{"id":"670f2be2-9256-4eaf-95ec-2f07019eb9be","arxiv_id":"2411.18291","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Keevash proves the existence of designs with a shorter self-contained proof, using a new absorber construction and improving the bound on n0.","lead":"This mathematics paper gives a new, shorter proof of the long-standing existence theorem for Steiner systems and other designs. It introduces a simpler absorber construction that also yields better explicit bounds on the size of the ground set.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 6.3(4) omits cliques whose vertex intersection with F has size > r; such cliques occur in the constructed Ω, so the second-moment bound and the rainbow-to-monochromatic generation step are not established.","rationale":"The reader's weakest assumption was the admissibility property (ii) of Lemma 3.1, which is indeed load-bearing for the random greedy steps. My concern is different and, in my reading, more acute: the proof of Lemma 6.3(4), a prerequisite for the integral absorber, appears to contain an unhandled second-moment case. The construction of Ω genuinely produces cliques intersecting F in more than r vertices when q−r>r, and these cliques are retained in the final Υ±. Such cliques create positive correlations between pairs of extensions because their shared fixed edges are counted once in the pair probability but twice in the square of the mean. The text's case split s2. Recompute EX^2 in Lemma 6.3(4) for this Q', counting pairs of extensions with the 3 fixed shared vertices and the binom(3,2)=3 shared edges counted once in the colour copy. If the resulting variance bound is still small enough for Chebyshev followed by the 20qα^{-1}-fold colour boost, the gap is repairable; otherwise Lemma 6.3(4) requires a new proof or a modified construction of Ω.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Lemma 6.3(4) estimates EX^2 by splitting pairs according to s = |V(Qhat')∩F|, with cases s < r and s = r. This assumes every relevant clique Qhat' ∈ Υ± outside F satisfies |V(Qhat')∩F| ≤ r. That assumption is false for the Ω built in Lemma 3.1 when q−r > r. In the blow-up Ω0, take the designated minus clique Qhat^-_0 = v0, which has coordinates 0 on [r] and 1 on [q]\\[r], and the plus clique Q' = Mw with w = (1,0,...,0). For the Vandermonde rows used, row_i·w = 1 for every i, so Q' and Qhat^-_0 agree on the q−r coordinates in [q]\\[r]. Thus |V(Q')∩V(Qhat^-_0)| = q−r > r. In the final Ω, Qhat^-_0 of the second copy in round e is exactly Qhat_e, and Q' is retained as a plus clique in Υ+. Hence Q' satisfies |V(Q')∩F| = q−r > r. For such a clique, the event that two extensions both land in the same colour copy requires the binom(q−r,r) shared edges of Q'∩Qhat_e to be in K* only once. The square of the first moment counts those edges twice, so the second-moment contribution is larger by a factor n^{Ω(α)} than the bound claimed. The stated estimate EX^2 ≤ (1+8|Ω|n^{-0.1α})(EX)^2 is therefore unsupported. This matters because property (4) is used in Section 6.5 to pass from arbitrary rainbow cliques to cliques generated over Γ by the monochromatic set Q0. Without a valid proof of (4), Lemma 6.1 and hence the absorber Lemma 2.2 are not established by the text.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper gives a new proof of Theorem 1.1, the existence of K_q^r-decompositions of K_r^n under the necessary divisibility condition, with a claimed bound log n0 = O(k^2 q^{r+1} log q). The proof follows the absorption framework: it reserves a sparse random subgraph, constructs an omni-absorber via a new clique-exchange gadget Ω, boosts regularity of the remaining graph, covers the leave by the reserve, and decomposes the absorber plus leave. Most of the paper is devoted to proving the auxiliary lemmas in a self-contained way, including Chernoff/Freedman concentration, local decoders, random greedy extension processes, the integral absorber via coloured random permutations, and the clique removal process for the nibble step.","tokens_in":29259,"tokens_out":23979,"duration_ms":226480,"significance":"If correct, this would be a fourth and substantially shorter proof of the existence conjecture, with the best explicit bound on n0, and it would make the whole argument self-contained rather than relying on previous iterative-absorption papers. The paper is well structured and gives full proofs of the supporting tools, and the parameter bookkeeping is explicit. However, the proof of Lemma 6.3(4) has a load-bearing gap that is not merely cosmetic; until that lemma is repaired, the integral absorber Lemma 6.1 and hence the main theorem are not established by the text.","major_comments":[{"comment":"The second-moment estimate in the proof of Lemma 6.3(4) is not justified as written. The proof splits pairs of extensions according to s = |V(Qhat')∩F| and treats only the cases s < r and s = r. But in the configuration Ω of Lemma 3.1, when q > 2r there are retained plus cliques Qhat' = Mw with w = (1,0,...,0) in the same Ω0-copy as a designated minus clique Qhat_e; these satisfy |V(Qhat')∩V(Qhat_e)| = q−r > r, hence |V(Qhat')∩F| > r. For such a clique, two extensions that agree on F share C(q−r,r) edges of the colour copy, so the probability that both are monochromatic in the same colour is about n^{−α(2k−C(q−r,r))}, not n^{−2kα}; this introduces a factor n^{α C(q−r,r)} into EX². The claimed bound EX² ≤ (1+8|Ω|n^{−0.1α})(EX)² is therefore unsupported, and Chebyshev no longer gives the polynomial concentration on which the subsequent union bound and the '20qα^{−1} disjoint colour sets' amplification depend. A similar, milder discrepancy already appears in the s = r case, where the shared edge contributes a factor n^α. Since Lemma 6.3(4) is the step in §6.5 that replaces arbitrary rainbow cliques by monochromatic unsaturated cliques, the proof of Lemma 6.1, and hence of the absorber Lemma 2.2 and Theorem 1.1, is incomplete as written. A repair would require either a different Ω construction avoiding cliques with large intersection with F, or a genuinely different second-moment argument.","section":"§6.3, proof of Lemma 6.3(4)"}],"minor_comments":[{"comment":"In the definitions of Υ±_0, the parameter vector should be u ∈ F_p^r, since M is a q×r matrix; the text writes u ∈ F_p^q in two places.","section":"§3.3"},{"comment":"The abstract contains the typo 's horter' instead of 'shorter'.","section":"Abstract"},{"comment":"The quantity M is introduced as 'there are M < 2|Ω|/k such Qhat''; it would be clearer to define M explicitly as the number of non-special cliques Qhat' in Υ±, since the second-moment product is taken over this set.","section":"§6.3, proof of Lemma 6.3(4)"}],"recommendation":"major_revision","confidential_remarks":"The paper is otherwise strong and the gap, while load-bearing, appears local and plausibly repairable; I would not reject if the authors can supply a corrected proof of Lemma 6.3(4) or modify the clique-exchange construction so that the omitted case disappears. As it stands, however, the main theorem is not proven."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Peter,\n\nThis is a serious new proof of the existence of designs: the absorber built from clique exchanges on Vandermonde blow-ups is much simpler than the earlier constructions, and the bound log n0 = O(k^2 q^{r+1} log q) is the best known. The paper is also genuinely self-contained, proving its own Chernoff and martingale ingredients and reducing the whole argument to five clean lemmas. The exposition is clear enough to use in a graduate course.\n\nI have to flag one soft spot that I think is load-bearing. The proof of Lemma 6.3(4) (rainbow cliques are generated by monochromatic unsaturated cliques) estimates the second moment of the number of extensions by splitting pairs of embeddings according to s = |V(Q') ∩ F| for each non-designated clique Q' in Υ±. The text only treats s < r and s = r. But the Ω from Lemma 3.1 contains plus cliques with s = q−r. Concretely, take the all-ones clique Mw with w = (1,0,...,0) in the second copy of a round-e gluing; it is retained in the final Υ+, and it agrees with the designated minus clique Qhat_e on the q−r coordinates of [q]\\[r], all of which lie in F. When q−r > r, such Q' share more than one r-edge with the fixed part, so the probability that two independent extensions both land in the monochromatic clique set is n^{-α(2k − C(s,r))}, larger than the n^{-2αk} used in the proof by a factor n^{α C(s,r)}. That factor swamps the claimed (1+8|Ω|n^{-0.1α}) bound. Consequently EX^2 ≤ (1+8|Ω|n^{-0.1α})(EX)^2 is not established, and property (4)—which Section 6.5 explicitly uses to pass from arbitrary rainbow cliques to monochromatic generators—does not have a valid proof in the text. For Steiner triple systems (q=3, r=2) the issue disappears, but for any q ≥ 2r+1 it is real.\n\nI did not find a contradiction in the theorem itself, and the gap may be repairable (a more careful second moment that tracks C(s,r), or a modification of Ω to avoid large intersections). But it is not a typo; as written, the absorber lemma 2.2 lacks a key ingredient.\n\nThis paper deserves a serious referee. The core idea is good and the presentation is strong; the referee should be asked to focus on Lemma 6.3(4) and the second-moment calculation before acceptance. I would take it to a reading group—it is a good example of a clean proof with a subtle quantitative flaw.\n\nRecommendation: send to peer review, but flag the gap explicitly; expect heavy revision or a substantial new calculation.","headline":"Clean short proof of design existence, but a real gap in Lemma 6.3(4) when q−r > r; worth refereeing with that point in mind.","tokens_in":29826,"tokens_out":11629,"would_cite":false,"duration_ms":101542,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05B05","05C65","05D40"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper gives a fourth, much shorter proof that every divisible complete r-uniform hypergraph has a decomposition into q-cliques for large n, with a logarithmic bound on the threshold.","keywords":["design theory","Steiner systems","hypergraph decompositions","absorption method","clique exchange","random greedy algorithm","nibble","existence conjecture"],"falsifier":"Enumerate the gadget produced by the gluing in Lemma 3.1 for $q=3,r=2$ with $p=3$ and check property (ii) edge by edge: if any edge $e'$ of $\\Omega$ meets the union $F$ of the designated cliques in a set not contained in $V(\\hat{Q}^+)$ or in some $V(\\hat{Q}_e)$, the absorber construction cannot be run. This is a finite computation, so it settles the structural premise independently of the rest of the proof.","tokens_in":28648,"feed_emoji":"","tokens_out":9934,"duration_ms":85074,"temperature":0.7,"pith_summary":"The paper claims a new proof of a classical existence theorem: for every $q>r\\ge 1$, once $n$ is large enough, the complete $r$-graph $K_r^n$ can be partitioned into copies of $K_q^r$, which is the same as a Steiner system (a set of $q$-subsets containing every $r$-subset exactly once), whenever the necessary divisibility conditions hold. The proof is intended to be self-contained and much shorter than the three earlier proofs, and it improves the quantitative threshold to $\\log n_0 = O(k^2 q^{r+1}\\log q)$, where $k=\\binom{q}{r}$. The main novelty is a small clique-exchange gadget that makes the absorber, the traditionally hard ingredient, easy to build through a few randomized cleaning steps. If the proof is right, the existence of Steiner systems gets a fourth independent route with the best known bound on $n_0$.","feed_headline":"Designs exist: fourth proof is shorter and sharpens the bound","feed_subtitle":"All complete r-graphs satisfying the divisibility conditions split into q-cliques once n is large, with a logarithmic bound.","key_machinery":"The load-bearing object is the clique-exchange configuration $\\Omega$ of Lemma 3.1: an $r$-graph made by gluing copies of a $p$-blowup of $K_q^r$, with two $K_q^r$-decompositions $\\Upsilon^+$ and $\\Upsilon^-$ and designated cliques $\\hat{Q}^+$ and $\\hat{Q}_e$ such that each $\\hat{Q}_e$ meets $\\hat{Q}^+$ in exactly one edge $e$. Its decisive property (ii) is that any edge of $\\Omega$ touching the union $F$ of the designated cliques is contained in $\\hat{Q}^+$ or in one $\\hat{Q}_e$; this admissibility is exactly what lets the random-greedy splitting, elimination, and further-elimination steps run. The two decompositions come from Vandermonde matrices over $\\mathbb{F}_p$, which give a unique clique through each edge. Applying the gadget repeatedly converts a signed collection of cliques with arbitrary multiplicities into one with controlled edge multiplicities, and the absorber sets $A=\\bigcup \\mathcal{Q}^-$ equal to the negative cliques produced by this process.","core_discovery":"The central claim is Theorem 1.1: for all $q>r\\ge 1$, $K_r^n$ has a $K_q^r$-decomposition whenever $K_r^n$ is $K_q^r$-divisible, provided $n\\ge n_0(q,r)$. The proof organizes the decomposition into five steps: reserve, absorber, regularity boost, nibble, and cover, and its contribution is a considerably simpler construction of the absorber. The absorber is built from a new configuration $\\Omega$ with two $K_q^r$-decompositions, which is used to transform an integral decomposition of a sparse divisible leave $L\\subseteq R$ into a signed decomposition whose positive and negative cliques are separated. The same construction supports the regularity boost and the cover step, and the quantitative section shows $\\log n_0 = O(k^2 q^{r+1}\\log q)$.","pith_inferences":["Beyond the paper, the fixed finite size of the clique-exchange gadget suggests the absorber construction should work for any host hypergraph that is pseudorandom enough to support the random-greedy steps, so the same proof may give $F$-design decompositions in typical graphs, not only in $K_r^n$.","The quantitative bottleneck is the gadget size $|\\Omega|\\le 3(2q)^r k^2$; if a smaller or more efficient gadget can be built, the threshold bound $\\log n_0$ would improve directly, since the paper's remaining inequalities are weaker.","A direct check of Lemma 3.1 for small parameters, say $q=3,r=2$, would independently confirm the structural premise of the whole absorber, because the construction is explicit enough to be computer-verified."],"forward_implications":["The existence conjecture for Steiner systems holds for every $q>r\\ge 1$ with a threshold satisfying $\\log n_0 = O(k^2 q^{r+1}\\log q)$, so the qualitative theorem now comes with an explicit, though large, bound.","The proof's five-step skeleton transfers to the more general decomposition problems the author sketches, such as replacing the host $K_r^n$ by a typical $r$-multigraph and the guest $K_q^r$ by a general $r$-graph.","Because the absorber is edge-disjoint from the reserve and absorbs every $K_q^r$-divisible subgraph of the reserve, divisibility alone remains sufficient for the final decomposition.","The paper's Remark 7.1 shows the usual degree-divisibility conditions are equivalent to $K_q^r$-divisibility, so the statement of Theorem 1.1 matches the classical formulation of Steiner system existence.","The clique-removal analysis proves a stronger nibble bound, with a leave of boundedness exponent $-\\varepsilon/3k$, than standard semi-random arguments, and this is what allows the sparse reserve to cover the leftover edges."],"supporting_citations":[{"why":"Supplies the original existence-of-designs proof, the integral absorber idea, and the nibble formulation used in Step 4.","marker":"[5]"},{"why":"Supplies the regularity-boosting lemma used to make clique counts nearly equal in the host left after removing the reserve and absorber.","marker":"[4]"},{"why":"Supplies the absorption framework and the omni-absorber notion that the five-step proof outline follows.","marker":"[2]"},{"why":"Supplies the alternative integral-absorber method via rotated colour classes, adapted in Section 6 to control edge multiplicities.","marker":"[6]"},{"why":"Supplies the clique-removal-process analysis used to prove the nibble lemma with the stronger boundedness of the leave.","marker":"[1]"},{"why":"Supplies the inclusion-matrix argument for local decoders used to convert integral decompositions into bounded signed ones.","marker":"[7]"}],"fun_headline_variants":["Designs: a much shorter proof with logarithmic bounds","Existence of designs via a shorter construction","Short proof of designs existence sharpens the bound","New proof for designs is shorter and gives better bounds"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof hinges on one structural property of its small building block: any edge of the block that touches the central designated cliques must be wholly contained in one of those cliques, since otherwise the random-greedy filling steps could get stuck.","fun_headline_variants_meta":{"raw":{"variants":["Designs: a much shorter proof with logarithmic bounds","Existence of designs via a shorter construction","Short proof of designs existence sharpens the bound","New proof for designs is shorter and gives better bounds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000531,"raw_usage":{"total_tokens":2446,"prompt_tokens":720,"completion_tokens":1726,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":336,"completion_tokens_details":{"reasoning_tokens":1666}},"tokens_in":336,"tokens_out":1726,"duration_ms":11261,"temperature":1.0,"reasoning_tokens":1666,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:20:27.653841+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate the gadget produced by the gluing in Lemma 3.1 for $q=3,r=2$ with $p=3$ and check property (ii) edge by edge: if any edge $e'$ of $\\Omega$ meets the union $F$ of the designated cliques in a set not contained in $V(\\hat{Q}^+)$ or in some $V(\\hat{Q}_e)$, the absorber construction cannot be run. This is a finite computation, so it settles the structural premise independently of the rest of the proof.","supporting_citations":[{"cited_title":"The existence of subspace designs","cited_arxiv_id":"2212.00870","evidence_quote":"Supplies the alternative integral-absorber method via rotated colour classes, adapted in Section 6 to control edge multiplicities."},{"cited_title":"A natural barrier in random greedy hypergraph matching","cited_arxiv_id":"1210.3581","evidence_quote":"Supplies the clique-removal-process analysis used to prove the nibble lemma with the stronger boundedness of the leave."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the inclusion-matrix argument for local decoders used to convert integral decompositions into bounded signed ones."}],"review_version":1}