{"id":"04a41a72-34be-493f-b7b1-bf83c1f5754e","arxiv_id":"2602.06893","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Cs2KInI6's cubic phase is phonon-unstable; a genetic-algorithm search finds 42 candidate polymorphs, several confirmed stable by DFT, with distorted phases having wider, more indirect band gaps.","lead":"DFT calculations show that the cubic form of the lead-free double perovskite Cs2KInI6 is dynamically unstable, and a machine-learned potential plus genetic algorithm finds 42 candidate lower-energy polymorphs. The most stable candidate lacks perovskite-like octahedral coordination; distortions widen the band gap and often make it indirect.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claim of 42 dynamically stable polymorphs and the identification of Cmc21 as the most stable phase rest on unvalidated MACE predictions; only 11 of 42 were DFPT-checked, 2 of which failed, and MACE-vs-DFT energy ordering is not reported.","rationale":"The reader correctly identifies the reliance on MACE as the weakest link. I agree that the DFPT validation of only 11 of 42 structures, with 2 false positives, leaves the status of the remaining 31 uncertain. My concern goes slightly further: the paper's headline 'most stable phase' claim does not depend only on dynamic stability but also on the relative energy ordering of all 42 polymorphs. The energies in Table S1 appear to mix MACE and DFT values without explicit labeling, and no cross-validation of MACE vs DFT energies is shown. Thus the GA selection of Cmc21 as the lowest-energy structure could be an artifact of MACE inaccuracy. That said, the core finding that the cubic phase is dynamically unstable is solidly established by DFPT, and the DFPT validation of 9 polymorphs (including Cmc21) provides credible evidence for several stable phases. The paper is transparent about the 2 failures, which is a sign of good practice. Nevertheless, the abstract's phrasing 'identify 42 dynamically stable structures' overstates what is actually known, and the missing energy comparison is a concrete gap. The proposed test—full DFPT validation of the remaining 31—would settle both the count and the energy ordering. Until then, conditional acceptance is appropriate: the central qualitative conclusions are plausible, but the quantitative claims about 42 structures and the uniqueness of Cmc21 as most stable are not fully supported.","tokens_in":14039,"tokens_out":8581,"duration_ms":79952,"concrete_test":"Perform DFT (PBE, same pseudopotentials and k-grid as the paper) structural relaxation and DFPT phonon calculations for all 31 MACE-stable structures in Table S1 that were not previously checked. Determine (a) which of these 31 are dynamically stable at the DFPT level, and (b) their relaxed DFT total energies relative to cubic. If all 31 are DFPT-stable and none has an energy below the Cmc21 value (-41.9 meV/atom), the abstract's '42 dynamically stable structures' and 'most stable phase' claims are supported; if any is unstable or lower in energy, those claims must be revised.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central quantitative claims are (i) that the genetic-algorithm search with MACE-OMAT-0 found 42 dynamically stable polymorphs, and (ii) that the most stable phase has Cmc21 symmetry with ΔE = -41.9 meV/atom (abstract, Table I). These claims rest on the assumption that MACE-OMAT-0 accurately reproduces both the phonon stability and the relative energies of Cs-K-In-I structures. The paper provides only partial validation: DFPT phonons were run for 11 of the 42 structures, and 2 of those 11 (18%) turned out to be dynamically unstable despite MACE predicting stability (main text; Table S1). The remaining 31 polymorphs have only MACE-based stability. Moreover, the relative energies in Table S1 are evidently a mix of MACE (unvalidated) and DFT (validated) values, but the paper does not state which, nor does it present a MACE-vs-DFT energy comparison for the 11 validated cases. If the MACE energy ranking is inaccurate for this chemical space, a different polymorph among the 31 could have a lower DFT energy than Cmc21, undermining the 'most stable phase' claim. The 2/11 false-positive rate also makes it likely that several of the 31 are not dynamically stable at the DFT level, so the advertised '42' overcounts. The absence of deposited structures/code further impedes independent verification.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper investigates the structural and electronic stability of the lead-free halide double perovskite Cs2KInI6. Using DFT/DFPT, the authors show that the cubic Fm-3m phase is dynamically unstable. They then use a genetic algorithm accelerated by a pretrained MACE machine-learned potential (MACE-OMAT-0) to search for low-energy polymorphs, reporting 42 candidate structures that are dynamically stable according to MACE finite-displacement phonon calculations. Eleven of these are selected for DFPT validation; 9 are confirmed dynamically stable, while 2 are false positives. The paper focuses on four phases (P-3, I-42m, Cmc21, P-1), reporting their structures, phonon dispersions, electronic band structures, and effective masses. The authors find that structural distortions generally widen the band gap and tend to change it from direct to indirect. They identify Cmc21 as the most stable phase (41.9 meV/atom below the cubic reference) and P-3 as the most stable perovskite-like phase, and highlight the trade-off between octahedral connectivity and electronic properties.","tokens_in":14366,"tokens_out":7042,"duration_ms":65519,"significance":"If the central claims are established, the paper offers a useful demonstration of combining a genetic algorithm with a machine-learned potential to explore the energy landscape of a dynamically unstable double perovskite, and it provides a structure-property map for Cs2KInI6. The DFPT phonon calculations for the cubic phase and the 11 selected structures are standard and credible, and the paper is transparent in reporting the two MACE false positives. However, the headline claim of '42 dynamically stable structures' is not supported: only 9 of the 42 have been confirmed by DFPT, and the MACE-vs-DFT energy ordering is not reported, so the identity of the true most-stable phase remains uncertain. These issues are load-bearing for the main message and require additional analysis or a substantial rewording of the claims.","major_comments":[{"comment":"The abstract states 'we identify 42 dynamically stable structures' and the main text states 'This process yielded 42 new polymorphs ... found to be dynamically stable based on frozen-phonon calculations using MACE.' Only 11 of these were checked with DFPT, and 2 of the 11 (id=03 and id=09 in Table S1) turned out to be dynamically unstable at the DFPT level. Thus, the claim '42 dynamically stable' is an overstatement; at most 9 are DFPT-confirmed, and the remaining 31 are only MACE-predicted stable. Given the 18% false-positive rate in the validated subset, the paper should either rephrase the claim as '42 MACE-predicted stable candidates' or validate more structures, and it should discuss the expected number of true positives among the unvalidated set.","section":"Abstract and main text (paragraph after Fig. 2; Table S1)"},{"comment":"The ΔE values in Table S1 are not labeled with the level of theory. For the 31 structures not validated by DFPT, the energies must be MACE values, but this is not stated. For the four structures in Table I, it seems the energies are from DFT, but again the method is not specified. More importantly, no MACE-vs-DFT energy comparison is provided for the 11 structures that were both predicted by MACE and evaluated by DFPT. Without such a comparison, the claim that Cmc21 (id=37) is the most stable phase at -41.9 meV/atom relative to cubic is not fully supported: if the energy ranking in Table S1 comes from MACE, another unvalidated polymorph could have a lower DFT energy. Please report both MACE and DFT ΔE for the validated set and clearly state the level of theory in both tables.","section":"Table S1 and Table I (energy columns)"},{"comment":"The selection of the 11 structures for DFPT is described only qualitatively ('all polymorphs with 10 atoms, two with 20 atoms, one with 40, and one with 80'). This selection is not random and may be biased toward low-energy or specific structural families. Since 2 of the 11 were false positives, the selection does not provide a reliable estimate of the false-positive rate across the full set of 42. The paper should either validate a more representative subset (e.g., a random sample) or at least discuss the selection bias and its implications for the '42' claim. A simple statistical argument would show that several of the 31 unvalidated structures are likely to be DFPT-unstable.","section":"Selection of 11 structures for DFPT validation (main text, after Table S1)"},{"comment":"The central output of the work is the set of 42 polymorphs and their relative stabilities, yet the crystal structures are not deposited in a public repository (e.g., CIF files), and the input/output files for the DFT/DFPT calculations are not provided. The VibroML code is mentioned only as a GitHub repository without a version or DOI. Independent verification of the claimed polymorphs, energy orderings, and phonon stabilities is therefore impossible. The authors should deposit the structures of all 42 polymorphs, the 11 DFPT-validated structures, and the relevant calculation inputs/outputs, or at minimum provide CIF files in the SI.","section":"Data availability (main text and SI)"}],"minor_comments":[{"comment":"The abstract at the top of the manuscript (as provided in the review package) states a 'calculated direct band gap of 1.94 eV', while the abstract in the full text and the band-structure plot (Fig. S10) give 1.24 eV. This numerical discrepancy must be corrected.","section":"Abstract vs. full text (first sentence)"},{"comment":"The abstract in the review package says Cmc21 'lies 13 meV/atom above the convex hull', whereas the full-text abstract says '41.9 meV/atom below the cubic reference'. These are not contradictory if the convex hull is ~55 meV/atom below the cubic phase, but they should be reconciled and both statements (or a single one) presented consistently.","section":"Abstract vs. full text (stability number)"},{"comment":"'vibrational dynamical stability' is an awkward phrase; consider simply 'dynamical stability'.","section":"Main text, first paragraph of results"},{"comment":"The sentence 'For the P-3 structure, our ChemEnv analysis gives 100% octahedral similarity with a slightly distorted octahedron for the K cations Finally, in the P-1 structure...' is missing a period after 'cations'. Also, 'a double perovskites structure' should be 'a double-perovskite structure'.","section":"Main text, paragraph on ChemEnv analysis"},{"comment":"The effective-mass table reports values for certain directions but does not explain how the effective masses were computed (e.g., parabolic fit, k·p method). Please add a brief methodological note in the text or caption.","section":"Table II"}],"recommendation":"major_revision","confidential_remarks":"The paper has a solid core of DFPT phonon calculations and a well-defined workflow, but the abstract and main text oversell the '42 dynamically stable structures' claim. The lack of a MACE-vs-DFT energy comparison is the most serious technical gap, as it directly affects the identification of the most stable phase. The abstract discrepancy (1.94 vs 1.24 eV) and the missing data deposition also need attention. I would be willing to see a revised version that addresses these points, but the current manuscript does not fully support its central quantitative claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful part of this paper is the first half: the cubic Fm-3m phase of Cs2KInI6 is genuinely dynamically unstable at the harmonic level, which corrects the implicit assumption in earlier high-throughput screens that it is a stable perovskite. The four phases they analyze in detail (P-3, I-42m, Cmc21, P-1) are all DFPT-validated, and the connection between coordination environments, octahedral connectivity, band-gap widening, and direct-to-indirect transitions is clearly laid out. That is solid, publishable work.\n\nThe soft spots are about scale and precision. The abstract and several statements say '42 dynamically stable structures' without qualification. What the paper actually shows is 42 candidates stable at the MACE level; only 11 were checked with DFPT, and two of those were false positives. The remaining 31 have no DFPT confirmation. That doesn't kill the paper—the main text is more careful—but the abstract needs to say '42 MACE-stable candidates, 11 DFPT-confirmed' or similar.\n\nThe energy ranking, including the claim that Cmc21 is the most stable phase, rests entirely on MACE. The paper never reports a MACE-vs-DFT energy comparison for the 11 validated structures, so if MACE misorders energies even slightly, a different polymorph could be lower in DFT. Given the 2/11 false-positive rate for phonons, this is a live possibility. A table comparing MACE and DFT energies for the 11 would close the gap.\n\nMinor but annoying: the arXiv abstract quotes the cubic band gap as 1.94 eV while the body and Fig. S10 give 1.24 eV. Absolute gaps are PBE only, so they shouldn't be used for PV assessment without a hybrid or SOC check. Also, no structures or code are deposited, which makes the 42-polymorph set hard to reuse independently.\n\nThese are fixable. The core instability, the four main phases, and the electronic-structure trends are all reliable. For anyone working on lead-free double perovskites, this is a useful correction and a good example of combining MLIPs with first-principles validation. Send it to peer review; a competent referee will ask for the energy comparison, the tighter abstract, and the data deposition, and the revision will be stronger for it.","headline":"Core finding (cubic Cs2KInI6 is dynamically unstable) is solid and the four highlighted polymorphs are DFPT-validated, but the '42 stable' count is a MACE claim that gets oversold in the abstract.","tokens_in":14914,"tokens_out":3189,"would_cite":true,"duration_ms":32760,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The cubic phase of the lead-free double perovskite Cs2KInI6 is dynamically unstable; a machine-learned genetic search finds 42 stable polymorphs whose distortions widen the band gap and mostly flip it from direct to indirect.","keywords":["Cs2KInI6","halide double perovskite","phonon instability","dynamical stability","polymorph discovery","genetic algorithm","machine-learned interatomic potential","band gap tuning"],"falsifier":"Synthesize Cs2KInI6 and collect a powder X-ray diffractogram: the paper's Fig. S8 shows the candidate polymorphs give clearly distinguishable diffraction patterns, so a measured pattern matching the Cmc21 or P-3 phases would confirm the central claim, while a pattern matching the cubic phase, or matching none of the predicted phases, would put it in doubt. A computational check of comparable force: run first-principles phonon calculations on the 31 candidates never validated beyond the machine-learned potential — if a substantial fraction of those also prove unstable, the '42 stable polymorphs","tokens_in":13914,"feed_emoji":"☀️","tokens_out":16945,"duration_ms":148882,"temperature":0.7,"pith_summary":"Cs2KInI6 had been flagged by earlier high-throughput screens as a lead-free halide double perovskite with a direct band gap well matched to sunlight. This paper establishes that the cubic structure those screens evaluated is dynamically unstable: its phonon spectrum contains imaginary frequencies, so the perfect double perovskite is a saddle point, not a resting state. Because unstable phonons are precisely the directions in which a crystal wants to distort, the authors seed an evolutionary search with those vibrational eigenvectors, use a machine-learned interatomic potential to relax and rank candidates, and harvest 42 polymorphs judged dynamically stable by the potential; first-principles phonon calculations on 11 of them confirm 9 and expose 2 false positives. The distortions are not cosmetic: the lowest-energy phases abandon octahedral coordination of indium, the perovskite-like phases trade 3D corner-sharing octahedra for 1D face-sharing strips, the band gap widens from 1.24 eV to as much as 2.26 eV, the direct gap usually becomes indirect, and the bands flatten. The practical conclusion is that the 'ideal' cubic material is probably not the material that would actually form, and the phase that keeps a direct gap (a low-symmetry 80-atom structure at 1.35 eV) is far removed from the structure originally screened.","feed_headline":"42 stable phases emerge from an unstable cubic perovskite","feed_subtitle":"Cs2KInI6's cubic form, once touted for solar cells, distorts into wider-gap, mostly indirect structures.","key_machinery":"The argument is carried by an instability-guided evolutionary search. First the soft (imaginary-frequency) phonon modes of the unstable cubic phase are identified; displacing atoms along those eigenvectors, at various amplitudes and in combinations, seeds an initial population of distorted crystals. A genetic algorithm then iterates: candidates are relaxed and ranked by energy with a machine-learned interatomic potential, low-energy parents breed through crossover and mutation, and each generation ends with a phonon check whose residual instabilities feed new distortion directions back into the population. This loop converts the cubic phase's instabilities into a systematic generator of lowe","core_discovery":"The central claim, stated on the paper's own terms, is that the cubic Fm-3m phase of Cs2KInI6 — the exact structure earlier high-throughput studies promoted for photovoltaics — is dynamically unstable, as shown by imaginary phonon frequencies from density functional perturbation theory. A genetic algorithm seeded with the eigenvectors of those unstable modes and evaluated with a machine-learned interatomic potential enumerates 42 dynamically stable polymorphs in 10-, 20-, 40-, and 80-atom cells; of the 11 re-examined with first-principles phonon calculations, 9 are confirmed stable and 2 are exposed as false positives. The most stable polymorph (Cmc21, 41.9 meV/atom below cubic) is not a dou","pith_inferences":["Editorial note: the abstract quotes a 1.94 eV direct gap for the cubic phase, while the body and the supplementary material consistently report 1.24 eV (Fig. S10, Table S1); the body value appears to be the operative one, but the manuscript leaves the mismatch unresolved.","Because only 11 of the 42 polymorphs received first-principles phonon checks and two of those failed, the stability census is an extrapolation; running density functional perturbation theory on the remaining 31 candidates is the direct test of the enumeration.","The instability is assessed at zero temperature; anharmonicity could heal it at operating temperatures, so finite-temperature molecular dynamics or self-consistent phonon calculations on this specific material would reveal whether the cubic form can be synthesized or stabilized — a scenario that would reframe which phase is the relevant solar-cell absorber.","The face-sharing 1D octahedral strips of the P-3 phase suggest strongly anisotropic transport; computing carrier mobility along versus across the strips would test whether the connectivity collapse is the bottleneck that the band flattening implies."],"forward_implications":["The cubic form of Cs2KInI6 that earlier screens advertised with a 1.24 eV direct gap is not a static ground state; at cryogenic temperatures, where anharmonicity cannot rescue it, the material will adopt one of the distorted polymorphs.","The lowest-energy structures are not double perovskites in the octahedral sense — indium prefers tetrahedral coordination — so the double-perovskite picture that motivated the material does not describe its ground state.","Distortions widen the band gap to 1.68–2.26 eV in the tested phases and in most cases make it indirect, so the direct-gap advantage survives only in the 80-atom P-1 phase (1.35 eV).","Flattened bands imply heavier carriers: the hole effective mass along one direction of the I-42m phase is about 16.6 electron masses, a transport penalty that follows directly from the stabilization.","The instability-seeded genetic search plus first-principles phonon validation is a transferable recipe for other halide double perovskites whose 'ideal' cubic phases may also be saddle points."],"fun_headline_variants":["Unstable cubic perovskite yields 42 stable forms","Cubic Cs2KInI6 spawns 42 stable structures","Distortion widens band gap, toppling cubic symmetry","42 stable phases hide inside an unstable cubic","Machine learning maps cubic instability to 42 phases"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the machine-learned interatomic potential faithfully represents the true energy landscape of this material: only 11 of the 42 reported stable polymorphs were checked with first-principles phonon calculations, and two of those eleven failed the check, so the stability of the other thirty-one is an extrapolation from the potential.","fun_headline_variants_meta":{"raw":{"variants":["Unstable cubic perovskite yields 42 stable forms","Cubic Cs2KInI6 spawns 42 stable structures","Distortion widens band gap, toppling cubic symmetry","42 stable phases hide inside an unstable cubic","Machine learning maps cubic instability to 42 phases"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000942,"raw_usage":{"total_tokens":3864,"prompt_tokens":746,"completion_tokens":3118,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":490,"completion_tokens_details":{"reasoning_tokens":3041}},"tokens_in":490,"tokens_out":3118,"duration_ms":23845,"temperature":1.0,"reasoning_tokens":3041,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T03:43:40.757928+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Synthesize Cs2KInI6 and collect a powder X-ray diffractogram: the paper's Fig. S8 shows the candidate polymorphs give clearly distinguishable diffraction patterns, so a measured pattern matching the Cmc21 or P-3 phases would confirm the central claim, while a pattern matching the cubic phase, or matching none of the predicted phases, would put it in doubt. A computational check of comparable force: run first-principles phonon calculations on the 31 candidates never validated beyond the machine-learned potential — if a substantial fraction of those also prove unstable, the '42 stable polymorphs","supporting_citations":[],"review_version":1}