).

41 The point: Ω Ψ0 (λ) L x x=λ 1 Sp Ψ 0 V L (x, λ)...v 1 (x, λ) Ψ 0, L x x=λ can be evaluated explicitly for matrix product states!!! e.g. for product states we only need to diagonalize a 4x4 matrix. Results: (short-distance) correlation function of spins in the GGE for quenches from a variety of initial states. Interaction quenches from large included!

42 Comparsion to numerics (TDMRG): Fagotti, Calabrese, Collura, Essler x(1) y(1) z(1) transverse longitudinal GGE thermal L=64 Néel π/6 =2 initial state

43 Conclusions 1. In the thermodynamic limit can describe local physics through a single eigenstate of all conservation laws. 2. Late-time dynamics given by small excitations around this state. 3. The GGE gives both static and dynamic correlators at stationarity. 4. The most local conservation laws are most important for describing the stationary state. 5. Have determined local correlators in stationary state of XXZ after quenching from certain initial states.

44 Physical interpretation of the distance It measures the average mean relative difference of all expectation values of local operators: Average defined by: