Refactor `Ansatz` to be a concrete type with lattice information #223

mofeing · 2024-10-26T22:14:00Z

As discussed offline, we decided to slice #204 into smaller PRs. This first PR makes Ansatz a concrete type with lattice information contained in a graph.

The important thing you should review is the API, not the details; we can change the graph type in the future but we want the API to be stable.

Changelog

Make Ansatz a concrete type in the type hierarchy.
- Its abstract counterpart is AbstractAnsatz.
- It adds Graphs and MetaGraphsNext as dependencies.
Delete Chain, Product, Grid and Dense
- These will be re-added in subsequent PRs, adapted to the new Ansatz layer.
Make Form as the trait canonical form information.
Refactor some evolve! methods based on simple-update to simple_update! and generalize it for any AbstractAnsatz.

Notes

Currently there is a bug in Julia makes Enzyme crash on an Ansatz due to its field lattice being abstract active_reg_inner crashes on Tuple{X} where {X} EnzymeAD/Enzyme.jl#1923
- It's fixed in Julia 1.11.0 and should be fixed in Julia ~~1.10.6~~ 1.10.7
- I temporarily fixed it by fixing Ansatz to just 1D connectivity (i.e. Site{1}) but I'm removing it now -> For the time being you will need Julia 1.11 if you want to diff an Ansatz
There are other bug fixes (like Base.in or GraphMakie.graphplot! on AbstractTensorNetwork) that I'm gonna remove now and move to other PRs.

jofrevalles · 2024-10-29T09:13:05Z

src/Tenet.jl

@@ -26,24 +26,9 @@ include("Ansatz/Ansatz.jl")
 export Ansatz
 export socket, Scalar, State, Operator
 export boundary, Open, Periodic
+export form



Suggested change

why not export it?

jofrevalles · 2024-10-29T09:17:24Z

@mofeing @starsfordummies I think that when we talked about splitting the large PR into small ones, we meant that this PR would only contain the new Ansatz struct, and I see here that you pushed the commits regarding the canonical form, which IMO they should be in another PR. In this way, we can refactor the tests that are failing (like Chain, Dense, ...) in another PR.

mofeing · 2024-10-29T09:19:11Z

@mofeing @starsfordummies I think that when we talked about splitting the large PR into small ones, we meant that this PR would only contain the new Ansatz struct, and I see here that you pushed the commits regarding the canonical form, which IMO they should be in another PR. In this way, we can refactor the tests that are failing (like Chain, Dense, ...) in another PR.

Sorry, but I don't agree. The Form trait is part of Ansatz, even if it doesn't have a field for that.

The other ansatzes will be reimplemented in subsequent PRs because tests had to be changed.

…top of recent changes

…ent a Julia bug

… circunvent a Julia bug" This reverts commit 7c67f58.

jofrevalles

I prefer to have @test_skip instead of commenting all the tests.

jofrevalles · 2024-10-29T10:48:37Z

test/integration/YaoBlocks_test.jl

@@ -8,35 +8,35 @@
    @test issetequal(sites(tn), [site"1", site"2", site"1'", site"2'"])
    @test Tenet.ntensors(tn) == 2

-    @testset "GHZ Circuit" begin
-        circuit_GHZ = chain(n_qubits, put(1 => Yao.H), Yao.control(1, 2 => Yao.X), Yao.control(2, 3 => Yao.X))
+    # @testset "GHZ Circuit" begin


Suggested change

# @testset "GHZ Circuit" begin

@test_skip begin #@testset "GHZ Circuit"

jofrevalles · 2024-10-29T10:49:19Z

test/Product_test.jl

@@ -1,33 +1,33 @@
-@testset "Product ansatz" begin
-    using LinearAlgebra
+# @testset "Product ansatz" begin


Suggested change

# @testset "Product ansatz" begin

@test_skip begin # @testset "Product ansatz"

jofrevalles · 2024-10-29T10:49:47Z

test/Ansatz_test.jl

@@ -0,0 +1,2 @@
+@testset "Ansatz" begin


jofrevalles · 2024-10-29T10:54:16Z

src/Ansatz/Ansatz.jl

+  - Either `threshold` or `maxdim` must be provided. If both are provided, `maxdim` is used.
+  - The bond must contain the Schmidt coefficients, i.e. a site canonization must be performed before calling `truncate!`.
+"""
+function truncate!(tn::AbstractAnsatz, bond; threshold=nothing, maxdim=nothing)


IMO we should have truncate only for Canonical or MixedCanonical now

mofeing · 2024-10-29T11:46:08Z

I prefer to have @test_skip instead of commenting all the tests.

That's a valid request and a good idea. Will change it now and recover the deleted tests as @test_skip... The only thing is I'm not sure if it will still try to evaluate the code, which will obviously crash because we're breaking the API. So maybe @test_broken but doesn't matter which.

mofeing added refactor Change internals triage Needs group consensus breaking-change labels Oct 26, 2024

mofeing requested review from a team, arturgs, jofrevalles and Todorbsc October 26, 2024 22:14

jofrevalles requested changes Oct 29, 2024

View reviewed changes

mofeing force-pushed the concretize-ansatz branch from 2742acd to 7a870b7 Compare October 29, 2024 09:32

mofeing added 19 commits October 29, 2024 10:35

Introduce Canonical Form trait

0db0a74

Refactor Ansatz into a concrete type

c8ef1ce

Fix typos in Ansatz

0305aee

Use Graphs.neighbors

6ff6b23

Implement copy, similar, zero for Ansatz

4d101bf

Relax sites condition on Ansatz construction for lanes

22b719d

Remove some exports

d068413

Move Chain code to AbstractAnsatz and MPS

d0fbfa6

Force Graphs to be a strong dependency

1d0e0b7

Implement Graphs.neighbors, isneighbor methods

53115bc

Refactor adapt_structure method to support additional types

6a388d4

Refactor Reactant.make_tracer, Reactant.create_result methods on …

03cb5f2

…top of recent changes

Refactor ChainRules methods on top of new types

7fa4004

Refactor ProjectTo for Ansatz

62d0156

Export canonize_site, canonize_site! methods

75df367

Add Graphs, MetaGraphsNext as test dependencies

0fd6cad

Fix tensors(; bond)

cd06fa2

Fix truncate!

52a6fb6

Fix truncate! extension when using threshold

e82c6cd

mofeing and others added 10 commits October 29, 2024 10:36

Format code

7d1fa16

Comment renormalize kwarg of evolve!

fcef0ec

Fix simple_update! on single site gates

1653ea1

Fix indexing problems in simple_update_1site!

ef4f029

Some fixes on simple_update!

ac5cea3

Prototype tests for Ansatz

e739d07

Fix reference to lattice in adapt_structure for Ansatz

ca5ad64

Stop using IdDict on Reactant extension

9e32ffd

Set temporarily a more concrete type of lattice in graph to circunv…

7c67f58

…ent a Julia bug

Delete Chain, Dense, Grid, Product types

0bef4a9

mofeing force-pushed the concretize-ansatz branch from 7a870b7 to 0bef4a9 Compare October 29, 2024 09:36

Revert "Set temporarily a more concrete type of lattice in graph to…

87bbbb0

… circunvent a Julia bug" This reverts commit 7c67f58.

jofrevalles requested changes Oct 29, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Refactor `Ansatz` to be a concrete type with lattice information #223

Refactor `Ansatz` to be a concrete type with lattice information #223

mofeing commented Oct 26, 2024 •

edited

Loading

jofrevalles Oct 29, 2024

mofeing Oct 29, 2024

jofrevalles commented Oct 29, 2024

mofeing commented Oct 29, 2024 •

edited

Loading

jofrevalles left a comment

jofrevalles Oct 29, 2024

jofrevalles Oct 29, 2024

jofrevalles Oct 29, 2024

mofeing Oct 29, 2024

jofrevalles Oct 29, 2024

mofeing commented Oct 29, 2024

	# @testset "GHZ Circuit" begin
	@test_skip begin #@testset "GHZ Circuit"

	# @testset "Product ansatz" begin
	@test_skip begin # @testset "Product ansatz"

Refactor Ansatz to be a concrete type with lattice information #223

Are you sure you want to change the base?

Refactor Ansatz to be a concrete type with lattice information #223

Conversation

mofeing commented Oct 26, 2024 • edited Loading

Changelog

Notes

jofrevalles Oct 29, 2024

Choose a reason for hiding this comment

mofeing Oct 29, 2024

Choose a reason for hiding this comment

jofrevalles commented Oct 29, 2024

mofeing commented Oct 29, 2024 • edited Loading

jofrevalles left a comment

Choose a reason for hiding this comment

jofrevalles Oct 29, 2024

Choose a reason for hiding this comment

jofrevalles Oct 29, 2024

Choose a reason for hiding this comment

jofrevalles Oct 29, 2024

Choose a reason for hiding this comment

mofeing Oct 29, 2024

Choose a reason for hiding this comment

jofrevalles Oct 29, 2024

Choose a reason for hiding this comment

mofeing commented Oct 29, 2024

Refactor `Ansatz` to be a concrete type with lattice information #223

Refactor `Ansatz` to be a concrete type with lattice information #223

mofeing commented Oct 26, 2024 •

edited

Loading

mofeing commented Oct 29, 2024 •

edited

Loading