Changelog

2026-10-21

integrate/batch-3 ranks the stock rows from the finished board branches. Each ranked row is one code, one version, and one stock method or preset. A finite-difference Hessian counts as one force call per coordinate. PAR-2 charges each failure twice the cap. hTST Al states no cap, so those rows leave par2_mean unset.

eOn develop 500350d9 on neb/Pt-heptamer, 13 starts, 0 failures:

• CG PAR-2 1141.076923076923

• FIRE PAR-2 678.9230769230769

• LBFGS PAR-2 319.53846153846155

• QuickMin PAR-2 1322.0

EON SVN r2549 on the same problem, 13 starts, 0 failures: CG 929.3846153846154, FIRE 678.9230769230769, LBFGS 336.7692307692308, QuickMin 1322.0. eOn 3.6.0 LBFGS stays ranked at PAR-2 3517.6923076923076.

eOn develop 500350d9 on double-ended-saddle-search/lj38, 50 starts: doubly nudged CI-NEB PAR-2 1228.96, 6 failures.

eOn develop 500350d9 on double-ended-saddle-search/lj38_2ts_separation, 51 starts:

• Lanczos improved PAR-2 536.6666666666666, 0 failures

• Lanczos PAR-2 573.7450980392157, 0 failures

• dimer PAR-2 3236.3529411764707, 0 failures

• energy-weighted CI-NEB PAR-2 15260.333333333334, 0 failures

• CI-NEB lj/cut PAR-2 38080.529411764706, 0 failures

• CI-NEB ljcluster PAR-2 50000.450980392154, 2 failures

• doubly nudged CI-NEB PAR-2 72306.94117647059, 4 failures

• OCI-NEB PAR-2 95094.3137254902, 6 failures

n2p2 2.3.0 Kalman energy: mlopt/C2H4 PAR-2 17397.12 (40/50 failures), mlopt/emt-pt38 PAR-2 25.02 (0/50). n2p2 2.3.0 Kalman force: mlopt/C2H4 and mlopt/emt-pt38 are 0/50 at PAR-2 60000.0.

hTST Al dimer, 639 windows, 0 failures. PAR-2 is unset. The mean of the recorded window costs is 45626.85758998435 for EON SVN r2549 and 48128.758998435056 for eOn develop 500350d9.