Conversation
Reproduces QPOD's neutral V_B formation energy (10.18 eV, standard states) for the h-BN vacancy tutorial pair, mirroring the graphene substitution simulation page's structure. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Review blockers: correct n_i to the signed atom-count change (the page had E_f short by 2*mu_B), stop promising the four Total Energy jobs always share one model (the notebook reuses an existing job per material instead), state the 0.02 eV finite-size offset without implying it bounds the tolerance, and register the page in mkdocs-guide.yml (Tutorials site), which was missing it entirely. Should-fix: numeric cutoff/k-grid in the settings table, npj (not Nature) Computational Materials in the body, reconcile the nitrogen reference job with its zero coefficient, $$...$$ for the equation, and name the compute defaults in what the default run does. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The notebook's model is GBRV ultrasoft PBE at 40 Ry / 200 Ry, not PAW PseudoDojo JTH -- ultrasoft PBE sets for B and N do exist. Update the settings table and the surrounding sentence, and keep the pseudopotential family as one of the differences the 0.5 eV tolerance covers. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Round-2 review: the "clean project" sentence described a project selector the notebook doesn't have -- replace it with the actual consistency check the results cell runs (recompute E_f from three owned total energies, warn on mismatch). Note that PAW and ultrasoft cutoffs aren't comparable numbers, and fold cutoff into the tolerance's unestimated differences. Correct "jobs already finished are reused" -- only the three reference jobs are; the defect job always runs. Fix the references.bib journal to npj Computational Materials (was Nature). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Round-3 review: the reference-reuse and cluster-fallback paragraphs described notebook behaviour 64b562aa changed away from. Reuse is now keyed to a per-material, per-model job name, so a cutoff change can't silently reuse a stale reference -- the residual hazard is the workflow's own elemental-energy resolution, which the results cell's consistency check catches; a missing cluster now raises and lists the available ones instead of falling back silently. Also: fix the inverted "skipped otherwise" for reference jobs, add the third verdict line form (reference mismatch), add k-point density to the unestimated differences, and make the atom counts (pristine -> defective) explicit in the settings table. Prerequisite page: journal name and a line noting both materials (h-BN supercell, B-vacancy h-BN) are saved by name, matching its own notebook's cell 15. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
api-examples 144f2c2f lets CLUSTER_NAME be None to use the first listed cluster; a name that is set but not found still raises and lists the available ones. State both in the compute paragraph. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Round-4 blocker: the prerequisite page's generic create_nanoribbon / create_point_defect route does not save two named materials -- only the combined notebook embedded in its own section 6 does. Say so on both pages, so the new page's prerequisite step names the right notebook. Should-fix: CLUSTER_NAME's shipped default is "cluster-001", not unset; match cell 44's verdict suffix byte for byte (ASCII "--", not an em dash); expand QPOD, GPAW, PBE, PAW at first use; state once that the elemental references are the platform's own seeded materials (why mu_B is alpha-boron specifically) and that an unseeded account stops the notebook at that step. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
… ref QPOD expands to "Quantum Point Defect database" (singular, lowercase "database") per the manuscript itself, not "Quantum Point Defects Database". Reflow that paragraph to the surrounding line width. The prerequisite notebook lives in its section 6, not "its last section" (References). Expand GBRV once (Garrity-Bennett-Rabe-Vanderbilt). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
✅ Deploy Preview for mat3ra-documentation ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
5 tasks
Physics decision (VB, 2026-09-15): the notebook now fixes the defect cell's total magnetization to 1 uB, the doublet QPOD reports, instead of a starting-magnetization guess. An unconstrained SCF on this cell finds a quartet (3 uB) 0.22 eV lower -- state both, and that the doublet is what is compared with QPOD. Add the measured production result (10.46 eV vs QPOD's 10.18 eV, +0.28 eV, inside tolerance, verdict yes; four jobs at 2-6 min each on 40 cores), and note in the offsets paragraph that the MACE estimates were spin-blind and that the 0.28 eV gap is dominated by the doublet's relaxation energy, which QPOD's cell includes and this SCF-only workflow does not. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…n timing Round-6 review: the offsets paragraph said skipping relaxation costs 0.04 eV and, two sentences later, that relaxation dominates the measured 0.28 eV gap -- both cannot hold. Retire the 0.04 eV figure for the doublet (it was a spin-blind MACE estimate) and back the "largest single contributor" claim with the actual forces (2.0 eV/A on the vacancy's N neighbours vs 0.01 eV/A pristine and QPOD's own threshold) rather than stating an unmeasured relaxation energy as fact. Also: the timing sentence now describes the doublet run that actually produced 10.46 eV (one ~8 min defect job, three references reused from an earlier run at 2-4 min each), not the earlier quartet run's range; one sentence on why the doublet -- not the lower-energy quartet -- is the state compared with QPOD (QPOD's entry is the relaxed doublet, and relaxation's effect on the ordering is undetermined here); and Bohr magneton written out at first use so muB (magnetization) and mu_B (boron's chemical potential) are no longer one underscore apart. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Physics decision (VB, 2026-09-15): RELAX_DEFECT is now an optional notebook parameter (default False). Document both configurations -- unrelaxed SCF (10.46 eV, +0.28 eV, no) and relaxed defect (10.12 eV, -0.07 eV, yes) -- where the page describes what the run does, and the 12-hour wall-time limit that the relaxation needs. Add the relaxation geometry (three N neighbours move 0.07-0.13 A outward, a small Jahn-Teller distortion of the doublet) and its 0.35 eV energy. With the relaxed value now measured, retire the spin-blind 0.02/0.04 eV MACE estimates and the "most likely relaxation" hedge; tighten the verdict tolerance from 0.5 to 0.2 eV, which the relaxed result and its pseudopotential/cell-size differences still clear. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Round-7 review: on the recommended (relaxed) path, the notebook prints a residual total force about 3x the page's stated 0.01 eV/A criterion. State why -- it is Quantum ESPRESSO's norm over every atom, not the per-atom maximum, which this run met at 0.009 eV/A. Also: justify the 0.2 eV tolerance from the known differences (0.02 eV finite-size estimate, unquantified pseudopotential set) rather than from the relaxed result itself; state the unrelaxed run's timing from the measured per-job table (~15 min cold, ~6 min once references are reused) instead of an unsourced "~7 minutes"; make the section heading and job count cover both configurations (up to five jobs with RELAX_DEFECT); note the mismatch suffix is appended to the verdict line, not substituted for it; merge the Geometry/Relaxation table rows and note only the defective cell is relaxed here, against QPOD relaxing every structure; and round 10.12 - 10.18 consistently to -0.06 eV. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The notebook is losing its data-presence guards and the "-- reference mismatch, see warning above" suffix (VB: unnecessary). Remove every mention of it and the consistency-check warning from the page; keep the description that the results cell prints the workflow's E_f next to the E_f recomputed from the notebook's own three total energies, so the reader can see they agree. The two verdict forms are now exactly "yes (relaxed defect)" / "no (unrelaxed SCF)". Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
VsevolodX
marked this pull request as ready for review
September 16, 2026 18:30
The tutorials index and its full-overview page still showed the h-BN vacancy row's Properties cell as "(Coming Soon)". Link it to the new simulation page, matching the Fujimoto and Liu rows' form. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
VB: the page must look like the existing tutorials, not invent its own way, and must not explain physics the paper already explains. Rewrote against SOF-8043's Gr/Ni simulation page (primary template, unmerged) and the merged MoS2 / graphene-substitution simulation pages: same section shape (Introduction + "what is being reproduced" subsection, Prerequisites, Workflow overview with a one-line formula, Calculation parameters table, How to run, Expected results, embed), same terse per-row/per-bullet treatment of settings instead of paragraphs. Deleted: the displayed formation-energy equation and its DeltaN derivation, the doublet/quartet discussion, the Jahn-Teller / displacement paragraph, the pseudopotential-cutoff-comparability sentences, the total-force-norm explanation, the tolerance-rationale paragraphs, and acronym expansions the sibling pages don't carry (GBRV, GPAW, PBE, PAW, QPOD). Kept every fact needed to run the notebook and read its output: the manuscript, the target values and source, the parameters table, the RELAX_DEFECT toggle, the two measured results, and the verdict strings. 107 lines, down from 185; closer to the sibling pages' proportions of manuscript / parameters / results content. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
VB pointed at the two merged pages -- defect-point-substitution-graphene- simulation.md and the MoS2 simulation page -- not SOF-8043's unmerged one, and said the earlier "collapse step-by-step and customization into one How-to-run section" was wrong. Undo that: restore the full section list, order and heading wording the graphene page (the closest match, another defect SE) uses -- Introduction (+ manuscript note and "What is being reproduced"), Prerequisites, Workflow overview, Calculation parameters (4.1 DFT parameters, 4.2 Relaxation settings), Step-by-step instructions (5.1-5.5, with the notebook's actual top-cell parameter values at api-examples 1f26d397), Expected results (+ 6.1 Comparison with published results), Customization options (7.1-7.3), Troubleshooting (8.1-8.3), embed, References. Still no physics exposition per VB's standing rule: every fact is one line or one table row, as the sibling pages state theirs. 237 lines, matching the graphene page's (235) proportions and depth. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
api-examples e205404c: the parameter is RELAX (was RELAX_DEFECT). RELAX = True now looks for an already-relaxed structure matched by content hash and model before relaxing again, and saves a new relaxation as "B-vacancy h-BN relaxed" for reuse by this or any other notebook. Update every RELAX_DEFECT mention and describe both timing regimes: default ~15 min (~6 min once references are reused); RELAX first run ~70 min, later runs ~6 min once the relaxed structure is found. Note the DEFECTIVE_NAME + RELAX = False shortcut for a structure already relaxed elsewhere. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…timing Round-9 review: the notebook resolves DEFECTIVE_NAME from ./uploads, not the platform, so pointing it at an already-relaxed platform material raises -- delete that paragraph. RELAX = True is the only route to a relaxed structure and finds it automatically. Also: "later runs ... take about 6 minutes" described a path that has not been executed yet. State it as what the notebook does (finds the relaxed structure and skips the relaxation) rather than as a timing measurement, and correct the relaxation time itself to the measured ~52 minutes (PLAN Sec19), not the unmeasured ~70 used previously. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
api-examples e912516f: load_material now tries the uploads folder, then the user's materials collection by name, before raising -- same shape as workflows/defect_formation_energy.ipynb. Reinstate the Customization note this enables: DEFECTIVE_NAME = "B-vacancy h-BN relaxed" with RELAX = False picks up an already-relaxed structure from the account instead of the uploads folder. Update Sec2 and Sec5's load descriptions to name both sources. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…abelling Review comments-6915843 SS2.5-2.7: SS4.2's reuse key now says "hash together with the model and relaxation settings" instead of "hash and model" -- the settings the notebook keys on include spin and k-point density, and RELAX_SETTINGS is being folded in too. SS7.1's "reusable by this notebook or any other" overstated it: the automatic hash-based lookup exists in this notebook only, so say "reused by later runs of this notebook" and keep "loadable by name elsewhere" for the separate, real capability. Also note that pointing the DEFECTIVE_NAME shortcut at an already-relaxed structure prints the "(relaxed defect)" verdict label, not "(unrelaxed SCF)". Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
api-examples 032f846a: the invented-metadata reuse design (structure hash + model + settings tag) is gone. The notebook now resolves an already-relaxed structure through the relaxation job recorded for that material under those settings, waiting for one still running instead of starting a second. Restate SS4.2 in user terms: reused if already relaxed with these settings, waited on if still running, repeated only when the model, spin or relaxation settings change. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
032f846a dropped the isRelaxed check (isRelaxed = False even on a genuinely relaxed structure, so it never fired) -- config_label is keyed on RELAX alone again. The shortcut still prints (unrelaxed SCF); point the reader at the provenance lines instead. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…rson
SS4.2's list of settings that trigger a fresh relaxation omitted
k-point density, which the notebook keys on alongside model, spin and
relaxation settings. Also replace the page's four "your" instances
with impersonal phrasing ("the account's materials collection"),
matching WRITING-STYLE.md and the rest of this folder.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
SS7.2's code block showed CLUSTER_NAME = None as if it were the default, contradicting SS5.2's cluster-001. State the default and what None does instead of just showing the value. SS8.3 said the relaxation "needs the full 12-hour time limit" while SS5.4 gives the measured ~52 minutes -- 12 hours is the notebook's default ceiling, not a requirement; state the measured time instead. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
api-examples 2dbb6779: find_relaxed_material matches on the structure's content hash only, with no workflow-name or settings filter -- a deliberate decision (any relaxed version of this structure is better than none). SS4.2 described a settings-keyed key that no longer exists; state instead that a relaxed version of the structure is reused whatever settings produced it, so changing the model, spin or k-point density does not trigger a fresh relaxation, and a run that needs one must remove or rename the existing relaxed structure (or use a different project/account). Also drop "matching" from SS5.3's mention of the same lookup. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The lookup matches on content hash and owner account only -- renaming or deleting the relaxed material does not help (project isn't in the query, and the material is reached through the job's final_structure, so deleting it would dangle). State what actually works: a different account, or relaxing separately and pointing DEFECTIVE_NAME at that structure with RELAX = False. Also: the found-branch prints the structure's name and id, not a job. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…1 default Merged origin/main (SOF-8043 PR #400) first. Its SIMULATION tutorial was rewritten twice after the shape this page originally copied ("as instructions, not physics", "remove verdict wording") and is now the settled reference for this page's family. Flattened SS4 to match -- one table, prose, no DFT/Relaxation subsections -- since our comparison table already carries the this-tutorial-vs-QPOD divergences a bullet list would restate. Kept the verdict wording: unlike Gr/Ni (whose notebook stopped printing a pass/fail verdict, which is why its page dropped the wording), this notebook still prints one, so the page describing it is accurate, not physics narrative. Kept Troubleshooting: it is mechanical (material not found, missing reference, job errors), the same kind of content the graphene-substitution sibling both pages follow still carries, not the physics-adjacent asides Gr/Ni's rewrite removed from its own. Replaced cluster-001 (our own test cluster) with CLUSTER_NAME = None (the account's first listed cluster) in SS5.2 and SS7.2, matching the notebook's own upcoming default; scanned the page for other environment-specific values (account names, job ids, project names, absolute paths) and found none. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This was referenced Sep 18, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Tutorial page for the h-BN boron-vacancy formation-energy Specific Example (api-examples PR: mat3ra/api-examples#367), Jira SOF-8044, epic SOF-7994.
What changed
tutorials/materials/specific/defect-point-vacancy-boron-nitride-simulation.md: what is reproduced (QPOD1BN-1.2d.v_B.0.1, 10.18 eV, standard states; only q = 0), the formation-energy definition, settings and where they differ from QPOD (GBRV ultrasoft vs PAW, 48- vs 84-atom cell), the optional relaxation (RELAX_DEFECT: fast SCF 10.46 eV / relaxed 10.12 eV vs QPOD 10.18) and the 0.2 eV tolerance, theory vs experiment, what the default run submits, the JupyterLite embed.mkdocs.ymlandmkdocs-guide.yml: one nav line each, next to the structure page;index-guide.mdandtutorials/materials/specific/overview.md: the Bertoldo row now links the page instead of "Coming Soon".references.bib:Bertoldo2022journal corrected to npj Computational Materials.Verified
scripts/serve-all.sh --build+scripts/links/check-links.py: no broken internal links (332 664 checked); the page appears in the Guide nav next to its prerequisite; the citation renders.Manual checks (merging accepts anything unticked)
jupyterlite_embedopensspecific_examples/defect_point_vacancy_boron_nitride_SIMULATION.ipynbon the preview — requires the api-examples PR merged and the JupyterLite build refreshed first.Merge after mat3ra/api-examples#367.