Skip to content

docs: molecular-GW benchmark findings for this code - #36

Open
mkakcl wants to merge 1 commit into
masterfrom
docs/mgw-benchmark-findings
Open

docs: molecular-GW benchmark findings for this code#36
mkakcl wants to merge 1 commit into
masterfrom
docs/mgw-benchmark-findings

Conversation

@mkakcl

@mkakcl mkakcl commented Aug 22, 2026

Copy link
Copy Markdown
Owner

Companion to molecular-mGW-testing#170 (and cayley-gw#55). Adds a top-level BENCHMARK_FINDINGS.md recording what four weeks of benchmarking established about running this package: the quote-at-nmom_max≤15 rule and the high-order turnaround (with no delivered-order diagnostic to reveal it), nmom_max=11 as a floor at def2-TZVPP scale, odd-orders-only (even orders bit-identical), one-build-many-orders with the se_static requirement, the npoints story (dRPA-only; −60.6 meV on MgO at the default 48; LiF has no valid datum; cost-flat so use 256), the silent compression='ia' default, the fork-vs-baseline result with its two attribution corrections (η⁰ algorithm difference, dyson provenance), the MBLSE/MLSE reconstruct_moments self-check defect, and the open LiF/MgO deviation against published values at matched basis.

Docs-only. Written per this repo's CLAUDE.md: targeted at origin (mkakcl) only — nothing here touches or targets BoothGroup. Since this repo is public, the note stays method-level and names no infrastructure.


Generated by Claude Code

Quote at nmom_max <= 15 (the error turns around above the conditioning
minimum; no delivered-order diagnostic exposes it); nmom_max=11 is a
floor at def2-TZVPP scale; even orders are bit-identical to the odd
order below; one build serves every order if se_static is stored;
npoints is dRPA-only (inert under dtda), -60.6 meV wrong on MgO at the
default 48, and LiF has no valid npoints datum; compression='ia'
applies silently unless set to None; fork-vs-baseline agreement is
3.3 ueV with the residual an eta0 algorithm difference; the
MBLSE/MLSE reconstruct_moments self-check defect; and the open
LiF/MgO deviation against published values at matched basis.

Co-Authored-By: Claude <[email protected]>
Claude-Session: https://claude.ai/code/session_01LK2cf5TpEw2JMXZ7PMaWd9
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants