Versions

Versions of every component that the records of the d-SEAMS 2.0 paper built or called. The paper describes seams-core v2.12.0. A rerun at a later release adds a row here and leaves the v2.12.0 rows in place.

Engine and front ends

component

version

notes

seams-core

v2.12.0

v2.11.0 (9f963b72) with two patches; the first holds the ring, cage-affiliation, update, device and command-line changes with their tests, and the second sets the version to 2.12.0

d-SEAMS 1 baseline

seams-core commit 1c8efa38 (2026-05-28)

between v2.0.1 and v2.1.0; walk enumeration, the pair-scan HC search followed by the DDC search, and the all-pairs index-ordered neighbour builder

PydSEAMSlib (Python, import pydseamslib)

2.11.1 on PyPI

the wheel wraps seams-core v2.11.0; the label tables and the head-to-head table use its sdist built with v2.12.0 as the seams-core subproject

yodaStruct (Lua, require("dseams"))

v2.10.0

a separately versioned front end that builds seams-core v2.10.0

dseams-plumed (PLUMED action DSEAMS_CAGES)

builds seams-core commit bdc9b74b

separate MIT module, d-SEAMS/dseams-plumed

LAMMPS compute compute dseams

in seams-core

extras/lammps/compute_dseams.cpp and compute_dseams.h

Lean development of the supplement

Lean toolchain v4.15.0

the lean/ folder of the supplementary archive, no other library

The lean/ directory of seams-core is a separate development (Lean 4.15.0 with Mathlib v4.15.0, see Lean proofs), and the paper does not use it.

Reproducibility package lock

config/ecosystem-lock.json of dseams2_repro clones these revisions.

component

revision

seams-core

9f963b725b6665b162a7487840574446a47fe408 (v2.11.0)

seams-base (d-SEAMS 1 baseline)

1c8efa382c4ae88ebc68e40a991260c4d57edbd5

linkcell

2e26b83c0c9bbc07d09eb056011dd8e2cec6f4a5

PydSEAMSlib

445fc5820c3c5a2172ccfecb7ed330c73599fbf8

yodaStruct

a905ca57b2eb030902b307d93c8e72f2cf077eda

Build dependencies

The wrap files under subprojects/ of v2.12.0 pin these sources.

wrap

pin

Catch2

v3.14.0

gpulite

1e90edb6d573289a7678c18d25b5dbf62f374abe

linkcell

2e26b83c0c9bbc07d09eb056011dd8e2cec6f4a5

minimage

v0.1.4

nanobind

v2.14.0

readcon-core

91181ef2699ef4c46fd7827cee1127776a32aa55

readcon-db

0001a2c5a931bdde6bc8fe9b308f666dc28f4fe5

robin-map

v1.4.0

vesin

v0.5.7

meson.build of v2.12.0 asks for these versions.

dependency

constraint

build option

Eigen

>= 3.4.0

required

linkcell

>= 0.3.11

with_linkcell (auto)

minimage

>= 0.1.4

with_minimage (auto)

readcon-core

>= 0.16.2

optional

readcon-db

>= 0.1.8

optional

Highway

>= 1.3.0

optional SIMD paths

vesin, chemfiles

any

optional

gpulite

any

with_gpulite (auto), needs CUDA and NVRTC

IRA, sphericart, nauty

any

with_ira, with_sphericart, with_nauty (auto)

MPI

any

with_mpi (disabled)

OpenMP

any

optional; target offload is with_openmp_offload (disabled)

The pixi environment pins the conda-forge compilers, Eigen 3.4.0, Catch2 and Meson, and the library is C++20. PydSEAMSlib needs Python 3.11 or later.

Comparison codes

The head-to-head table of the paper (record h2h-2026-10-31) and the label tables call these versions. The pixi lock of dseams2_repro pins OVITO, freud, GenIce2, cycless, pyscal3 and Snakemake. genice2-cage comes from pip, and R.I.N.G.S. is built from its SourceForge archive.

code

version

used for

OVITO Python module

3.15.5

CHILL+ and polyhedral template matching

freud

3.5.0

Steinhardt \(q_6\) and \(\bar q_6\) on the four nearest neighbours

GenIce2

2.2.13.3

test lattices; its ring formatter calls cycless

cycless

0.7

shortcut-free rings and polyhedral cages

genice2-cage

2.5.2 (pip)

the cage formatter, run once untimed on sI

analice2

2.2.13.3

not used: it stops with an AttributeError on both the mW and the sI input

R.I.N.G.S.

rings-code 1.3.5 (SourceForge)

primitive ring census

pyscal3

3.3.2

Steinhardt cross-check

Snakemake

9.6.2

the reproducibility workflow

Machines

machine

stack

records

cloud VM, 8 vCPU KVM guest, Intel Xeon (family 6, model 207)

Ubuntu 24.04.5, GCC 13.3.0, Python 3.12.3, NumPy 2.5.4, SciPy 1.18.1

bench-2026-10-31, cli-carry-2026-10-31, h2h-2026-10-31 and the label records

Elja, dual-socket Intel Xeon Platinum 8358 (32 cores per socket, 2.60 GHz)

conda-forge GCC of the pixi environment

paper_manifest-elja-1785712.json, elja-1788072, elja-production

Elja, two AMD EPYC 7713 (64 cores per socket)

conda-forge toolchain of the pixi environment, Python 3.12.14

elja-1798666

Elja, two A100-PCIE-40GB

CUDA and NVRTC

tip-gpu-batch-elja-1787449.txt

AMD Ryzen 9 9950X workstation, 16 cores, 32 threads

conda-forge GCC, OpenMP on (terra-2026-09-02)

terra-2026-09-01, terra-2026-09-02