material.codes
Get Free Access
Get Free Access

Report

Screening 32 Li-ion cathodes: MACE ordering against Materials Project, with one Quantum ESPRESSO verification

Eugène Filin · 23 September 2026

Thirty-two Materials Project insertion electrodes — Li–transition-metal oxides and phosphates — screened for average voltage with MACE-MP-0 (the 0b3 medium checkpoint) through the production pipeline, compared pair by pair with MP's own DFT voltages, and one candidate (LiFePO4) verified with Quantum ESPRESSO PBE+U. The input, MACE-relaxed and QE-verified geometries, the QE parameters as the runner echoed them, the MACE checkpoint identity and the run specification are in the bundle; the diagnostic leg's free-lattice host cell is not, only its energy.

The question

A machine-learned interatomic potential relaxes a cathode cell in seconds where DFT takes hours. That makes it the obvious first filter for a list of candidate cathodes — provided the ordering it produces is the ordering DFT would produce. This report asks exactly that of MACE-MP-0 on the chemistry we screen: does a MACE average-voltage screen order Li-ion cathodes the way Materials Project's DFT does, and what does one Quantum ESPRESSO verification add to the number?

It is also a sample. Everything below — the table, the geometries, the parameters as the runners echoed them, the run specification — is what a client of a material.codes screening study receives, produced by the same pipeline on the same stand.

What was screened

Thirty-two insertion electrodes from Materials Project, all Li–transition-metal oxides and phosphates. They were drawn blind, in a hash order over battery ids, from the 829 MP electrodes inside our operating window — average voltage between 1.5 and 5.0 V, at most 0.1 eV/atom above the hull, lithium with the transition metals titanium to copper, oxygen and phosphorus. Forty were drawn; the eight with more than 60 sites in the discharged cell were dropped, leaving 32. The seed list in run.yaml carries those 32 sorted by MP's average voltage, the order of the earlier measurement's table, not the order they were drawn in. They are the same 32 electrodes we had measured earlier with an operator script outside the product, and this run repeats that measurement through the production screening pipeline so that it comes with a record a reader can re-run from. All 32 are known materials by construction: every relaxed cell matched an MP entry by structure, or for five of them by composition (the novelty columns of the table).

The inputs are MP's discharged, lithiated structures, resolved from the pinned snapshot named in instruments.txt; they are the input geometries in the bundle.

The protocol

For each electrode the screen runs four MACE-MP-0 legs (the 0b3 revision of the medium checkpoint, mace-mp-0b3-medium.model, float64; the checkpoint file and its SHA-256 are in instruments.txt, and a citation of this run names that revision, not the 2023 release):

  1. the lithiated cell is relaxed, lattice and ions;
  2. every Li is removed from that relaxed cell — not from the MP cell — and the resulting host is relaxed with its lattice held fixed;
  3. the average voltage per removed Li follows from the two energies and a MACE energy for bcc lithium (mp-135), which is the zero of the mace_metric_v column;
  4. a fourth, diagnostic leg relaxes the same host with its lattice free. The energy it releases, per removed Li, is host_relaxation_v: the amount by which the fixed lattice inflates that candidate's voltage — as long as the free host stays the structure the fixed leg priced, which the volume_change_fraction column beside it can only rule out, not confirm: in one row (mp-756601_Li) the free host moved its volume by more than half, and there the value prices a different structure, not the lattice. It is reported and never subtracted, so that the DFT tier confirms a number the screen actually produced; a candidate above this run's ceiling of 0.2 V is flagged host_relaxation_exceeded in the table (four of the 32) and stays in the ranking.

This is the pipeline's protocol for a screening study — the same protocol an autonomous search campaign steers on. It differs from the earlier script measurement in one respect that matters for reading the comparison: that measurement also ran a "true endpoints" protocol, relaxing MP's charged and discharged structures separately. The discussion returns to the difference.

One candidate, LiFePO4 (mp-19017), was then selected for verification with Quantum ESPRESSO. The selection was the operator's, not the screen's: in screening mode the pipeline verifies nothing on its own — run.yaml carries verify_top_n: 0 — and a group runs only because an operator names a candidate. It was made outside the screen's ranking (LiFePO4 sits at rank 21 of 32), for a candidate whose plateau is measured: the value quoted below is what the verified number is read against. The same electrode had been computed before by this deployment's Quantum ESPRESSO voltage group, in stand runs outside this bundle and under a different protocol; this report does not compare with them. The verification: PBE+U with U(Fe) = 4 eV, ferromagnetic order, PseudoDojo PBE pseudopotentials, a k-point spacing of 0.3 Å⁻¹ resolved to a grid per cell, cutoffs as the runner echoed them in physics.txt. The lithiated cell and the delithiated host are each relaxed with the ions free and the lattice held at the MACE-relaxed lithiated cell's; a bcc-Li cell computed under the same k-spacing rule is the zero of dft_voltage_v. One verified candidate is a demonstration of the record, not a statistic; the study packages verify three to ten.

Results

input_refcandidate_idstructure_sourceformulaatoms (count, sites of the lithiated cell)validation_okvalidation_reasoncharge_balancevolume_per_atom_a3 (ų, input cell as validated, or the relaxed cell when that was the one refused)novelty_relaxednovelty_refmace_metric_v (V, vs MACE bcc-Li mp-135 on the screen checkpoint in instruments.txt)screen_frame_keyhost_relaxation_v (V, per removed Li; the fixed-lattice host's bias measured by the free-host leg)volume_change_fraction (dimensionless, unsigned |V of the free-lattice fourth-leg host − V of the relaxed lithiated cell| over the latter; that host cell is not in cif/)screen_labelstatusledger_seqdft_voltage_v (V, vs bare QE bcc-Li mp-135; cathode legs on the row's applied frame)applied_frame_keyk_grid_lithiatedverify_statusverify_failure_reasonverify_job_ids
mp:mpbat_mp-754753_Li66a02951-2191-4bcb-9373-6fcaaa0050c5materials_project:mp-1663233Li3MnO8P256truebalanced10.358knownmp:mp_mp-16632335.0134mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.03780.0148screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-26489_Li847be382-6669-4bde-80f9-a49d07c4d5d3materials_project:mp-26489Cr2Li3O12P340truebalanced10.116knownmp:mp_mp-264894.6748mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.00790.0107screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-759133_Lib6880af0-4cb4-4455-b07f-21825f04f35fmaterials_project:mp-760785Li8Ni3O24P6V42truebalanced10.954knownmp:mp_mp-7607854.6162mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.03530.0483screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-759831_Lida126723-38f3-470c-b9bc-ddf880154c60materials_project:mp-759831CrLi4Ni5O24P640truebalanced11.132known_compositionmp:mp_mp-7598314.4632mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.00970.0112screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-765972_Li8b30e425-706f-49be-9d0e-12c8f04ea559materials_project:mp-1176577LiO12P4V36truebalanced12.071knownmp:mp_mp-11765774.4493mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.33670.0715host_relaxation_exceededscreenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-757753_Lif96a444d-e318-48da-9b78-3eb772d574ebmaterials_project:mp-757753Fe3Li6O24P6V40truebalanced11.711known_compositionmp:mp_mp-7577534.3549mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.04990.0493screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-756601_Li5168c785-fdf7-4019-a58d-c155d7c35368materials_project:mp-1277563Li4Ni3O8Ti32truebalanced9.009knownmp:mp_mp-12775634.2875mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.18380.6324screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-753104_Li2032c827-e9c4-4556-b8df-0e42b9b0b0b7materials_project:mp-753104CuFe3Li4O816truebalanced9.320knownmp:mp_mp-7531044.1179mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.00050.0128screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-26071_Li51e61c7d-3257-4166-8bf7-8660d1b3dba7materials_project:mp-26071LiMnO4P42truebalanced12.875knownmp:mp_mp-260714.0824mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.09000.0730screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-752986_Lidd261030-8911-4042-8436-a92a102100ecmaterials_project:mp-752986CuLi4Ni3O816truebalanced8.875knownmp:mp_mp-7529864.0759mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.08530.1008screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-753151_Li0f007f45-9460-494d-a78a-dd9b2696ebcematerials_project:mp-753151CoLiO4V14truebalanced11.197knownmp:mp_mp-7531513.9109mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.20030.1173host_relaxation_exceededscreenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-757489_Li54e1a983-f408-44b5-b3ef-38f541b2388amaterials_project:mp-775260CoCu3Li6O16P430truebalanced11.074known_compositionmp:mp_mp-7752603.8390mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.04490.0446screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-755163_Li527084bd-eddc-4ff7-92aa-7265b2da2fc9materials_project:mp-769962Fe3Li4O8Ti16truebalanced9.500knownmp:mp_mp-7699623.8216mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.03120.1291screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-755063_Li53765df5-3ef0-4439-a2c4-f6e5878cc4e2materials_project:mp-1177549Fe3Li3O8V30truebalanced9.637knownmp:mp_mp-11775493.7665mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.08490.0693screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-757069_Lib95fb498-92c7-415f-b921-6db21c366318materials_project:mp-760357Fe3Li4O8V16truebalanced9.475knownmp:mp_mp-8015783.7124mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.00020.0057screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-757895_Lid7c694c2-3ca2-4799-bbd1-045518c0a2c1materials_project:mp-757895CoCrLi2O8P228truebalanced10.790knownmp:mp_mp-7578953.7092mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.13760.0985screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-773672_Lif3142b95-5b2f-4afa-a99d-f0dfc773a7c3materials_project:mp-1177464Fe3Li4Mn3O24P640truebalanced11.865known_compositionmp:mp_mp-11774643.6926mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.04630.0325screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-26795_Liecf6fc77-1417-4cd3-a2d2-d4a034209009materials_project:mp-26795Cu5LiO13P344truebalanced11.584knownmp:mp_mp-267953.6709mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.04950.0067screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-767440_Li40971e9d-7f19-425e-856e-1fcdbe911944materials_project:mp-767440Fe2Li2MnO12P340truebalanced11.807knownmp:mp_mp-7674403.6475mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.20040.0428host_relaxation_exceededscreenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-781664_Li80022790-9639-4f3b-8812-f1c296df004bmaterials_project:mp-1177850CoLi2O4Ti32truebalanced9.385knownmp:mp_mp-11778503.5903mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.03930.0645screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-19017_Li988d4190-50df-4105-b6d2-0194724e24f1materials_project:mp-19017FeLiO4P28truebalanced10.161knownmp:mp_mp-190173.5550mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.10160.0562selected03.6848pseudo=pseudodojo-v0.4-r2 / family=pseudodojo_pbe / xc=PBE / u=Fe=4 / spin=fm / dk=0.33x4x5baselineverify_delithiated=b7278849-fa00-46c1-9491-4d91ba020e60 verify_lithiated=1f81492a-477b-4737-a766-1f8b5304966a verify_reference=2820474f-45fa-453f-881e-cab5124e138fRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-772144_Lif08c5152-00f5-430a-a1ca-76245eaf893ematerials_project:mp-775798Li4Mn5O24P6V40truebalanced12.096known_compositionmp:mp_mp-7757983.4963mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.04300.0257screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-22526_Li68318b14-3bd0-4704-8ca7-bdde49b92825materials_project:mp-22526CoLiO24truebalanced7.933knownmp:mp_mp-225263.3757mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.02180.0598screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-755562_Lif54ce0e5-5336-488b-8e3d-75677dda30d0materials_project:mp-2506805LiMnO216truebalanced9.886knownmp:mp_mp-25068053.3710mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.13470.0506screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-753473_Lic56b1ba0-df76-4b17-824d-6b3eb0d8a4bbmaterials_project:mp-753473CoLiO28truebalanced9.217knownmp:mp_mp-7534733.3513mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.00870.0324screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-758706_Li63a3d8c4-0afd-4fb7-827a-e47da385fc67materials_project:mp-758706CrLi6O24P6Ti340truebalanced11.910knownmp:mp_mp-7587063.0494mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.06030.0073screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-759589_Li0a662ae4-c823-48e8-8496-af4a27e05972materials_project:mp-771262Li5Mn5O12Ti224truebalanced10.040knownmp:mp_mp-7712623.0280mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.06190.0545screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-26727_Li6c37e0dc-5cc7-44ff-a3fa-a1797fb5aa94materials_project:mp-26731Li2O7P2V24truebalanced11.330knownmp:mp_mp-267312.9757mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.03910.0128screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-504336_Li06077d4a-102f-4f07-b9c7-9a4bffe035e9materials_project:mp-1101544Cr5Li2O16P427truebalanced12.475knownmp:mp_mp-11015442.8511mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.22140.0437host_relaxation_exceededscreenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-18896_Lif13bd10b-6072-4341-b9b3-d1ef2ad46ef6materials_project:mp-542665Li3O13V622truebalanced11.142knownmp:mp_mp-313722.7724mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.15550.0561screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-757012_Li806ca889-fd00-4aae-aa3e-001e64886464materials_project:mp-757012Li3O8TiV330truebalanced9.766knownmp:mp_mp-7570122.7686mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.05120.0581screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)
mp:mpbat_mp-752510_Liee1e1cd4-339f-4a47-8393-4da2214bc419materials_project:mp-760954Li5O10Ti2V340truebalanced9.415knownmp:mp_mp-13055272.5466mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a0.03540.0171screenedRepeat this candidate in material.codes (chat recipe protocol: fixed-cell relax of the topotactically stripped host — a replication of the question, not a repeat of this row's run)

Comparison with MP average voltage

input_refformulamace_metric_v (V, vs MACE bcc-Li mp-135 on the screen checkpoint in instruments.txt)MP average voltage (V, MP GGA/GGA+U with MP2020 corrections over MP's own relaxed endpoints)our rankreference rank
mp:mpbat_mp-754753_LiLi3MnO8P25.01344.47612
mp:mpbat_mp-26489_LiCr2Li3O12P34.67484.42023
mp:mpbat_mp-759133_LiLi8Ni3O24P6V4.61624.18937
mp:mpbat_mp-759831_LiCrLi4Ni5O24P64.46324.27144
mp:mpbat_mp-765972_LiLiO12P4V4.44934.79851
mp:mpbat_mp-757753_LiFe3Li6O24P6V4.35494.21565
mp:mpbat_mp-756601_LiLi4Ni3O8Ti4.28753.675714
mp:mpbat_mp-753104_LiCuFe3Li4O84.11793.775810
mp:mpbat_mp-26071_LiLiMnO4P4.08244.20696
mp:mpbat_mp-752986_LiCuLi4Ni3O84.07593.6811013
mp:mpbat_mp-753151_LiCoLiO4V3.91093.920118
mp:mpbat_mp-757489_LiCoCu3Li6O16P43.83903.7251212
mp:mpbat_mp-755163_LiFe3Li4O8Ti3.82162.5521329
mp:mpbat_mp-755063_LiFe3Li3O8V3.76652.6951427
mp:mpbat_mp-757069_LiFe3Li4O8V3.71242.7321525
mp:mpbat_mp-757895_LiCoCrLi2O8P23.70923.5701615
mp:mpbat_mp-773672_LiFe3Li4Mn3O24P63.69263.5441716
mp:mpbat_mp-26795_LiCu5LiO13P33.67093.7461811
mp:mpbat_mp-767440_LiFe2Li2MnO12P33.64753.4011919
mp:mpbat_mp-781664_LiCoLi2O4Ti3.59033.0362023
mp:mpbat_mp-19017_LiFeLiO4P3.55503.4992117
mp:mpbat_mp-772144_LiLi4Mn5O24P6V3.49633.3252220
mp:mpbat_mp-22526_LiCoLiO23.37573.790239
mp:mpbat_mp-755562_LiLiMnO23.37103.1502421
mp:mpbat_mp-753473_LiCoLiO23.35133.4102518
mp:mpbat_mp-758706_LiCrLi6O24P6Ti33.04943.1152622
mp:mpbat_mp-759589_LiLi5Mn5O12Ti23.02802.7152726
mp:mpbat_mp-26727_LiLi2O7P2V2.97572.7802824
mp:mpbat_mp-504336_LiCr5Li2O16P42.85112.2612931
mp:mpbat_mp-18896_LiLi3O13V62.77242.5883028
mp:mpbat_mp-757012_LiLi3O8TiV32.76861.8863132
mp:mpbat_mp-752510_LiLi5O10Ti2V32.54662.2823230

This comparison is ordinal: the two columns are absolute numbers in different frames, and their difference is a frame gap, not an error. Median offset (ours − reference): 0.21 V.

345 of 391 pairs with |Δ MP average voltage| ≥ 0.30 V agree in order (0.882) — our campaign protocol (b): discharged cell relaxed with MACE-MP-0, every Li stripped in the relaxed cell, host relaxed at fixed lattice, MACE bcc-Li (mp-135) as the zero against MP GGA/GGA+U with MP2020 corrections over MP's own relaxed endpoints.

No rows excluded.

Reference frame: MP GGA/GGA+U with MP2020 corrections over MP's own relaxed endpoints

Our protocol: campaign protocol (b): discharged cell relaxed with MACE-MP-0, every Li stripped in the relaxed cell, host relaxed at fixed lattice, MACE bcc-Li (mp-135) as the zero

Source: Materials Project insertion-electrode records, field average_voltage, snapshot mp-insertion-electrodes-2025-09-25 (CC BY 4.0; A. Jain et al., APL Materials 1, 011002, 2013)

Caveats

Applied physics

campaign bad01bb4-eaba-4481-9167-88a8e29c1e11 — applied physics

requested frame (immutable config): functional PBE, pseudos pseudodojo_pbe, spin fm, calculation relax, k-spacing 0.3 1/Å
calculation relax is what the config requests for the two cathode legs; this line does not state the Li reference leg's, which depends on the release that queued it (scf since #155) — each leg's line below prints the calculation the runner reported it applied
screen preset: fmax 0.05 eV/Å, max_steps 500
structlab pinned at create: structlab-1+pymatgen-2026.5.4+spglib-2.7.0

mp:mpbat_mp-754753_Li (66a02951-2191-4bcb-9373-6fcaaa0050c5)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-26489_Li (847be382-6669-4bde-80f9-a49d07c4d5d3)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-759133_Li (b6880af0-4cb4-4455-b07f-21825f04f35f)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-759831_Li (da126723-38f3-470c-b9bc-ddf880154c60)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-765972_Li (8b30e425-706f-49be-9d0e-12c8f04ea559)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-757753_Li (f96a444d-e318-48da-9b78-3eb772d574eb)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-756601_Li (5168c785-fdf7-4019-a58d-c155d7c35368)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-753104_Li (2032c827-e9c4-4556-b8df-0e42b9b0b0b7)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-26071_Li (51e61c7d-3257-4166-8bf7-8660d1b3dba7)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-752986_Li (dd261030-8911-4042-8436-a92a102100ec)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-753151_Li (0f007f45-9460-494d-a78a-dd9b2696ebce)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-757489_Li (54e1a983-f408-44b5-b3ef-38f541b2388a)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-755163_Li (527084bd-eddc-4ff7-92aa-7265b2da2fc9)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-755063_Li (53765df5-3ef0-4439-a2c4-f6e5878cc4e2)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-757069_Li (b95fb498-92c7-415f-b921-6db21c366318)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-757895_Li (d7c694c2-3ca2-4799-bbd1-045518c0a2c1)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-773672_Li (f3142b95-5b2f-4afa-a99d-f0dfc773a7c3)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-26795_Li (ecf6fc77-1417-4cd3-a2d2-d4a034209009)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-767440_Li (40971e9d-7f19-425e-856e-1fcdbe911944)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-781664_Li (80022790-9639-4f3b-8812-f1c296df004b)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-19017_Li (988d4190-50df-4105-b6d2-0194724e24f1)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
  ledger row 0 — applied frame pseudo=pseudodojo-v0.4-r2 / family=pseudodojo_pbe / xc=PBE / u=Fe=4 / spin=fm / dk=0.3
    verify_delithiated: calculation relax, U Fe=4 eV, spin fm, functional PBE, pseudo family pseudodojo_pbe, pseudo image pseudodojo-v0.4-r2, k-grid 3x4x5, ecutwfc 118 Ry, ecutrho 472 Ry, smearing gaussian, degauss 0.01 Ry
    verify_lithiated: calculation relax, U Fe=4 eV, spin fm, functional PBE, pseudo family pseudodojo_pbe, pseudo image pseudodojo-v0.4-r2, k-grid 3x4x5, ecutwfc 118 Ry, ecutrho 472 Ry, smearing gaussian, degauss 0.01 Ry
    verify_reference: calculation relax, U none, spin none, functional PBE, pseudo family pseudodojo_pbe, pseudo image pseudodojo-v0.4-r2, k-grid 9x9x9, ecutwfc 118 Ry, ecutrho 472 Ry, smearing gaussian, degauss 0.01 Ry
mp:mpbat_mp-772144_Li (f08c5152-00f5-430a-a1ca-76245eaf893e)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-22526_Li (68318b14-3bd0-4704-8ca7-bdde49b92825)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-755562_Li (f54ce0e5-5336-488b-8e3d-75677dda30d0)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-753473_Li (c56b1ba0-df76-4b17-824d-6b3eb0d8a4bb)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-758706_Li (63a3d8c4-0afd-4fb7-827a-e47da385fc67)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-759589_Li (0a662ae4-c823-48e8-8496-af4a27e05972)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-26727_Li (6c37e0dc-5cc7-44ff-a3fa-a1797fb5aa94)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-504336_Li (06077d4a-102f-4f07-b9c7-9a4bffe035e9)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-18896_Li (f13bd10b-6072-4341-b9b3-d1ef2ad46ef6)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-757012_Li (806ca889-fd00-4aae-aa3e-001e64886464)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
mp:mpbat_mp-752510_Li (ee1e1cd4-339f-4a47-8393-4da2214bc419)
  screen checkpoint frame: mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a

Instruments

campaign bad01bb4-eaba-4481-9167-88a8e29c1e11 — instruments

MACE screen checkpoint frame keys:
  mace-mp-0b3-medium.model / float64 / 0.3.16 / sha256:2f2be696351a
    checkpoint sha256: 2f2be696351ac9e94fbe01cdfb6f017679acdbd2db7645209ef55fec9826b012
  CARRIED WEAKNESS: this key is the checkpoint file name, precision and library version. It
  carries no image digest, so a carrier image rebuilt under the same MACE_MODEL_VERSION reads as a
  repeat where it is a replication under a new instrument.

QE applied frame keys (bundle, family, functional, U as applied, spin, k-spacing):
  pseudo=pseudodojo-v0.4-r2 / family=pseudodojo_pbe / xc=PBE / u=Fe=4 / spin=fm / dk=0.3

QE pseudo image versions:
  pseudodojo-v0.4-r2

StructureLab versions the candidates were analysed under:
  structlab-1+pymatgen-2026.5.4+spglib-2.7.0
  pinned on the campaign row at create: structlab-1+pymatgen-2026.5.4+spglib-2.7.0

Dataset instruments — pinned snapshots:
  mp / mp-insertion-electrodes-2025-09-25
  mp / mp-summary-2025-09-25

Dataset instruments — the resolved entry id of every input:
  mp:mpbat_mp-765972_Li                    external mpbat_mp-765972_Li               entry faba5f3f-83d4-412a-a440-26c8893101f1
  mp:mpbat_mp-754753_Li                    external mpbat_mp-754753_Li               entry 6fc16000-274c-4700-94ac-46a63e3acd44
  mp:mpbat_mp-26489_Li                     external mpbat_mp-26489_Li                entry fb3f67c2-2391-45db-9906-9fe4ca708cd6
  mp:mpbat_mp-759831_Li                    external mpbat_mp-759831_Li               entry 1fa52b3c-a961-4ca8-8920-81be45ec6db4
  mp:mpbat_mp-757753_Li                    external mpbat_mp-757753_Li               entry 0b0deeba-0c9c-4b81-a586-474fb8a48e56
  mp:mpbat_mp-26071_Li                     external mpbat_mp-26071_Li                entry e24eaa5f-d38e-42ca-82f2-2415f25323e1
  mp:mpbat_mp-759133_Li                    external mpbat_mp-759133_Li               entry 5368d3fd-bed0-4926-b25d-4fffc5234b0d
  mp:mpbat_mp-753151_Li                    external mpbat_mp-753151_Li               entry 6a766aab-04d6-4545-8c35-a7b3fce2875f
  mp:mpbat_mp-22526_Li                     external mpbat_mp-22526_Li                entry efcf85d0-84b6-4ce2-98c5-da8fb090c5cf
  mp:mpbat_mp-753104_Li                    external mpbat_mp-753104_Li               entry 4a62ba22-b699-49a0-88a7-53074fb5f29d
  mp:mpbat_mp-26795_Li                     external mpbat_mp-26795_Li                entry e0a6f29a-9edd-4298-a9ca-a53cc5abb2e6
  mp:mpbat_mp-757489_Li                    external mpbat_mp-757489_Li               entry a70ac004-bb5b-4e11-9953-b1fc615dec26
  mp:mpbat_mp-752986_Li                    external mpbat_mp-752986_Li               entry 24ca20a0-c243-4770-863f-8295c5dc2343
  mp:mpbat_mp-756601_Li                    external mpbat_mp-756601_Li               entry d295c805-9079-4385-83dd-82bd9d7dee6b
  mp:mpbat_mp-757895_Li                    external mpbat_mp-757895_Li               entry 44d32ad4-bbd0-4d7e-9d56-115ea095ace3
  mp:mpbat_mp-773672_Li                    external mpbat_mp-773672_Li               entry 805a9f12-ce3f-4057-b66d-08a2b5fadfd4
  mp:mpbat_mp-19017_Li                     external mpbat_mp-19017_Li                entry aaa14168-39aa-43ac-ab51-353e03213e7c
  mp:mpbat_mp-753473_Li                    external mpbat_mp-753473_Li               entry e55467e3-5d27-49d0-82b8-970bb0459a35
  mp:mpbat_mp-767440_Li                    external mpbat_mp-767440_Li               entry e06ad2b0-58b2-4324-b6ea-50ca4f17ba8c
  mp:mpbat_mp-772144_Li                    external mpbat_mp-772144_Li               entry ce867b50-5511-4bb8-9810-7653b4a205a8
  mp:mpbat_mp-755562_Li                    external mpbat_mp-755562_Li               entry 7b65ed9e-69ff-4fab-bd72-de9836ce0521
  mp:mpbat_mp-758706_Li                    external mpbat_mp-758706_Li               entry 689c919d-62f5-4053-b4ee-2c20b41f1d09
  mp:mpbat_mp-781664_Li                    external mpbat_mp-781664_Li               entry 73b1896e-2d8e-46f5-a0ed-012e13b7969e
  mp:mpbat_mp-26727_Li                     external mpbat_mp-26727_Li                entry 0d3239f2-6839-4e13-9ef5-367c0b1cc54e
  mp:mpbat_mp-757069_Li                    external mpbat_mp-757069_Li               entry 3f8c32af-ebd2-481e-96b6-c39ccd134cd4
  mp:mpbat_mp-759589_Li                    external mpbat_mp-759589_Li               entry d0e7f372-f048-4b18-93ee-35d1a8edceee
  mp:mpbat_mp-755063_Li                    external mpbat_mp-755063_Li               entry 663ee1cb-5667-44f4-a811-2513f2980c89
  mp:mpbat_mp-18896_Li                     external mpbat_mp-18896_Li                entry e577eb75-fb6f-4f2c-9aa4-4a17c9bcb29b
  mp:mpbat_mp-755163_Li                    external mpbat_mp-755163_Li               entry 27922af3-5055-4368-9543-99103441cc6c
  mp:mpbat_mp-752510_Li                    external mpbat_mp-752510_Li               entry dab3537c-4c91-445c-97e6-00c2e0e8762e
  mp:mpbat_mp-504336_Li                    external mpbat_mp-504336_Li               entry 42d67add-0c65-4522-be39-da5f8b5d05e3
  mp:mpbat_mp-757012_Li                    external mpbat_mp-757012_Li               entry b434e992-c8a4-4424-91b9-56164985b977

Reproduce

A repeat is the same run.yaml on an instrument whose identity agrees with instruments.txt; a different instrument is a replication, not a repeat.

Download the bundle (zip)

Geometries (CIF)

Reading the comparison

The comparison above is ordinal by construction. MP's average_voltage is a GGA/GGA+U number over MP's own relaxed endpoints with MP2020 corrections; mace_metric_v is a MACE-MP-0 number against a MACE lithium reference. The median offset between the two columns is therefore a difference of protocol and frame, not an error to be corrected, and what the columns can be asked is whether they order the 32 electrodes the same way — over pairs far enough apart on MP's side that the order is a meaningful question. At 0.3 V that is 391 of the 496 possible pairs.

MACE-MP-0 agrees with MP on 345 of those 391 pairs (0.882). For the purpose the screen serves — cutting a long list down to the candidates worth DFT — that is a usable gradient: a candidate MP places a volt above another is placed above it by the screen in almost every pair, and most of the disagreements cluster where the two protocols differ rather than at random.

Where they cluster is instructive. The three Fe₃O₈-framework oxides (MP electrodes mp-755163_Li, mp-757069_Li and mp-755063_Li, screened as their discharged phases mp-769962, mp-760357 and mp-1177549: iron with titanium or vanadium) sit about a volt above MP's value in this screen, at MACE ranks 13–15 against MP's 25–29, while their host_relaxation_v is at most 0.09 V — so the fixed lattice is not what moved them. Under the earlier true-endpoints protocol on the same checkpoint — the operator-script measurement named above, whose record is not part of this bundle — the same three land within 0.4 V of MP and at ranks 24–26. The difference is the reaction being measured: stripping every Li from the relaxed discharged cell leaves a host that is not MP's charged endpoint for these frameworks, and the screen prices that host, not MP's. Layered LiCoO₂ (mp-22526) is the opposite case — the screen places it 0.4 V below MP under both protocols — which is MACE-MP-0's own error on that compound rather than the protocol's.

This is also why this report's figure is lower than the 0.96 of that earlier measurement on the same 32 electrodes, the figure the screening study is described with. That figure is the true-endpoints protocol's; the screening pipeline runs the stripped-host protocol, and 0.88 is its number on the same sample. Both are stated here for the same reason the frames are named next to the statistic above: a reader who sees only one of them reads a difference of protocol as a regression.

The offset is positive — MACE sits a median 0.21 V above MP over the included rows. It belongs to the stripped-host protocol, not to the lithium reference: in the earlier true-endpoints measurement the same checkpoint against the same MACE lithium reference showed no such offset from MP. The fourth leg estimates the fixed-lattice share of it per candidate rather than removing it — on the one row where the free host left the structure the fixed leg priced (mp-756601_Li), not even that; the rest is the reaction being measured.

What one verification says

The Quantum ESPRESSO group for LiFePO4 gives 3.68 V. The caveats list says this number is not comparable to Materials Project or to experiment, and it means comparable as an accuracy claim: both differ from ours in the Hamiltonian and the corrections, so they are set beside it here for scale, not as a score. MP's average voltage for the same electrode is 3.50 V (GGA+U with U(Fe) = 5.3 eV and MP2020 corrections), and the olivine plateau is measured at about 3.4–3.5 V versus Li/Li⁺ (Padhi, Nanjundaswamy and Goodenough, J. Electrochem. Soc. 144, 1188, 1997). Ours sits 0.2 V above both, and the caveats name what it is conditioned on: a lithium reference sampled on the k-spacing rule's grid rather than a converged dense one, ferromagnetic order, a host relaxed at the lithiated lattice, and U(Fe) = 4 eV rather than MP's 5.3 eV. Not all four can supply the excess: the reference-grid term is a few millivolts — measured on this reference frame, the one-atom bcc cell at this run's 9×9×9 grid against a 15×15×15 one, in a stand run whose record is not part of this bundle — and a lower U lowers the olivine voltage rather than raising it, so against MP's U it pushes our number down. What can is the host held at the lithiated lattice — the screen's fourth leg puts that bias at 0.10 V on this candidate, a MACE estimate rather than a bound on the DFT number — and the rest of the frame, PseudoDojo norm-conserving pseudopotentials without MP2020 corrections. One group cannot apportion the 0.2 V among them, and this report does not try. A screening study reports the verified number in exactly this form, with its frame, its cutoffs and grids per leg and its caveats, because a bare 3.68 V read against experiment looks like a 0.2 V error, while 3.68 V on a stated frame is a defined quantity a reader can move to their own.

The bundle carries no timings, and this report quotes none for the run. The ratio the opening paragraph states — seconds for a MACE relaxation, hours for a DFT one — is why the screen exists, and why a study is priced by verifications and not by screens.

One line of physics.txt deserves a note. The Li-reference leg of this run was submitted as a relax — the campaign's verify_calculation applied to all three legs when it ran — on a one-atom bcc cell with its lattice held fixed, where a relaxation moves nothing. The pipeline now submits the reference as an scf, which is why the header line of physics.txt states the requested calculation for the two cathode legs only and leaves the reference's to that leg's own line, where this run's relax is recorded. A repeat whose operator selects the same candidate, on the current release, produces the same reference energy under the new name.

Limitations

Repeating this run

The bundle's run.yaml is a specification the campaign engine loads after the two edits it deliberately leaves to the operator: an owner — an account on the deployment, not a fact about the run, so the export leaves it empty — and a deadline_at in the future, since this run's has passed. That specification reproduces the screen and nothing else. The verification is a third operator step outside run.yaml — campaign verify -id <run> -candidate <id>, with the candidate id the new run assigns to mp-19017 (its own export's table.csv carries it) — the same act that produced this run's one DFT row, recorded in table.csv as status selected, ledger seq 0. That step has to land while the campaign is still running: campaign verify refuses a campaign that has stopped, and a screening campaign stops at its deadline_at or at a spent ceiling, so the repeat's deadline is also the window in which the operator exports the screen's table, reads the candidate id and selects; the verb also refuses a candidate that is not in status screened or carries the implausible_voltage label. A run on an instrument whose identity agrees with instruments.txt — the same MACE checkpoint by SHA-256, the same pseudopotential image, the same StructureLab version, the same dataset snapshot entries — is a repeat of this one; a run on a different instrument is a replication under that instrument. That list is as far as instruments.txt reaches, and its CARRIED WEAKNESS line marks the edge: the MACE key is the checkpoint file, its precision and the library version, with no digest of the image that carried them, so a rebuilt image under the same checkpoint reads as a repeat where it is a replication. The QE side has the same edge and no line for it: the bundle records the pseudopotential image and the applied frame, not the pw.x version or the runner image's digest, so this record cannot tell two QE builds apart. Within those limits, a later checkpoint or pseudopotential set that moves a number is read as an instrument change and not as an error in this record.

Your own candidates

The same report for your own candidate list is the screening study material.codes offers. Write to info@material.codes, or request access to run it on the platform.

Input structures from the Materials Project (CC BY 4.0; A. Jain et al., APL Materials 1, 011002, 2013). Relaxed cells, the table, the text and the bundle by material.codes, CC BY 4.0. The MP average voltage series is not covered by this licence; its terms are those of its source: Materials Project insertion-electrode records, field average_voltage, snapshot mp-insertion-electrodes-2025-09-25 (CC BY 4.0; A. Jain et al., APL Materials 1, 011002, 2013).