Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

58
Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Transcript of Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Page 1: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Introduction to Tensor Network StatesSukhwinder Singh

Macquarie University (Sydney)

Page 2: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contents

• The quantum many body problem.

• Diagrammatic Notation

• What is a tensor network?

• Example 1 : MPS

• Example 2 : MERA

Page 3: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

}D

1 2 N

Total Hilbert Space : NV

Quantum many body system in 1-D

dim( )V D

Page 4: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

1 2

1 2

1 2N

N

i i i Ni i i

i i i

NV

!

Dimension = ND

Huge

Page 5: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

How many qubits can we represent with 1 GB of memory?

Here, D = 2.

To add one more qubit double the memory.

302 8 2

27

N

N

Page 6: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

But usually, we are not interested in arbitrary states in the Hilbert space.

Typical problem : To find the ground state of a local

Hamiltonian H,

12 23 34 1,... N NH h h h h

Page 7: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Ground states of local Hamiltonians are special

( ) logi ii

S l

Limited Correlations and Entanglement.

( ) x x lC l O O

Page 8: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

1) Gapped Hamiltonian

2) Critical Hamiltonian

( ) log( )S l l

( )S l const l /( ) lC l e

( ) 0aC l l a

Properties of ground states in 1-D

Page 9: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

We can exploit these properties to represent ground states more

efficiently using tensor networks.

Page 10: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Ground states of local Hamiltonians

NV

Page 11: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contents

• The quantum many body problem.

• Diagrammatic Notation

• What is a tensor network?

• Example 1 : MPS

• Example 2 : MERA

Page 12: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Multidimensional array of complex numbers

Tensors

1 2 ki i iT

1

2

3

:Ket

* * *1 2 3

: Bra

11 12

21 22

31 32

Matrix

M M

M M

M M

a

a

a

b

a

b

c

11 12

21 22

31 32

11 12

21 22

31 32

1

2

Rank-3 TensorM M

c M M

M M

N N

c N N

N N

Page 13: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contraction

=

a ab bb

M

M

a ba

Page 14: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contraction

=P QR

ac ab bcb

R P Q

contraction cost a b c

b caa c

Page 15: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contraction

= P

Q

R

S

b

ca

b

cae

f g

abc afe fbg egcefg

S P Q R

Page 16: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Trace

=

=

Maa

a

z M

P Rab abcc

c

P R

a

b

a

b

a

c

Page 17: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Tensor product

a be a b

ab

dcf c d

e a b

(Reshaping)

Page 18: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Decomposition

=M Q D

1Q

=M U S V

=TU S V

Page 19: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Decomposing tensors can be useful

=M QP

d d d d

d

Number of components in M = 2d

Number of components in P and Q = 2 d

Rank(M) =

Page 20: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contents

• The quantum many body problem.

• Diagrammatic Notation

• What is a tensor network?

• Example 1 : MPS

• Example 2 : MERA

Page 21: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

1 2

1 2

1 2N

N

i i i Ni i i

i i i

Many-body state as a tensor

1i 2i Ni

Page 22: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Expectation values

O1 2 1 2

1 2

*

N N

N

i i i k i i ii i i

O

O

contraction cost = NO D

Page 23: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlators

1 2OO

1O 2O

contraction cost = NO D

Page 24: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Reduced density operators

contraction cost = NO D

Trs block

Page 25: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Tensor network decomposition of a state

Page 26: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Essential features of a tensor network

1) Can efficiently store the TN in memory

2) Can efficiently extract expectation values of local observables from TN

Total number of components = O(poly(N))

Computational cost = O(poly(N))

Page 27: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Number of tensors in TN = O(poly(N)) is independent of N

1

Page 28: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contents

• The quantum many body problem.

• Diagrammatic Notation

• What is a tensor network?

• Example 1 : MPS

• Example 2 : MERA

Page 29: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Matrix Product States

MPS

Page 30: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

1

2Total number of components = N D

Page 31: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Recall!

O1 2 1 2

1 2

*

N N

N

i i i k i i ii i i

O

O

contraction cost = NO D

Page 32: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Expectation values

Page 33: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Expectation values

Page 34: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Expectation values

Page 35: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Expectation values

Page 36: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Expectation values

4contraction cost = O N D

Page 37: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

But is the MPS good for representing ground states?

Page 38: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

But is the MPS good for representing ground states?

Claim: Yes!Naturally suited for gapped systems.

Page 39: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Recall!

1) Gapped Hamiltonian

2) Critical Hamiltonian

( ) log( )S l l

( )S l const l /( ) lC l e

( ) 0aC l l a

Page 40: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

In any MPS

Correlations decay exponentially

Entropy saturates to a constant

MPS

Page 41: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Recall!

1 2OO

1O 2O

contraction cost = NO D

Page 42: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlations in a MPS

l

0 1l

Page 43: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlations in a MPS

l

Page 44: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlations in a MPS

l

Page 45: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlations in a MPS

l

Page 46: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlations in a MPS

M M M

l

Page 47: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Correlations in a MPS

lM

0 1l

1l l l lL M R L QD Q R L D R

Page 48: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Entanglement entropy in a MPS

l

( )

S const

const

rank

Page 49: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Entanglement entropy in a MPS

Page 50: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Entanglement entropy in a MPS

Page 51: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Entanglement entropy in a MPS

Page 52: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Entanglement entropy in a MPS

Page 53: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Entanglement entropy in a MPS

2

ld

ld

2( ) rank

2log( )S

logi ii

S

Page 54: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

1. Variational optimization by minimizing energy

2. Imaginary time evolution

MPS as an ansatz for ground states

MPS

lim Htground state random

te

minMPS MPS MPSH

gs

0

Page 55: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Contents

• The quantum many body problem.

• Diagrammatic Notation

• What is a tensor network?

• Example 1 : MPS

• Example 2 : MERA

Page 56: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)
Page 57: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Summary

• The quantum many body problem.

• Diagrammatic Notation

• What is a tensor network?

• Example 1 : MPS

• Example 2 : MERA

Page 58: Introduction to Tensor Network States Sukhwinder Singh Macquarie University (Sydney)

Thanks !