Entanglement Entropy in integrable models

Francesco Ravanini

Collaboration with

Davide Bianchini, Elisa Ercolessi, Fabio Franchini, Paul Pearce

arXiv:0905.4000, 1008.3892, 1201.6367, 1205.6426, 1301.6758
& work in progress
Seoul, 17 dic 2013

F. Ravanini

EE in Integrable Models

Outline
Introduction: Von Neumann and Renyi entropies as a measure
of Entanglement
Entanglement entropy in 1+1 dim CFT
Entanglement entropy in 1D lattice spin chains: the Corner
Transfer Matrix (CTM) method
XYZ chain exact Entanglement entropy
Entanglement entropy in bipartite sine-Gordon
Entanglement entropy in spin chains related to ABF and FB
models
Hamiltonians of ABF-FB chains from an algebraic point of
view
Conclusions

F. Ravanini

EE in Integrable Models

Introduction
Entanglement: fundamental quantum property
Dierent reasons for interest:

Quantum Information

Quantum Phase Transitions

Condensed matter physics

Integrable models

Black holes

Quantum computers

universality

non-local correlators

new playground

Information paradox & Quantum Gravity

NON LOCALITY intrinsic in Quantum Mechanics?

EPR paradox: uncompleteness of QM or non-locality?
Bell inequalities local hidden variables exist only if a
certain correlation

P<2

Clauser Friedmann & Aspect

experiments P > 2 =

possible non-locality of QM

F. Ravanini

EE in Integrable Models

Quantum systems and sub-systems

Quantum system with unique pure ground state |0i composed
of two subsystems, A and B.
If a state has wavefunction

|iAB = |iA |iB

Separable = No entanglement
(i.e. measurements on B do not aect A state)

If instead

|iAB =
with d

> 1 =

d
X
j =1

j |j iA |j iB

Entangled (measurements of B do aect A

state)

F. Ravanini

EE in Integrable Models

Entanglement and density matrix

Dene

reduced density matrix for subsystem A

A = TrB

Quantum entropy (Von Neumann) of Entanglement (E-Entropy)

SA

= TrA (A log A ) = SB

[Bennett, Bernstein, Popescu, Schumacher 1996]

For a separable state SA
maximal

= SA

Area law

= 0,

for a maximally entangled state it is

is a measure of Entanglement

[Srednicki 1993]

SA

Area( A)

F. Ravanini

EE in Integrable Models

Renyi Entropy

Rnyi Entropy
S

1
1

log

TrA A

Introduced by hungarian mathematician Rnyi in probability theory

It reduces to Von Neumann for

Contains higher momenta and for

reduced density matrix

the spectrum of the

can be read

link with replica trick la Calabrese Cardy

F. Ravanini

EE in Integrable Models

Entanglement entorpy in CFT

[Holzhey, Larsen, Wilczek 1994 - Calabrese, Cardy 2004]
S (`)

'

c
3

log

`
a

+U

c = central charge of CFT, U = non-universal constant,

a = UV-cuto

F. Ravanini

EE in Integrable Models

Entanglement entopy in CFT: generalizations

Finite temperature T

SA

c
3

log

= 1 ,

innite size

`
sinh
a


+U

(
`
`

Finite size L
SA

c
3


log

sin

+ U = SB

Open systems: same formulae with dierent U

U
2

=boundary

entropy

= log g

[Aeck, Ludwig 1991]

F. Ravanini

EE in Integrable Models

c `
3

3 log a

Entanglement in a Spin Chain

Hamiltonian
H

N
X
k =1

Consider the ground state with

Hk ,k +1

= |0ih0|

Block of spins in the space interval [1, `] is subsystem A

The rest of the ground state is subsystem B

Entanglement of a block of spins in the space interval

[1, `]
of `

with the rest of the ground state as a function

F. Ravanini

EE in Integrable Models

E-Entropy and Universality

Powers of

easily accessible in CFT (replica trick)

S (`)

1
1

TrA (A )

= scaling dimension of the operator responsible for the correction

[Calabrese, Cardy 2004-2010]
h

nonuniversal
S (`)

}|
{
h
+ c0 + b (`)` + ...
{z
}
|
z

log `

Conjectured
6

Close to criticality

c
6

a 1


1

and


log

[Calabrese, Cardy, Peschel 2010]

F. Ravanini

}|
  x

+ C0 + B

EE in Integrable Models

{
+ ...

Corner Transfer Matrix

CTM is a very useful tool

A
s ,s 0

[Baxter (1972)]

XY

wi

and analogously B , C , D with 90 rotations.

F. Ravanini

EE in Integrable Models

Partition function and CTM

Now we can build up the whole lattice by using the 4 CTM's

Partition function

Z=

0 B
0
00 C
00
000 D
000

= Tr(ABCD )

,
0 ,
00 ,
000

F. Ravanini

EE in Integrable Models

Reduced density matrix and CTM

Now suppose to divide the spins in two subsystems A:

A = (1 , ..., p )

= (
A ,
B )

and B:

B = (p+1 , ..., L ),

i.e.

Reduced density matrix of subsystem A

A (
A ,
A0 ) =

h
A ,
B |0ih0|
A0 ,
B i = TrB h
A |0ih0|
A0 i

A = (ABCD ) ,0

F. Ravanini

1
1

log TrA A

EE in Integrable Models

XYZ model
Hamiltonian

HXYZ

N
X
(Jx kx kx +1 + Jy ky ky +1 + Jz kz kz +1 )

k =1

commutes with transfer matrix of 8-vertex model

Useful parametrization

Jy
Jx

+ k 2 sn2 (i ; k )
1 k 2 sn2 (i ; k )

<k <1

and 0

F. Ravanini

Jz
Jx

cn(i ; k ) dn(i ; k )
1 k 2 sn2 (i ; k )

I (k 0 )

EE in Integrable Models

Diagonalization of CTM
One can prove that, in XYZ, A

=C

and B

= D.

In the thermodynamic limit the following formula holds for the

diagonalized CTM

[Baxter (1977)]


A = ABCD = (AB ) =
where

 

 

1
0

0
x


...

 = I
(k )

= e I (k ) = e 

depends on the XYZ parameters through elliptic

functions and I (k ) is the elliptic integral of 1st kind of

modulus k

= e HCTM

where


HCTM =

HCTM


is a operator with integer spectrum

F. Ravanini

EE in Integrable Models


...

Diagonalization of CTM
One can prove that, in XYZ, A

=C

and B

= D.

In the thermodynamic limit the following formula holds for the

diagonalized CTM

[Baxter (1977)]


A = ABCD = (AB ) =
where

 

 

1
0

0
x


...

 = I
(k )

= e I (k ) = e 

depends on the XYZ parameters through elliptic

functions and I (k ) is the elliptic integral of 1st kind of

modulus k

= e HCTM

where


HCTM =

HCTM


is a operator with integer spectrum

F. Ravanini

EE in Integrable Models


...

Entanglement entropy of XYZ model

The trace of the reduced density matrix

Y
Z = TrA =
(1 + x j )

and

j =1

SA

= 

log Z

+ log Z

leads to the nal formula for Von Neumann

SA

X
=

(1 + e j  )

j =1

and for Rnyi entropy (q

+ e j  )

j =1

x =nome)

log(1

X
2j
log(1 + q ) +
1
j =1

1
1

log(1

+ q 2j )

j =1

that can also be written in theta function terms

S

22 (0, q )
4 (0, q )3 (0, q )
+
log
=
log
6(1 )
3 (0, q )4 (0, q )
22 (0, q )
1

F. Ravanini

EE in Integrable Models

Phase diagram of XYZ model

Approaching criticality the
SA

[Calabese - Cardy (2004)] formula holds

=

c
6

log

+ cost.

everywhere but at the E1,2 points

Jy
u-constant line

l-constant line

C2

E2

Jz

E1

F. Ravanini

C1

EE in Integrable Models

Isoentropic lines

F. Ravanini

EE in Integrable Models

Link with Ising characters

Z(x ) =

Y
(1 + x j ) = x Ising
1,2 (i /)
1

12

j =1

and

Ising
1,2 (i /)
i
Tr = h
Ising
(
i
/)
1,2

Critical XXZ line: approached for x

1.

Use modular

transformation to get this limit x

Tr

Similar to
not c

=1

=2

1
2

=e

Ising
Ising
1,1 (i /) 2,1 (i /)
h
i
Ising
Ising
1,1 (i /) 2,1 (i /)

[Calabrese, Cardy, Peschel (2011)] but with c = 12

(!):

HCTM

characters,

is the hamiltonian of a free fermion

property of many integrable models

F. Ravanini

[Peschel (2010)]

EE in Integrable Models

The sine-Gordon limit

Scaling limit of the XYZ model:
lattice spacing a

while

mass gap kept constant


M

= 8

sin

  /
`

where

2
1
8


1974]


= arccos (Jz )

sine-Gordon theory

`2

Jy2
1 Jz2
1

[Johnson, Krinsky, McCoy 1973 - Luther

Jz is connected to the parameter

Jy scales in the scaling limit a
F. Ravanini

of sine-Gordon

Jy

EE in Integrable Models

Entanglement in bipartite SG
Exact entanglement entropy of a bipartite XYZ model in the
sine-Gordon limit
SsG

or, taking

SsG

1
6

k ) 3 ln 2
r

12
1 + Jz
arctan
1 Jz



Jz = cos 1
8
=

ln(1

ln 2

+ O (1/ ln(a))


ln

1
Ma

h 
i

sin 1
8
1
+ O (1/ ln(a))


+ ln

6
1
8
2

F. Ravanini

EE in Integrable Models

Minimal models + 13
Continuum limit of ABF models on square lattice. CTM
diagonalization is given and the calculation of
straightforward

[Franchini, De Luca (2012)]

Can be generalized to FB non-unitary models

Ercolessi, FR, Pearce, work in progress]

[Bianchini,

Renyi entropy

1
1

TrA A

1
1

log Z

log Z1

expression very complicated in terms of theta functions, but

expanding:

+ 1 ce
12

F. Ravanini

log

+ U + A ()

EE in Integrable Models

Open problems
Taking p , p

in

Mp , p 0

but with their ratio xed, one

accesses E-Entropy for log-CFT and their o-critical lattice

realizations (percolations,...)

[Pearce, Seaton (2011)]

Is ce the correct leading coecient predicition from CFT?

c-theorem like arguments do not work for non-unitary cases.
One has to compute 4-pt correlators with twist elds

h|T T |i

instead of

hT T i

to compute with replica trick. Progresses have been done

[Sierra et al. (2013)]

What is the interpretation of E-Entropy in non-unitary

theories?
The answer is yes, but... what was the question?

[W. Allen]:

We know the 2D classical lattice model, we can compute

formally S , but what is the quantum Hamiltonian we are
dealing with?
F. Ravanini

EE in Integrable Models

Open problems
Taking p , p

in

Mp , p 0

but with their ratio xed, one

accesses E-Entropy for log-CFT and their o-critical lattice

realizations (percolations,...)

[Pearce, Seaton (2011)]

Is ce the correct leading coecient predicition from CFT?

c-theorem like arguments do not work for non-unitary cases.
One has to compute 4-pt correlators with twist elds

h|T T |i

instead of

hT T i

to compute with replica trick. Progresses have been done

[Sierra et al. (2013)]

What is the interpretation of E-Entropy in non-unitary

theories?
The answer is yes, but... what was the question?

[W. Allen]:

We know the 2D classical lattice model, we can compute

formally S , but what is the quantum Hamiltonian we are
dealing with?
F. Ravanini

EE in Integrable Models

Open problems
Taking p , p

in

Mp , p 0

but with their ratio xed, one

accesses E-Entropy for log-CFT and their o-critical lattice

realizations (percolations,...)

[Pearce, Seaton (2011)]

Is ce the correct leading coecient predicition from CFT?

c-theorem like arguments do not work for non-unitary cases.
One has to compute 4-pt correlators with twist elds

h|T T |i

instead of

hT T i

to compute with replica trick. Progresses have been done

[Sierra et al. (2013)]

What is the interpretation of E-Entropy in non-unitary

theories?
The answer is yes, but... what was the question?

[W. Allen]:

We know the 2D classical lattice model, we can compute

formally S , but what is the quantum Hamiltonian we are
dealing with?
F. Ravanini

EE in Integrable Models

Open problems
Taking p , p

in

Mp , p 0

but with their ratio xed, one

accesses E-Entropy for log-CFT and their o-critical lattice

realizations (percolations,...)

[Pearce, Seaton (2011)]

Is ce the correct leading coecient predicition from CFT?

c-theorem like arguments do not work for non-unitary cases.
One has to compute 4-pt correlators with twist elds

h|T T |i

instead of

hT T i

to compute with replica trick. Progresses have been done

[Sierra et al. (2013)]

What is the interpretation of E-Entropy in non-unitary

theories?
The answer is yes, but... what was the question?

[W. Allen]:

We know the 2D classical lattice model, we can compute

formally S , but what is the quantum Hamiltonian we are
dealing with?
F. Ravanini

EE in Integrable Models

Quantum critical hamiltonian

Uq (sl (2))

At criticality

invariant XXZ model

(1988) - Pasquier, Saleur (1990)]

H

= J

N
1
X
n=1

(nx nx+1 +ny ny+1 +

[Alcaraz, Barber, Batchelor

1
+ q 1 z z
q q
z)
n n+1 )+
(1z N
2

Can be rewritten in terms of Temperley-Lieb operators

= J

N
1
X
n =1

en

= (q +q 1 )en

en en1 en

F. Ravanini

= en

en

en em

= em en

EE in Integrable Models

if |n m |

>1

Quantum o-critical hamiltonian

What happens o-criticality? Exercise done so far for the r

=4

case. Try to write the hamiltonian using tile operators Fn and

exploit tile algebra to show that

= J

N
1
X
n =1

Fn

with
Fn Fn+1 Fn
where

= Fn

(F2n )2 = 1 F2n

1 = 2 = (q + q 1 )

(F2n+1 )2 = 2 F2n+1

at the critical point t

It resembles TL-algebra, but with

algebras?)

F. Ravanini

= 0.

two parameters (= elliptic

EE in Integrable Models

Summary
Von Neumann and Rnyi E-Entropies are crucial tools to study
entanglement in quantum systems. In integrable models, they
can be calculated using integrable techniques.
Corner Transfer Matrix technique allows the exact calculation
of bipartite E-Entropy in spin chains. Having the exact formula
at hand, one can test some of the open issues about
entanglement in these models.
A suitable scaling limit yields the sine-Gordon model and
E-entropy can be calculated for its bipartition. It is the rst
calculation of this kind for an interacting eld theory.
An integrable way to compute nite size E-Entropy is to be
developed. It would complement the present knowledge by
new precious information.

F. Ravanini

EE in Integrable Models

Conclusions

Entanglement entropy is a new way to approach interesting

problems in theoretical physics and it should be better
understood in (integrable) QFT, as it seems crucial in the
solution of challenging paradoxes, like the information loss in
black holes.
It also stimulates progresses in mathematics, in the best
tradition of the integrability approach.

Thank you!!!

F. Ravanini

EE in Integrable Models