SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.10216 · ICLR · 2024

Enabling Efficient Equivariant Operations in the Fourier Basis via Gaunt Tensor Products

Shengjie Luo, Aditi Krishnapriyan, Tianlang Chen

arXiv · PDF · Open in the Atlas

Code that ran

20 of 24 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 6 violated it and 2 have no contract verdict; 11 failures of our instrument; 4 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2401.10216.json · badge: /api/ran/2401.10216.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
lsj2408/Gaunt-Tensor-Product canonical (classifier label)
classifier basis: census
13 ran · 4 no recorded run
e3nn/e3nn canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
lsj2408/gaunt-tensor-product community
Papers with Code archive
6 ran · 0 no recorded run
FunctionStatusWhere it lives
ClebschGordan Ran
ran on a synthesized input · constructed an object; no result was computed
lsj2408/gaunt-tensor-product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("a0b88a548830cf98")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, ClebschGordan: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from sympy.physics.wigner import clebsch_gordan
from sympy import S

ARGS = {}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
sympy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ClebschGordan — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ClebschGordan output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
FFT Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/fft.py
code served (permissive licence) · get_code("05fbb48f432304eb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (6, 6) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.fft as fft

torch.manual_seed(0)

# Create two small 2D coefficient tensors
# The function expects square 2D tensors (matrices)
fourier_coef1 = torch.randn(3, 3, dtype=torch.complex64)
fourier_coef2 = torch.randn(4, 4, dtype=torch.complex64)

ARGS = {
    "fourier_coef1": fourier_coef1,
    "fourier_coef2": fourier_coef2,
    "return_real": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (6, 6) — shape (6, 6) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
FFT_batch_channel Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/fft.py
code served (permissive licence) · get_code("ce1ffb6956fa7c7b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 8, 8) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.fft as fft

torch.manual_seed(0)

# Create two small tensors with batch and channel dimensions
# fourier_coef1: shape (B=2, C=3, spatial=4, spatial=4)
fourier_coef1 = torch.randn(2, 3, 4, 4, dtype=torch.complex64)

# fourier_coef2: shape (B=2, C=3, spatial=5, spatial=5) - different spatial size
fourier_coef2 = torch.randn(2, 3, 5, 5, dtype=torch.complex64)

# return_real: boolean flag
return_real = False

ARGS = {
    "fourier_coef1": fourier_coef1,
    "fourier_coef2": fourier_coef2,
    "return_real": return_real,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 8, 8) — shape (2, 3, 8, 8) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
FFT_channel Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/fft.py
code served (permissive licence) · get_code("f18e28a7c0cf9e04")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 3, dtype family float, shape (2, 6, 6) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.fft as fft

torch.manual_seed(0)

# Create small test tensors for fourier coefficients
# fourier_coef1 and fourier_coef2 should have shape (C, H, W)
# where C is number of channels, H and W are spatial dimensions

C = 2  # number of channels
H1 = 3  # height/width for first coefficient
H2 = 4  # height/width for second coefficient (different from H1 for distinctness)

fourier_coef1 = torch.randn(C, H1, H1, dtype=torch.complex64)
fourier_coef2 = torch.randn(C, H2, H2, dtype=torch.complex64)

ARGS = {
    "fourier_coef1": fourier_coef1,
    "fourier_coef2": fourier_coef2,
    "return_real": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 6, 6) — shape (2, 6, 6) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Rot3DCoeffs Ran
ran on a synthesized input · constructed an object; no result was computed
lsj2408/gaunt-tensor-product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("5a51dad6fe773f9c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Rot3DCoeffs: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math

torch.manual_seed(0)

ARGS = {
    "max_N": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
sympy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Rot3DCoeffs — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Rot3DCoeffs output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
cg_conditions Ran
ran, violated the contract we drafted
lsj2408/gaunt-tensor-product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("e8488725c2821c4e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# These are Clebsch-Gordan coefficient conditions
# The function checks angular momentum coupling rules
# Parameters are angular momentum quantum numbers (j, m, J, M)

# Valid angular momentum quantum numbers: can be integers or half-integers
# For simplicity, using small integer values

j1 = 1  # first angular momentum quantum number
m1 = 0  # first magnetic quantum number (must satisfy |m1| <= j1)
j2 = 1  # second angular momentum quantum number
m2 = 0  # second magnetic quantum number (must satisfy |m2| <= j2)
J = 1   # total angular momentum (must satisfy |j1 - j2| <= J <= j1 + j2)
M = 0   # total magnetic quantum number (must equal m1 + m2)

ARGS = {
    "j1": j1,
    "m1": m1,
    "j2": j2,
    "m2": m2,
    "J": J,
    "M": M,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
cg_l_condition Ran
ran, violated the contract we drafted
lsj2408/gaunt-tensor-product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("1d1c344056d67391")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function cg_l_condition takes three parameters: j1, j2, and J
# It checks if the triangle inequality condition holds for Clebsch-Gordan coefficients
# All three parameters should be numeric values (int or float)

j1 = 1
j2 = 2
J = 3

ARGS = {
    "j1": j1,
    "j2": j2,
    "J": J
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
cg_m_condition Ran
ran, violated the contract we drafted
lsj2408/gaunt-tensor-product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("c6a6415c93989edb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 1, dtype family float, shape (4,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function checks if M == m1 + m2
# m1, m2, and M should be values that support the + operator and == comparison
# They could be scalars, tensors, numpy arrays, etc.

# Using small tensors for clarity
m1 = torch.tensor([1, 2, 3, 4], dtype=torch.float32)
m2 = torch.tensor([5, 6, 7, 8], dtype=torch.float32)
M = m1 + m2

ARGS = {
    "m1": m1,
    "m2": m2,
    "M": M,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (4,) — shape (4,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
clebschgordan Ran
ran, honoured the contract we drafted
lsj2408/gaunt-tensor-product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("2ceb0025f1a3db12")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from fractions import Fraction

torch.manual_seed(0)

# The clebschgordan function takes 6 parameters: j1, m1, j2, m2, J, M
# All are angular momentum quantum numbers (can be integers or half-integers)
# The function checks conditions and returns a float Clebsch-Gordan coefficient

# Valid quantum numbers must satisfy:
# - |j1 - j2| <= J <= j1 + j2
# - M = m1 + m2
# - |m1| <= j1, |m2| <= j2, |M| <= J

# Using simple integer quantum numbers for testing
j1 = 1
m1 = 0
j2 = 1
m2 = 1
J = 2
M = 1

# Verify conditions:
# |1 - 1| <= 2 <= 1 + 1  => 0 <= 2 <= 2 ✓
# 1 = 0 + 1 ✓
# |0| <= 1, |1| <= 1, |1| <= 2 ✓

ARGS = {
    "j1": j1,
    "m1": m1,
    "j2": j2,
    "m2": m2,
    "J": J,
    "M": M,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sympy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
coco_filter Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("ad67339b74a878f4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Analyzing the function:
# - ll: used in sum(ll), so needs to be iterable of numbers
# - mm: used in sum(mm), so needs to be iterable of numbers
# - invariant: boolean flag
# The function checks if sum(ll) is even and sum(mm) equals 0

# Create valid arguments
ll = [2, 4, 6]  # sum is 12, which is even
mm = [0, 0, 0]  # sum is 0

ARGS = {
    "ll": ll,
    "mm": mm,
    "invariant": True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
coco_filter2 Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot.py
code served (permissive licence) · get_code("629660a18e55d63e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Construct arguments for coco_filter2
# The function expects:
# - ll: an iterable (list) of numbers that sum to an even number
# - mm: an iterable (list) of numbers that sum to 0
# - kk: an iterable (list) of numbers that sum to 0
# - invariant: boolean flag (default True)

ll = [2, 4, 6]  # sum is 12, which is even
mm = [1, -1, 0]  # sum is 0
kk = [5, -5]  # sum is 0
invariant = True

ARGS = {
    "ll": ll,
    "mm": mm,
    "kk": kk,
    "invariant": invariant
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_Jx Ran
ran, violated the contract we drafted
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/LieCG/Lie_groups/Rotations.py
code served (permissive licence) · get_code("662f5627f0873e13")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (5, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "j": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (5, 5) — shape (5, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_Jy Ran
ran, violated the contract we drafted
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/LieCG/Lie_groups/Rotations.py
code served (permissive licence) · get_code("67bb14caad016db4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (5, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function create_Jy takes a single parameter 'j' which is used to:
# 1. Create a range from -j to j
# 2. Compute diagonal elements based on j
# 3. Create diagonal matrices
# The parameter j should be a numeric value (int or float) representing
# a quantum angular momentum quantum number

j = 2

ARGS = {
    "j": j
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (5, 5) — shape (5, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
f2sh Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/f2sh.py
code served (permissive licence) · get_code("74dca1841d91deed")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (3, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Determine L from the function's logic
# fourier_coef has shape (2L-1, 2L-1)
# f2sh_bases has shape (L, 2L-1, 2L-1, 2)
# Let's use L=3, so 2L-1 = 5

L = 3
fourier_size = 2 * L - 1  # 5

# Create fourier_coef with shape (2L-1, 2L-1) = (5, 5)
fourier_coef = torch.randn(fourier_size, fourier_size)

# Create f2sh_bases with shape (L, 2L-1, 2L-1, 2) = (3, 5, 5, 2)
f2sh_bases = torch.randn(L, fourier_size, fourier_size, 2)

ARGS = {
    "fourier_coef": fourier_coef,
    "f2sh_bases": f2sh_bases,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (3, 5) — shape (3, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
f2sh_batch_channel Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/f2sh.py
code served (permissive licence) · get_code("055b503cdc6b0c1b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 4, 7) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Set up dimensions
B = 2  # batch size
C = 3  # channels
L = 4  # spherical harmonic band limit

# fourier_coef: shape (B, C, 2L-1, 2L-1)
fourier_coef = torch.randn(B, C, 2*L-1, 2*L-1)

# f2sh_bases: shape (L, 2L-1, 2L-1, 2)
# The function accesses f2sh_bases[:, :, :, 0] and f2sh_bases[:, :, :, 1]
# which are the positive and negative components
f2sh_bases = torch.randn(L, 2*L-1, 2*L-1, 2)

ARGS = {
    "fourier_coef": fourier_coef,
    "f2sh_bases": f2sh_bases,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 4, 7) — shape (2, 3, 4, 7) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
f2sh_channel Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/f2sh.py
code served (permissive licence) · get_code("8c4389223f782822")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Parameters based on function signature and usage:
# fourier_coef: shape (C, 2L-1, 2L-1) - coefficients of 2D Fourier bases
# f2sh_bases: shape (L, 2L-1, 2L-1, 2) - precomputed spherical harmonics coefficients

# Let's choose L=3, C=2 for small tensors
# Then:
# fourier_coef shape: (2, 5, 5)  where 2L-1 = 5
# f2sh_bases shape: (3, 5, 5, 2)  where L=3, 2L-1=5

L = 3
C = 2

# Create fourier_coef with shape (C, 2L-1, 2L-1) = (2, 5, 5)
fourier_coef = torch.randn(C, 2*L-1, 2*L-1)

# Create f2sh_bases with shape (L, 2L-1, 2L-1, 2) = (3, 5, 5, 2)
f2sh_bases = torch.randn(L, 2*L-1, 2*L-1, 2)

ARGS = {
    "fourier_coef": fourier_coef,
    "f2sh_bases": f2sh_bases,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 5) — shape (2, 3, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sh2f Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/sh2f.py
code served (permissive licence) · get_code("27dc61986a394a2a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (5, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Determine L from the expected shapes
# sh_coeff shape: (L, 2L-1)
# sh2f_bases shape: (L, 2L-1, 2L-1, 2)
# res shape: (2L-1, 2L-1)

# Let's use L=3 as a small example
L = 3
height = 2 * L - 1  # 5

# sh_coeff: shape (L, 2L-1) = (3, 5)
sh_coeff = torch.randn(L, height)

# sh2f_bases: shape (L, 2L-1, 2L-1, 2) = (3, 5, 5, 2)
sh2f_bases = torch.randn(L, height, height, 2)

ARGS = {
    "sh_coeff": sh_coeff,
    "sh2f_bases": sh2f_bases,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (5, 5) — shape (5, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sh2f_batch_channel Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/sh2f.py
code served (permissive licence) · get_code("31a695005a03793b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 7, 7) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create small tensors that satisfy the function's requirements
# sh_coeff: shape (B, C, L, 2L-1)
# sh2f_bases: shape (L, 2L-1, 2L-1, 2)

B = 2  # batch size
C = 3  # channels
L = 4  # spherical harmonics degree

# sh_coeff shape: (B, C, L, 2L-1) = (2, 3, 4, 7)
sh_coeff = torch.randn(B, C, L, 2*L - 1)

# sh2f_bases shape: (L, 2L-1, 2L-1, 2) = (4, 7, 7, 2)
sh2f_bases = torch.randn(L, 2*L - 1, 2*L - 1, 2)

ARGS = {
    "sh_coeff": sh_coeff,
    "sh2f_bases": sh2f_bases,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 7, 7) — shape (2, 3, 7, 7) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sh2f_channel Ran
ran; our contract draft was wrong, not the code
lsj2408/Gaunt-Tensor-Product/efficiency-comparison/sh2f.py
code served (permissive licence) · get_code("d382cba70545c430")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 3, dtype family float, shape (2, 7, 7) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Set up parameters for sh2f_channel function
# sh_coeff: shape (C, L, 2L-1) where C is channels, L is spherical harmonics degree
# sh2f_bases: shape (L, 2L-1, 2L-1, 2)

# Let's use L=4 for reasonable dimensions
L = 4
C = 2  # number of channels

# sh_coeff shape: (C, L, 2L-1) = (2, 4, 7)
sh_coeff = torch.randn(C, L, 2*L - 1)

# sh2f_bases shape: (L, 2L-1, 2L-1, 2) = (4, 7, 7, 2)
sh2f_bases = torch.randn(L, 2*L - 1, 2*L - 1, 2)

ARGS = {
    "sh_coeff": sh_coeff,
    "sh2f_bases": sh2f_bases
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 7, 7) — shape (2, 7, 7) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
t_or_f Ran
ran, violated the contract we drafted
e3nn/e3nn/examples/tensor_product_benchmark.py
pointer only (licence: NOASSERTION) · get_code("440a098dda4f24f0")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, e3nn/e3nn (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

U_matrix_real Not yet run
unverified (harvested, no recorded run)
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/LieCG/CG_coefficients/CG_rot_real.py
code served (permissive licence) · get_code("cee2ed5ee0d679ee")
How it ran

no invocation recorded.

create_J Not yet run
unverified (harvested, no recorded run)
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/LieCG/Lie_groups/Rotations.py
code served (permissive licence) · get_code("ae70e599fd8a7ff6")
How it ran

no invocation recorded.

linear_out_irreps Not yet run
unverified (harvested, no recorded run)
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/MACE/modules/irreps_tools.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from ACEsuit/mace, is pointer only (licence: NOASSERTION) · get_code("3ded1daa0770bc97")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from ACEsuit/mace, and GitHub could not classify the licence of that repository (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

tp_out_irreps_with_instructions Not yet run
unverified (harvested, no recorded run)
lsj2408/Gaunt-Tensor-Product/force-field-modeling-3bpa/MACE/MACE/modules/irreps_tools.py
code served (permissive licence) · get_code("18d9a7b0f17155e6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 24 samples lifted from this paper's linked repositories, 24 carry a function identity and 0 do not. Of the 24 distinct functions with an identity, 5 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 4 later) and 19 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
FFT_batch_channel 1 other paper
1 later
2607.01408 (2026-07)
Spin-Weighted Spherical Harmonics Enable Complete and Scalable E(3)-Equivariant Networks
f2sh_batch_channel 1 other paper
1 later
2607.01408 (2026-07)
Spin-Weighted Spherical Harmonics Enable Complete and Scalable E(3)-Equivariant Networks
linear_out_irreps 1 other paper
1 later
2412.11569 (2024-12)
The dark side of the forces: assessing non-conservative force models for atomistic machine learning
sh2f_batch_channel 1 other paper
1 later
2607.01408 (2026-07)
Spin-Weighted Spherical Harmonics Enable Complete and Scalable E(3)-Equivariant Networks
t_or_f 1 other paper
1 dated earlier
2207.09453 (2022-07)
e3nn: Euclidean Neural Networks

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Developing equivariant neural networks for the E(3) group plays an important role in modeling 3D data across real-world applications. Enforcing this equivariance primarily involves the tensor products of irreducible representations (irreps). However, the computational complexity of such operations increases significantly as higher-order tensors are used. In this work, we propose a systematic approach to substantially accelerate the computation of the tensor products of irreps. We mathematically connect the commonly used Clebsch-Gordan coefficients to the Gaunt coefficients, which are integrals of products of three spherical harmonics. Through Gaunt coefficients, the tensor product of irreps becomes equivalent to the multiplication between spherical functions represented by spherical harmonics. This perspective further allows us to change the basis for the equivariant operations from spherical harmonics to a 2D Fourier basis. Consequently, the multiplication between spherical functions represented by a 2D Fourier basis can be efficiently computed via the convolution theorem and Fast Fourier Transforms. This transformation reduces the complexity of full tensor products of irreps from O(L 6 ) to O(L 3 ), where L is the max degree of irreps. Leveraging this approach, we introduce the Gaunt Tensor Product, which serves as a new method to construct efficient equivariant operations across different model architectures. Our experiments on the Open Catalyst Project and 3BPA datasets demonstrate both the increased efficiency and improved performance of our approach.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2401.10216")
get_code_for_paper("2401.10216")
have("2401.10216")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2401.10216.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2401.10216.svg)](https://syntology.ai/paper/2401.10216)

Connect an agent — have() is free.