Thermal TNR#172
Conversation
|
As a Physics based test. I also reproduced the diagrams 6(a) and 6(d) (roughly for smaller bond dimensions) from https://arxiv.org/pdf/2602.13124. The way to obtain them is to perform ThermalTNR on the initial PEPO/TNO of the Z2 and Z3 gauge theory tensors that represent the partition function for Lz -1 number of iterations. Then trace out the physical legs and perform LoopTNR to obtain the Universal data. |
Codecov Report❌ Patch coverage is
🚀 New features to boost your workflow:
|
Codecov Report❌ Patch coverage is
🚀 New features to boost your workflow:
|
VictorVanthilt
left a comment
VictorVanthilt
left a comment
There was a problem hiding this comment.
Some minor comments about a docstring and the exports.
Some major comments about the tests. Please write new tests that test the validity of the scheme.
| """ | ||
| $(TYPEDEF) | ||
|
|
||
| Minimal storage object for thermal tensor network renormalization on a square-lattice |
There was a problem hiding this comment.
This line is a stylistic break from all the other scheme's docstrings. Please fix
| export TNO, TNOTensor | ||
| export ThermalTNR, apply! |
There was a problem hiding this comment.
Rethink exports. Does the user need apply!, TNO, TNOTensor ?
| @@ -0,0 +1,74 @@ | |||
| @testset "TNO" begin | |||
There was a problem hiding this comment.
This AI slop test needs to be removed
|
|
||
| scheme_top = ThermalTNR(TNO(local_tensor; unitcell = (1, 1))) | ||
| scheme_bottom = ThermalTNR(TNO(local_tensor; unitcell = (1, 1))) | ||
| @test apply!(scheme_top, scheme_bottom, truncrank(8)) isa ThermalTNR |
| @test apply!(scheme_top, scheme_bottom, truncrank(8)) isa ThermalTNR | ||
| end | ||
|
|
||
| @testset "ThermalTNR finalize!" begin |
| @test all(norm(@tensor scheme.T[i, j][1 1; 2 3 2 3]) ≈ 1 for i in 1:2, j in 1:2) | ||
| end | ||
|
|
||
| @testset "ThermalTNR run!" begin |
There was a problem hiding this comment.
this doesn't really test the validity of the scheme. Please write new tests that test the scheme on the validity of its outcomes.
2 participants
This PR adds Thermal Tensor Network Renormalization (ThermalTNR) to TNRKit, following Ueda et al. (2025).
It introduces a workflow for coarse-graining in the z-direction, centered TNO container and a custom ThermalTNR scheme. Two operator layers are composed sitewise, with virtual-bond truncation performed using the oblique projector method.
What’s included
TNO: a matrix-like container for local tensor-network-operator tensors (basically PEPOs from PEPSKit)ThermalTNR <: TNRSchemeapply!(top, bottom, trunc)for composing two TNO/PEPO-like operator layersrun!methods for evolving a target operator with an auxiliary layerfinalize!(::ThermalTNR)support with geometric-mean normalization output. (product(norms)^(1/N))-tests for container behavior,
apply!,finalize!, andrun!Implementation notes
The local tensors are treated in PEPO convention, with two physical legs and four virtual legs.
TNO composition contracts the physical output/input legs between layers.
The merged virtual bonds are compressed using oblique projectors derived from neighbouring tensors, following Wang-Verstraete.(2011).
References
Thermal TNR: https://arxiv.org/abs/2508.05406
Oblique projector method: https://arxiv.org/abs/1110.4362