Cluster expansions #47

gsommers · 2025-11-17T03:48:23Z

Added code for cluster expansion and cluster cumulant expansion. Calculation of one- and two-point functions. Example code in examples/clustercorrections.jl.

Note: generalized loop finding in cluster expansion relies on nauty which must be downloaded here.

Other changes:

Renamed freenergy to free_energy.
message_diff no longer returns negative numbers.
Added packages to Project.toml.

merge from hpc/clusters

JoeyT1994 · 2025-11-17T13:33:43Z

Project.toml

 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 Combinatorics = "861a8166-3701-5b0c-9a16-15d98fcdc6aa"
+DataGraphs = "b5a273c3-7e6c-41f6-98bd-8d7f1525a36a"


Is this dependency used? I removed anything requiring it a few version ago

No, it's not used, I'll remove it.

JoeyT1994 · 2025-11-17T13:34:02Z

Project.toml


 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"


This shouldn't be a hard dependency (requires a GPU)

JoeyT1994 · 2025-11-17T13:34:57Z

Project.toml

 ITensors = "9136182c-28ba-11e9-034c-db9fb085ebd5"
 KrylovKit = "0b1a1467-8014-51b9-945f-bf0ae24f4b77"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+MKL = "33e6dc65-8f57-5167-99aa-e5a354878fb2"


Suggested change

MKL = "33e6dc65-8f57-5167-99aa-e5a354878fb2"

JoeyT1994 · 2025-11-17T13:35:14Z

Project.toml

 KrylovKit = "0b1a1467-8014-51b9-945f-bf0ae24f4b77"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+MKL = "33e6dc65-8f57-5167-99aa-e5a354878fb2"
+NPZ = "15e1cf62-19b3-5cfa-8e77-841668bca605"


Suggested change

NPZ = "15e1cf62-19b3-5cfa-8e77-841668bca605"

JoeyT1994 · 2025-11-17T13:35:33Z

Project.toml

+NPZ = "15e1cf62-19b3-5cfa-8e77-841668bca605"
 NamedGraphs = "678767b0-92e7-4007-89e4-4527a8725b19"
 PauliPropagation = "293282d5-3c99-4fb6-92d0-fd3280a19750"
+ProgressMeter = "92933f4c-e287-5a05-a399-4b506db050ca"


Suggested change

ProgressMeter = "92933f4c-e287-5a05-a399-4b506db050ca"

JoeyT1994 · 2025-11-17T13:38:55Z

src/MessagePassing/abstractbeliefpropagationcache.jl

 function update(alg::Algorithm"bp", bpc::AbstractBeliefPropagationCache)
    compute_error = !isnothing(alg.kwargs.tolerance)
+    if !compute_error
+        println("ALERT, not computing error")


This forces tolerance to always be defined. But it maybe be the case the user just wants to run BP for a fixed number of iterations and doesn't want to know convergence info

It doesn't force it to be defined, just prints a message. But sure I can remove it.

JoeyT1994 · 2025-11-17T13:41:58Z

src/MessagePassing/loopcorrection.jl

 end

 #Get the all edges incident to the region specified by the vector of edges passed
+# Optionally pass in a vector of vertices in the region, to handle the corner case where the region is just one vertex. 


If you region is vertex based instead, this can be handled by boundary_edges(bpc::BeliefPropagationCache, verts::Vector) which should be an existing function (as there is a NamedGraphs version of it)

OK, fixed it

JoeyT1994 · 2025-11-18T20:37:05Z

src/MessagePassing/abstractbeliefpropagationcache.jl

-            end
-            break
+	if compute_error
+	    diffs[i] = diff.x


diffs[i] is not being used ?

It gets printed out in line 222. In a previous version I was returning it along with bpc, but that seemed like too much of a breaking change.

There is a print statement at the end of the function where the entire vector is printed. In an earlier version I also returned diffs, but that affects too many other functions so I opted not to.

JoeyT1994 · 2025-11-18T20:39:07Z

src/MessagePassing/clustercorrections.jl

+end
+
+struct Loop
+    vertices::Vector{Int}


This assumes the vertices are Int in the loop? But can't we label them arbitrarily like we do with a NamedGraph

In the cluster finding, we convert to a SimpleGraph first, which has integer vertices.

But I just changed to Vector anyway.

JoeyT1994 · 2025-11-18T20:41:22Z

src/MessagePassing/cumulant-clustercorrections.jl

+    is_proper_subset(a::RegionKey, b::RegionKey)::Bool
+Return true if region `a` is a proper subset of `b`.
+"""
+function is_proper_subset(a::RegionKey, b::RegionKey)::Bool


Can the Julia function issubset not handle this?

Good point. I just copied this from the code you shared with me before. Technically, issubset(a,b) also returns true if a==b, but I can just add that as a separate check.

Yeah could have

is_proper_subset(a::RegionKey, b::RegionKey) = issubset(a,b) && !issetequal(a,b)

JoeyT1994 · 2025-11-18T22:48:00Z

Thanks Grace! How much work is the hypernumbers doing here? Is it just a nice way to evaluate (d/depsilon)(<psi|(1+epsilon *O)|psi> with messages incident to the region of support, or is it actually doing work to automatically cancel out certain terms in the expansion?

gsommers · 2025-11-18T22:55:53Z

HyperDualNumbers is mainly just a nice way to take the derivative. Technically every term in the expansion could be written as a ratio (i.e. evaluating the derivative explicitly), which perhaps would be more efficient than using HyperDualNumbers.

JoeyT1994 · 2025-11-20T00:45:41Z

Yeah maybe it's best to drop the Hyperdualnumbers dependency then and go with the ratio approach? It's probably more efficient like you said and clearer from a code perspective (also removes a dependency)

JoeyT1994 · 2025-11-24T14:29:05Z

src/MessagePassing/clustercorrections.jl

+
+This is overkill as it finds ALL subgraphs first, but my other implementation had bugs
+"""
+function enumerate_clusters(ng::NamedGraph, max_weight::Int; min_v::Int = 4, triangle_free::Bool = true, must_contain = [], min_deg::Int = 2, verbose::Bool = false)


So this should definitely return the clusters with vertex labelling that ng has.

This is how I do it for edge_color (basically define the function on a SimpleGraph and then overload it on the NamedGraph by unwrapping the graph and re-wrapping the result:

function SimpleGraphAlgorithms.edge_color(g::AbstractNamedGraph, k::Int64) pg, vs = position_graph(g), collect(vertices(g)) ec_dict = SimpleGraphAlgorithms.edge_color(UndirectedGraph(pg), k) # returns k vectors of edges which each contain the colored/commuting edges return [ [edgetype(g)(vs[first(first(e))], vs[last(first(e))]) for e in ec_dict if last(e) == i] for i in 1:k ] end end

I think I already handle the vertex labeling in the function generalized_loop_named? Is there an issue with the existing approach?

gsommers added 30 commits October 28, 2025 16:40

local version PEPO

9ad09bf

minor changes test

27229fa

split main() into many functions, added bp

dd56cb2

expect_bmps

09db0c6

fixed typo expect_bmps

56db073

adding print statements

14bb36e

added time checks, use_gpu for expect_bmps

8ae6305

fixed small bugs

1f44ba2

merged from big fork

9b3c8d0

Project.toml

8576d83

merge main

06469eb

PEPO_2d_ising no more ITensorNetworks

29634a8

fixed dependencies in PEPO_2d_ising.jl

1608c31

fixed another bug

5a43d66

merged from local

bf71a37

cluster folder

7dfad00

started implementing expect-corrected

bcc34b7

fixed typos in compatibility

bf30dee

joey's version

bec5ebf

gpu for bp

aa7fdcb

time evolution kicked Ising

5a30bb9

linkdim is now virtualdim

2c99aa0

update Project, add error tracking

dd68ee8

use_gpu option for boundary mps

402a5b8

nMerge remote-tracking branch 'hpc/clusters' into clusters

76fe4b1

merge from hpc/clusters

tracking BP message diffs

2bd6ccb

coeffs and op_strings in free_energy

48ba539

norm_factors with kwargs

ef1d623

cluster expansions

7f73463

merge from origin

d7dc2a0

gsommers added 9 commits November 15, 2025 23:41

merged from clusters branch

24f81e5

bp message diffs

6354948

merge from laptop branch

29fd0f6

added comments

ad05ebe

added comments

93f6283

removed finite temp evolution

f6c4654

example with cluster expansions

36b51b8

cluster expansion examples

e6bd010

removed extraneous packages from Project

ce6c49e

JoeyT1994 reviewed Nov 18, 2025

View reviewed changes

gsommers added 5 commits November 18, 2025 20:48

cleaned up Project.toml

00370f5

changes from clusters branch

1c2f735

merge from clusters

26a1ec2

cluster cumulant seems buggygit st!

9fb5c49

maybe fixed?

0356915

fixed boundary_edges

d04aefa

JoeyT1994 reviewed Nov 24, 2025

View reviewed changes

Cluster expansions #47

Are you sure you want to change the base?

Cluster expansions #47

Uh oh!

Conversation

gsommers commented Nov 17, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

JoeyT1994 commented Nov 18, 2025

Uh oh!

gsommers commented Nov 18, 2025

Uh oh!

JoeyT1994 commented Nov 20, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

gsommers Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

gsommers Nov 24, 2025 •

edited

Loading