SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.17747 · 2024

Probing the effects of broken symmetries in machine learning

arXiv · PDF · Open in the Atlas

Code that ran

15 of 17 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; 4 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 11 failures of our instrument; 2 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/2406.17747.json · badge: /api/ran/2406.17747.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
spozdn/pet canonical (classifier label)
classifier basis: census
15 ran · 2 no recorded run
FunctionStatusWhere it lives
batch_to_dict Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/molecule.py
code served (permissive licence) · get_code("55931e9283fdb1b3")
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 dict. 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 numpy as np

torch.manual_seed(0)

class Batch:
    def __init__(self):
        self.x = torch.randn(2, 4)
        self.central_species = torch.randint(0, 5, (2,))
        self.neighbor_species = torch.randint(0, 5, (2, 3))
        self.mask = torch.ones(2, 3, dtype=torch.bool)
        self.batch = torch.tensor([0, 1])
        self.nums = torch.tensor([3, 3])
        self.neighbors_index = torch.randint(0, 2, (2, 3)).T
        self.neighbors_pos = torch.randn(2, 3, 4)
        self.target_id = torch.tensor([10, 20])
        self.neighbor_scalar_attributes = torch.randn(2, 3, 2)
        self.central_scalar_attributes = torch.randn(2, 2)
        self.k_vectors = torch.randn(2, 5, 3)
        self.k_mask = torch.ones(2, 5, dtype=torch.bool)
        self.positions = torch.randn(2, 3)
        self.volume = torch.tensor([100.0, 150.0])

batch = Batch()

ARGS = {
    'batch': batch
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
combine_hypers_shallow Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/hypers.py
code served (permissive licence) · get_code("4bc4baf4b5f5ccd8")
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 dict. 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)

# Construct test data for combine_hypers_shallow function
provided_hypers = {
    'learning_rate': 0.01,
    'batch_size': 32
}

default_hypers = {
    'learning_rate': 0.001,
    'batch_size': 16,
    'epochs': 100,
    'optimizer': 'adam'
}

duplicated_params = [
    ('lr', 'learning_rate'),
    ('batch', 'batch_size')
]

ARGS = {
    'provided_hypers': provided_hypers,
    'default_hypers': default_hypers,
    'duplicated_params': duplicated_params
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
convert_atomic_throughput Ran
ran, honoured the contract we drafted
spozdn/pet/src/analysis.py
code served (permissive licence) · get_code("4bec4c57a9f5a050")
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 int, ndim 0, dtype family int, 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 numpy as np
import math

class MockStructure:
    """Stand-in for atomic structure object with get_positions() method."""
    def __init__(self, num_atoms):
        self.num_atoms = num_atoms
    
    def get_positions(self):
        """Return positions array with shape (num_atoms, 3)."""
        return np.random.rand(self.num_atoms, 3)

# Create train_structures with distinct sizes
train_structures = [
    MockStructure(num_atoms=5),
    MockStructure(num_atoms=7),
    MockStructure(num_atoms=3),
]

# Create atomic_throughput value
atomic_throughput = 100

ARGS = {
    "train_structures": train_structures,
    "atomic_throughput": atomic_throughput,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
cutoff_func Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/pet.py
code served (permissive licence) · get_code("95ddf4e6c591ac91")
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, 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)

# Create a small grid tensor with values in a range that will exercise all branches
# of the cutoff function
grid = torch.linspace(0, 10, 8).reshape(2, 4)

r_cut = 5.0
delta = 2.0

ARGS = {
    "grid": grid,
    "r_cut": r_cut,
    "delta": delta,
}
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 (2, 4) — shape (2, 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
get_activation Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/pet.py
code served (permissive licence) · get_code("e8687ce6441116b2")
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 Mish. 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
from torch import nn

# Create a small stand-in class for hypers that has ACTIVATION attribute
class HyperConfig:
    def __init__(self, activation):
        self.ACTIVATION = activation

# Test with 'mish' activation
hypers = HyperConfig('mish')

ARGS = {
    'hypers': hypers
}
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 Mish — a Mish output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Mish 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
get_all_k_from_reciprocal Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
spozdn/pet/src/long_range.py
code served (permissive licence) · get_code("1a94a922640cc3c0")
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 list, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 numpy as np
import math

# Analyze the function to understand what arguments are needed
# get_all_k_from_reciprocal takes: w_1, w_2, w_3 (reciprocal lattice vectors), k_cut (cutoff)
# Each w_i is used in np.cross and vector operations, so they should be 1D arrays
# k_cut is a scalar used in comparison

# Create small 3D vectors for reciprocal lattice vectors
w_1 = np.array([1.0, 0.0, 0.0])
w_2 = np.array([0.0, 1.0, 0.0])
w_3 = np.array([0.0, 0.0, 1.0])

# Set a reasonable k_cut value
k_cut = 2.5

ARGS = {
    "w_1": w_1,
    "w_2": w_2,
    "w_3": w_3,
    "k_cut": k_cut
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, dtype family float — a list output has no array shape of its own; none is stated
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
get_all_species Ran
ran, honoured the contract we drafted
spozdn/pet/src/data_preparation.py
code served (permissive licence) · get_code("b20572fb9eed36e9")
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 ndarray, ndim 1, dtype family int, shape (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 numpy as np

class MockStructure:
    """Mock structure object that has get_atomic_numbers() method."""
    def __init__(self, atomic_numbers):
        self.atomic_numbers = atomic_numbers
    
    def get_atomic_numbers(self):
        return self.atomic_numbers

# Create sample structures with different atomic numbers
structure1 = MockStructure([1, 1, 8])  # H, H, O
structure2 = MockStructure([6, 8, 8])  # C, O, O
structure3 = MockStructure([1, 6])     # H, C

structures = [structure1, structure2, structure3]

ARGS = {
    "structures": structures
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (3,) — shape (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
get_forces Ran
ran, honoured the contract we drafted
spozdn/pet/src/data_preparation.py
code served (permissive licence) · get_code("e16a920d4ecca23d")
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 list, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 a stand-in Structure class that mimics ASE or similar structure object
class Structure:
    def __init__(self, arrays_dict):
        self.arrays = arrays_dict

# Create two sample structures with forces arrays
structure1 = Structure({
    'forces': np.array([[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]], dtype=np.float32)
})

structure2 = Structure({
    'forces': np.array([[0.5, 1.5, 2.5, 3.5], [4.5, 5.5, 6.5, 7.5]], dtype=np.float32)
})

structures = [structure1, structure2]
FORCES_KEY = 'forces'

ARGS = {
    'structures': structures,
    'FORCES_KEY': FORCES_KEY
}
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 list, ndim 3, dtype family float — a list output has no array shape of its own; none is stated
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
get_pyg_graphs Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/data_preparation.py
code served (permissive licence) · get_code("d7df1a51b8ac9361")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 numpy as np
import torch

torch.manual_seed(0)

class MockMolecule:
    def __init__(self, structure, R_CUT, USE_ADDITIONAL_SCALAR_ATTRIBUTES, USE_LONG_RANGE, K_CUT, MULTI_TARGET, TARGET_INDEX_KEY):
        self.structure = structure
        self.R_CUT = R_CUT
        self.USE_ADDITIONAL_SCALAR_ATTRIBUTES = USE_ADDITIONAL_SCALAR_ATTRIBUTES
        self.USE_LONG_RANGE = USE_LONG_RANGE
        self.K_CUT = K_CUT
        self.MULTI_TARGET = MULTI_TARGET
        self.TARGET_INDEX_KEY = TARGET_INDEX_KEY
        self._max_num = np.random.randint(5, 15)
        self._num_k = np.random.randint(3, 8) if USE_LONG_RANGE else 0
    
    def get_max_num(self):
        return self._max_num
    
    def get_num_k(self):
        return self._num_k
    
    def get_graph(self, max_num, all_species, max_k_num):
        return {"nodes": np.random.rand(max_num, 4), "edges": np.random.rand(max_num, 3)}

# Monkey patch the Molecule class in the function's module
import sys
module = sys.modules[__name__]
module.Molecule = MockMolecule

# Create mock structures
structures = [
    {"atoms": ["C", "H", "O"], "positions": np.random.rand(3, 3)},
    {"atoms": ["N", "C"], "positions": np.random.rand(2, 3)}
]

# Create all_species list
all_species = ["C", "H", "O", "N"]

# Set parameters
R_CUT = 5.0
USE_ADDITIONAL_SCALAR_ATTRIBUTES = True
USE_LONG_RANGE = True
K_CUT = 3.0
MULTI_TARGET = False
TARGET_INDEX_KEY = "energy"

ARGS = {
    "structures": structures,
    "all_species": all_species,
    "R_CUT": R_CUT,
    "USE_ADDITIONAL_SCALAR_ATTRIBUTES": USE_ADDITIONAL_SCALAR_ATTRIBUTES,
    "USE_LONG_RANGE": USE_LONG_RANGE,
    "K_CUT": K_CUT,
    "MULTI_TARGET": MULTI_TARGET,
    "TARGET_INDEX_KEY": TARGET_INDEX_KEY
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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
get_structural_batch_size Ran
ran, honoured the contract we drafted
spozdn/pet/src/analysis.py
code served (permissive licence) · get_code("072569b89a08575e")
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 int, ndim 0, dtype family int, 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 numpy as np
import math

class MockStructure:
    """Mock ASE-like structure object with get_positions() method"""
    def __init__(self, num_atoms):
        self.num_atoms = num_atoms
    
    def get_positions(self):
        """Return positions array with shape (num_atoms, 3)"""
        return np.random.randn(self.num_atoms, 3)

# Create mock structures with different numbers of atoms
structures = [
    MockStructure(num_atoms=5),
    MockStructure(num_atoms=7),
    MockStructure(num_atoms=3),
]

atomic_batch_size = 32

ARGS = {
    "structures": structures,
    "atomic_batch_size": atomic_batch_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
get_volume Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
spozdn/pet/src/long_range.py
code served (permissive licence) · get_code("07ab60b0a26fa0b2")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 numpy as np

# Create three vectors for computing volume of parallelepiped
# Each vector should be a numpy array of shape (3,) for 3D vectors
v_1 = np.array([1.0, 2.0, 3.0])
v_2 = np.array([4.0, 5.0, 6.0])
v_3 = np.array([7.0, 8.0, 10.0])

ARGS = {
    "v_1": v_1,
    "v_2": v_2,
    "v_3": v_3,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
propagate_duplicated_params Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/hypers.py
code served (permissive licence) · get_code("d10705f1d479d240")
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, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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)

# Simple test case 1: first_key provided in provided_hypers
provided_hypers_1 = {'learning_rate': 0.001}
default_hypers_1 = {'batch_size': 32}
first_key_1 = 'learning_rate'
second_key_1 = 'lr'

# Simple test case 2: second_key provided in provided_hypers
provided_hypers_2 = {'lr': 0.002}
default_hypers_2 = {'batch_size': 16}
first_key_2 = 'learning_rate'
second_key_2 = 'lr'

# Simple test case 3: first_key in default_hypers
provided_hypers_3 = {}
default_hypers_3 = {'learning_rate': 0.0005}
first_key_3 = 'learning_rate'
second_key_3 = 'lr'

# Simple test case 4: second_key in default_hypers
provided_hypers_4 = {}
default_hypers_4 = {'lr': 0.0003}
first_key_4 = 'learning_rate'
second_key_4 = 'lr'

# Use the first test case for the ARGS
ARGS = {
    'provided_hypers': provided_hypers_1,
    'default_hypers': default_hypers_1,
    'first_key': first_key_1,
    'second_key': second_key_1
}
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, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
q_func_exp Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/sp_frames_calculator.py
code served (permissive licence) · get_code("bb7a0d45da617036")
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, 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

torch.manual_seed(0)

# Create a small grid tensor with values
grid = torch.randn(2, 4)

# Define scalar parameters
w = 0.5
delta = 0.1
add_linear_multiplier = True

ARGS = {
    "grid": grid,
    "w": w,
    "delta": delta,
    "add_linear_multiplier": add_linear_multiplier,
}
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, 4) — shape (2, 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
smooth_max Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/sp_frames_calculator.py
code served (permissive licence) · get_code("ebdd8aa78165df48")
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 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

torch.manual_seed(0)

# The function smooth_max takes:
# - values: an iterable of numeric values (used in zip with weights, iterated over)
# - beta: a numeric scalar (used in multiplication with value)

# Create small test values as a list of scalars
values = [torch.tensor(0.5), torch.tensor(1.2), torch.tensor(-0.3)]

# Beta is a scalar multiplier
beta = 2.0

ARGS = {
    "values": values,
    "beta": beta,
}
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 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
smooth_max_weighted Ran
ran; our contract draft was wrong, not the code
spozdn/pet/src/sp_frames_calculator.py
code served (permissive licence) · get_code("86484398825c248c")
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 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

torch.manual_seed(0)

# The function expects:
# - values: iterable of scalars/tensors that will be multiplied by beta
# - weights: iterable of scalars/tensors that will be multiplied by exp(value * beta)
# - beta: scalar that scales the values

values = [torch.tensor(0.5), torch.tensor(1.2), torch.tensor(-0.3)]
weights = [torch.tensor(0.2), torch.tensor(0.5), torch.tensor(0.3)]
beta = torch.tensor(2.0)

ARGS = {
    "values": values,
    "weights": weights,
    "beta": beta
}
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 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
combine_hypers Not yet run
unverified (harvested, no recorded run)
spozdn/pet/src/hypers.py
code served (permissive licence) · get_code("2d4f94e828d7243b")
How it ran

no invocation recorded.

get_upper_bound Not yet run
unverified (harvested, no recorded run)
spozdn/pet/src/long_range.py
code served (permissive licence) · get_code("4d07a8d02807a64e")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 17 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.

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

Symmetry is one of the most central concepts in physics, and it is no surprise that it has also been widely adopted as an inductive bias for machine-learning models applied to the physical sciences. This is especially true for models targeting the properties of matter at the atomic scale. Both established and state-of-the-art approaches, with almost no exceptions, are built to be exactly equivariant to translations, permutations, and rotations of the atoms. Incorporating symmetries -- rotations in particular -- constrains the model design space and implies more complicated architectures that are often also computationally demanding. There are indications that non-symmetric models can easily learn symmetries from data, and that doing so can even be beneficial for the accuracy of the model. We put a model that obeys rotational invariance only approximately to the test, in realistic scenarios involving simulations of gas-phase, liquid, and solid water. We focus specifically on physical observables that are likely to be affected -- directly or indirectly -- by symmetry breaking, finding negligible consequences when the model is used in an interpolative, bulk, regime. Even for extrapolative gas-phase predictions, the model remains very stable, even though symmetry artifacts are noticeable. We also discuss strategies that can be used to systematically reduce the magnitude of symmetry breaking when it occurs, and assess their impact on the convergence of observables.

For agents

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

get_harvested_code_for_paper("2406.17747")
get_code_for_paper("2406.17747")
have("2406.17747")

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

curl https://syntology.ai/api/ran/2406.17747.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.17747.svg)](https://syntology.ai/paper/2406.17747)

Connect an agent — have() is free.