# Borazine inverted-gap reproduction

Source: Shizu, Ishihara, Uratani, Kaji, "Inorganic benzenes with inverted singlet-triplet gaps",
Communications Chemistry (2026), doi:10.1038/s42004-026-02141-0 (CC BY 4.0).
SI1.pdf SHA-256: 2106a4fd1f0ebf32c4b2b4442bef9b000461f1aabb031dec274bbe13c4d9802f.

## Question and boundary

- Post type: research
- Status: 2b ADC stopped without a gap; 2b EOM-CCSD complete; Amendment 1 complete, with a non-decisive verdict
- Question: does independent PySCF DF-ADC(2)/def2-TZVP on the published PBE0/6-31G(d)
  geometries reproduce the ADC(2) gaps for borazine (1), boroxine (11), 9, 10, 12 and 2a?
- Research falsifier: [PREREGISTRATION.md](PREREGISTRATION.md), P1/P2/S1.
- What this experiment can establish: agreement with the published ADC(2) column under the stated settings.
- What it cannot establish: SCS-CC2 or kF (not run by us), experimental ordering, or S1-geometry behaviour.
- Traceability: resource figures, EOM-CCSD gaps and basis shift in the follow-up are traceable; the original note remains unbound
- Highest reproduction level: none
- Archived-evidence or rerun constraints: the source SI is identified in sources.json;
  SCS-CC2 remains the authors' evidence. The initial queue excluded 2b and 2c.
  The expanded queue includes 2b and the Amendment 1 EOM-CCSD/def2-TZVP checks;
  2c remains excluded. The original frozen text is preserved with Amendment 1 appended.

## Layout

- `xyz/`: all 57 S0 geometries from SI Tables 9–65, unchanged; index and basis counts included.
- `expected/published_SI_tables_1-3.csv`: the authors' published values, not our results.
- `tier1/`: supplied PySCF ADC(2), EOM-CCSD and optional ORCA inputs.
- `tier2/`: supplied Turbomole inputs; not run by us.
- `bench/`: supplied timing smoke tests used to size the jobs; not canonical results.
- `results/`: unchanged JSONs for the completed initial queue and a journal snapshot.
- `prompts/`: the archived launch prompt and its provenance note.

## Run

Calculations run outside the repository in `~/Molecules/ist-borazine/package/`,
using `~/Molecules/ist-borazine/.venv/`. The local launchd queue is started only
once the supplied preregistration has been frozen in the scaffold commit.
Its label is `com.pvjohnston.ist-borazine-tier1` and its runner is
`~/Molecules/ist-borazine/run_queue.sh` (wraps the queue in `caffeinate -ims`).

```sh
bash ~/Molecules/ist-borazine/run_queue.sh
```

The sequential order is ADC(2) 1, 11; EOM-CCSD 1, 11; ADC(2) 9, 10, 12;
EOM-CCSD 9, 10, 12; ADC(2) 2a; EOM-CCSD 2a. The expanded queue then runs
ADC(2) and EOM-CCSD/def2-SVP on 2b, followed by EOM-CCSD/def2-TZVP on 1 and 11.
Each process uses
`OMP_NUM_THREADS=8` and `--mem-mb 24000`; ADC(2) uses `--basis def2-tzvp
--aux def2-tzvp-ri`. EOM-CCSD uses `--basis def2-svp` for the initial checks
and `--basis def2-tzvp` for Amendment 1.
Outputs remain in `package/tier1/results/`; `status.json` and `JOURNAL.md`
next to the runner record progress. Existing result JSON files are skipped.
Failed jobs are recorded without editing the frozen scripts or retrying them.

## Generate publication metrics

The completed initial outputs are archived in results/ and reported in the first note,
which has no active experiment binding. The follow-up uses `metrics.json` and
`generate-metrics.mjs` for the resource figures, completed EOM-CCSD gaps and basis shift.

## Data and publication

The coordinates are attributed to the authors' CC BY supplementary information.
The supplied source manifest and parsers record acquisition and extraction.
[PUBLIC_FILES.txt](PUBLIC_FILES.txt) lists the reader-facing files: the protocol,
prompts, completed JSONs and journal, both Tier 1 scripts, environment, source
manifest, published comparison values and the six geometries used in the draft.
The first note's raw runtime logs, the venv and caches remain outside this publication bundle.

## Follow-up: stopped 2b attempt

[2b-handoff.md](2b-handoff.md) collects the prepared calculation, exact script version,
output schema, falsifier and resource estimate. The stopped attempt yielded no
ADC(2) gap. The 2026-09-30 journal/status snapshots and raw logs are allowlisted.
`node research/borazine-inverted-gap/generate-metrics.mjs --check` verifies the
follow-up resource figures, EOM-CCSD gaps and basis shift from the retained inputs. This does not
establish a completed 2b ADC(2) calculation or change the original post's traceability status.

The completed A1 and 2b EOM-CCSD JSONs have raw PySCF, timing and stdout companions
archived as `*-completed.{pyscf,time,stdout}.txt`; their bytes are unchanged from
the run directory. The journal/status snapshots now record the completed queue.
The first note's only update is a dated paragraph; its displayed A1 value was
read from the retained JSON, while the follow-up uses metric references.