generalized fourier transforms with the spectral tensor network
TRANSCRIPT
![Page 1: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/1.jpg)
Generalized Fourier transforms with the Spectral Tensor Network
Andy Ferris ICFO – Ins>tute of Photonic Science (Spain)
Recently jointly with: Max Planck Ins>tute for Quantum Op>cs
![Page 2: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/2.jpg)
What am I talking about today?
• Overall goal: use tensor networks to study fermionic models in 2D and beyond
• Introduce a generalized Fourier transform for bosons and fermions (and more!)
• This suggests a new tensor network ansatz to use varia>onally with interac>ng systems
![Page 3: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/3.jpg)
Tensor networks and area law
Many low-‐energy systems have entanglement bounded by boundary area, not volume. E.g. in 2D,
Can use tensor networks to accurately and efficiently describe such states
But: some “cri>cal” 2D systems obey:
S / L logL
S / L
R. Orus, arXiv:1306.2164 (2013)
![Page 4: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/4.jpg)
Free-‐fermions • Simple quadra>c / bi-‐linear Hamiltonian
• Represents metals, band insulators, etc
• Could be more complicated – Spins/orbitals, longer range interac>ons, etc – Anamolous pair terms
H =X
i
tc†i ci+1 + h.c.
c†i c†j + h.c.
![Page 5: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/5.jpg)
Entanglement in free-‐fermion systems
• Free-‐fermions exactly diagonalized by Fourier transform
• No entanglement in momentum space! • Lots of entanglement in real space, more than area law (depending on Fermi surface)
1D: 2D:
c†k
=1pn
X
x
c†x
eikx
S / logL S / L logL
![Page 6: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/6.jpg)
Where to?
• Given the large amount of entanglement in rela>vely simple systems, tensor networks like PEPS might not offer very efficient descrip>on of the state
• Here we make use of the fact that the state has no entanglement in momentum space
![Page 7: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/7.jpg)
The rest of this talk…
Fourier transform for quantum many-‐body systems
![Page 8: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/8.jpg)
Transla>on invariance
Transla>onally invariant states don’t change under transla>on: The Fourier transform is a unitary that diagonalizes . Eigenvalues of are .
T1| i = eik| i
T1 T1 eik
![Page 9: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/9.jpg)
Fast Fourier transform
Fourier transform of vector of numbers, . Linear transforma>on represented by matrix:
xj
2
664
1 1 1 11 i �1 �i1 �1 1 �11 �i �1 i
3
775F4 =
ak
=X
x
ax
eikx
![Page 10: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/10.jpg)
Fast Fourier transform
Matrix can be decomposed into product of sparse matrices (prime factors).
2
664
1 1 1 11 i �1 �i1 �1 1 �11 �i �1 i
3
775 = (F2 ⌦ I2)W (I ⌦ F2)P
F2 =
1 11 �1
� Diagonal phases “twiddle factors”
Permuta>on matrix
![Page 11: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/11.jpg)
Classical Fourier decomposi>on
The Fourier transform can be re-‐wrieen as a sum of two smaller Fourier transforms: Can be applied itera>vely for 2N sites
Even sites Odd sites Phase factor
FT over all N sites FT over N/2 even sites FT over N/2 odd sites
N�1X
x=0
e2⇡ikx
N ax
=
N/2�1X
x
0=0
e2⇡ikx
0N/2 a2x0 + e
2⇡ik
N
N/2�1X
x
0=0
e2⇡ikx
N/2 a2x0+1
![Page 12: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/12.jpg)
Classical Fourier decomposi>on
The Fourier transform can be re-‐wrieen as a sum of two smaller Fourier transforms: Can be applied itera>vely for 2N sites
Even sites Odd sites Phase factor
FT over all N sites FT over N/2 even sites FT over N/2 odd sites
N�1X
x=0
e2⇡ikx
N ax
=
N/2�1X
x
0=0
e2⇡ikx
0N/2 a2x0 + e
2⇡ik
N
N/2�1X
x
0=0
e2⇡ikx
N/2 a2x0+1
![Page 13: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/13.jpg)
Apply decomposi>on successively
!ab = �12a/b
The trick: use one-‐ and two-‐body linear elements
Momentum space
Real space
![Page 14: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/14.jpg)
Quantum unitary circuit for Fourier transform
Can use graphical iden>>es to manipulate it, in addi>on to . (Similar freedom in FFT) FT
n = Fn
![Page 15: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/15.jpg)
Gates for fermions, bosons…
• Gates are linear: • For fermions these are 4 x 4
• When wires cross, mul>ply by -‐1 when two fermions are exchanged.
c01 =c1 + c2p
2
c02 =c1 � c2p
2
U =
2
664
1 0 0 00 1/
p2 1/
p2 0
0 1/p2 �1/
p2 0
0 0 0 �1
3
775
c0† = c†ei�
![Page 16: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/16.jpg)
Fourier transform quantum circuit
!ab = �12a/b
![Page 17: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/17.jpg)
Spectral tensor network
![Page 18: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/18.jpg)
Expecta>on values
![Page 19: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/19.jpg)
Tensor contrac>ons • The remaining tensor network isomorphic to a tree – Fold through middle
• Can be contracted with cost .
• Everything with
• Two body
O(�5 n)
O(�8 n log n)
O(�8)
![Page 20: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/20.jpg)
2D Fourier transform
Actually, the structure of the Fourier transform is very similar in 2D, 3D, etc…
No change in cost.
![Page 21: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/21.jpg)
Results: 1D
Filling the lowest momentum states.
H =X
i
tc†i ci+1 + h.c.
g
(1)(�x) =hc†
i
c
i+�x
ihni
−6 −4 −2 0 2 4 6−0.5
0
0.5
1
1.5
6 x
Cor
rela
tion
func
tion
g(1)
g(2)
g
(2)(�x) =hn
i
n
i+�x
ihni2
![Page 22: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/22.jpg)
Results: 2D
• Start with 512 x 512 lagce (1/4 million sites) • Low filling factor (approximates free space)
6 x
6 y
g(1)
−4 −2 0 2 4
−4
−3
−2
−1
0
1
2
3
4
0
0.2
0.4
0.6
0.8
1
∆ x
∆ y
(b)
−4 −3 −2 −1 0 1 2 3 4
−4
−3
−2
−1
0
1
2
3
4 0
0.2
0.4
0.6
0.8
1g(2)
![Page 23: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/23.jpg)
Bogoliubov transforma>ons H =
X
i
tc†i ci+1 +�c†i c†i+1 + h.c.
H =X
i
�x
i
�x
i+1 + h�z
i
Correla>ons between modes
±k
⌘k = uk ck + vk c†�k
0.8 0.9 1 1.1 1.20
0.5
1
1.5
2
2.5
h
r
1024 sites (c-‐cyclic approxima>on)
![Page 24: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/24.jpg)
Free fermions: Other things to calculate
• Few-‐site correla>ons, finite temperature, 3D systems, entanglement entropy of blocks
• Mul>ple bands or species per site – Conductors (metals) with arbitrary Fermi-‐ surface or Dirac points
– Band-‐insulators – Topological phases (chiral, Majorana, SPTO…) 0
0
2
4
6
8
+kLkLQuasimomentum q
Energy�(E
R)
![Page 25: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/25.jpg)
Interac>ng systems?
![Page 26: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/26.jpg)
Varia>onal approach
The “spectral tensor network” can be used as a wave-‐func>on ansatz
Fill with arbitrary unitary
![Page 27: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/27.jpg)
Generalized Fourier transform The operator that diagonalizes the shil operator is not unique
Notably: a simple constraint on the gates leads to the same decomposi>on as standard FFT.
![Page 28: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/28.jpg)
2-‐site Fourier transform
The gate must diagonalize – Eigenvalues (symmetric or an>symmetric)
T1 = ˆSWAP±1
Parity of state in real space = number parity of -‐momentum state ⇡
(real space)
(momentum space)
*
k=⇡k=0 k=⇡k=0
![Page 29: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/29.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 30: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/30.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 31: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/31.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 32: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/32.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 33: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/33.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 34: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/34.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 35: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/35.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 36: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/36.jpg)
4-‐site Fourier transform
Using any such a gate we can construct a Fourier transform!
!04 = 1
!14 = i
!24 = �1
!34 = �i
![Page 37: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/37.jpg)
Larger systems The paeern con>nues… Generally, can decompose any non-‐prime system size.
![Page 38: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/38.jpg)
The meaning of the phase factors
We can interpret the phase factors as applying a half-‐quanta momentum-‐boost to the right system
T.I. system 1 T.I. system 2
![Page 39: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/39.jpg)
Generalized fast Fourier transform
![Page 40: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/40.jpg)
Spectral tensor network (1D)
![Page 41: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/41.jpg)
The possibili>es
• We now have a tool for crea>ng a wide variety of transla>onally invariant states. Known cases:
• Bosons: problem -‐ large bond dimension • Fermions: small Fock space – Free fermions include beyond area-‐law, chiral states, Majorana/SPTO, etc…
– Interac>ng fermions?
• Abelian anyons: E.g. parafermions
![Page 42: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/42.jpg)
A simple example
Generalized Bose “condensate”
![Page 43: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/43.jpg)
A simple example
Generalized Bose “condensate” Coherent state
MoE state (repulsive)
Collapsed state/“Bosanova” (aEracMve)
| i = |↵i|↵i . . . |↵i
| i = |1111 . . . i| i = |101010 . . . i+ |010101 . . . i
| i = |N00 . . . i+ |0N0 . . . i+ . . .
![Page 44: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/44.jpg)
Beeer example: Hubbard model
Each site has two fermion orbitals (spin up/down).
Unitary has significant freedom.
H =X
i,s
�ta†i,sai+1,s + h.c.+ Uni,"ni,#
![Page 45: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/45.jpg)
Arbitary unitary
Beeer example: Hubbard model Unitary must preserve number & sort momenta:
n = 0
n = 1
n = 2
n = 3
n = 4
k = 0 k = ⇡|00i
| " 0i+ |0 "i | " 0i � |0 "i| # 0i � |0 #i| # 0i+ |0 #i
|22i
| ""i| ##i| "#i+ | #"i | "#i � | #"i|20i � |02i|20i+ |02i
|2 "i+ | " 2i|2 "i � | " 2i|2 #i � | # 2i
|2 #i+ | # 2i
,
![Page 46: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/46.jpg)
|2 "i+ | " 2i|2 "i � | " 2i|2 #i � | # 2i
|2 #i+ | # 2i
Arbitary unitary
Beeer example: Hubbard model Unitary must preserve number & sort momenta:
n = 0
n = 1
n = 2
n = 3
n = 4
k = 0 k = ⇡|00i
| " 0i+ |0 "i | " 0i � |0 "i| # 0i � |0 #i| # 0i+ |0 #i
|22i
| ""i| ##i| "#i+ | #"i | "#i � | #"i|20i � |02i|20i+ |02i,
![Page 47: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/47.jpg)
Z3 parafermions • Non-‐interac>ng parafermions. No more than two par>cles per site.
• Operators get phase when commuted:
• Local representa>on of operator:
a† 3i = 0
aiaj = e2⇡i/3aj ai (i < j)
a†i aj = e2⇡i/3aj a†i (i < j)
a†i =
2
40 1 00 0
p2
0 0 0
3
5 Like bosons with truncated Fock
space
![Page 48: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/48.jpg)
Z3 parafermions
• Can find a two-‐body gate such that
• This gate sa>sfies the condi>ons for the spectral tensor network. Implica>ons: – Can easily perform (linear) Fourier transform on these parafermions
– Exactly solve transla>onally invariant, quadra>c Hamiltonians (1D, 2D, long-‐range hopping + chemical poten>al)
U a1U† =
a1 + a2p2
U a2U† =
a1 � a2p2
![Page 49: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/49.jpg)
Parafermion “Jordan-‐Wigner” transforma>on
• Can perform a more general Jordan-‐Wigner transforma>on to a spin-‐1 chain.
• Doesn’t work for anomalous terms. • Gives a very messy spin-‐1 Hamiltonian – I won’t even write it here!
• (previously known result)
ai = exp(2⇡i/3X
j<i
(
ˆZj + 1)
ˆMi)
![Page 50: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/50.jpg)
Discussion
• We are just scratching the surface of what is possible
• Many ways to implement / extend the ansatz – TN geometry, system type, Bogoliubov, MPS…
• Many open ques>ons – Efficient for interac>ng models? Fermi liquids? – Open boundary condi>ons? (DST/DCT) – Impurity problems or other non-‐transla>onally invariant systems?
![Page 51: Generalized Fourier transforms with the Spectral Tensor Network](https://reader031.vdocument.in/reader031/viewer/2022022000/58a1b0561a28abd8728bfb1e/html5/thumbnails/51.jpg)
Thank you!