Pith. sign in

REVIEW 2 major objections 6 minor 34 references

How many ways a GaN surface can be tiled predicts which adatom arrangement is most stable.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 07:01 UTC pith:LQ57NF3W

load-bearing objection Exhaustive EC catalog plus a geometry-only multiplicity that actually ranks stability within Ga classes; the rhombus-block restriction is real but already scoped by the authors. the 2 major comments →

arxiv 2607.11105 v1 pith:LQ57NF3W submitted 2026-07-13 cond-mat.mtrl-sci physics.comp-ph

Tiling decomposition multiplicity predicts stability of GaN(0001) surface reconstructions

Tetsuji Kuboyama , Akira Kusaba , Karol Kawka , Pawel Kempisty This is my paper
classification cond-mat.mtrl-sci physics.comp-ph
keywords GaN(0001)surface reconstructionelectron counting ruletiling decomposition multiplicityexhaustive enumerationadatom configurationmachine-learning interatomic potential
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Finding the stable way Ga and H atoms sit on a GaN surface has usually meant sampling a huge configuration space with first-principles energies. This paper shows that, under the electron-counting rule on a (6 imes6) cell with fixed stoichiometry, the problem can be rewritten as an exhaustive discrete tiling task. The authors enumerate every rhombus tiling of the lattice and every compatible adatom pattern on top of it, producing a complete catalog of 416,683 configurations grouped into 14 Ga-placement classes. Within each class, the configuration that admits the largest number of compatible tilings—the tiling decomposition multiplicity n_til—is the most stable. The rule holds strictly in 13 of 14 classes; in the remaining class the n_til-max configuration sits only 8.5 meV above the true minimum, a gap that is negligible at growth temperature and is reproduced by independent DFT. The local mechanism is the avoidance of adjacent bare surface sites. Enumeration therefore supplies a coverage guarantee that sampling cannot, and shrinks the set that needs first-principles ranking from hundreds of thousands of structures to 24.

Core claim

Within each of the 14 symmetry classes of Ga adatom placements on the GaN(0001)-(6 imes6) surface at fixed electron-counting stoichiometry (3 Ga adatoms and 18 H atoms), the configuration that maximizes the number of compatible rhombus tilings, n_til, is the most stable. The rule is strict in 13 classes; in the remaining class the n_til-max configuration lies only 8.5 meV (MLIP) / 17.5 meV (DFT) above the true minimum. Exhaustive enumeration of all 416,683 EC-compatible configurations makes the claim free of sampling incompleteness and reduces the first-principles candidate set to 24 structures.

What carries the argument

Tiling decomposition multiplicity n_til: the number of distinct rhombus tilings of the (6 imes6) torus that are compatible with a given adatom configuration under the local electron-counting rule. It is an enumerative invariant computed solely from geometry and the 456 tilings; higher n_til within a Ga-placement class predicts lower energy.

Load-bearing premise

The count of tilings is defined only for one fixed rhombus block shape taken from the two standard (2 imes2) reconstructions; if other electron-counting local units matter for stability, the ranking and the shortlist can change.

What would settle it

On a larger cell such as (12 imes12), or at another electron-counting stoichiometry, enumerate (or fully sample) the EC-compatible space, group by Ga-placement class, and check whether the configuration that maximizes n_til under the same rhombus model is still the lowest-energy member of each class; a clear, reproducible inversion outside the single documented 8.5 meV exception would falsify the rule.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. The manuscript recasts the search for stable GaN(0001)-(6×6) adatom reconstructions under the electron-counting (EC) rule at fixed stoichiometry (3 Ga adatoms, 18 H) as an exhaustive discrete tiling problem. All 456 rhombus tilings of the torus are enumerated and reduced to 8 symmetry classes; EC-compatible colorings on those tilings, after full G′ reduction, yield a catalog of 416,683 configurations in 14 Ga placement classes. The authors define the tiling decomposition multiplicity n_til of a configuration and report an intra-class rule: the configuration maximizing n_til is the most stable. The rule holds strictly in 13 of 14 classes; in Class 2 the n_til-max structure lies 8.5 meV (MLIP) / 17.5 meV (DFT) above the true minimum, a gap negligible at growth temperature. Screening uses a UMA MLIP validated against 685 fixed-placement DFT structures and 25 cross-class DFT structures; the rule reduces the first-principles candidate set from 416,683 to 24 configurations, all DFT-evaluated. A local frustration count n_adj (adjacent bare-site pairs) is identified as the principal energy-carrying mechanism, with tileability itself carrying an additional ~100 meV cost near the minima.

Significance. If the result holds within its stated model, the work is a genuine methodological advance for surface reconstruction prediction: it replaces DFT-steered sampling with a coverage-guaranteed combinatorial catalog and defers first-principles input to a final ranking of 24 candidates. Strengths that should be credited explicitly include (i) exact enumeration with Burnside consistency checks and an openly released catalog of 416,683 configurations with identifiers, coordinates, and n_til; (ii) dual independent DFT validation (RS-DFT fixed-placement set and SIESTA cross-class set) that preserves within-class orderings; (iii) honest treatment of the single Class-2 exception with DFT confirmation; and (iv) a clear local mechanism (n_adj) that largely mediates the n_til–energy correlation. The structural parallel to Kekulé counts is interpretive rather than essential, but the reduction from hundreds of thousands of candidates to a DFT-tractable shortlist is of practical value for III-nitride growth modeling.

major comments (2)
  1. [Sec. II B; Abstract; Sec. V] Sec. II B and the abstract/conclusions: the phrases “complete catalog,” “every EC-compatible configuration,” and “coverage guarantee” are accurate only inside the modeling choice that EC decompositions are restricted to nine non-overlapping 4+4 rhombi of the three orientations that encode Ga-adatom (2×2) and 3Ga-H (2×2) units. The manuscript states this restriction explicitly (“other EC-satisfying local units are not considered”), yet the central practical claim—that the shortlist of 24 is exhaustive for first-principles evaluation—depends on it. Configurations that are stable via other local EC motifs would be labeled n_til = 0 or fall outside the 14 classes and would never enter the shortlist. The higher energies of the 88 untileable n_adj = 0 arrangements (Sec. III F) mitigate but do not close this gap, because those checks still use the same rhombus definition of tileability. The aut
  2. [Sec. III D; Table III; Sec. III C] Sec. III D and Table III: in seven of the 13 rule-conforming classes the margin of the n_til-max configuration over the next competitor is 16–48 meV, comparable to the calibrated low-energy MLIP MAE (~16–27 meV after scale correction; Sec. III C). The manuscript correctly notes that identity of those class minima therefore rests on the DFT evaluation of the 24 candidates. That evaluation is reported only as “in 13 of 14 classes the lowest DFT energy computed for the class is attained by an n_til-max candidate.” For the seven tight-margin classes, the paper should state explicitly the DFT energy gaps to the nearest non-max competitors that were relaxed (or acknowledge that only n_til-max candidates were DFT-evaluated in those classes). Without that, the load-bearing claim that DFT confirms the rule in those classes is under-specified.
minor comments (6)
  1. [Fig. 10] Fig. 10: the light-gray abundance bars are log-scaled and panel-normalized and “do not refer to the energy axis,” but this is easy to misread against the dense scatter. A brief axis annotation or a separate abundance panel would help.
  2. [Sec. II A, Eq. (1)] Eq. (1) and the stoichiometry line: it would help readers outside the GaN community to state once, near Eq. (1), that bare-face count is fixed at 9 for every EC composition, since n_adj is defined on those bare sites.
  3. [Sec. III E] Sec. III E: the three geometric descriptors fail globally; the text already says so. Consider moving the failed I_iso / V_HH / d_GaH correlations to a short SI table so the main text can go directly from n_til to n_adj.
  4. [Sec. IV B] Sec. IV B: the ln n_til vs energy slope (−195 meV per e-fold) is interesting but secondary; the Kekulé analogy is already clear without the functional-form comparison, which is only marginally stronger (r = −0.60 vs −0.54).
  5. [Data availability] Data availability: the GitHub URL is given; please also deposit a frozen snapshot (Zenodo or similar) with a DOI so the 416,683-entry catalog remains citable if the repository moves.
  6. [Abstract; Sec. III D–F] Typographical: “itstiling” in the abstract (missing space); “then til-max” / “then adj” spacing inconsistencies appear in several places in the compiled text (e.g., near Table III and Sec. III F).

Circularity Check

0 steps flagged

No circularity: n_til is a pure combinatorial count independent of energy; the n_til-max rule is an observed correlation against independent MLIP/DFT rankings.

full rationale

The derivation chain is self-contained and non-circular. n_til is defined combinatorially (Eq. 2) solely by testing each of the 416,683 EC-compatible (G,H) configurations against the fixed set of 456 rhombus tilings of the adatom-free lattice; the count uses no energy model, no fitted parameters, and no DFT/MLIP input. Energies used to rank configurations and to verify the rule are obtained independently: UMA MLIP relaxations (validated on 710 DFT structures from two codes, RS-DFT and SIESTA) plus direct DFT on the 24 n_til-max candidates and the Class-2 exception. The n_til-max rule is an empirical intra-class observation (holds strictly in 13/14 classes; 8.5 meV MLIP / 17.5 meV DFT gap in the remaining class), not a quantity forced by normalization, by construction of the catalog, or by any self-cited uniqueness theorem. Prior self-citations ([6,7]) supply the Bayesian-optimization sampling database used only for MLIP validation and the Ising-model comparison; they are not load-bearing for the enumeration, the definition of n_til, or the ranking rule itself. The explicit modeling restriction to a single rhombus block shape (Sec. II B) is a stated scope limitation, not a circular step that equates the prediction to its inputs. Cell-averaged geometric descriptors and the later n_adj count are checked against the same independent energies and do not redefine n_til. No fitted-input-called-prediction, self-definitional loop, or ansatz smuggled via citation appears in the load-bearing chain.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 2 invented entities

The central claim is an empirical combinatorial correlation on a fixed lattice model. It inherits the electron-counting rule and the (2×2)-anchored rhombus block geometry from prior surface-science literature, uses a pretrained MLIP for screening, and introduces two new combinatorial descriptors (n_til, n_adj). No free parameters are fitted to produce the n_til-max rule itself; energies enter only as an independent ranking oracle.

free parameters (2)
  • UMA MLIP relative-energy scale factor (~1.17–1.18 vs DFT)
    Observed post-hoc linear compression of MLIP energies relative to two DFT setups; not used to define n_til, but affects quantitative meV gaps quoted from MLIP. Rankings are invariant under positive rescaling.
  • Stratified-sample composite score weights (n_til + I_iso + V_HH + d_GaH)
    Used only to choose which of the 416k configurations to relax with the MLIP; the paper checks that within-class correlations survive median-per-n_til and random-per-n_til resampling, so the rule is not an artifact of these weights.
axioms (5)
  • domain assumption The electron-counting rule (global balance 3 n_Ga + n_H = 27 on the 36-face cell) is the correct stoichiometric constraint for stable MOVPE GaN(0001) reconstructions.
    Taken from the established III-nitride surface literature (Pashley, Kusaba et al.) and used to define the configuration space in Sec. II A.
  • ad hoc to paper EC decompositions may be restricted to nine non-overlapping rhombus blocks of the three orientations that encode the Ga-adatom (2×2) and 3Ga-H (2×2) local units.
    Explicit modeling choice in Sec. II B; other EC-satisfying local units are excluded by construction and therefore do not contribute to n_til.
  • domain assumption The sublattice-preserving symmetry group G′ = T ⋊ D3 (|G′|=216) is the correct equivalence for physically distinct configurations.
    Ga and H occupy distinct sublattices, so sublattice-exchanging operations of the full D6 are discarded (Sec. II A).
  • domain assumption The UMA MLIP preserves within-class energy orderings of low-energy EC-compatible configurations sufficiently for screening.
    Justified by the two DFT validation sets in Sec. II E / III C (Spearman ρ ≥ 0.85 in low-energy windows; exact within-class order agreement on multi-config classes).
  • standard math Standard group-action orbit counting (Burnside) and exact-cover SAT enumeration correctly generate the tiling and configuration catalogs.
    Used in Sec. II C; orbit counts are cross-checked by Burnside’s lemma.
invented entities (2)
  • tiling decomposition multiplicity n_til independent evidence
    purpose: Combinatorial stability descriptor: number of rhombus tilings compatible with a fixed adatom configuration (G,H).
    Defined in Eq. (2); the paper’s central predictor. Independent handle is the exhaustive catalog and the DFT/MLIP energy correlation; not a new physical field or particle.
  • local frustration count n_adj independent evidence
    purpose: Number of nearest-neighbor pairs of bare surface sites; proposed local mechanism mediating the n_til–energy correlation.
    Defined in Eq. (5); emerges from analysis of why n_til works. Correlates with both MLIP and DFT energies; still a combinatorial count on the same lattice model.

pith-pipeline@v1.1.0-grok45 · 27667 in / 3695 out tokens · 48868 ms · 2026-07-14T07:01:12.196340+00:00 · methodology

0 comments
read the original abstract

The stable adatom configurations of a semiconductor surface have traditionally been sought by sampling: density functional theory (DFT) energies steer a heuristic or Bayesian search through a configuration space far too large to cover. Here we show that, for the GaN(0001)-$(6\times6)$ surface under the electron counting (EC) rule, the search can instead be posed as a discrete tiling problem and solved exhaustively. Enumerating all rhombus tilings of the surface lattice, together with all EC-compatible adatom arrangements built on them, yields the complete catalog of 416,683 configurations at fixed stoichiometry (3 Ga adatoms and 18 H atoms), organized by symmetry into 14 Ga placement classes. The number of tilings compatible with a given configuration, its tiling decomposition multiplicity $n_\mathrm{til}$, predicts stability. Within each class, the configuration maximizing $n_\mathrm{til}$ is the most stable. The rule holds strictly in 13 of the 14 classes; in the remaining class the minimum is itself among the highest-multiplicity configurations, with the $n_\mathrm{til}$-max configuration only 8.5 meV above it; this ordering is reproduced by independent DFT calculations, and the difference is negligible at growth temperature. Stability screening uses a machine-learning interatomic potential validated against 710 DFT-computed structures. The rule reduces the candidate set for first-principles evaluation from 416,683 to 24 configurations, all of which have been evaluated with DFT. Analysis of the rule identifies the local mechanism, the avoidance of adjacent bare surface sites, while the existence of a compatible tiling remains a separate requirement with an energy cost of its own. Enumeration thus provides what sampling cannot: a coverage guarantee, and a route to stable-structure prediction in which first-principles input enters only at the final ranking step.

Figures

Figures reproduced from arXiv: 2607.11105 by Akira Kusaba, Karol Kawka, Pawel Kempisty, Tetsuji Kuboyama.

Figure 1
Figure 1. Figure 1: FIG. 1. From the atomic arrangement to the puzzle model. (a) Atomic arrangement of an EC-compatible (Sec. II B) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Symmetry operations of the sublattice-preserving [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. All eight rhombus tilings compatible with a single EC-compatible configuration ( [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. The enumeration procedure (Sec. II C). Boxes give [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Representatives of the 8 tiling classes of the adatom-free (6 [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Canonical representatives of the 14 Ga placement classes (Table II): the three Ga adatom positions (dark red [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Distribution of the tiling decomposition multiplicity [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. MLIP versus DFT relative energies for the two validation sets of Sec. II E. The dashed lines indicate perfect agreement. [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. MLIP energy versus tiling decomposition multiplicity [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11. Energy versus the local frustration count [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references

  1. [1]

    Zywietz, J

    T. Zywietz, J. Neugebauer, and M. Scheffler, Appl. Phys. Lett.73, 487 (1998)

    1998

  2. [2]

    C. G. Van de Walle and J. Neugebauer, J. Vac. Sci. Tech- nol. B20, 1640 (2002)

    2002

  3. [3]

    Neugebauer, T

    J. Neugebauer, T. K. Zywietz, M. Scheffler, J. E. Northrup, H. Chen, and R. M. Feenstra, Phys. Rev. Lett. 90, 056101 (2003)

    2003

  4. [4]

    Akiyama and T

    T. Akiyama and T. Kawamura, Cryst. Growth Des.24, 5906 (2024)

    2024

  5. [5]

    Kangawa, A

    Y. Kangawa, A. Kusaba, T. Kawamura, P. Kempisty, K. Ishisone, and M. Boero, Cryst. Growth Des.25, 740 (2025)

    2025

  6. [6]

    Kusaba, Y

    A. Kusaba, Y. Kangawa, T. Kuboyama, and A. Os- hiyama, Appl. Phys. Lett.120, 021602 (2022)

    2022

  7. [7]

    Kawka, P

    K. Kawka, P. Kempisty, K. Sakowski, S. Krukowski, M. Bo´ ckowski, D. Bowler, and A. Kusaba, J. Appl. Phys. 135, 225302 (2024)

    2024

  8. [8]

    M. D. Pashley, Phys. Rev. B40, 10481 (1989)

    1989

  9. [9]

    Zhang, E

    L. Zhang, E. G. Wang, Q. K. Xue, S. B. Zhang, and Z. Zhang, Phys. Rev. Lett.97, 126103 (2006)

    2006

  10. [10]

    Kangawa, T

    Y. Kangawa, T. Akiyama, T. Ito, K. Shiraishi, and T. Nakayama, Materials6, 3309 (2013)

    2013

  11. [11]

    Kusaba, Y

    A. Kusaba, Y. Kangawa, P. Kempisty, H. Valencia, K. Shiraishi, Y. Kumagai, K. Kakimoto, and A. Koukitu, Jpn. J. Appl. Phys.56, 070304 (2017)

    2017

  12. [12]

    W. C. Herndon, J. Am. Chem. Soc.95, 2404 (1973)

    1973

  13. [13]

    Swinborne-Sheldrake, W

    R. Swinborne-Sheldrake, W. C. Herndon, and I. Gutman, Tetrahedron Lett.16, 755 (1975)

    1975

  14. [14]

    Randi´ c, Chem

    M. Randi´ c, Chem. Rev.103, 3449 (2003)

    2003

  15. [15]

    Mora-Fonz, T

    D. Mora-Fonz, T. Lazauskas, M. R. Farrow, C. R. A. Catlow, S. M. Woodley, and A. A. Sokol, Chem. Mater. 29, 5306 (2017)

    2017

  16. [16]

    G. L. W. Hart and R. W. Forcade, Phys. Rev. B77, 224115 (2008)

    2008

  17. [17]

    A. R. Smith, R. M. Feenstra, D. W. Greve, J. Neuge- bauer, and J. E. Northrup, Phys. Rev. Lett.79, 3934 (1997)

    1997

  18. [18]

    Rapcewicz, M

    K. Rapcewicz, M. Buongiorno Nardelli, and J. Bernholc, Phys. Rev. B56, R12725 (1997)

    1997

  19. [19]

    Fritsch, O

    J. Fritsch, O. F. Sankey, K. E. Schmidt, and J. B. Page, Phys. Rev. B57, 15360 (1998)

    1998

  20. [20]

    Audemard and L

    G. Audemard and L. Simon, inProceedings of the 21st International Joint Conference on Artificial Intelligence (IJCAI-09)(2009) pp. 399–404

    2009

  21. [21]

    Kuboyama and A

    T. Kuboyama and A. Kusaba, J. Cryst. Growth650, 127927 (2025)

    2025

  22. [22]

    Hjorth Larsenet al., J

    A. Hjorth Larsenet al., J. Phys.: Condens. Matter29, 273002 (2017)

    2017

  23. [23]

    B. M. Wood, M. Shuaibi,et al., UMA: A family of univer- sal models for atoms (2025), arXiv:2506.23971 [cs.LG]

    arXiv 2025

  24. [24]

    Iwata, D

    J.-I. Iwata, D. Takahashi, A. Oshiyama, T. Boku, K. Shi- raishi, S. Okada, and K. Yabana, J. Comput. Phys.229, 2339 (2010)

    2010

  25. [25]

    J. M. Soler, E. Artacho, J. D. Gale, A. Garc´ ıa, J. Jun- quera, P. Ordej´ on, and D. S´ anchez-Portal, J. Phys.: Con- dens. Matter14, 2745 (2002)

    2002

  26. [26]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett.77, 3865 (1996)

    1996

  27. [27]

    L. S. Pedroza, A. J. R. da Silva, and K. Capelle, Phys. Rev. B79, 201106 (2009)

    2009

  28. [28]

    M. M. Odashima, K. Capelle, and S. B. Trickey, J. Chem. Theory Comput.5, 798 (2009)

    2009

  29. [29]

    Duzik, J

    A. Duzik, J. C. Thomas, A. Van der Ven, and J. M. Millunchick, Phys. Rev. B87, 035313 (2013)

    2013

  30. [30]

    Kusaba, PyAPX: Python toolkit for atomic configu- ration pattern exploration,https://github.com/a-ksb/ PyAPX(2025)

    A. Kusaba, PyAPX: Python toolkit for atomic configu- ration pattern exploration,https://github.com/a-ksb/ PyAPX(2025)

    2025

  31. [31]

    Kawahara, T

    J. Kawahara, T. Inoue, H. Iwashita, and S. Minato, IE- ICE Trans. FundamentalsE100.A, 1773 (2017)

    2017

  32. [32]

    B. D. McKay, J. Algorithms26, 306 (1998)

    1998

  33. [33]

    A. R. Smith, R. M. Feenstra, D. W. Greve, M.-S. Shin, M. Skowronski, J. Neugebauer, and J. E. Northrup, J. Vac. Sci. Technol. B16, 2242 (1998)

    1998

  34. [34]

    V. M. Bermudez, Surf. Sci. Rep.72, 147 (2017)

    2017