mbe-tools 0.2.2

A practical toolkit for Many-Body Expansion (MBE) workflows: cluster design, MBE input generation, output parsing, and analysis.

mbe-tools

mbe-tools is a Python toolkit for the Many-Body Expansion (MBE) workflow:

• Cluster handling: read .xyz, fragment (water heuristic or connectivity + labels), and sample fragments (random/spatial, ion-aware).
• Job prep: generate subset geometries, render Q-Chem/ORCA inputs, and emit PBS/Slurm scripts (supports chunked submission with run-control).
• Parsing: read ORCA/Q-Chem outputs, auto-detect program, infer method/basis/grid metadata, emit JSONL.
• Analysis: inclusion–exclusion MBE(k), summaries, CSV/Excel export, and quick plots.

Status: 0.2.2 release — backend syntax (e.g., ghost atoms) can be customized per site. License: MIT.

Install (editable for development)

cd mbe-tools
python -m pip install -e .[analysis,cli]

Settings precedence (P0)

Configure default commands/modules/scratch once and reuse across CLI calls. Precedence (low → high): env vars → ~/.config/mbe-tools/config.toml → ./mbe.toml → explicit load_settings(path=...).

Keys: qchem_command, orca_command, qchem_module, orca_module, scratch_dir, scheduler_queue, scheduler_partition, scheduler_account.

Env map: MBE_QCHEM_CMD, MBE_ORCA_CMD, MBE_QCHEM_MODULE, MBE_ORCA_MODULE, MBE_SCRATCH, MBE_SCHED_QUEUE, MBE_SCHED_PARTITION, MBE_SCHED_ACCOUNT.

Minimal mbe.toml example:

qchem_command = "/opt/qchem/bin/qchem"
orca_command  = "/opt/orca/bin/orca"
qchem_module  = "qchem/6.2.2"
orca_module   = "orca/5.0.3"
scratch_dir   = "/scratch/${USER}"
scheduler_queue = "normal"
scheduler_partition = "work"
scheduler_account = "proj123"

Quickstart (Python API)

1. Fragment an XYZ

from mbe_tools.cluster import read_xyz, fragment_by_water_heuristic, fragment_by_connectivity

xyz = read_xyz("Water20.xyz")
frags = fragment_by_water_heuristic(xyz, oh_cutoff=1.25)
frags_conn = fragment_by_connectivity(xyz, scale=1.2)

2. Sample and write XYZ

from mbe_tools.cluster import sample_fragments, write_xyz

picked = sample_fragments(frags, n=10, seed=42)
write_xyz("Water10_sample.xyz", picked)

3. Generate subset geometries

from mbe_tools.mbe import MBEParams, generate_subsets_xyz

params = MBEParams(max_order=3, cp_correction=True, backend="qchem")
subset_jobs = list(generate_subsets_xyz(frags, params))  # (job_id, subset_indices, geom_text)

4. Build inputs

mbe build-input water.geom --backend qchem --method wb97m-v --basis def2-ma-qzvpp --out water_qchem.inp
mbe build-input water.geom --backend orca  --method wb97m-v --basis def2-ma-qzvpp --out water_orca.inp

5. Emit PBS/Slurm templates (run-control included; PBS can local-run)

mbe template --scheduler pbs   --backend qchem --job-name mbe-qchem --chunk-size 20 --local-run --builtin-control --out qchem.run
mbe template --scheduler slurm --backend orca  --job-name mbe-orca  --partition work --chunk-size 10 --out orca.sbatch

6. Parse outputs to JSONL

mbe parse ./Output --program auto --glob "*.out" --out parsed.jsonl

7. Analyze JSONL

mbe analyze parsed.jsonl --to-csv results.csv --to-xlsx results.xlsx --plot mbe.png

CLI cheat sheet

• mbe fragment <xyz>: water-heuristic fragmentation + sampling → XYZ. Options: --out-xyz [sample.xyz], --n [10], --seed, --require-ion, --mode [random|spatial], spatial extras --prefer-special, --k-neighbors, --start-index, --oh-cutoff.
• mbe gen <xyz>: generate subset geometries. Options: --out-dir [mbe_geoms], --max-order [2], --order/--orders, --cp/--no-cp, --scheme, --backend [qchem|orca], --cluster-name (filename prefix, fallback to backend), --oh-cutoff; --monomers-dir DIR + --monomer-glob "*.geom" can also reuse monomer .geom files instead of fragmenting.
• mbe gen_from_monomer <dir>: generate subsets directly from existing monomer .geom files; options mirror mbe gen monomer mode: --order/--orders/--max-order, --cp/--no-cp, --scheme, --backend, --monomer-glob, --out-dir, --cluster-name.
• mbe build-input <geom>: render Q-Chem/ORCA input. Options for backend, method, basis (required), charge/multiplicity; Q-Chem adds --thresh, --tole, --scf-convergence, --xc-grid, --rem-extra, --sym-ignore/--no-sym-ignore, embeddings --giee elem=charge (repeatable per element) or --gdee file for $external_charges; ORCA adds --grid, --scf-convergence, --keyword-line-extra; batch mode: point geom to a directory and add --glob "*.geom" --out-dir outputs/ to render many at once.
• mbe template: PBS/Slurm scripts with run-control wrapper. Shared: --scheduler [pbs|slurm], --backend [qchem|orca], --job-name, --walltime, --mem-gb, --chunk-size, --module, --command, --out; PBS+qchem adds --ncpus, --queue, --project, --local-run (emit local bash runner), --control-file (external TOML), --builtin-control (write default control TOML); Slurm+orca adds --ncpus (cpus-per-task), --ntasks, --partition, --project (account), --qos; --wrapper emits a bash submitter (bash job.sh) that writes hidden ._*.pbs/.sbatch and submits via qsub/sbatch.
• mbe parse <root>: outputs → JSONL. Options: --program [qchem|orca|auto] (default qchem), --glob-pattern, --out, --infer-metadata, geometry search controls (--cluster-xyz, --geom-mode first|last, --geom-source singleton|any, --geom-max-lines, --geom-drop-ghost, --nosearch). If no singleton metadata is available, it falls back to the first parsable geometry as monomer 0 for embedding.
• mbe where: print default data/config/cache/state paths and the runs archive root.
• mbe analyze <parsed.jsonl>: summaries/exports. Options: --to-csv, --to-xlsx, --plot, --scheme [simple|strict], --max-order.
• mbe show <jsonl>: options: optional JSONL_PATH (uses default selection if omitted); --monomer N (0-based) to print monomer geometry and include it in participation/CPU summaries. Output includes cluster info, CPU totals, per-order energy stats, and strict inclusion–exclusion MBE(k) totals with per-order ΔE.
• mbe calc <jsonl>: options: optional JSONL_PATH; --scheme [simple|strict] (default simple); --to K (upper order); --from K0 (lower bound for ΔE K0→K); --monomer N (report monomer energy); --unit [hartree|kcal|kj] (default hartree); --interaction i,j[,k] (0-based, repeatable) to report subset interaction energy E(subset) − ΣE(monomers). Strict scheme uses inclusion–exclusion; simple scheme uses ΔE vs mean monomer.

Use mbe <command> --help for full flags.

Definitions (CLI & API)

Area	Item	What it does	Key options/args	Notes	Implementation
CLI	mbe fragment <xyz>	Water-heuristic fragmentation and sampling → XYZ	--n, --seed, --mode random/spatial, --require-ion, --prefer-special, --k-neighbors, --start-index, --oh-cutoff, --out-xyz	Spatial mode can force special fragment; writes sampled XYZ	src/mbe_tools/cli.py
CLI	mbe gen <xyz>	Generate subset geometries up to chosen orders	--max-order or repeatable --order/--orders, --cp/--no-cp, --scheme, --backend qchem/orca, --oh-cutoff, --out-dir	Orders can be explicit list; CP toggles ghost atoms	src/mbe_tools/cli.py
CLI	mbe build-input <geom>	Render Q-Chem/ORCA input from .geom	Required --method, --basis; Q-Chem: --thresh, --tole, --scf-convergence, --xc-grid, --rem-extra, --sym-ignore/--no-sym-ignore, embedding via --giee elem=charge (repeatable) or --gdee file; ORCA: --grid, --scf-convergence, --keyword-line-extra; --out; batch: --glob, --out-dir	With --glob, geom must be a directory; outputs named after stems	src/mbe_tools/cli.py
CLI	mbe template	Emit PBS/Slurm scripts (with run-control wrapper)	Shared: --scheduler pbs/slurm, --backend qchem/orca, --job-name, --walltime, --mem-gb, --chunk-size, --module, --command, --out; PBS extras: --ncpus, --queue, --project, --local-run, --control-file, --builtin-control; Slurm extras: --ncpus(per task), --ntasks, --partition, --project(account), --qos; --wrapper	--wrapper writes a bash submitter that generates hidden ._*.pbs/.sbatch then submits; run-control autodetects control files	src/mbe_tools/cli.py → src/mbe_tools/hpc_templates.py
CLI	mbe parse <root>	Parse Q-Chem/ORCA outputs to JSONL	--program qchem/orca/auto (default qchem), --glob-pattern, --out, --infer-metadata, --cluster-xyz, --nosearch, --geom-mode first/last, --geom-source singleton/any, --geom-drop-ghost, --geom-max-lines	Infers method/basis/grid from names/inputs; can embed cluster geometry	src/mbe_tools/cli.py → src/mbe_tools/parsers/io.py
CLI	mbe analyze <parsed.jsonl>	Summaries/exports/plots	--to-csv, --to-xlsx, --plot, --scheme simple/strict, --max-order	strict uses inclusion–exclusion; simple computes ΔE vs mean monomer	src/mbe_tools/cli.py → src/mbe_tools/analysis.py
CLI	mbe show <jsonl>	Quick cluster/CPU/energy view plus strict MBE(k) totals with per-order ΔE	--monomer N (0-based) prints geometry and participation/CPU; default JSONL selection if path omitted	Uses default JSONL selection; prints inclusion–exclusion MBE rows	src/mbe_tools/cli.py
CLI	mbe info <jsonl>	Coverage + CPU summary	Filters: --program, --method, --basis, --grid, --cp, --status; --scheme; --max-order; --json	Status counts by subset_size	src/mbe_tools/cli.py
CLI	mbe calc <jsonl>	CPU totals + MBE energies (simple/strict) and subset interaction ΔE vs monomer sums	--scheme simple/strict, --to, --from, --monomer, --unit hartree/kcal/kj, --interaction i,j[,k] (0-based, repeatable)	Warns on mixed program/method/basis/grid/cp combos	src/mbe_tools/cli.py
CLI	mbe save <jsonl>	Archive JSONL to timestamped folder	--dest DIR, --order, --no-include-energy	Uses cluster_id/stamp subfolders	src/mbe_tools/cli.py
CLI	mbe compare <dir or glob>	Compare multiple JSONL runs	--cluster ID, --scheme simple/strict, --order K, --ref latest/first/PATH	Lists cpu_ok, record counts, combo labels; ΔCPU/ΔE vs reference	src/mbe_tools/cli.py
API	Cluster	read_xyz, write_xyz, fragment_by_water_heuristic, fragment_by_connectivity, sample_fragments, spatial_sample_fragments	See function args for cutoffs, scaling, seeds	Supports ion retention and special-fragment preference	src/mbe_tools/cluster.py
API	MBE generation	MBEParams, generate_subsets_xyz	Args: max_order, orders, cp_correction, backend, scheme	Yields (job_id, subset_indices, geom_text) for each subset	src/mbe_tools/mbe.py
API	Input builders	render_qchem_input, render_orca_input, build_input_from_geom	Method/basis required; optional thresh/tole/scf/grid/extra lines	Used by CLI build-input; accepts .geom path	src/mbe_tools/input_builder.py
API	Templates	render_pbs_qchem, render_slurm_orca	Scheduler resources + chunking + run-control wrapper	wrapper flag mirrors CLI behavior	src/mbe_tools/hpc_templates.py
API	Parsing	detect_program, parse_files, infer_metadata_from_path, glob_paths	Program auto-detect; metadata inference from names/inputs	Companion inputs help fill method/basis/grid	src/mbe_tools/parsers/io.py
API	Analysis	read_jsonl, to_dataframe, summarize_by_order, compute_delta_energy, strict_mbe_orders, assemble_mbe_energy, order_totals_as_rows	Convenience helpers for MBE tables and plots	strict_mbe_orders builds inclusion–exclusion rows	src/mbe_tools/analysis.py

CLI details with examples

Command	Option(s)	Meaning	Example
mbe fragment <xyz>	--mode random/spatial, --n, --require-ion	Fragment and sample XYZ	mbe fragment water3.xyz --mode spatial --n 2
mbe gen <xyz>	--max-order, --order, --cp/--no-cp	Generate subset geometries	mbe gen big.xyz --max-order 3 --out-dir geoms
mbe build-input <geom>	--backend qchem/orca, --method, --basis, Q-Chem extras --sym-ignore/--no-sym-ignore, --giee elem=charge (repeatable) or --gdee file	Render Q-Chem/ORCA input from geom	mbe build-input frag.geom --backend qchem --giee O=0.2 --giee H=0.1 --out a.inp
mbe template	--scheduler pbs/slurm, --backend, --wrapper	Emit PBS/Slurm script (optional wrapper submitter)	mbe template --scheduler pbs --backend qchem --wrapper
mbe parse <root>	--program auto/qchem/orca, --glob-pattern, geometry search flags	Parse outputs to JSONL (can embed cluster geometry)	mbe parse ./Output --glob "*.out" --geom-source any
mbe analyze <jsonl>	--scheme simple/strict, --to-csv, --plot	Summaries, exports, plots	mbe analyze parsed.jsonl --scheme strict
mbe show <jsonl>	--monomer N (0-based) plus default selection if path omitted	Quick cluster/CPU/energy view plus strict MBE(k) totals with per-order ΔE	mbe show parsed.jsonl --monomer 0
mbe info <jsonl>	Filters: --program, --method, --basis, --grid, --cp, --status; --scheme; --max-order; --json	Coverage + CPU + optional MBE summary	mbe info --program qchem --json
mbe calc <jsonl>	--scheme simple/strict; --to; --from; --monomer; --unit hartree/kcal/kj; --interaction i,j[,k]	CPU totals + MBE energies; interaction ΔE for specified subset; monomer energy reporting	mbe calc parsed.jsonl --scheme strict --unit kcal --interaction 0,1 --monomer 0
mbe save <jsonl>	--dest DIR, --order, --no-include-energy	Archive JSONL to <dest>/<cluster>/<stamp>__<method>__<basis>__<grid>__<cp>/run.jsonl with run.meta.json	mbe save parsed.jsonl --dest runs/
mbe set-library <dir>	none	Persist default archive directory used by mbe save	mbe set-library ~/mbe_runs
mbe compare <dir or glob>	--cluster, --scheme simple/strict, --order K, --ref latest/first/PATH	Compare runs; shows cpu_ok, counts, combo labels, and ΔCPU/ΔE vs reference	mbe compare runs/**/*.jsonl --cluster water20 --ref latest

CLI option notes

• mbe fragment <xyz>: --mode random|spatial (sampling strategy); --n (samples); --require-ion (retain ions); spatial extras --prefer-special, --k-neighbors, --start-index; --oh-cutoff (bond cutoff); --out-xyz (write sampled XYZ).
• mbe gen <xyz>: --max-order or repeatable --order/--orders (subset orders); --cp/--no-cp (counterpoise ghosts); --scheme (naming scheme); --backend [qchem|orca] (job_id style); --oh-cutoff (connectivity for water heuristic); --out-dir (geom output dir).
• mbe build-input <geom>: required --backend, --method, --basis; --charge, --multiplicity; Q-Chem: --thresh, --tole, --scf-convergence, --xc-grid, --rem-extra, --sym-ignore/--no-sym-ignore, embedding --giee elem=charge (repeatable; bare value applies to O/H) or --gdee file for $external_charges; ORCA: --grid, --scf-convergence, --keyword-line-extra; batch with --glob + --out-dir.
• mbe template: --scheduler [pbs|slurm], --backend [qchem|orca], --job-name, --walltime, --mem-gb, --chunk-size, --module, --command, --out; PBS extras --ncpus, --queue, --project, --local-run, --control-file, --builtin-control; Slurm extras --ncpus(per task), --ntasks, --partition, --project(account), --qos; --wrapper emits a submitter script.
• mbe parse <root>: --program qchem/orca/auto (default qchem); --glob-pattern; --out; --infer-metadata; geometry search --cluster-xyz, --geom-mode first|last, --geom-source singleton|any, --geom-drop-ghost, --geom-max-lines, --nosearch.
• mbe where: show default data/config/cache/state/runs paths.
• mbe analyze <jsonl>: --scheme simple|strict; --to-csv, --to-xlsx, --plot; --max-order (trim orders).
• mbe show <jsonl>: optional path (defaults apply); --monomer N (0-based) prints geometry, CPU share, participation; output also shows CPU totals, per-order energy stats, strict MBE(k) totals with per-order ΔE.
• mbe info <jsonl>: filters --program/method/basis/grid/cp/status; --scheme; --max-order; --json for machine output; reports coverage by subset_size plus CPU.
• mbe calc <jsonl>: --scheme simple|strict (simple: ΔE vs mean monomer; strict: inclusion–exclusion); --to K (upper order); --from K0 (lower bound for ΔE K0→K); --monomer N (report monomer energy); --unit hartree|kcal|kj; --interaction i,j[,k] (0-based, repeatable) gives subset interaction E − ΣE(monomers).
• mbe save <jsonl>: --dest DIR (override default library/env); --order (filter subsets); --no-include-energy (skip energies).
• mbe set-library <dir>: persist default archive root for save/compare.
• mbe compare <dir|glob>: --cluster ID filter; --scheme simple|strict; --order K; --ref latest|first|PATH sets reference; outputs ΔCPU/ΔE vs ref.

Run-control (templates)

• Control file discovery: prefer <input>.mbe.control.toml, else mbe.control.toml, else run-control disabled.
• Attempt logging: write job._try.out; on failure rename to job.attemptN.out; on success rename to job.out. confirm.log_path can override temp log location.
• Confirmation: confirm.regex_any (must match) and confirm.regex_none (must not match) on the temp log; success also requires exit code 0.
• Retry: retry.enabled, max_attempts, sleep_seconds, cleanup_globs, write_failed_last (copy last attempt to failed_last_path).
• Delete safeguards: delete.enabled + allow_delete_outputs=true to delete outputs; inputs removed only if matched by delete_inputs_globs.
• State: .mbe_state.json records status, attempts, matched regex, log paths; skip_if_done skips reruns when marked done.

Subset naming

• Recommended: {backend}_k{order}_f{i1}-{i2}-{i3}_{cp|nocp}_{hash} with 0-based fragment indices (zero-padding allowed), e.g., qchem_k2_f000-003_cp_deadbeef.out.
• Legacy (still parsed): {backend}_k{order}_{i1}.{i2}..._{hash} treated as 1-based in the name but converted to 0-based internally. JSON always exposes subset_indices as 0-based.

JSONL schema (parse output)

{
  "job_id": "qchem_k2_f000-003_cp_deadbeef",
  "program": "qchem",
  "program_detected": "qchem",
  "status": "ok",
  "error_reason": null,
  "path": ".../job.out",
  "energy_hartree": -458.7018184,
  "cpu_seconds": 1234.5,
  "wall_seconds": 1234.5,
  "method": "wB97M-V",
  "basis": "def2-ma-QZVPP",
  "grid": "SG-2",
  "subset_size": 2,
  "subset_indices": [0, 2],
  "cp_correction": true,
  "extra": {}
}

API highlights

• Cluster (src/mbe_tools/cluster.py): read_xyz, write_xyz, fragment_by_water_heuristic, fragment_by_connectivity, sample_fragments, spatial_sample_fragments.
• MBE generation (src/mbe_tools/mbe.py): MBEParams, generate_subsets_xyz, qchem_molecule_block, orca_xyz_block.
• Input builders (src/mbe_tools/input_builder.py): render_qchem_input, render_orca_input, build_input_from_geom.
• HPC templates (src/mbe_tools/hpc_templates.py): render_pbs_qchem, render_slurm_orca (both embed run-control wrapper).
• Parsing (src/mbe_tools/parsers/io.py): detect_program, parse_files, infer_metadata_from_path, glob_paths.
• Analysis (src/mbe_tools/analysis.py): read_jsonl, summarize_by_order, compute_delta_energy, strict_mbe_orders.

Notebook

See notebooks/sample_walkthrough.ipynb for an end-to-end demo: build inputs, generate templates, and assemble MBE(k) energies from synthetic data.

Contributing and Contact

Contributions are welcome—feel free to open issues or send pull requests. For questions or collaboration, reach out to Jiarui Wang at [e-mail].

License

MIT