Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Concretize Ansatz type with "lattice"/connectivity information and refactor subtypes on top of it #204

Draft
wants to merge 75 commits into
base: master
Choose a base branch
from
Draft

Concretize Ansatz type with "lattice"/connectivity information and refactor subtypes on top of it #204

wants to merge 75 commits into from

Conversation

mofeing
Copy link
Member

@mofeing mofeing commented Sep 16, 2024 •
edited
Loading

Inspired by ITensorNetworks.jl (but not equal), I experimented with using a graph for representing the connectivity between different sites.

In "ITensorNetworks.jl", they use the graph for representing the structure of the TensorNetwork and the connectivity graph of the different sites, which is useful when dealing with lattices or hamiltonians. Unlike them, the hypergraph structure is already implicit in TensorNetwork, but it doesn't comfortably accommodate for fixed structured Tensor Networks like MPS, PEPS, TTN, ...

By adding a graph field in Ansatz that accounts for the connectivity pattern between sites, a lot of the software design problems can be mitigated.

I've also refactored Chain into MPS and MPO and Grid into PEPS and PEPO (actually, I've never seen a PEPO in real life).

@jofrevalles @starsfordummies you might want to reconsider refactoring your PR on top of this.

@mofeing mofeing added refactor Change internals triage Needs group consensus breaking-change labels Sep 17, 2024
@mofeing mofeing self-assigned this Sep 17, 2024
@mofeing mofeing force-pushed the refactor/ansatz-lattice branch 4 times, most recently from 678c3af to 9b22b13 Compare September 19, 2024 17:55
@mofeing
Prototype MPS, MPO
9d8f4ef
@mofeing
Introduce Canonical Form trait
e48e2c6
@mofeing
Implement rand, adjoint, defaultorder, boundary, form for `…
0382f57 
…MPS`, `MPO`
@mofeing
Implement conversion from Product to MPS, MPO
d29ff7d
@mofeing
Implement hassite method and alias to in
0f0e5dd
@mofeing
Refactor Ansatz into a concrete type
555440a
@mofeing
Fix typos in Ansatz
425df9e
@mofeing
Use Graphs.neighbors
d2c8ce4
@mofeing
Refactor Product on top of new Ansatz type
e492846
@mofeing
Implement copy, similar, zero for Ansatz
d334d13
@mofeing
Format code
86fdd7a
@mofeing
Refactor Dense on top of new Ansatz type
1937f80
@mofeing
Refactor MPS, MPO on top of new Ansatz type
ba817a2
@mofeing
Relax sites condition on Ansatz construction for lanes
3370244
@mofeing
Implement PEPS, PEPO types
8c05563
@mofeing
Remove some exports
c21db5e
@mofeing
Move Chain code to AbstractAnsatz and MPS
c9d351f
@mofeing
Force Graphs to be a strong dependency
c76fdb3
@mofeing mofeing added the on hold Blocked due to some reason label Sep 23, 2024
@mofeing
Copy link
Member Author

mofeing commented Sep 23, 2024

@jofrevalles Seems like the problem is (again) related to the way indices are replaced when doing merge!. I thought it was a good idea, but we might want to rethink how to deal with indices on that case.

@jofrevalles
Copy link
Member

@jofrevalles Seems like the problem is (again) related to the way indices are replaced when doing merge!. I thought it was a good idea, but we might want to rethink how to deal with indices on that case.

In my opinion, it would be much better to revert the code to what we had previously (so indices are not repeated, and just have an uuid). If we want to have "nice symbols" when doing plots, we can do some conversion or something before plotting in order to fix the labels.

@mofeing
Copy link
Member Author

mofeing commented Sep 23, 2024

If we want to have "nice symbols" when doing plots, we can do some conversion or something before plotting in order to fix the labels.

I wouldn't change the indices for plotting because it will lie to the users.

@jofrevalles
Copy link
Member

I wouldn't change the indices for plotting because it will lie to the users.

I mean, you can take any tn, and replace the indices with any mapping you want. For example, we could modify the internals so you could choose to use uuid for some situations, and "nice symbols" (using some kind of index counter as we had) in other situations.

mofeing and others added 22 commits September 26, 2024 12:33
@mofeing
Some fixes for PEPS constructor
54312c5
@mofeing
Remove check in PEPS constructor
9781717
@mofeing
Fix reference to lattice in adapt_structure for Ansatz
5b8da11
@mofeing
Stop orthogonalization to index on mixed_canonize!
63edfc2
@mofeing
Aesthetic name fix
7921a49
@mofeing
Stop using IdDict on Reactant extension
eb95d07
@mofeing
Fix create_result on MPS, MPO
22824d8
@mofeing
Refactor lattice generation in constructors of Dense, Product, `M…
dfb5c2e 
…PS`, `MPO`, `PEPS`
@mofeing
Fix make_tracer, create_result from Reactant on Product, Dense
ef25de5
@mofeing
Implement rand, normalize!, overlap for Dense states
9ffddf3
Set temporarily a more concrete type of lattice in graph to circunv…
7cd5fae 
…ent a Julia bug
@mofeing
Fix pairwise contract between TracedRArray and Array
71964b0
@mofeing
Small fix to promote_to of Tensor
20d7183
@mofeing
Refactor by importing Reactant.TracedRArray
4121133
@mofeing
Try remove ambiguity on contract with TracedRArray
4aa2830
@mofeing
Dispatch det, logdet, tr methods to underlying array on matrix …
2579137 
…`Tensor`s
@mofeing
Implement Eigendecomposition for Tensor
1c65921
@mofeing
Implement eigen! for TensorNetwork
3670dfe
@mofeing
small fix
cd937af
@Todorbsc @jofrevalles @mofeing
Implement an MPS method initializing the tensors to identity (copy-te…
fee3b07 
…nsors) (#218)

* Format code

* Implement MPS identity initialization

* Add tests for all dispatches of MPS identity init

* Format julia code

* Rename function header & add docstring

* Fix test set for identity MPS

* Format code

* Rewrite MPS identity init function to nsites instead of arrays' dimensions

* Format julia code

* Update docstring of identity

* Clean code in test (suggested by Jofre)

Co-authored-by: Jofre Vallès Muns <[email protected]>

* Format julia code

* Refactor virtualdims in identity (suggested by Sergio)

* Update src/Ansatz/MPS.jl

* Restrict to default order in identity MPS

* Update src/Ansatz/MPS.jl

* Remove order parameter in identity

---------

Co-authored-by: Jofre Vallès Muns <[email protected]>
Co-authored-by: Sergio Sánchez Ramírez <[email protected]>
@mofeing
Implement traced_type for AbstractTensorNetwork
f39b6f8 
Fixes some problems when calling `make_tracer` on an object that wraps an `Ansatz`, like `Tuple{MPS}` for example.
@mofeing
Format code
f5aa926
@mofeing mofeing marked this pull request as ready for review October 21, 2024 10:17
@mofeing mofeing mentioned this pull request Oct 29, 2024
Merged
@mofeing mofeing marked this pull request as draft November 3, 2024 00:36
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jofrevalles jofrevalles Awaiting requested review from jofrevalles

@arturgs arturgs Awaiting requested review from arturgs

@starsfordummies starsfordummies Awaiting requested review from starsfordummies

@Todorbsc Todorbsc Awaiting requested review from Todorbsc

At least 1 approving review is required to merge this pull request.

Assignees

@mofeing mofeing

Labels
breaking-change on hold Blocked due to some reason refactor Change internals triage Needs group consensus
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

Move Ansatz type-hierarchy to a trait system
3 participants
@mofeing @jofrevalles @Todorbsc