Pre-registration (code frozen 17 Nov 2026): is the +0.06 eV Co-Ni-V CO-binding shift from short-range order specific to the MACE-MP-0 potential?

Frozen protocol and code, deposited before any production arm was run. BACKGROUND: in 12 paired Co40Ni40V20(111) slabs, layer-conserving Monte Carlo annealing weakened CO binding by +0.0598 eV at identical surface composition under MACE-MP-0 medium (sign-flip p = 0.00055); a later dose-response test found the shift does not track the degree of ordering (rho = -0.154). All numbers come from one machine-learned potential whose published CO-adsorption error on high-entropy alloys is five to seven times the shift. QUESTION: does the paired, composition-identical CO shift survive fresh seeds under the same potential and an independently trained potential? ARMS: M0 (MACE-MP-0 medium, fresh seeds, 12 pairs), MPA (MACE-MPA-0 medium, 20 pairs), and MH1 (MACE-MH-1, OC20 head oc20_usemppbe, fixed from the model's own head list, 20 pairs; counts toward the verdict only if its licence is cleared). Pair k of every arm starts from the same slab. GATES checked before any adsorption energy is read: stationarity, exact surface-composition identity within pairs, and an ordering contrast; a potential that does not order is INCONCLUSIVE, not a null. DECISION RULE: per arm, REPLICATED (gates pass, positive control detected, positive shift, one-sided sign-flip p < 0.05), ABSENT (gates pass, positive control detected, 95 percent CI upper bound below +0.030 eV) or UNRESOLVED; overall ROBUST, POTENTIAL-DEPENDENT, SEED-FRAGILE or INCONCLUSIVE as defined in the protocol. SMOKE TEST: three one-pair 200-step CPU runs under a throwaway seed confirmed each model loads and per-pair seeding works; their adsorption values were discarded unread. A pre-existing label bug (config field shadowed by the pair index) was fixed; the analysis keys on pair and is unaffected. SCOPE: agreement among MACE potentials shows robustness to the choice of machine-learned potential, not agreement with DFT or experiment; it does not test ordering as the cause, CO2 adsorption, or selectivity.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24306950
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Pre-registration (code frozen 17 Nov 2026): is the +0.06 eV Co-Ni-V CO-binding shift from short-range order specific to the MACE-MP-0 potential?

John Goodman
Zenodo (CERN European Organization for Nuclear Research)
Machine Learning in Materials Science
article

Pre-registration (code frozen 17 Nov 2026): is the +0.06 eV Co-Ni-V CO-binding shift from short-range order specific to the MACE-MP-0 potential?

John Goodman
article en

Abstract

Frozen protocol and code, deposited before any production arm was run. BACKGROUND: in 12 paired Co40Ni40V20(111) slabs, layer-conserving Monte Carlo annealing weakened CO binding by +0.0598 eV at identical surface composition under MACE-MP-0 medium (sign-flip p = 0.00055); a later dose-response test found the shift does not track the degree of ordering (rho = -0.154). All numbers come from one machine-learned potential whose published CO-adsorption error on high-entropy alloys is five to seven times the shift. QUESTION: does the paired, composition-identical CO shift survive fresh seeds under the same potential and an independently trained potential? ARMS: M0 (MACE-MP-0 medium, fresh seeds, 12 pairs), MPA (MACE-MPA-0 medium, 20 pairs), and MH1 (MACE-MH-1, OC20 head oc20_usemppbe, fixed from the model's own head list, 20 pairs; counts toward the verdict only if its licence is cleared). Pair k of every arm starts from the same slab. GATES checked before any adsorption energy is read: stationarity, exact surface-composition identity within pairs, and an ordering contrast; a potential that does not order is INCONCLUSIVE, not a null. DECISION RULE: per arm, REPLICATED (gates pass, positive control detected, positive shift, one-sided sign-flip p < 0.05), ABSENT (gates pass, positive control detected, 95 percent CI upper bound below +0.030 eV) or UNRESOLVED; overall ROBUST, POTENTIAL-DEPENDENT, SEED-FRAGILE or INCONCLUSIVE as defined in the protocol. SMOKE TEST: three one-pair 200-step CPU runs under a throwaway seed confirmed each model loads and per-pair seeding works; their adsorption values were discarded unread. A pre-existing label bug (config field shadowed by the pair index) was fixed; the analysis keys on pair and is unaffected. SCOPE: agreement among MACE potentials shows robustness to the choice of machine-learned potential, not agreement with DFT or experiment; it does not test ordering as the cause, CO2 adsorption, or selectivity.

Zenodo (CERN European Organization for Nuclear Research)
Machine Learning in Materials Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon