{"id":"236ac7c7-0618-42d7-890d-e3032f0fdb31","arxiv_id":"2502.08396","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For periodic tilings of the plane with two different tile areas, the minimal-interface configurations are exactly three shapes: a pair of hexagons, a curved rectangle with a chipped parallelogram, or a Reuleaux triangle with a chipped hexagon.","lead":"This mathematics paper classifies the most efficient ways to tile the plane with two periodic shapes of different areas, finding exactly three optimal configurations depending on the area ratio. It also computes the exact minimal interface length for any ratio, including the two transition points where the optimal shape changes.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The classification rests on Theorem 2.1(4), whose proof for the fixed-lattice problem moves a G-invariant family by an arbitrary translation, which is not an admissible competitor; this gap could allow disconnected minimizers outside the three configurations.","rationale":"The reader's weakest_assumption is exactly the fixed-lattice regularity extension in Theorem 2.1, and my reading of the proof of property (4) identifies a concrete defect: the 'translation' of a G-invariant family is not a valid competitor for the fixed-lattice variational problem unless the translation is in G, in which case nothing changes. This makes the connectedness of the interface genuinely unproven, and the later classification depends on that connectedness through Proposition 2.2. If disconnected interfaces or non-simply-connected components were possible, the three-configuration dichotomy would not exhaust all minimizers, and the explicit profile I(x) would miss additional branches. The paper has independent support in the explicit geometric computations and the three candidate configurations are natural, but the regularity foundation is imported from a variable-lattice setting and the fixed-lattice extension is the point where correctness risk concentrates. Other issues I noticed are less consequential: the proof of Theorem 3.5 contains a typo in case (5) where it states I(x)=q2 instead of q3, and Theorem 3.4's Step 2 relies on a geometric assertion about signed distances between an ellipse and a hyperbola that is stated without proof. Neither of these threatens the central claim as directly as the regularity gap. Since my concern is the same as the reader's and does not change the conditional verdict, no verdict adjustment is needed.","tokens_in":21338,"tokens_out":3586,"duration_ms":42638,"concrete_test":"Re-derive Theorem 2.1(4) for the fixed-lattice problem (2.1) without invoking translations by vectors outside G. Concretely: take a connected component Γ of ∂T, let C be the union of all tile components enclosed by Γ together with their G-translates, and try to construct an admissible cut-and-paste inside a fundamental domain D that preserves |E1| and |E2| and strictly decreases perimeter whenever ∂T is disconnected. If such an exchange cannot be constructed, run a numerical search for fixed-lattice minimizers—for example a phase-field or Surface Evolver computation on a fixed square lattice with area fraction x=0.1—and check whether the minimizer exhibits a disconnected E2 or a non-simply-connected tile; either outcome would refute Theorem 1.1 as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central classification (Theorem 1.1) and the explicit profile (Theorem 1.2) both depend on Theorem 2.1, which imports regularity from [30] and adds property (4): the interface ∂T is connected and every tile component is simply connected. The proof of property (4) is the weakest link. It argues that if ∂T were disconnected, one takes a connected component of ∂T together with all components contained in it and all their translations by G, and then 'moves these components with a translation' to obtain a new tiling with the same volumes and perimeter. For the fixed-lattice problem (2.1), this competitor is not admissible unless the translation lies in G: translating a G-invariant family {C+g:g∈G} by a vector t∉G produces {C+g+t}, which is not a union of translates of a single generator by G, so the resulting tiling is not periodic with the same lattice. If t∈G, no component actually moves. The sketch also does not address the simply-connectedness part of (4). The diameter bound mentioned for the fixed-lattice extension does not control the number of connected components of a tile. If property (4) fails, Proposition 2.2 and the subsequent dichotomy into (3;9), (4;8), (6;6) could miss valid minimizers with disconnected interfaces or non-simply-connected tiles, and the explicit isoperimetric profile would be incomplete. This is a genuine gap in a load-bearing step, not a stylistic issue.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies periodic tilings of the plane by two tiles with prescribed areas, minimizing interface length either for a fixed lattice or with the lattice free. It claims a classification into three curvilinear configurations, labelled (3;9), (4;8), and (6;6), and an explicit isoperimetric profile with two transition points. The proofs combine regularity theory imported from the authors' earlier work [30] with planar graph arguments and trigonometric computations. The main statements are Theorem 1.1, the fixed-lattice classification, and Theorem 1.2, the explicit profile for the optimal lattice.","tokens_in":21675,"tokens_out":15671,"duration_ms":186864,"significance":"If the regularity and connectivity assumptions are fully justified, this is a substantial explicit solution of a natural unequal-cell variant of the planar Kelvin problem. The formulas are parameter-free, the transition points x1 and x2 are computed rather than fitted, and the paper gives uniqueness statements, a stationary non-minimizing configuration, and clearly formulated open problems. The main caveat is that the classification rests on a regularity/connectivity theorem whose proof is partly sketched and on an imported extension from a companion paper; these points are load-bearing and need to be completed.","major_comments":[{"comment":"The proof of connectivity of the interface and simple connectivity of the tiles is only a sketch. It says to take a connected component of ∂T together with all enclosed components and their G-translates and move them by a translation, but it does not construct the new generators for the fixed-lattice problem (2.1), does not prove that the new partition has the same areas and perimeter, and does not explain how the moved components eventually produce an illegal vertex. The simple-connectivity assertion is not proved at all. Since Proposition 2.2, Proposition 2.3, and the dichotomy in Theorems 2.4–2.6 all use property (4), this is load-bearing. I do not think the issue is simply that a translation by t∉G breaks G-periodicity, because translating the entire G-orbit of a block can still yield a G-periodic configuration; rather, the argument as written is not a complete proof.","section":"Section 2, Theorem 2.1(4)"},{"comment":"The extension of [30, Theorem 5.2] to the fixed-lattice variational problem is asserted with 'readily checked' and a diameter bound. Because Theorem 1.1 and the later profile depend on fixed-lattice regularity, this step needs a real proof: the authors should explain how [30, Lemma 4.2 and Proposition 5.1] adapt when the lattice is fixed and why the diameter bound is compatible with the cell area constraints. As written, the regularity foundation of the classification is an unverified import.","section":"Section 2, paragraph after (2.1)"},{"comment":"The statement that 'the smaller signed distance between E and H is obtained by taking the points on the axis y=x' is used to conclude a=b and u=v, and hence to derive the square-lattice formula (3.7). No proof is supplied for this comparison between a level ellipse of r(x,y) and a level hyperbola of xy. Since the middle branch of I(x) in Theorem 3.5 depends on (3.7), this is another load-bearing point that should be justified.","section":"Section 3, proof of Theorem 3.4, Step 2"}],"minor_comments":[{"comment":"The text says I(x)=q2 on the sixth interval; from the definitions and Theorem 1.2 the constant should be q3.","section":"Theorem 3.5, item (5)"},{"comment":"There are repeated typos: 'preassure' should be 'pressure', and 'Reauleaux' should be 'Reuleaux'.","section":"Throughout"},{"comment":"The phrase 'the curvilinear rectangle, E1 is a curvilinear square' is missing a verb or comma; it should say that the tile E1 is a curvilinear square.","section":"Theorem 3.4 proof"},{"comment":"The plain-text rendering '4√12' is ambiguous; the typeset fourth-root notation should be used consistently so that q1 and q3 are not confused with multiples of square roots.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and the main contribution is strong if the regularity and connectivity theorem can be completed. I would encourage the authors to make Theorem 2.1 self-contained for the fixed-lattice case, or to cite a complete proof of property (4). The unresolved point is technical but central, so major revision rather than rejection is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real advance over the authors' earlier qualitative work, but the classification is built on a regularity import whose proof has a hole. The hole is in Theorem 2.1(4), and it is load-bearing. I'd send it to review, but the authors need to fix it.\n\nWhat's new: the complete classification of periodic planar 2-tilings with two prescribed areas (three configurations, (3;9), (4;8), (6;6)) and the explicit isoperimetric profile with transition points x1 ≈ 0.062 and x2 ≈ 0.317. The profile is derived from explicit geometry, no fitted parameters, and the optimal-lattice computations are concrete. That's a substantive step beyond the qualitative existence and regularity results in [30].\n\nWhat it does well: the geometric arguments in Section 2 are careful and, assuming the regularity properties, the dichotomy into the three configurations is convincing. The authors also clearly separate the fixed-lattice and variable-lattice problems, and the open problems section is honest and useful.\n\nThe soft spots, in order of seriousness.\n\n1. Theorem 2.1(4) is not proved correctly for the fixed-lattice problem. The proof says: if ∂T is disconnected, take a connected component plus its G-translates and 'move these components with a translation.' For problem (2.1), the competitor must remain a union of G-translates of the two generators. A translation by t ∉ G destroys that; a translation by t ∈ G moves nothing. So the argument doesn't rule out disconnected interfaces. The same proof also skips the simply-connectedness claim in (4). Since Theorem 1.1 and the explicit profile both depend on this property, this is a genuine gap, not a stylistic issue.\n\n2. Theorem 3.4 contains an algebraic simplification that is delegated to numerical checking. That's acceptable in principle, but a rigorous derivation would be better.\n\n3. Theorem 3.5 has a proof typo: item (5) in the proof says I(x)=q2 on (x2,1/2], while the statement correctly says q3. Minor, but should be fixed.\n\nThe reliance on [30] for regularity is reasonable — that's a published paper by two of the authors — but the extension to the fixed-lattice problem is asserted with a sketch, and that's exactly where the gap lives.\n\nOverall: the central result is likely correct, but the proof as written is incomplete. The paper deserves a serious referee, but I would not accept it as-is. The authors should give a correct proof of Theorem 2.1(4) (or state connectedness and simple-connectedness as assumptions and discuss the consequences), fix the typo, and tighten the numerical check. Once that's done, this is a valuable paper for anyone working on isoperimetric problems, periodic tilings, or foam simulations.","headline":"A genuine classification result with an explicit isoperimetric profile, but the proof of the connectedness property rests on an inadmissible competitor for the fixed-lattice problem.","tokens_in":22151,"tokens_out":3876,"would_cite":false,"duration_ms":40435,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q05","52C20","58E12"],"pacs":[],"model":"deepseek-v4-flash","headline":"The least-perimeter periodic tiling of the plane by two unequal-area tiles is exactly one of three shapes, with an explicit interface-length formula.","keywords":["periodic tilings","Kelvin problem","isoperimetric profile","unequal cell areas","perimeter minimization","Reuleaux triangle","curvilinear polygons","honeycomb tiling"],"falsifier":"For a fixed square lattice and area ratio $x = 0.1$, solve the two-tile perimeter-minimization problem numerically with no shape restrictions: the theorem predicts the minimizer is the $(4;8)$ curvilinear square with a chipped square, so finding any admissible tiling with smaller interface length, or one whose interface contains a vertex where three edges do not meet at 120 degrees, would refute the classification.","tokens_in":21047,"feed_emoji":"🐝","tokens_out":8852,"duration_ms":89442,"temperature":0.7,"pith_summary":"This paper determines the least possible interface length for a periodic tiling of the plane made of two kinds of tiles with prescribed unequal areas, and it writes the answer as explicit formulas. It proves that only three shapes can ever be optimal relative to any fixed lattice: two flat-edged hexagons, a curvilinear square paired with a chipped square, or a Reuleaux triangle paired with a nine-sided chipped hexagon. Which shape wins depends only on the ratio of the two tile areas, with two sharp transitions at approximately $x=0.062$ and $x=0.317$. Because the resulting isoperimetric profile is explicit, any candidate unequal-cell partition can be checked against a closed-form benchmark rather than by numerical search alone.","feed_headline":"Three shapes solve the least-perimeter two-tile tiling problem","feed_subtitle":"For every ratio of tile areas, the optimal interface length is a closed formula with two sharp transitions.","key_machinery":"The load-bearing mechanism is the pressure-vector description of minimizers, imported from the regularity theory of earlier work and stated as Theorem 2.1. It says that every interface is a finite union of circular arcs and straight segments that meet only at triple points with equal 120-degree angles, that the signed curvature of an arc equals the difference of two pressures $p_1=-p_2$, and that tiles are connected and simply connected. With periodicity and the two-tile assumption, Proposition 2.2 and Proposition 2.3 turn this into a short list of possible edge counts, and the planar-graph bound in Lemma A.5 (each vertex has degree at most 6) cuts the list to the three claimed configurations. The explicit formulas then come from elementary trigonometric identities: the turning-angle relation $\\sum_k \\alpha_k = (6-n)60^\\circ$ (Lemma A.1), the area and perimeter of a Reuleaux triangle (Lemma A.2), the Steiner-tripod length formula (Lemma A.3), and the quadrangular curvilinear polygon computation (Lemma A.4), together with Lagrange-multiplier optimizations over lattice parameters in Section 3.","core_discovery":"The central claim is that the isoperimetric problem for periodic two-tile tilings of the plane is completely solvable. Theorem 1.1 states that, for any fixed lattice $G$ and any prescribed positive areas of the two generators, a perimeter-minimizing tiling must be one of three configurations: two hexagons with straight edges $(6;6)$, a strictly convex curvilinear quadrangle with four circular arcs paired with an octagonal 'chipped parallelogram' $(4;8)$, or a Reuleaux triangle paired with a nine-sided chipped hexagon $(3;9)$, with all edges meeting at 120 degrees and curvatures governed by a pressure difference. Theorem 1.2 combines the three perimeter formulas with the optimal lattice for each regime and gives the isoperimetric profile $I(x)$ explicitly: $\\sqrt{2}\\sqrt{\\pi-\\sqrt{3}}\\sqrt{x}+\\sqrt[4]{12}$ on $[0,x_1)$, $2\\sqrt{\\pi/3+1-\\sqrt{3}}\\sqrt{x}+2$ on $[x_1,x_2)$, $2\\sqrt[4]{3}$ on $(x_2,1/2]$, and $I(x)=I(1-x)$ on $(1/2,1]$, with $x_1\\approx 0.062$ and $x_2\\approx 0.317$.","pith_inferences":["One consequence the authors leave implicit is that the fixed-lattice problem reduces to a finite check: for a given lattice $G$, only the admissibility intervals of the $(3;9)$, $(4;8)$, and $(6;6)$ shapes need to be determined, so their Problem 4.1 could be settled by explicit inequalities rather than by new regularity theory.","The same counting argument suggests that for $N$-tilings with one large cell and $N-1$ small cells, the optimal shapes should be a large polygon with small Reuleaux-triangle caps on some vertices, at least when the small cells are sufficiently small; this is exactly the candidate the authors mention in Section 4.2.","Because the profile is explicit and $(I(x)-I(0))^2$ is concave, one can bound the perimeter of any periodic partition with many area ratios by a weighted average of these three formulas; an interested reader could test whether such a Jensen-type bound is sharp."],"forward_implications":["For $0 < x < x_1$, the unique optimal tiling is a Reuleaux triangle of area $x$ together with a nine-sided chipped regular hexagon on a honeycomb lattice, with interface length $\\sqrt{2}\\sqrt{\\pi-\\sqrt{3}}\\sqrt{x}+\\sqrt[4]{12}$ per unit fundamental-domain area.","For $x_1 < x < x_2$, the unique optimum is a curvilinear square with four circular arcs plus a chipped square, on a square lattice, with interface length $2\\sqrt{\\pi/3+1-\\sqrt{3}}\\sqrt{x}+2$.","For $x_2 < x \\le 1/2$, two adjacent regular hexagons sharing an edge are optimal, with constant interface length $2\\sqrt[4]{3}$; at $x = 1/2$ this is the honeycomb partition into equal hexagons.","At the two transition points $x_1 \\approx 0.062$ and $x_2 \\approx 0.317$ the adjacent configurations tie, so minimizers are not unique exactly there; for every other area ratio the optimal tiling is unique up to isometry.","The squared shifted profile $(I(x)-I(0))^2$ is concave on $[0,1]$, a structural property that may be useful in comparison arguments."],"supporting_citations":[{"why":"Supplies the regularity theory, imported as Theorem 2.1, that minimizers are made of finitely many circular or straight edges meeting at 120 degrees and have connected, simply connected tiles.","marker":"[30]"},{"why":"Provides the honeycomb theorem that fixes the degenerate x=0 case and the periodic one-tile benchmark used in Theorem 3.1.","marker":"[17]"},{"why":"Its Lemma 5.1 is used in Lemma A.4 to prove the symmetries of the curvilinear quadrangular tile.","marker":"[42]"},{"why":"Provides the equilateral-triangle construction behind the Steiner-tripod length formula in Lemma A.3.","marker":"[23]"}],"fun_headline_variants":["Three shapes solve least-perimeter two-tile tilings","Two-tile tilings: optimal interface has closed form","Isoperimetric two-tile tilings: three shapes suffice","Least-perimeter tilings with two tiles classified","Optimal two-tile tiling: three cases, explicit formula"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The classification presupposes that, for a fixed lattice, every area-minimizing tiling has interfaces made of finitely many circular arcs and straight segments that meet only in threes at 120 degrees, with connected and simply connected tiles; if minimizers could have more complicated or disconnected interfaces, the three-configuration list could miss real minimizers.","fun_headline_variants_meta":{"raw":{"variants":["Three shapes solve least-perimeter two-tile tilings","Two-tile tilings: optimal interface has closed form","Isoperimetric two-tile tilings: three shapes suffice","Least-perimeter tilings with two tiles classified","Optimal two-tile tiling: three cases, explicit formula"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00017,"raw_usage":{"total_tokens":1251,"prompt_tokens":914,"completion_tokens":337,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":530,"completion_tokens_details":{"reasoning_tokens":255}},"tokens_in":530,"tokens_out":337,"duration_ms":3623,"temperature":1.0,"reasoning_tokens":255,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T05:14:50.624471+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed square lattice and area ratio $x = 0.1$, solve the two-tile perimeter-minimization problem numerically with no shape restrictions: the theorem predicts the minimizer is the $(4;8)$ curvilinear square with a chipped square, so finding any admissible tiling with smaller interface length, or one whose interface contains a vertex where three edges do not meet at 120 degrees, would refute the classification.","supporting_citations":[{"cited_title":"Nobili and M","cited_arxiv_id":null,"evidence_quote":"Supplies the regularity theory, imported as Theorem 2.1, that minimizers are made of finitely many circular or straight edges meeting at 120 degrees and have connected, simply connected tiles."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the honeycomb theorem that fixes the degenerate x=0 case and the periodic one-tile benchmark used in Theorem 3.1."},{"cited_title":"Wichiramala, Proof of the planar triple bubble conjecture , J","cited_arxiv_id":null,"evidence_quote":"Its Lemma 5.1 is used in Lemma A.4 to prove the symmetries of the curvilinear quadrangular tile."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the equilateral-triangle construction behind the Steiner-tripod length formula in Lemma A.3."}],"review_version":1}