We prove which phase is a ground state of the Ising model on with nearest-neighbour ferromagnetic coupling and an antiferromagnetic tail, among all configurations, for every except one window of width ; inside each interval it is the only translation-invariant one. The phases are the Néel state, checkerboards and stripes; ten couplings where two phases tie are certified to at least 17 decimals. We prove that the Melko–den Hertog–Gingras (MHG) state is the unique translation-invariant ground state of tin-foil Ewald-summed dipolar spin ice with nearest-neighbour exchange only, for . Numerical studies had located these phases 1, 3, 8, 10, 12, 18; we know of no earlier proof, apart from the Néel state at .
The question
The 2D model has spins on , a ferromagnetic nearest-neighbour coupling and an antiferromagnetic tail . For , the dipolar case, it describes an ultrathin magnetic film with out-of-plane anisotropy 18. The energy is
where is the ratio of Fonseca et al. 2, whose dipolar unit sums over all distinct pairs. Giuliani and Seiringer proved stripes just below the ferromagnetic transition for and wrote in 2015 that the dipolar case had "eluded any rigorous treatment so far" 4. Fermi and Giuliani proved the stripe conjecture for , with unspecified, among periodic states with straight domain walls 5.
Dipolar spin ice, the model used for DyTiO and HoTiO 3, 19, has Ising spins on the pyrochlore lattice along local axes. With , the axis, and the nearest-neighbour distance, the energy is
Here is the dipolar strength, is or for first or second neighbours, and for third neighbours if the midpoint of the pair is a lattice site and if not. favours all-in-all-out (AIAO). The nearest-neighbour couplings and of 3 obey and ; for DyTiO, K and K give K and . Energies are per spin, the dipolar sum being taken by Ewald's method with metallic (tin-foil) boundary. Monte Carlo found the MHG state, a zero-moment state with ordering vector 3; that it is the ground state had rested on numerics 10, 12.
What we found
A periodic state is a ground state if no configuration of has lower energy per site along any sequence of boxes. Images are translates, lattice rotations and reflections, and spin flips. The checkerboard is , the square wave of period , so is the Néel state; stripe is .
| Name | Phases | Images | |
|---|---|---|---|
| Néel and | 0.4152030558751828190 | 10 | |
| and | 0.43941990286594504453 | 20 | |
| and | 0.44027907033310511597 | 28 | |
| and | 0.44031096003405425963 | 36 | |
| and | 0.44031218581380430686 | 44, SAT | |
| stripes 1 and 2 | 1.2585385182612923583 | 12 | |
| stripes 2 and 3 | 2.172454495378555558 | 20 | |
| stripes 3 and 4 | 2.81449513817075321 | 28 | |
| stripes 4 and 5 | 3.3042419453427474952 | 36 | |
| stripes 5 and 6 | 3.69902470137631926895 | 44, SAT |
Theorem 1 (Ground states of the 2D model). Let and . (a) For the checkerboard is a ground state for , and is one at . (b) Stripe 1 is a ground state for , and stripe for , . (c) In the interior of each interval of (a) and (b), the translation-invariant ground states are exactly the mixtures of the images of that phase. (d) At and both neighbouring phases are ground states, is the crossing of their energy lines, and the translation-invariant ground states are the mixtures of their images; for and the last assertion rests on the SAT enumeration.
Theorem 2 (MHG ground state). For and every , the MHG state is the unique translation-invariant ground state, up to its 12 images. The same holds for with , and . At the energy is with , enclosed to , the dipolar part of the eigenvalue defined below. At , the rounded DyTiO ratio of the nearest-neighbour model, K.
Proposition 3 (AIAO crossing). For , with , that is . For the AIAO state is lower than every MHG state.
Proposition 4 (Double-chain comparison state). Let S be one fixed ice-rule configuration of 64 spins on a block of cubic cells, with Fourier amplitudes at and its images; it is not published. Then in units of , with , independent of and . Where , S is below MHG, so MHG is not a ground state there.
Result 5 (Energy window). At , rounded from Yavors'kii et al. with K 14 (not 1.41 K), the lowest energy per spin of a translation-invariant state obeys to . The width is , or 8.9 mK per spin; the upper end is the MHG energy. The ground state there is undecided: S is 11.3 mK per spin above MHG, and no ice-rule state in a cell of up to 32 spins is lower by more than (Exhaustive computation).
Why it is true
Dual. Translation-invariant (TI) states suffice, since . A TI state has correlations with of mass 1 on the torus. For real , one per class of displacements under the lattice symmetries, set , with and the cosine sums of and of . Then , with and the correlation on . A slack on all sign patterns of a small window gives . If then for every TI state, with equality for a periodic state when on the windows of its images and at its stars, the atoms of its spectrum.
Crossing and uniqueness. Two phases share one dual when with zero gradient at all their stars: a square linear system in the dual parameters and . Both energies then equal at the solved , the crossing of their energy lines. The cover proves off the stars, so equality forces onto the stars and on all windows of the images, and a finite enumeration returns the images. Energies are affine in , so a ground state at two couplings is one between them, and the only TI one inside if the enumeration holds at one end.
Spin ice. The energy per spin is with of trace 4 and a real symmetric Bloch matrix. Hence the Luttinger–Tisza (LT) bound 11: if for all then , with equality only on the kernels. Let be the eigenvalue at on . Of the Ising states of the 16-site cubic cell, 12 have Fourier amplitudes in the kernels at and its images and zero at the origin; they obey the ice rule (two spins in, two out of every tetrahedron) and attain . So the bound is tight, as 10 observed numerically, and an atomic spectrum forces periodicity and uniqueness. The computer-assisted step is on the whole Brillouin zone: Ewald sums in ball arithmetic with explicit tails, a cover by boxes closed by a second-order matrix lemma (the box lemma), and a separate lemma at , where the minimum is degenerate. Lowering raises , since for the exchange matrix , with equality on at ; so covers all smaller , and affinity makes the prism cost its eight vertices.
How it was checked
Theorems 1 and 2 and the lower end of Result 5 are computer-assisted proofs, Propositions 3 and 4 certified computations, the enumerations exhaustive computations. The 2D checker evaluates the dual slack on all sign patterns of the small window, exactly on the windows of the two phases and in ball arithmetic elsewhere (for the larger windows after a floating-point screen with a rigorous error bound), and the star system and the cover of in ball arithmetic. A second program, written from the proof without the checker's code, checked the slack, the star system and positivity with its own cover, which for the four staircase couplings uses floating point with an explicit margin on one region. Bessel and Hurwitz-zeta sums for the energy lines reproduced all ten couplings. A second reader re-derived the lemmas without finding a gap; the second program shares the statement of the method. Tampered certificates, and certificates re-used past a crossing, were rejected, except modifications that are valid alternative duals. The enumerations were repeated with CP-SAT and, at to and , a second SAT solver. Shared trust is python-flint, mpmath and the solvers.
For spin ice the cover was re-run in full for , the prism and the window, with code written from the proof. A 45-digit Ewald evaluation reproduced and to every printed digit, and a damped direct sum without Ewald splitting reproduced to . Separately written floating-point code reproduced to 13 digits. Both covers follow the proof's route, so a flaw in the box lemma would be shared. The proofs have not been peer reviewed.
What it does not show
The 2D results are for classical Ising spins at zero temperature and field, not for films. Uniqueness is among TI states only. Open are the window and , except that stripe 7 is certified as a ground state on and stripe 8 at .
For spin ice the energy is the tin-foil Ewald energy of TI states; samples of definite shape are not treated for states with net moment, which MHG and AIAO lack. Temperature, quantum effects and real materials are not addressed. The interval has no certificate. The curvature of the lowest LT band at vanishes at and for and (numerical); the proof stops near on the negative side. S is one configuration, identified with the double-chain state of 13 by propagation vector only, and not shown to be a ground state anywhere; identifying with of 13 is an inference.
Earlier and parallel work
2D model. At the Néel state is known from reflection positivity 6, 7. Giuliani, Lebowitz and Lieb 6 proved matching large- energy asymptotics for , without the ground state. MacIsaac et al. 8 found stripe ground states numerically; Cannas et al. 9 quote the transitions between stripes 1 and 2, 2 and 3, 3 and 4 near 1.26, 2.2 and 2.8, in the convention of 2, and a Néel-to-stripe threshold near 0.425, inside our interval; Pighin and Cannas 18 repeat it and mark the stripe crossings in a figure without printed values. Rastelli, Regina and Tassi 1 reported the checkerboards , our . Fonseca, Rizzi and Alves 2 give the ranges (Néel), , (stripe 1) and (stripe 2) with data from 1; 0.4403 appears earlier in 20. Our , and agree with the digits printed in 2. We found no printed value for , or , but read only the abstract of 1 and could not read 8, 17, the 1995 Booth et al. letter cited in 9 or the 2006 Rastelli et al. paper cited in 2; they may have some. We add the proof that each phase is a ground state among all configurations, the intervals of Theorem 1 and the certified digits.
Spin ice. Den Hertog and Gingras 19 put the boundary to AIAO at about by Monte Carlo, and Melko et al. 12 at by mean-field theory and Ewald energies. Gingras and den Hertog 21 had located a weak maximum of the mean-field spectrum at numerically; 12 also has the soft mode with its two-in two-out eigenvectors and the mean-field order at that vector for . McClarty et al. 10 showed numerically, for and , that the lowest LT eigenvectors at are Ising-realisable, so MHG is a ground state for . They also have the MHG and double-chain energies, the boundary and on 128 and 1024 sites, and DyTiO with the parameters of 14 in the MHG phase. Henelius et al. 13 gave the approximate boundary . We add the proof, for and on the prism including positive , the energy and crossing to 17 or more digits, the 13-digit coefficients of Proposition 4 with separated from , and the window. Paul and Refael 16 give a numerically solved semidefinite hierarchy of bounds that contains the LT bound; Shah et al. 15 used the LT bound with an Ising-realisable minimiser for a pyrochlore Rydberg-array model with a tail, without interval arithmetic.
Searches of arXiv, Zenodo, Crossref and the web on 9 and 15 November 2026 found no earlier proof or certified bound.
References
- 1.E. Rastelli, S. Regina and A. Tassi, "Phase diagram of a square Ising model with exchange and dipole interactions: Monte Carlo simulations", Phys. Rev. B 76, 054438 (2007). doi.org/10.1103/PhysRevB.76.054438
- 2.J. S. M. Fonseca, L. G. Rizzi and N. A. Alves, "Stripe-tetragonal phase transition in the two-dimensional Ising model with dipole interactions: Partition function zeros approach", Phys. Rev. E 86, 011103 (2012), arXiv:1206.4501. arxiv.org/abs/1206.4501
- 3.R. G. Melko, B. C. den Hertog and M. J. P. Gingras, "Long-range order at low temperatures in dipolar spin ice", Phys. Rev. Lett. 87, 067203 (2001), arXiv:cond-mat/0009225. arxiv.org/abs/cond-mat/0009225
- 4.A. Giuliani and R. Seiringer, "Periodic striped ground states in Ising models with competing interactions", Commun. Math. Phys. 347, 983 (2016), arXiv:1509.00057. arxiv.org/abs/1509.00057
- 5.D. Fermi and A. Giuliani, "Periodic striped states in Ising models with dipolar interactions", in The Physics and Mathematics of Elliott Lieb, EMS Press (2022), pp. 269–293, arXiv:2203.01249. arxiv.org/abs/2203.01249
- 6.A. Giuliani, J. L. Lebowitz and E. H. Lieb, "Ising models with long-range antiferromagnetic and short-range ferromagnetic interactions", Phys. Rev. B 74, 064420 (2006), arXiv:cond-mat/0604668. arxiv.org/abs/cond-mat/0604668
- 7.A. Giuliani, J. L. Lebowitz and E. H. Lieb, "Checkerboards, stripes, and corner energies in spin models with competing interactions", Phys. Rev. B 84, 064205 (2011), arXiv:1106.0922. arxiv.org/abs/1106.0922
- 8.A. B. MacIsaac, J. P. Whitehead, M. C. Robinson and K. De'Bell, "Striped phases in two-dimensional dipolar ferromagnets", Phys. Rev. B 51, 16033 (1995). doi.org/10.1103/PhysRevB.51.16033
- 9.S. A. Cannas, P. M. Gleiser and F. A. Tamarit, "Two dimensional Ising model with long-range competing interactions", Recent Research Developments in Physics 5, 751 (2004), arXiv:cond-mat/0502403. arxiv.org/abs/cond-mat/0502403
- 10.P. A. McClarty, O. Sikora, R. Moessner, K. Penc, F. Pollmann and N. Shannon, "Chain-based order and quantum spin liquids in dipolar spin ice", Phys. Rev. B 92, 094418 (2015), arXiv:1410.0451. arxiv.org/abs/1410.0451
- 11.J. M. Luttinger and L. Tisza, "Theory of dipole interaction in crystals", Phys. Rev. 70, 954 (1946). doi.org/10.1103/PhysRev.70.954
- 12.R. G. Melko, M. Enjalran, B. C. den Hertog and M. J. P. Gingras, "Thermodynamic properties of the dipolar spin ice model", arXiv:cond-mat/0308282 (2003). The arXiv listing gives J. Phys.: Condens. Matter 16, R1277 (2004), whose Crossref record is R. G. Melko and M. J. P. Gingras, "Monte Carlo studies of the dipolar spin ice model"; we read the arXiv version only. arxiv.org/abs/cond-mat/0308282
- 13.P. Henelius, T. Lin, M. Enjalran, Z. Hao, J. G. Rau, J. Altosaar, F. Flicker, T. Yavors'kii and M. J. P. Gingras, "Refrustration and competing orders in the prototypical Dy2Ti2O7 spin ice material", Phys. Rev. B 93, 024402 (2016), arXiv:1512.05361. arxiv.org/abs/1512.05361
- 14.T. Yavors'kii, T. Fennell, M. J. P. Gingras and S. T. Bramwell, "Dy2Ti2O7 spin ice: a test case for emergent clusters in a frustrated magnet", Phys. Rev. Lett. 101, 037204 (2008), arXiv:0707.3477. arxiv.org/abs/0707.3477
- 15.J. Shah, G. Nambiar, A. V. Gorshkov and V. Galitski, "Quantum spin ice in three-dimensional Rydberg atom arrays", Phys. Rev. X 15, 011025 (2025), arXiv:2301.04657. arxiv.org/abs/2301.04657
- 16.N. Paul and G. Refael, "Bootstrapping ground state properties of classical frustrated magnets", arXiv:2605.06784 (2026). arxiv.org/abs/2605.06784
- 17.K. De'Bell, A. B. MacIsaac and J. P. Whitehead, "Dipolar effects in magnetic thin films and quasi-two-dimensional systems", Rev. Mod. Phys. 72, 225 (2000). doi.org/10.1103/RevModPhys.72.225
- 18.S. A. Pighín and S. A. Cannas, "Phase diagram of an Ising model for ultrathin magnetic films: comparing mean field and Monte Carlo predictions", Phys. Rev. B 75, 224433 (2007), arXiv:cond-mat/0611260. arxiv.org/abs/cond-mat/0611260
- 19.B. C. den Hertog and M. J. P. Gingras, "Dipolar interactions and origin of spin ice in Ising pyrochlore magnets", Phys. Rev. Lett. 84, 3430 (2000), arXiv:cond-mat/0001369. arxiv.org/abs/cond-mat/0001369
- 20.L. G. Rizzi and N. A. Alves, "Phase transitions and autocorrelation times in two-dimensional Ising model with dipole interactions", arXiv:0903.4084 (2009). arxiv.org/abs/0903.4084
- 21.M. J. P. Gingras and B. C. den Hertog, "Origin of spin-ice behavior in Ising pyrochlore magnets with long-range dipole interactions: an insight from mean-field theory", Can. J. Phys. 79, 1339 (2001), arXiv:cond-mat/0012275. arxiv.org/abs/cond-mat/0012275