SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2607.01408 · 2026

Spin-Weighted Spherical Harmonics Enable Complete and Scalable E(3)-Equivariant Networks

Shuiwang Ji, Yuchao Lin, Jianwen Xie, Xiaofeng Qian, Chuan Li, Wendi Yu, Chenxing Liang, Andrii Kryvenko

arXiv · PDF · Open in the Atlas

Code that ran

8 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 5 failures of our instrument; 2 no recorded run), measured 2026-11-22; 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/2607.01408.json · badge: /api/ran/2607.01408.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
divelab/SpinGTP canonical (classifier label)
classifier basis: census
8 ran · 2 no recorded run
FunctionStatusWhere it lives
FFT_batch_channel Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
divelab/SpinGTP/spingtp/spingtp/modules/efficient_utils.py
code served (permissive licence) · get_code("ca82bd4ded61acb5")
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, 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 Fourier coefficient tensors
# Shape: (B, C, spatial_dim, spatial_dim)
# Using B=2, C=3 for batch and channel dimensions
# Using spatial dims of 4 and 3 to make them distinct

fourier_coef1 = torch.randn(2, 3, 4, 4, dtype=torch.complex64)
fourier_coef2 = torch.randn(2, 3, 3, 3, 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 4, dtype family float, shape (2, 3, 6, 6) — shape (2, 3, 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
cart_to_l1_swsh Ran
ran, violated the contract we drafted
divelab/SpinGTP/spingtp/spingtp/modules/swsh_spin_head.py
code served (permissive licence) · get_code("72eaeb235460632c")
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, 4, 3) 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)

# Create a small tensor with shape (..., 3) where the last dimension has 3 components [x, y, z]
# Using batch size 2 and an extra dimension to make it (..., 3)
v = torch.randn(2, 4, 3)

ARGS = {
    "v": v
}
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, 4, 3) — shape (2, 4, 3) 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
compute_per_edge_frame Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
divelab/SpinGTP/spingtp/spingtp/modules/swsh_reference.py
code served (permissive licence) · get_code("822b72a8e99b7a55")
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 tuple. 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 warnings

# Set seed for reproducibility
torch.manual_seed(0)

# Helper function from the code
def _edge_frame_from_per_edge_reference(edge_vec, frame):
    """Stand-in implementation to avoid import errors"""
    # This function is called but not defined in the snippet
    # We need to provide a minimal implementation that returns expected types
    batch_size = edge_vec.shape[0]
    F_e = frame  # Return frame as-is
    good_proj = torch.ones(batch_size, dtype=torch.bool)
    return (F_e, good_proj)

# Build test arguments for different modes
# We'll create arguments that work for mode='global' as a baseline

# Number of edges and nodes
num_edges = 2
num_nodes = 3
num_globals = 1

# edge_vec: (num_edges, 3) - edge direction vectors
edge_vec = torch.randn(num_edges, 3)

# frame: For mode='edge', should be (num_edges, 3, 3)
# For other modes, it's gathered by index, so shape depends on what we're gathering
# For mode='edge_from_node' or 'edge_from_global', frame is per-node or per-global
frame_edge = torch.randn(num_edges, 3, 3)
frame_node = torch.randn(num_nodes, 3, 3)
frame_global = torch.randn(num_globals, 3, 3)

# frame_good: Optional boolean tensor, same first dimension as frame
frame_good_edge = torch.ones(num_edges, dtype=torch.bool)
frame_good_node = torch.ones(num_nodes, dtype=torch.bool)
frame_good_global = torch.ones(num_globals, dtype=torch.bool)

# edge_dst: indices into node dimension for edge_from_node mode
edge_dst = torch.tensor([0, 1], dtype=torch.long)

# edge_batch: indices into global dimension for edge_from_global or global mode
edge_batch = torch.tensor([0, 0], dtype=torch.long)

# Test case 1: mode='edge'
ARGS_edge = {
    'mode': 'edge',
    'edge_vec': edge_vec,
    'frame': frame_edge,
    'frame_good': frame_good_edge,
    'edge_dst': None,
    'edge_batch': None,
    'already_z_aligned': True,
}

# Test case 2: mode='edge_from_node'
ARGS_edge_from_node = {
    'mode': 'edge_from_node',
    'edge_vec': edge_vec,
    'frame': frame_node,
    'frame_good': frame_good_node,
    'edge_dst': edge_dst,
    'edge_batch': None,
    'already_z_aligned': True,
}

# Test case 3: mode='edge_from_global'
ARGS_edge_from_global = {
    'mode': 'edge_from_global',
    'edge_vec': edge_vec,
    'frame': frame_global,
    'frame_good': frame_good_global,
    'edge_dst': None,
    'edge_batch': edge_batch,
    'already_z_aligned': True,
}

# Test case 4: mode='global'
ARGS_global = {
    'mode': 'global',
    'edge_vec': edge_vec,
    'frame': frame_global,
    'frame_good': frame_good_global,
    'edge_dst': None,
    'edge_batch': edge_batch,
    'already_z_aligned': True,
}

# Use the 'global' mode as the primary test case
ARGS = ARGS_global
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 tuple — a tuple output has no array shape of its own; none is 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
edge_frame_from_graph_frame Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
divelab/SpinGTP/spingtp/spingtp/modules/swsh.py
code served (permissive licence) · get_code("5e2f7932895ca997")
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 tuple. 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

torch.manual_seed(0)

# Build the arguments for edge_frame_from_graph_frame
# Parameters: edge_vec, graph_frame, edge_batch, eps

# edge_vec: tensor of edge direction vectors
# Shape: (num_edges, 3) - vectors in 3D space
edge_vec = torch.randn(5, 3)

# graph_frame: reference frames for each graph node
# Shape: (num_nodes, 3, 3) - 3x3 rotation matrices for each node
# We need 3 nodes for this example
graph_frame = torch.randn(3, 3, 3)

# edge_batch: indices mapping edges to their parent graph nodes
# Shape: (num_edges,) - each edge points to a node index
# Values should be in range [0, num_nodes)
edge_batch = torch.tensor([0, 1, 2, 0, 1], dtype=torch.long)

# eps: small numerical constant for stability
eps = 1e-12

ARGS = {
    "edge_vec": edge_vec,
    "graph_frame": graph_frame,
    "edge_batch": edge_batch,
    "eps": eps,
}
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 tuple — a tuple output has no array shape of its own; none is 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
edge_frame_from_node_frame Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
divelab/SpinGTP/spingtp/spingtp/modules/swsh.py
code served (permissive licence) · get_code("1fb9d5972ac903f4")
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 tuple. 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
torch.manual_seed(0)

def safe_normalize(v: torch.Tensor, eps: float=1e-12):
    n = v.norm(dim=-1, keepdim=True)
    good = n[..., 0] > eps
    vhat = v / n.clamp(min=eps)
    return (torch.where(good[..., None], vhat, torch.zeros_like(vhat)), good)

def _edge_frame_from_per_edge_reference(edge_vec: torch.Tensor, ref_frame_per_edge: torch.Tensor, eps: float=1e-12):
    """Re-project per-edge reference frames so z is the edge direction."""
    z, good_z = safe_normalize(edge_vec, eps)
    a = ref_frame_per_edge[..., :, 0]
    x = a - (a * z).sum(-1, keepdim=True) * z
    b = ref_frame_per_edge[..., :, 1]
    x_alt = b - (b * z).sum(-1, keepdim=True) * z
    bad = x.norm(dim=-1, keepdim=True) < eps
    x = torch.where(bad, x_alt, x)
    x, good_x = safe_normalize(x, eps)
    y, good_y = safe_normalize(torch.cross(z, x, dim=-1), eps)
    x, good_x2 = safe_normalize(torch.cross(y, z, dim=-1), eps)
    F = torch.stack([x, y, z], dim=-1)
    good = good_z & good_x & good_y & good_x2
    return (F, good)

# Create small tensors with distinct sizes for different dimensions
# edge_vec: shape (num_edges, 3) - direction vectors for edges
num_edges = 2
edge_vec = torch.randn(num_edges, 3)

# node_frame: shape (num_nodes, 3, 3) - frame matrices at each node
# We need 3 columns for the reference frame
num_nodes = 4
node_frame = torch.randn(num_nodes, 3, 3)

# edge_dst: shape (num_edges,) - indices into node_frame, values in [0, num_nodes)
edge_dst = torch.tensor([0, 2], dtype=torch.long)

# eps parameter
eps = 1e-12

ARGS = {
    "edge_vec": edge_vec,
    "node_frame": node_frame,
    "edge_dst": edge_dst,
    "eps": eps,
}
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 tuple — a tuple output has no array shape of its own; none is 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
evaluate_in_frame_for_mode Ran
ran, violated the contract we drafted
divelab/SpinGTP/spingtp/spingtp/modules/swsh_reference.py
code served (permissive licence) · get_code("b18892cd97bcbe1e")
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

ARGS = {
    'mode': 'global'
}
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 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
l1_swsh_to_cart Ran
ran, honoured the contract we drafted
divelab/SpinGTP/spingtp/spingtp/modules/swsh_spin_head.py
code served (permissive licence) · get_code("2f6ba43cdb7d8232")
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 (2, 3) 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)

# Create a tensor with shape [batch, dim] where dim=4
# The function indexes into the last dimension with [2, 0, 1]
# So we need at least 3 elements in the last dimension, we'll use 4
q = torch.randn(2, 4)

ARGS = {
    "q": q
}
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 (2, 3) — shape (2, 3) 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
safe_normalize Ran
ran; our contract draft was wrong, not the code
divelab/SpinGTP/spingtp/spingtp/modules/swsh.py
code served (permissive licence) · get_code("ef4b3563a85b0412")
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 tuple. 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

torch.manual_seed(0)

# The function takes two parameters:
# 1. v: torch.Tensor - a tensor that will be normalized
# 2. eps: float - epsilon value for numerical stability (default 1e-12)

# Create a small tensor with shape (2, 4) - batch size 2, dimension 4
v = torch.randn(2, 4)

# eps parameter - use default value
eps = 1e-12

ARGS = {
    "v": v,
    "eps": eps,
}
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 tuple — a tuple output has no array shape of its own; none is 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
f2sh_batch_channel Not yet run
unverified (harvested, no recorded run)
divelab/SpinGTP/spingtp/spingtp/modules/efficient_utils.py
code served (permissive licence) · get_code("400e038d70d314e2")
How it ran

no invocation recorded.

sh2f_batch_channel Not yet run
unverified (harvested, no recorded run)
divelab/SpinGTP/spingtp/spingtp/modules/efficient_utils.py
code served (permissive licence) · get_code("f631159ecaae3519")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 0 later) and 7 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 dated earlier
2401.10216 (2024-01)
Enabling Efficient Equivariant Operations in the Fourier Basis via Gaunt Tensor Products
f2sh_batch_channel 1 other paper
1 dated earlier
2401.10216 (2024-01)
Enabling Efficient Equivariant Operations in the Fourier Basis via Gaunt Tensor Products
sh2f_batch_channel 1 other paper
1 dated earlier
2401.10216 (2024-01)
Enabling Efficient Equivariant Operations in the Fourier Basis via Gaunt Tensor Products

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

E(3)-equivariant networks are promising for 3D atomistic system modeling, yet their scalability is limited by the O(L 6 ) complexity of the Clebsch-Gordan Tensor Product (CGTP). The recently proposed Gaunt Tensor Product (GTP) reduces the complexity but is unable to capture the antisymmetric paths, resulting in incomplete expressivity. In this work, we present SpinGTP, an approach to overcome the GTP incompleteness by generalizing from scalar functions to Spin-Weighted Spherical Harmonics (SWSH). By relying on the algebraic properties of SWSH, SpinGTP recovers the missing antisymmetric interactions while maintaining the asymptotic efficiency of GTP. It also allows for a more expressive equivariant basis that naturally accounts for the parity-odd components of tensor products. We evaluate SpinGTP across diverse benchmarks, including Tetris, 3BPA, SPICE-MACE-OFF, and OC20. Our results show that SpinGTP achieves accuracies comparable to full CGTP. Notably, by explicitly capturing antisymmetric paths, SpinGTP exhibits superior performance in tasks involving chiral materials and non-centrosymmetric geometries. This work provides a complete, scalable, and mathematically rigorous path toward high-order equivariance in large-scale 3D atomistic system simulations. Our code is publicly available at https://github.com/divelab/SpinGTP/.

For agents

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

get_harvested_code_for_paper("2607.01408")
get_code_for_paper("2607.01408")
have("2607.01408")

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

curl https://syntology.ai/api/ran/2607.01408.json

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

[![Syntology run record](https://syntology.ai/api/ran/2607.01408.svg)](https://syntology.ai/paper/2607.01408)

Connect an agent — have() is free.