{"id":"3d93ef89-65ed-4e61-ad10-a0c3ec0e327c","arxiv_id":"2510.23022","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For fixed h and large k, the possible sizes of h-fold sumsets of k-element integer sets form the full interval [hk−h+1, C(h+k−1,h)] minus C(h−1,2) specified numbers; the h=3 case is settled for all k>2.","lead":"Given any fixed number h, this paper determines which sizes an h-fold sumset of a k-element integer set can have once k is large: essentially every size in the interval from the minimum to the maximum, with exactly C(h−1,2) forbidden numbers. It also gives the exact answer for h=3 for every k>2, resolving a problem posed by Nathanson.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Gap between Propositions 4.1 and 4.2: Prop 4.1 reaches only ~h k^h/4^{(h^2+h)/2}, while Prop 4.2 starts at k^h/(42h^2); for h≥4 the two intervals are disjoint, so Theorem 1.4 does not follow from the stated arguments.","rationale":"The reader's weakest assumption was Lemma 4.4's unverified filling property. While that is a legitimate concern, the most load-bearing issue is even more immediate: the two main propositions as used in Theorem 1.4's proof do not cover the required interval. This is a quantitative mismatch that can be checked directly from the displayed formulas. It is not an attack on the conjecture's truth; the construction strategy may be salvageable by choosing ε appropriately. Still, the published proof is incomplete. Since the error is repairable in principle, I keep the CONDITIONAL verdict rather than calling for rejection. Agreement with reader: the reader did not flag the overlap gap, so our concerns differ; I mark disagreement.","tokens_in":19561,"tokens_out":11210,"duration_ms":101015,"concrete_test":"Verify the overlap condition: fix h=4 and take k large. From Prop 4.1's proof, the maximum size attained is at most h((k-1)^4/4^{10}-k) ≈ k^4/262144. Prop 4.2 with ε=1/(42·16)=1/672 starts at k^4/672. Since 1/262144 < 1/672, there are integers (e.g., k^4/1000) that lie in neither interval. This demonstrates the proof gap. More generally, compute the ratio (h/4^{(h^2+h)/2})/(1/(42h^2)) = 42h^3/4^{(h^2+h)/2} for h=4,5,6; if it is <1, the two propositions cannot be combined.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 4.1's proof (Section 4.2) constructs sets A={0}∪(i+B) with B an (h,d)-filling set of size k-1 and d ≤ (k-1)^h/4^{(h^2+h)/2}. It obtains |hA|=h(i+d)-i+2, and after varying i,d concludes R(h,k) ⊇ [m, h((k-1)^h/4^{(h^2+h)/2}-k)] \\ Δ_{h,k}. Thus the largest size guaranteed by the small-size construction is ~ h k^h / 4^{(h^2+h)/2}. Proposition 4.2 (Section 4.3) guarantees [ε k^h, C(h+k-1,h)] ⊆ R(h,k) for any fixed ε>0 and k sufficiently large. In the proof of Theorem 1.4, ε is set to 1/(42h^2). For h≥4, h/4^{(h^2+h)/2} < 1/(42h^2) (e.g., h=4: 4/4^{10}=1/262144 vs 1/672; h=5: 5/4^{15}≈4.9e-6 vs 1/1050). Hence the upper endpoint of Prop 4.1 lies strictly below the lower endpoint of Prop 4.2; the intervals do not overlap and there is an enormous uncovered middle range. The claim that Propositions 4.1 and 4.2 together imply Theorem 1.4 is therefore false as written, independent of the 'not hard to check' steps in Lemma 4.4. This is a concrete numerical obstruction, not merely a missing proof. A corrected proof would need to choose ε much smaller (e.g. ε = h/4^{(h^2+h)/2}) or strengthen Prop 4.1; as it stands the theorem is not established.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies R(h,k), the set of possible cardinalities of h-fold sumsets of k-element subsets of Z. Theorem 1.3 rules out a triangular set Δ_{h,k}; Theorem 1.4 asserts that for each fixed h and all sufficiently large k, R(h,k) is exactly the bounding interval minus Δ_{h,k}; Theorem 1.7 proves the h=3 case for all k>2. The proof of Theorem 1.7 uses a graph of pairs (|2A|,|3A|) and an inductive path argument. The proof of Theorem 1.4 is split into Proposition 4.1, which constructs sets with small h-fold sumsets via (h,d)-filling sets, and Proposition 4.2, which constructs large h-fold sumsets by combining dense and sparse pieces. The paper also contains a clean diameter-based proof of Theorem 1.3 using Freiman-type results.","tokens_in":20049,"tokens_out":14132,"duration_ms":130703,"significance":"If the main result is correct, it answers Nathanson's problem on the form of R(h,k) for fixed h and large k, and Theorem 1.7 confirms Conjecture 1.6 for h=3. The methods are novel and varied: the diameter argument for exclusions, the graph-connectivity argument for h=3, and the generating-function/disjoint-union technique for the large-size range are all interesting and worth publishing. The computational verification for h+k<12 and the explicit formulation of the excluded triangle are useful contributions. However, the current proof of the main theorem contains a concrete interval-overlap gap that must be repaired before the claims are established.","major_comments":[{"comment":"The display at the end of §4.2 gives R(h,k) ⊇ [hk-h+1, h((k-1)^h/4^{(h^2+h)/2} - k)] \\ Δ_{h,k}. The leading term of this upper endpoint is (h/4^{(h^2+h)/2}) k^h. Proposition 4.1 claims the upper endpoint k^h/(42h^2). For h≥4, h/4^{(h^2+h)/2} < 1/(42h^2) (e.g. h=4: 4/4^{10}=1/262144 vs 1/672), so the derived interval does not imply Proposition 4.1. In the proof of Theorem 1.4, ε is set to 1/(42h^2), hence Proposition 4.2's lower endpoint lies strictly above the largest size guaranteed by §4.2 for large k; the two intervals do not meet. The theorem is therefore not established as written. This is repairable by taking ε = h/(2·4^{(h^2+h)/2}) in Proposition 4.2, but the current statements and proof must be aligned.","section":"§4.2 / Proof of Proposition 4.1"},{"comment":"The proof asserts without details that A' = \\tilde A ∪ {i\\tilde d} ∪ (d-\\tilde d + \\tilde A) is (h,d)-filling. The displayed equality for (h-1)(\\tilde A ∪ (d-\\tilde d + \\tilde A)) does not by itself show hA'=[0,hd]; when d ≫ \\tilde d, the intervals j(d-\\tilde d)+[0,(h-1)\\tilde d] are widely separated, and the role of the grid {i\\tilde d} in filling the gaps is precisely the content of the filling claim. Since this filling property is the engine of Proposition 4.1 and hence of the small-sumset half of Theorem 1.4, a complete verification must be supplied.","section":"Lemma 4.4"},{"comment":"The path P rests entirely on the exact formulas for g(A_a) and g(B_b), which are asserted as 'explicit computations' without any derivation. The case k=4 requires a separate B'_2, and the endpoint A_{2k-2} lies outside the three listed ranges for A_a, so the formulas are not entirely routine. Because Theorem 1.7 depends on this lemma, the computations should be provided in full or in an appendix.","section":"Lemma 3.2"}],"minor_comments":[{"comment":"The notation O[[az]] and Θ+[[az]] is introduced but the multiplication of a monomial like z^m O[[ℓz]] is not formally defined. This should be clarified for readability.","section":"§4.3.2"},{"comment":"In the path P of Lemma 3.2, the index B_{k-3} is skipped. This is intentional, but the reader has to verify the adjacency across the jump from B_{k-4} to A_{k-3}; a remark would help.","section":"§3"},{"comment":"There is a typo in the question about possible tuples: '( |3A|,4A| )' should read '(|3A|,|4A|)'.","section":"§5.5"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know two things about this paper before deciding what to do with it. First, it is real work: Theorem 1.3 gives a clean new obstruction—a whole triangle of excluded sizes—and Theorem 1.7 determines R(3,k) exactly for all k>2. Second, the proof of the main large-k theorem has a concrete gap as written, not just a missing detail.\n\nThe genuinely new material is solid. Extending the known two gap intervals to the full triangle Δ_{h,k} is a natural step, and the nonconstructive strategy—building a graph on possible sumset-size tuples and proving connectivity—is fresh. It actually delivers the h=3 result, which is the first complete determination for fixed h beyond h=2. The generating-function and disjoint-union machinery in Section 4.3 is interesting, and the paper is honest about what is proved versus conjectured.\n\nThe soft spots are real but, I think, fixable. The most serious is in the proof of Theorem 1.4. Proposition 4.1 is stated with upper endpoint k^h/(42h^2), but its own proof in Section 4.2 only reaches about h k^h / 4^{(h^2+h)/2}, using Lemma 4.4 with d ≤ (k-1)^h/4^{...}. For h≥4, h/4^{(h^2+h)/2} is far smaller than 1/(42h^2). So the two propositions, with ε chosen as 1/(42h^2), do not overlap; there is an enormous uncovered middle interval. The good news is that Proposition 4.2 allows any fixed ε>0, so the proof can be repaired by taking ε much smaller, provided the rest holds. But as written, Theorem 1.4 does not follow from the stated arguments.\n\nTwo other assertions need checking: Lemma 4.4's 'not hard to check' that A' is (h,d)-filling, and Lemma 3.2's explicit values for g(A_a) and g(B_b). These are finite computations rather than structural gaps, but they are load-bearing. The paper would be much stronger with expanded proofs or a short verification script.\n\nThe central idea is sound and the h=3 result is likely correct. The paper deserves a serious referee, but the referee should ask for a repaired connection between Propositions 4.1 and 4.2, and for justification of the computational claims, before the main theorem is accepted. I would not cite it as a theorem until that happens, but I would bring it to the reading group; it is a good paper to think through.","headline":"A serious paper whose gap exclusion and h=3 theorem are new and likely correct, but whose main large-k theorem is not established as written because the two halves of the proof do not connect for h≥4.","tokens_in":20569,"tokens_out":5708,"would_cite":false,"duration_ms":48352,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11B13","11P70"],"pacs":[],"model":"deepseek-v4-flash","headline":"For each fixed h, the paper proves that once a set A of k integers is large enough, every possible size of the h-fold sumset occurs—except for a precisely specified triangular block of values near the bottom of the range.","keywords":["sumsets","h-fold sumsets","additive combinatorics","range of sumset sizes","sumset cardinalities","discrete intermediate value","nonconstructive proof","generating functions"],"falsifier":"Test the filling claim in Lemma 4.4 for a small concrete case—e.g., $h=3$, $d=10$, $k=5$—by explicitly listing the sumsets; if the union fails to cover $[0, hd]$, the proof's construction of the lower half of the interval collapses. Alternatively, a computer search for any $(h,k)$ with $k>h$ whose computed $R(h,k)$ contains a value from $\\Delta_{h,k}$ would disprove the more general conjecture.","tokens_in":19441,"feed_emoji":"➕","tokens_out":8241,"duration_ms":72829,"temperature":0.7,"texified_at":"2026-08-05T20:34:56.041841+00:00","pith_summary":"A long-standing problem asks which cardinalities can occur as the size of the h-fold sumset of a k-element set of integers. This paper answers that problem asymptotically in k: for every fixed h, once k exceeds a threshold, the set $R(h,k)$ of attainable sizes is the full interval from $hk-h+1$ to $C(h+k-1,h)$ with a single triangular block of $C(h-1,2)$ numbers removed. It further pins down the $h=3$ case completely, showing $R(3,k) = \\{3k-2\\} \\cup [3k, C(k+2,3)]$ for all $k>2$. The proof is nonconstructive: rather than exhibiting a set for each size, it shows that families of sets built from dense and sparse pieces sweep out a whole interval of sizes, via a discrete intermediate-value lemma.","texify_model":"deepseek-v4-flash","texify_usage":{"total_tokens":8802,"prompt_tokens":831,"completion_tokens":7971,"prompt_tokens_details":{"cached_tokens":0},"prompt_cache_hit_tokens":0,"prompt_cache_miss_tokens":831,"completion_tokens_details":{"reasoning_tokens":7189}},"feed_headline":"For large k, sumsets realize every size minus a triangle","feed_subtitle":"The range of h-fold sums is a full interval minus a triangle, for fixed h and large enough sets.","key_machinery":"The central engine is Lemma 4.5, a discrete intermediate-value lemma: an integer-valued function on a box that is nonincreasing in each coordinate, has unit maximum single-coordinate drop, and whose cumulative drops dominate the next coordinate's maximum drop, must have an interval for its image. The paper applies it to $|hA|$ where $A$ is a disjoint union of a dense arithmetic-progression-like set and sparse geometric sets, with parameters controlling the number of relations among sums. A second key object is the k-sumset graph, whose vertices are the possible pairs $(|2A|, |3A|)$ and whose edges join nearby pairs; the $h=3$ proof works by showing endpoints are connected by paths.","core_discovery":"The paper establishes that for each fixed $h$ there is a constant $k_h$ such that for all $k > k_h$, $R(h,k) = [hk-h+1, C(h+k-1,h)] \\setminus \\Delta_{h,k}$, where $\\Delta_{h,k}$ is a union of $\\min(h,k)-3$ disjoint intervals that form a triangle; Theorem 1.3 proves none of these $\\Delta$ values can occur, so the description is exact. For $h=3$, Theorem 1.7 proves the stronger identity $R(3,k) = \\{3k-2\\} \\cup [3k, C(k+2,3)]$ for all $k>2$, so the threshold $k_3$ can be taken to be 2. The proof combines a graph whose vertices are possible $(|2A|,|3A|)$ pairs with an inductive lifting argument, and for general $h$ a parameterized construction whose sumset size varies continuously over an interval.","pith_inferences":["The discrete intermediate-value lemma (Lemma 4.5) is stated as a standalone principle and could be reused in other combinatorial parameter-counting problems where monotone families have controlled variation, such as subgraph-density ranges or other additive statistics.","The h=3 graph-path strategy suggests that extending to h=4 or 5 would require controlling triples of sumset sizes simultaneously; the paper notes this is the main obstruction, so a natural testbed is to plot the possible (|3A|,|4A|) pairs for fixed |A| and look for connected paths.","If the conjecture in Section 5.2 (a diameter-based lower bound on |hA|) is true, then for h > k+1 there would be additional gaps beyond the triangle, which could be detected by brute-force computation for k=4 and small h.","The resemblance noted in the paper between the possible (|2A|,|3A|) plot and the edge/triangle density region from extremal graph theory could be made explicit; if an exact map exists, graph-theoretic methods might transfer to sumset sizes."],"forward_implications":["The range of h-fold sumset sizes is fully determined for fixed h and all sufficiently large k: it is an explicit interval minus an explicit triangle.","For h=3 the answer is complete for every k>2: R(3,k) = {3k−2} ∪ [3k, C(k+2,3)].","The possibility that the same description holds for all k>h is reduced to a finite verification for each h (though computationally intensive).","The nonconstructive interval-sweeping method provides a new tool for studying the joint distribution of (|2A|,...,|hA|).","All results transfer to any infinite torsion-free abelian group, as noted in Section 5.4."],"fun_headline_variants":["For each h, sumsets skip only a triangle of sizes","Sumset sizes: full interval minus a triangle for large k","Missing sumset sizes form a triangle, exactly","Nathanson's problem: large sumsets cover all but a triangle","Almost every sumset size occurs, gaps form a triangle"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof leans on a filling lemma (Lemma 4.4) whose core step—that a particular union of three translated copies of a smaller filling set is itself filling—is asserted with a 'not hard to check' rather than a demonstrated argument; if that filling property fails, the lower half of the main interval construction collapses.","fun_headline_variants_meta":{"raw":{"variants":["For each h, sumsets skip only a triangle of sizes","Sumset sizes: full interval minus a triangle for large k","Missing sumset sizes form a triangle, exactly","Nathanson's problem: large sumsets cover all but a triangle","Almost every sumset size occurs, gaps form a triangle"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000243,"raw_usage":{"total_tokens":1363,"prompt_tokens":739,"completion_tokens":624,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":483,"completion_tokens_details":{"reasoning_tokens":540}},"tokens_in":483,"tokens_out":624,"duration_ms":6991,"temperature":1.0,"reasoning_tokens":540,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T08:02:00.121199+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test the filling claim in Lemma 4.4 for a small concrete case—e.g., $h=3$, $d=10$, $k=5$—by explicitly listing the sumsets; if the union fails to cover $[0, hd]$, the proof's construction of the lower half of the interval collapses. Alternatively, a computer search for any $(h,k)$ with $k>h$ whose computed $R(h,k)$ contains a value from $\\Delta_{h,k}$ would disprove the more general conjecture.","supporting_citations":[],"review_version":1}