lecture 4: tensor product ansatz - part ii graduate lectures€¦ · lecture 4: tensor product...
TRANSCRIPT
![Page 1: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/1.jpg)
Lecture 4: Tensor Product Ansatz - part II
Dr Gunnar MöllerCavendish Laboratory, University of Cambridge
Cavendish Laboratory
January 2014
Graduate Lectures
slide credits: Philippe Corboz (ETH / Amsterdam)
![Page 2: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/2.jpg)
Summary: Tensor network algorithms
StructureVariational
ansatz
iterative optimization of individual tensors
(energy minimization)imaginary time
evolution
Contraction of the tensor network
exact / approximate
Find the best (ground) state
|��
Compute observables��|O|�⇥
MPSPEPS
2D MERA
1D MERA
TODAY
✓
![Page 3: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/3.jpg)
Outline, part II
‣ Optimization of tensor networks✦ Variational optimization
✦ Imaginary-time evolution
‣ Contraction of tensor networks ✦ Basics
✦ Approximate contraction of PEPS/iPEPS
‣ Outlook & Summary
![Page 4: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/4.jpg)
1. Select one of the PEPS tensors T
Variational optimization for PEPS
E =h |H| ih | i
tensor T reshaped as a vector
environment including all Hamiltonian terms environment from norm term
3. Take the next tensor (leave others fixed)
4. Repeat 2-3 iteratively until convergence is reached
2. Optimize tensor T (leaving all the others fixed) by minimizing the energy:
F x = E G x
solve generalized eigenvalue problem
minimize
Verstraete, Murg, Cirac, Adv. in Phys. 57, 143 (2008)
![Page 5: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/5.jpg)
Variational Optimization for MPSs
min[h |H| i � �h | i]minimize Energy E, enforcing normalization with a Lagrange multiplier λ
- λHh |
| ih |
minimize with respect to tensor T: @
@T lr�
⇤ ( % ) = 0
| i
![Page 6: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/6.jpg)
- λHh |
| ih |
@
@T lr�
⇤ ( % ) = 0
Variational Optimization for MPS’s
| iT T
T* removed
= 0
T* removed
read as matrix equation in linearized composite index
a b�
�0a0 b0
a0 b0
� = �0
µ = (a b�)
Fµ0µTµ = �Gµ0µTµ
F, G: remainders of tensor networks with both T and T* cut out
☛ solve for smallest eigenvalue λ0 and -vector T = new optimized tensor
(Gµ0µ / ���0)where
in pictures:
![Page 7: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/7.jpg)
Optimization via imaginary time evolution
• Get the ground state via imaginary time evolution (Trotter-Suzuki)
exp(�⌧ ˆHb)
• At each step: apply a two-site operator to a bond and truncate bond back to D
Keep D largest singular values
Ups
psV
SVD sU V †
Time Evolving Block Decimation (TEBD) algorithm
Note: Here I simplified. MPS needs to be in canonical form
⌧ = �/n
exp(�� ˆH) = exp(��X
b
ˆHb) =
exp(�⌧
X
b
ˆHb)
!n
⇡ Y
b
exp(�⌧ ˆHb)
!n
...
...
![Page 8: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/8.jpg)
Optimization via imaginary time evolution
• Get the ground state via imaginary time evolution (Trotter-Suzuki)
• However, SVD update is not optimal (because of loops in PEPS)!
simple update (SVD)
★ “local” update like in TEBD
★ Cheap, but not optimal(e.g. overestimates magnetization in S=1/2 Heisenberg model)
full update
★ Take the full wave function into account for truncation
★ optimal, but computationally more expensive
Cluster update Wang, Verstraete, arXiv:1110.4362 (2011)
• 2D: same idea: apply a two-site operator to a bond and truncate bond back to D at each step
exp(�⌧ ˆHb)
⌧ = �/n
exp(�� ˆH) = exp(��X
b
ˆHb) =
exp(�⌧
X
b
ˆHb)
!n
⇡ Y
b
exp(�⌧ ˆHb)
!n
![Page 9: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/9.jpg)
Contracting tensor networks
![Page 10: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/10.jpg)
Contracting a tensor network
contract tensors pairwise
![Page 11: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/11.jpg)
Pairwise contractions...
![Page 12: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/12.jpg)
Pairwise contractions...
![Page 13: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/13.jpg)
Pairwise contractions...
![Page 14: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/14.jpg)
Pairwise contractions...
![Page 15: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/15.jpg)
Pairwise contractions...
![Page 16: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/16.jpg)
Pairwise contractions...
done!
the order of contraction matters for the computational cost!!!
![Page 17: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/17.jpg)
Contracting a tensor network
★ Reshape tensors into matrices and multiply them with optimized routines (BLAS)
i
j
u
vA B w =
u
v
wT
= wT(uv)
dimension D
★ Computational cost: multiply the dimensions of all legs (connected legs only once)
cost D5
B w(uv) A(ij)
dimension D2
![Page 18: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/18.jpg)
Contracting an MPS��
|�⇥
= BAD!
��|�
⇥
= Good!
![Page 19: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/19.jpg)
Relation to Transfer Matrixh
| i
hvl|
d=D2 vector D2 x D2 matrix
⇥hv0l|
=
=
Norm
Correlation functions for two sites
hvl|
�zi
. . .
�zj
| {z }Tn
each rung acts as a transfer matrix T !
![Page 20: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/20.jpg)
MERA: Properties
��|
Let’s compute ��|O|�⇥ O : two-site operator
|��
O two-site operator
![Page 21: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/21.jpg)
Causal cone
MERA: Contraction��
|O|�
⇥
Isometries are isometric
= I
u
u†
= I
w
w†
Disentanglers are unitary
![Page 22: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/22.jpg)
MERA: Contraction��
|O|�
⇥
Causal cone
Efficient computation of expectation values of observables!
Isometries are isometric
= I
u
u†
= I
w
w†
Disentanglers are unitary
![Page 23: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/23.jpg)
Contracting the PEPS
€
Ψ
€
Ψ†
€
≡
��|�
⇥
reduced tensors
D2
![Page 24: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/24.jpg)
Contracting the PEPS
Problem: how do we contract this??
no matter how we contract, we will get intermediate
tensors with O(L) legs
number of coefficients DL
Exponentially increasing with L!
NOT EFFICIENT
dimension D2
![Page 25: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/25.jpg)
Contracting the PEPS
★ Exact contraction of an PEPS is exponentially hard!
need approximate contraction scheme
MPS-based approach
Corner transfer matrix method
TRG Tensor Renormalization Group
Murg,Verstraete,Cirac, PRA75 ’07Jordan,et al. PRL79 (2008)
Nishino, Okunishi, JPSJ65 (1996)Orus, Vidal, PRB 80 (2009)
Gu, Levin, Wen, B78, (2008)Levin, Nave, PRL99 (2007)Xie et al. PRL 103, (2009)
★Accuracy of the approximate contraction is controlled by “boundary dimension” �
★Convergence in needs to be carefully checked�
★Overall cost: with O(D10...12) � ⇠ D2
![Page 26: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/26.jpg)
Contracting the PEPS
★Convergence in needs to be carefully checked�
★Overall cost: with O(D10...12) � ⇠ D2
0 0.02 0.04 0.06 0.08 0.1−0.6695
−0.669
−0.6685
1/χ
E s
D=4D=5D=6
Example: 2D Heisenberg model
★Accuracy of the approximate contraction is controlled by “boundary dimension” �
![Page 27: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/27.jpg)
Contracting the PEPS
★ Exact contraction of an PEPS is exponentially hard!
need approximate contraction scheme
MPS-based approach
Corner transfer matrix method
TRG Tensor Renormalization Group
Murg,Verstraete,Cirac, PRA75 ’07Jordan,et al. PRL79 (2008)
Nishino, Okunishi, JPSJ65 (1996)Orus, Vidal, PRB 80 (2009)
Gu, Levin, Wen, B78, (2008)Levin, Nave, PRL99 (2007)Xie et al. PRL 103, (2009)
★Convergence in needs to be carefully checked�
★Overall cost: with O(D10...12) � ⇠ D2
★Accuracy of the approximate contraction is controlled by “boundary dimension” �
![Page 28: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/28.jpg)
this is an MPO (matrix product operator)
this is an MPS
Contracting the PEPS using an MPSdimension D2
Verstraete, Murg, Cirac, Adv. in Phys. 57, 143 (2008)
![Page 29: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/29.jpg)
Contracting the PEPS using an MPS
this is an MPS with bond dimension D2 xD2
dimension D2xD2
truncate the bonds to �
there are different techniques for the efficient MPO-MPS multiplication (SVD, variational optimization, zip-up algorithm...)Schollwöck, Annals of Physics 326, 96 (2011)Stoudenmire, White, New J. of Phys. 12, 055026 (2010).
Verstraete, Murg, Cirac, Adv. in Phys. 57, 143 (2008)
![Page 30: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/30.jpg)
Contracting the PEPS using an MPS
dimension �
proceed...
★ We can do this from several directions★ Similar procedure when computing an expectation value
Verstraete, Murg, Cirac, Adv. in Phys. 57, 143 (2008)
![Page 31: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/31.jpg)
Compute expectation values
Figure taken from Corboz, Orús, Bauer, Vidal, PRB 81, 165104 (2010)
environment
compute environment approximately
Connect two-body operator
Contract this network!
![Page 32: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/32.jpg)
Contracting the iPEPS using the corner transfer matrix
A A A AA A
A A A AA A
A A A AA A
A A A A AA
A A A A AA
A A A A AA
iPEPSinfinite projected entangled-pair state
open boundaries, but “infinitely” far away
![Page 33: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/33.jpg)
figure taken from Orus, Vidal, PRB 80 (2009)
Contracting the iPEPS using the corner transfer matrix Nishino, Okunishi, JPSJ65 (1996)Orus, Vidal, PRB 80 (2009)
★ Let the system grow in all directions.
★ Repeat until convergence is reached
★ The boundary tensors form the environment
★ Can be generalized to arbitrary unit cell sizes
Corboz, et al., PRB 84 (2011)
dimension �
![Page 34: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/34.jpg)
Contracting the PEPS/iPEPS using TRG Gu, Levin, Wen, B78, (2008)Levin, Nave, PRL99 (2007)Xie et al. PRL 103, (2009)
★ Contract PEPS with periodic boundary conditions★ Finite or infinite systems★ Related schemes: SRG, HOTRG, HOSRG, ...
Tensor Renormalization Group
SVD dimension �sublattice A:
SVDsublattice B:
![Page 35: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/35.jpg)
Contracting the PEPS
★ Exact contraction of an PEPS is exponentially hard!
need approximate contraction scheme
MPS-based approach
Corner transfer matrix method
TRG Tensor Renormalization Group
Murg,Verstraete,Cirac, PRA75 ’07Jordan,et al. PRL79 (2008)
Nishino, Okunishi, JPSJ65 (1996)Orus, Vidal, PRB 80 (2009)
Gu, Levin, Wen, B78, (2008)Levin, Nave, PRL99 (2007)Xie et al. PRL 103, (2009)
★Convergence in needs to be carefully checked�
★Overall cost: with O(D10...12) � ⇠ D2
�★Accuracy of the approximate contraction is controlled by
“boundary dimension” �
![Page 36: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/36.jpg)
H| i
Expressing the Hamiltonian as a MPO
![Page 37: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/37.jpg)
Summary: Tensor network algorithmsMPS
StructureVariational
ansatz
iterative optimization of individual tensors
(energy minimization)imaginary time
evolution
Contraction of the tensor network
exact / approximate
Find the best (ground) state
|��
Compute observables��|O|�⇥
PEPS2D MERA
1D MERA
✓✓✓
![Page 38: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/38.jpg)
0 0.05 0.1
0.3
0.32
0.34
0.36
0.38
0.4
0.42
1/L
stag
gere
d m
agne
tizat
ion
QMC from Ms2
QMC from C(L/2,L/2)QMC extrap.
0 0.2 0.4 0.61/D
iPEPSQMC extrap.linear fitsquadratic fits
Heisenberg model: Comparison QMC vs iPEPS
Energy: QMC (extrap.):
iPEPS (D=10): rel. error < 10-4
strong finite D effects
how to extrapolate?
0 0.05 0.1
0.3
0.32
0.34
0.36
0.38
0.4
0.42
1/L
stag
gere
d m
agne
tizat
ion
QMC from Ms2
QMC from C(L/2,L/2)QMC extrap.
0 0.2 0.4 0.61/D
iPEPSQMC extrap.linear fitsquadratic fits
strong finite size effects
accurate extrapolation
m = 0.30743(1)
m ⇡ 0.295� 0.314rel. error ~ 2%
-0.66939J
-0.669437(5)J
QMC study: Sandvik & Evertz, PRB82, 024407 (2010): system sizes up to 256x256
A. Sandvik, PRB56, 11678 (1997)
A Benchmark: Heisenberg model
![Page 39: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/39.jpg)
H = J�
�i,j⇥A
SiSj +�
�i,j⇥B
SiSj
. . .
. . .. . .
. . .
J0
Wenzel, Janke, PRB 79 (2009)
1
Phase diagram
Distinguish between ordered / disordered phase?
2nd
orde
r ph
ase
tran
sitio
n
0.3969Dimer phase
S S
S
SS
S S
SS
S
m=0
Néel phase
m>0
![Page 40: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/40.jpg)
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
1/D
m
J=0.39
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
1/Dm
J=0.3969
0 0.1 0.2 0.3 0.4 0.50
0.1
0.2
0.3
1/D
m
J=0.4
m=0 m~0 m>0
Distinguish between ordered / disordered phase?
★ Extrapolations in D are important to distinguish between ordered phase and a disordered one!
★A better understanding how to accurately extrapolate in D would be very useful
![Page 41: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/41.jpg)
Fermionic tensor networks
![Page 42: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/42.jpg)
Breakthrough in 2009: Fermions with 2D tensor networks
How to take fermionic statistics into account?
P. Corboz, G. Evenbly, F. Verstraete, G. Vidal, Phys. Rev. A 81, 010303(R) (2010)C. V. Kraus, N. Schuch, F. Verstraete, J. I. Cirac, Phys. Rev. A 81, 052338 (2010)C. Pineda, T. Barthel, and J. Eisert, Phys. Rev. A 81, 050303(R) (2010)P. Corboz and G. Vidal, Phys. Rev. B 80, 165129 (2009)T. Barthel, C. Pineda, and J. Eisert, Phys. Rev. A 80, 042333 (2009)Q.-Q. Shi, S.-H. Li, J.-H. Zhao, and H.-Q. Zhou, arXiv:0907.5520 P. Corboz, R. Orus, B.Bauer, G. Vidal, PRB 81, 165104 (2010)I. Pizorn, F. Verstraete, Phys. Rev. B 81, 245110 (2010)Z.-C. Gu, F. Verstraete, X.-G. Wen. arXiv:1004.2563
Different formulations:
fermionic operators anticommute
cicj = �cj ci
![Page 43: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/43.jpg)
Bosons vs Fermions
�B(x1, x2) = �B(x2, x1)
symmetric!
�F (x1, x2) = ��F (x2, x1)
antisymmetric!
+ --Crossings in a tensor network
ignore crossings take care!
--=+
operators anticommutebibj = bj bi
operators commute
cicj = �cj ci
3x3 PEPS
€
Ψ
€
؆��|�
⇥
![Page 44: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/44.jpg)
The swap tensor
RULE:
# Fermions
+
even even
+
evenodd
--oddodd
+
oddeven Parity
(even parity), even number of fermionsParity of a state: P
(odd parity), odd number of fermions
P = +1
P = �1{Replace crossing by swap tensor
i1 i2
j2 j1
B
S(P (i1), P (i2)) =��1+1
if P (i1) = P (i2) = �1otherwise
Bi1i2j2j1
= �i1,j1�i2,j2S(P (i1), P (i2))
Use parity preserving tensors: Ti1i2...iM = 0 if P (i1)P (i2) . . . P (iM ) �= 1
![Page 45: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/45.jpg)
ExampleBosonic tensor network Fermionic tensor network
Simple rules!Same computational cost!
![Page 46: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/46.jpg)
Outlook
![Page 47: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/47.jpg)
Improvements of tensor networks methods
Tensor networks
Monte Carlo sampling
Fixed-node Monte CarloNew tensor networkscombine w. variational WF
Symmetries
Parallelization
Better optimization algorithms
![Page 48: Lecture 4: Tensor Product Ansatz - part II Graduate Lectures€¦ · Lecture 4: Tensor Product Ansatz - part II Dr Gunnar Möller Cavendish Laboratory, University of Cambridge Cavendish](https://reader033.vdocuments.us/reader033/viewer/2022051602/5aeb38e97f8b9a36698e0e1a/html5/thumbnails/48.jpg)
‣ Variational ansatz where the accuracy can be systematically controlled
‣ Simulate bosonic, (frustrated) spin and fermionic systems
Conclusion:Tensor network algorithms
✓ 1D: State-of-the-art (MPS, DMRG)
✓ 2D:A lot of progress in recent years!★ iPEPS can outperform state-of-the-art variational methods!
★ cMPS (not discussed) evolving as state of the art for fractional QHE
Tensor networks yield promising routes tosolve challenging open problems in 2D