Cu h2o rdf by IsaacParker30 · Pull Request #470 · ddmms/ml-peg

IsaacParker30 · 2026-04-15T14:28:23Z

Pre-review checklist for PR author

PR author must check the checkboxes below when creating the PR.

I've confirmed the contribution guidelines.

Summary

Add's benchmarks for copper-water interface [rdfs,vdos,vacf,dipole fluctuations] and for water/ice [rdfs,vdos,vacf]

Linked issue

Resolves #299 #300

Progress

Calculations
Analysis
Application
Documentation
Currently using loaded AIMD trajectory in s3 bucket, so quite big in that regard.

Testing

Tested on mace-mp-0b3, orb-v3-consv-inf-omat

New decorators/callbacks

Have added new call backs, mainly to make interactive pdf score plots.

joehart2001 · 2026-04-15T14:42:59Z

Thanks for the pr @IsaacParker30 @xradvincula, excited to get these metrics in for everyone to use.

Initial thoughts:

could we add the AML package as a dependency so we don't need the aml.py script, or is this adapted/impr from AML? anyways, we can remove the two additional copies of it
are the decorator files copies of each other? we could also help you get this into the main decorator script if needed
can use janus for MD like Mandy added in PR Add water slab dipoles benchmark #349 (direct link to line). @ElliottKasoar could help out if you have any questions about this.

code cleanup:

rebase
get the latest version of main's uv.lock (unless we add aml as a dependcy in the pyproject.toml`
remove: changes to the notebook tutorial, et --soft HEAD~1‎

alinelena · 2026-04-28T19:43:24Z

@joehart2001 is aml still maintained? I know we used in a previous life, @ElliottKasoar can you see what is missing and add them in janus post-processing if aml is not working...

joehart2001 · 2026-06-18T13:51:27Z

@xradvincula @IsaacParker30 reminder

ElliottKasoar · 2026-06-23T10:56:57Z

@xradvincula @IsaacParker30 please also take a look at our new filtering guidelines: https://ddmms.github.io/ml-peg/developer_guide/filter.html

Since you have a single system, it should require pretty minimal changes.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

De-duplicate the three identical per-benchmark aml.py copies by vendoring a single shared copy at ml_peg/analysis/utils/aml_md_analysis.py (PR ddmms#470 review). The per-benchmark copies are removed in a later commit once imports are updated. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Extract the RDF/VDOS/VACF loader/scorer/bar-builder logic that is byte-identical across the bulk_water, ice and copper_water_interface analyses into a single shared module, parametrised by per-benchmark paths/models (PR ddmms#470 review). The analyse_*.py files are rewritten to delegate here in a later commit. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Consolidate the cell_to_bar decorator (identical across the three per-benchmark decorators.py copies) into the shared decorator module (PR ddmms#470 review). The histogram decorator is not duplicated: the analyses are switched onto the existing shared plot_hist instead. Local decorators.py copies are removed later. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The bulk_water, ice and copper_water_interface analyses shared ~700 lines of byte-identical RDF/VDOS/VACF loader/scorer/bar-builder logic. Rewrite each as config constants + thin pytest fixtures delegating to the shared md_water_analysis module and the shared decorators (PR ddmms#470 review). Copper's dipole histogram is unified onto the existing shared plot_hist: the analysis now keeps raw dipole arrays (which also better serve future structure sampling than the unused pre-binned indices) and lets plot_hist bin them, so the bespoke pre-binned histogram decorator is no longer needed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Restore docs/.../adding_benchmark.ipynb to main (the branch's tutorial edits were out of scope per the PR ddmms#470 review) and remove the personal CODEBASE_UNDERSTANDING.md ignore line, keeping the benchmark data-dir ignores. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Delete the three per-benchmark aml.py and decorators.py copies now that their content is vendored/consolidated in ml_peg/analysis/utils, and remove the stray 'et --soft HEAD~1' file accidentally committed earlier (PR ddmms#470 review). All recoverable from git history. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Replace the hand-rolled ASE Langevin loop with janus-core NVT, mirroring the water_slab_dipoles calc. Writes md-traj.extxyz (positions + momenta + masses, every step) instead of md.thermo/md-pos.xyz/md-velc.xyz; the per-frame water dipole is no longer computed during MD (moved to analysis, matscipy dropped). Keeps deuterated H, fixed bottom slab layers, 330 K / friction 0.05 / 1 fs, 5000 equilibration + 30000 production. Sets pbc=[True, True, False] for the slab. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Replace the hand-rolled ASE Langevin loops with janus-core NVT (as done for copper). Each writes md-traj.extxyz (positions + momenta + masses, every step) instead of md.thermo/md-pos.xyz/md-velc.xyz. Preserves bulk_water T=330 K and ice T=250 K, friction 0.05, 1 fs, 25000 equilibration + 300000 production, periodic in all directions. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Adapt create_rdfs/create_vdos/create_vacf to read the single janus-core md-traj.extxyz (positions + momenta) instead of the legacy md-pos.xyz/md-velc.xyz. New _load_positions/_load_velocities helpers rebuild the mdtraj trajectories with the same Å→nm scaling and init.pdb cell injection the old paths used, so the shipped reference pkls and the aml maths stay valid. Velocities come from atoms.get_velocities() (masses in-file, so deuteration is handled). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Add ml_peg/analysis/utils/dipoles.get_z_dipoles (point-charge total z-dipole per unit area). water_slab_dipoles now delegates to it (q=0.5562, output unchanged). Copper computes its dipole the same way from the janus md-traj.extxyz with the same q=0.5562 instead of reading a bespoke md.thermo column, keeping the stdev_dipole_z_deviation metric and histogram. NOTE: with the shared charge, copper's ref_dipole_data.npy must be regenerated at q=0.5562 for the deviation to be on a consistent scale (flagged in code). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Report the fraction of MD frames whose total z-dipole per unit area falls outside the stable interfacial band-gap window (below -0.019686563354143947 or above 0.011680694256792076 e/Å), mirroring the water_slab_dipoles "Fraction Breakdown Candidates" metric but with an asymmetric window on the shared q=0.5562 dipole scale. - New fraction_breakdown_candidates fixture reuses the raw_dipoles already computed by stdev_dipole_z_deviation (no extra trajectory read). - Added to the metrics table and metrics.yml (good 0 / bad 1). - App: clicking the new column shows the same dipole distribution histogram. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Bring the three MD benchmarks up to the element-filtering standard already used by physicality/water_slab_dipoles, so they can be excluded from scoring when a model lacks a required element. - Analysis: each test_* now writes OUT_PATH/info.json via write_struct_info, reading the mock model's md-final.extxyz (the janus final structure). - App: each app defines INFO_PATH = DATA_PATH / "info.json" and passes info_path to its *App constructor so BaseApp loads the elements and filters the metrics table. Calcs already catch MD errors -> NaN and the mock model already runs as part of ml_peg calc, so no calc changes and no extra runtime. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

De-duplicate the three identical per-benchmark aml.py copies by vendoring a single shared copy at ml_peg/analysis/utils/aml_md_analysis.py (PR ddmms#470 review). The per-benchmark copies are removed in a later commit once imports are updated. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Extract the RDF/VDOS/VACF loader/scorer/bar-builder logic that is byte-identical across the bulk_water, ice and copper_water_interface analyses into a single shared module, parametrised by per-benchmark paths/models (PR ddmms#470 review). The analyse_*.py files are rewritten to delegate here in a later commit. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Consolidate the cell_to_bar decorator (identical across the three per-benchmark decorators.py copies) into the shared decorator module (PR ddmms#470 review). The histogram decorator is not duplicated: the analyses are switched onto the existing shared plot_hist instead. Local decorators.py copies are removed later. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The bulk_water, ice and copper_water_interface analyses shared ~700 lines of byte-identical RDF/VDOS/VACF loader/scorer/bar-builder logic. Rewrite each as config constants + thin pytest fixtures delegating to the shared md_water_analysis module and the shared decorators (PR ddmms#470 review). Copper's dipole histogram is unified onto the existing shared plot_hist: the analysis now keeps raw dipole arrays (which also better serve future structure sampling than the unused pre-binned indices) and lets plot_hist bin them, so the bespoke pre-binned histogram decorator is no longer needed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Restore docs/.../adding_benchmark.ipynb to main (the branch's tutorial edits were out of scope per the PR ddmms#470 review) and remove the personal CODEBASE_UNDERSTANDING.md ignore line, keeping the benchmark data-dir ignores. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Delete the three per-benchmark aml.py and decorators.py copies now that their content is vendored/consolidated in ml_peg/analysis/utils, and remove the stray 'et --soft HEAD~1' file accidentally committed earlier (PR ddmms#470 review). All recoverable from git history. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

joehart2001 · 2026-07-07T14:49:30Z

i guess this is not only for water md analysis so could be renamed more generally

joehart2001 · 2026-07-07T14:51:00Z

Looking really good overall! @IsaacParker30 @xradvincula

TODO:

docs
send us processed input data to uplaod to the s3 bucket

joehart2001 · 2026-07-15T14:08:57Z

Could you revert this change?

Yep.
FYI I'm making some bigger changes to add plots from Fig 1 of https://arxiv.org/abs/2603.04228

Ok nice! Send me a message when you next want me to take a look

joehart2001 added the new benchmark Proposals and suggestions for new benchmarks label Apr 15, 2026

ElliottKasoar mentioned this pull request Apr 29, 2026

Add AML post-processing stfc/janus-core#707

Open

IsaacParker30 and others added 12 commits July 2, 2026 17:14

add core functionality

448f45d

added bulk water and ice

694da24

fixed some comms

4feaafa

sorting

adac747

fixing liqice

6cd3169

updated gitingore

167d2ab

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

IsaacParker30 force-pushed the cu-h2o-rdf branch from fded566 to 9c7f2e5 Compare July 2, 2026 16:36

IsaacParker30 and others added 7 commits July 2, 2026 18:09

Ignore personal local change log

3ee9625

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

joehart2001 mentioned this pull request Jul 3, 2026

feat: water rdf benchmark #660

Draft

5 tasks

joehart2001 force-pushed the cu-h2o-rdf branch from 3d527f3 to 3c2fec3 Compare July 7, 2026 14:25

joehart2001 reviewed Jul 7, 2026

View reviewed changes

joehart2001 marked this pull request as ready for review July 10, 2026 13:57

Tested app functionality and bug fixes

bf54094

IsaacParker30 force-pushed the cu-h2o-rdf branch from 7219707 to bf54094 Compare July 15, 2026 10:52

Unstashed changes

5d8df50

joehart2001 reviewed Jul 15, 2026

View reviewed changes

IsaacParker30 added 2 commits July 15, 2026 16:45

Added dipole profile test and changed notebook back to 3.12.8

df85043

Added dipole profile test and changed notebook back to 3.12.8

43a1b10

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Cu h2o rdf#470

Cu h2o rdf#470
IsaacParker30 wants to merge 23 commits into
ddmms:mainfrom
IsaacParker30:cu-h2o-rdf

IsaacParker30 commented Apr 15, 2026 •

edited by joehart2001

Loading

Uh oh!

joehart2001 commented Apr 15, 2026 •

edited

Loading

Uh oh!

alinelena commented Apr 28, 2026

Uh oh!

joehart2001 commented Jun 18, 2026

Uh oh!

ElliottKasoar commented Jun 23, 2026

Uh oh!

joehart2001 Jul 7, 2026

Uh oh!

joehart2001 commented Jul 7, 2026 •

edited

Loading

Uh oh!

joehart2001 Jul 15, 2026

Uh oh!

IsaacParker30 Jul 15, 2026

Uh oh!

joehart2001 Jul 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Uh oh!

Conversation

IsaacParker30 commented Apr 15, 2026 • edited by joehart2001 Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Pre-review checklist for PR author

Summary

Linked issue

Progress

Testing

New decorators/callbacks

Uh oh!

joehart2001 commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

alinelena commented Apr 28, 2026

Uh oh!

joehart2001 commented Jun 18, 2026

Uh oh!

ElliottKasoar commented Jun 23, 2026

Uh oh!

joehart2001 Jul 7, 2026

Choose a reason for hiding this comment

Uh oh!

joehart2001 commented Jul 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joehart2001 Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

IsaacParker30 Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

joehart2001 Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

IsaacParker30 commented Apr 15, 2026 •

edited by joehart2001

Loading

joehart2001 commented Apr 15, 2026 •

edited

Loading

joehart2001 commented Jul 7, 2026 •

edited

Loading