{"id":"e375906b-0d02-48a0-9800-aa6286301740","arxiv_id":"1908.09036","paper_version":3,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"For v_p(a_p) > floor((k-1)/p), the semisimple mod p reduction of the crystalline representation V_{k,a_p} is V_{k,0}.","lead":"The paper proves that certain two-dimensional p-adic Galois representations, once a parameter a_p is p-adically small enough, all reduce modulo p to the same representation. This improves a fifteen-year-old bound and is shown through explicit Kisin modules.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the descent algorithm's error-growth estimate is intricate but internally coherent.","rationale":"The stress-test pass examined the descent algorithm, the integrality step (Proposition 5.2.2), and the reduction argument (Corollary 5.2.3). The proof of Proposition 4.3.5(d) is terse but the estimates are sufficient; the exceptional case in Lemma 4.3.4 is handled by a second diagonal operation. The infinite-product convergence is justified by the linear growth of ε and completeness of R. No internal gap was found. The central claim is independent of the conclusion in the descent step. Therefore the reader's ACCEPT verdict stands unchanged.","tokens_in":20639,"tokens_out":44592,"duration_ms":443175,"concrete_test":"Recompute Lemma 4.3.4(a)-(c) from Proposition 4.2.2(a)(iii) with symbolic generic valuations for the case (i,j) = (1,1), (k,ℓ) = (2,2), and then run the two-step composition α22 ∘ α11 on a sample γ-allowable matrix where all four ε entries equal ε_C; verify every entry of the error matrix has valuation at least ε_C + min{γ, p−1}. This directly tests the only non-uniform step in Proposition 4.3.5(d).","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the descent algorithm (Section 4.3) as the load-bearing premise, as the reader did. The critical estimate is Proposition 4.3.5(d): after a finite sequence of allowed operations, the error ε_C increases by at least δ = min{γ, p−1}. I checked the exceptional case in Lemma 4.3.4: for α11, only ε22 is not covered by the uniform min bound, and part (c) gives ε22′ ≥ min{ε22, ε11}, so a subsequent α22 raises it by γ ≥ δ. For the off-diagonal pre-improvement step, applying each off-diagonal operation to an entry with error below ε_C + δ either raises that entry by γ (hence past ε_C + δ) or, through the min in Lemma 4.3.4(b), raises the complementary off-diagonal entry; at most two operations suffice. The telescoping product in Theorem 4.3.7 then converges because ε_C^{(m)} grows linearly, making A_m − 1 tend to 0 in the complete Gauss-valuation topology of R. I found no circular use of the theorem being proved, and the applications in Section 5 use the estimate exactly as stated. The main residual risks are the external inputs (Kisin's theory and [4, Théorème 3.2.1] for h < 2p), but these are standard and not internal inconsistencies.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the semisimple mod p reductions of the two-dimensional irreducible crystalline representations V_{k,a_p} of G_{Q_p} with Hodge–Tate weights 0 and k−1 and Frobenius characteristic polynomial X^2 − a_p X + p^{k−1}. The main theorem (Theorem 1.1.1, Corollary 5.2.3) states that for every k≥2 and every a_p with v_p(a_p)>⌊(k−1)/p⌋, the semisimple reduction V_{k,a_p} is isomorphic to V_{k,0}, and it is explicitly the induction Ind_{G_{Q_{p^2}}}^{G_{Q_p}}(ω_2^{k−1}χ). The proof is built on explicit Kisin modules: Section 3 determines the family of ϕ-modules satisfying the monodromy relation (Proposition 3.0.4), Section 4 develops a descent algorithm from a p-adic disc to the formal power series ring (Theorem 4.3.7) with quantitative error growth (Proposition 4.3.5), and Section 5 applies the algorithm to obtain an integral Kisin module under the stated slope condition (Proposition 5.2.2) and then reads off the reduction.","tokens_in":20923,"tokens_out":13804,"duration_ms":131392,"significance":"If correct, the main theorem is a genuine improvement over the Berger–Li–Zhu bound δ_p(k)≤⌊(k−2)/(p−1)⌋ and reaches the range suggested by global and computational evidence up to a known gap. The paper is also valuable as one of the few fully explicit calculations with Kisin modules beyond small Hodge–Tate weights. I found the central argument coherent: the monodromy relation is verified by direct calculation, the descent is reduced to a precise error-growth statement whose exceptional case (Lemma 4.3.4(c)) is handled correctly, and the integrality step (Lemma 5.1.3 and Proposition 5.2.2) uses only elementary coefficient estimates. The reduction constancy is not presupposed: the constructed Kisin module is produced independently of the desired isomorphism, and the appeal to [4] is confined to the small-weight range h<2p. The paper therefore appears to meet the standard for publication once the presentation issues below are addressed.","major_comments":[],"minor_comments":[{"comment":"The notation A:=∏_m A_m is ambiguous: the inductive definition C^{(m)}=A_m *_φ C^{(m−1)} means that the relevant limit is the right-to-left product lim_{n→∞} A_n⋯A_1, so the product ordering should be stated explicitly.","section":"§4.3 (Theorem 4.3.7)"},{"comment":"The proof of Lemma 4.3.4 handles the diagonal cases (1,1) and (2,2) in detail and leaves the off-diagonal cases to the translations in Remark 4.3.3; since Lemma 4.3.4 underpins Proposition 4.3.5(d), adding one sentence recording the parameter choices for α_{12} and α_{21} would make the verification of the off-diagonal bounds easier to check.","section":"§4.3 (Lemma 4.3.4)"},{"comment":"The displayed lower bound for v_R(P−T_{≤N}(a_p(λ_−/λ_++)^h)) is missing a closing parenthesis in both Theorem 5.2.1 and the proof of Proposition 5.2.2; the parenthesized expression should read T_{≤N}(a_p(λ_−/λ_++)^h).","section":"§5.2 (Theorem 5.2.1 and Proposition 5.2.2)"},{"comment":"The sentence about Arsovski's work contains a double negative: 'they do not recover neither the more specific Theorem 1.2.1 nor Theorem 5.2.1' should be 'they do not recover either the more specific Theorem 1.2.1 or Theorem 5.2.1'.","section":"§1.1"}],"recommendation":"minor_revision","confidential_remarks":"The paper relies on [25, Proposition 5.3] for the monodromy criterion in Corollary 2.2.5, and the second author of the present paper is a coauthor of [25]; the use is transparent, the result is published, and I do not see a conflict that affects the recommendation. The manuscript is a good fit for a number theory journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"John,\n\nYou should know two things before you spend time on this paper. First, the main theorem is real progress: it improves the fifteen-year-old Berger–Li–Zhu bound on slope stability for mod p reductions of two-dimensional crystalline representations from floor((k-2)/(p-1)) to floor((k-1)/p), and in that range identifies the reduction explicitly as the induction of omega_2^{k-1} chi. Second, the proof is a direct calculation, not a black box: the authors construct rational Kisin modules explicitly, then run a descent algorithm to get integral modules and read off the reduction. I read Section 4.3 carefully, and the error-growth estimate in Proposition 4.3.5(d) is load-bearing but appears sound. The reader's report and the stress-test both went through the same calculation and found no gap.\n\nWhat is actually new: the explicit family of phi-modules in Section 3 and the descent algorithm in Section 4. Previous explicit Kisin-module computations (Caruso–David–Mézard; Le–Le Hung–Levin–Morra) were limited to small weights. Here the descent works uniformly in the weight, with careful p-adic estimates. The integrality step (Prop 5.2.2) is clean, and the application to the reduction is immediate.\n\nSoft spots, in proportion: the descent algorithm is intricate and not machine-checked. The estimates in Lemma 4.3.4 have an exceptional case for alpha_11 that the authors handle with part (c); it is delicate but does not look wrong. The paper relies on external inputs—Kisin's theory for the equivalence, Berger's theorem for small weights h < 2p, and [25, Prop 5.3] for the monodromy criterion. The last is by one of the authors, but it is a published result in Inventiones and the use is legitimate; self-citation is not a problem here. The theorem itself is a bound, not a full classification, and the authors note that Arsovski later improved it under extra conditions. That does not undercut the result; the method here is different and gives the explicit reduction for all a_p in the range.\n\nWho this is for: anyone working on Serre weight conjectures, p-adic local Langlands, or explicit integral p-adic Hodge theory. It deserves a serious referee. My recommendation: send it to review. The mathematics is sound, and the paper would benefit from minor expository additions in Section 4.3, but I would not desk-reject.","headline":"Solid paper: proves a genuine improvement to the Berger–Li–Zhu bound by explicit Kisin-module calculations, and the descent algorithm appears to hold up on a careful read.","tokens_in":21464,"tokens_out":3228,"would_cite":true,"duration_ms":31633,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11F80","11F85"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every $k\\geq 2$, once $v_p(a_p)>\\lfloor (k-1)/p \\rfloor$, the semisimple mod $p$ reduction of $V_{k,a_p}$ is $V_{k,0}$.","keywords":["crystalline representations","mod p reduction","Kisin modules","monodromy condition","descent algorithm","p-adic Hodge theory","Galois representations","local constancy"],"falsifier":"Take a prime $p$ and a weight $k\\geq 2p+1$, choose $a_p$ with $v_p(a_p)$ just above $\\lfloor (k-1)/p \\rfloor$, and compute the semisimplification of $V_{k,a_p}$ modulo $p$ by an independent method (for instance, the construction of [6]); if the result is not isomorphic to $\\operatorname{Ind}_{G_{\\mathbb{Q}_{p^2}}}^{G_{\\mathbb{Q}_p}}(\\omega_2^{k-1}\\chi)$, the theorem is false.","tokens_in":20423,"feed_emoji":"🧮","tokens_out":13601,"duration_ms":121487,"temperature":0.7,"pith_summary":"The paper proves that, for every prime $p$ and every weight $k\\geq 2$, once the slope $v_p(a_p)$ of a two-dimensional crystalline Galois representation of $G_{\\mathbb{Q}_p}$ exceeds $\\lfloor (k-1)/p \\rfloor$, the semisimple reduction modulo $p$ no longer depends on $a_p$: it is the same representation as at $a_p=0$, namely the induction $\\operatorname{Ind}_{G_{\\mathbb{Q}_{p^2}}}^{G_{\\mathbb{Q}_p}}(\\omega_2^{k-1}\\chi)$. This extends the previous constancy range of [6] from $\\lfloor (k-2)/(p-1) \\rfloor$ to the larger $\\lfloor (k-1)/p \\rfloor$. The proof works by constructing explicit integral Kisin modules: for $k\\geq 2p+1$ and $v_p(a_p)>\\lfloor (k-1)/p \\rfloor$, the Frobenius can be put in the simple form $\\begin{pmatrix}P&-1\\\\E^{k-1}&0\\end{pmatrix}$ with a single polynomial $P$ of degree at most $k-1$, $P(0)=a_p$, and $P$ integral when the slope is large enough. Reducing that matrix modulo $p$ gives $\\begin{pmatrix}0&-1\\\\u^{k-1}&0\\end{pmatrix}$ independently of $a_p$, which forces the constant reduction. The method replaces the usual $p$-adic local Langlands route with a direct descent calculation on Kisin modules, and it is set up to work beyond the crystalline $\\mathrm{GL}_2(\\mathbb{Q}_p)$ setting.","feed_headline":"Past slope ⌊(k−1)/p⌋, mod p reduction is constant","feed_subtitle":"At such slopes the reduction is explicit and constant, no longer depending on a_p.","key_machinery":"The load-bearing objects are Kisin modules—integral $\\phi$-modules over $\\Lambda[[u]]$ of finite $E$-height—and a descent algorithm for them. The paper starts from a one-parameter family of $\\phi$-modules over the rigid-analytic ring $R=O_{F,[0,p^{-1/p}]}$ with Frobenius matrix $C_{a_p}=\\begin{pmatrix}a_p(\\lambda_-/\\lambda_{++})^{h}&-1\\\\E^h&0\\end{pmatrix}$, where $h=k-1$ and $\\lambda_\\pm$ are explicit infinite products defined from $E(u)=u+p$. It then defines $\\gamma$-allowable matrices and four allowed row operations $\\alpha_{ij}$ that replace $C$ by $A\\ast_\\phi C=AC\\phi(A)^{-1}$; the key estimate (Proposition 4.3.5(d)) is that each block of operations raises the numerical error $\\varepsilon_C$ by at least $\\min\\{\\gamma,p-1\\}$, so the errors go to infinity, the infinite product of the $A_m$ converges in $\\mathrm{GL}_2(R)$, and the conjugate matrix becomes polynomial with controlled integrality. This polynomial matrix gives the desired Kisin module, and the monodromy condition—checked via Corollary 2.2.5 on a single disc—guarantees it corresponds to the intended crystalline representation.","core_discovery":"The central theorem (Corollary 5.2.3) states that for every $k\\geq 2$ and every $a_p$ with $v_p(a_p)>\\lfloor (k-1)/p \\rfloor$, the semisimple reduction modulo $p$ of the crystalline representation $V_{k,a_p}$ is isomorphic to $V_{k,0}$; explicitly, it is the induction to $G_{\\mathbb{Q}_p}$ of the character $\\omega_2^{k-1}\\chi$ of the unramified quadratic extension of $\\mathbb{Q}_p$. The proof goes through an explicit family of rank-two Kisin modules. The base case $a_p=0$ has Frobenius matrix $\\begin{pmatrix}0&-1\\\\E^{k-1}&0\\end{pmatrix}$ and an explicit monodromy operator; deforming by $a_p$ gives a one-parameter family of $\\phi$-modules on a $p$-adic disc, and the descent algorithm of Section 4 transforms each such module into one whose Frobenius matrix is $\\begin{pmatrix}P&-1\\\\E^{k-1}&0\\end{pmatrix}$ with $P$ a polynomial of degree at most $k-1$ and $P(0)=a_p$. When $v_p(a_p)>\\lfloor (k-1)/p \\rfloor$ and $k\\geq 2p+1$, $P$ has integral coefficients, so reducing modulo the maximal ideal yields the $a_p$-independent matrix $\\begin{pmatrix}0&-1\\\\u^{k-1}&0\\end{pmatrix}$; the small-weight cases $k<2p+1$ follow from the earlier small-weight theorem of [4].","pith_inferences":["The Section 4 descent estimates are formulated for a general modulus $m$ and the initial setup allows a general Eisenstein polynomial $E(u)$, so the same row-reduction algorithm should produce explicit integral Kisin modules for crystalline representations of unramified extensions of $\\mathbb{Q}_p$; checking that is a direct translation of the paper's estimates.","The threshold $\\lfloor (k-1)/p \\rfloor$ is probably not optimal: the paper itself points to computational and global evidence for $\\lfloor (k-1)/(p+1) \\rfloor$. A natural test is to run the descent with the choice $a'=h/2-(p-1)/2$ replaced by values tuned to that smaller bound, and see where the error estimate breaks.","Because Theorem 5.2.1 gives an explicit polynomial $P$ for any $v_p(a_p)>0$ over $F$, the algorithm could in principle be implemented symbolically to tabulate Kisin modules and compare reductions with weight-elimination predictions.","The explicit polynomial $P$ is $p$-adically close to the truncation of $a_p(1+u^p/p)^{k-1}$; this proximity might connect the mod $p$ reduction to the slope filtration of overconvergent modular forms, though the paper does not explore that."],"forward_implications":["For every $k\\geq 2$, the reduction $\\overline{V}_{k,a_p}$ is constant on the slope interval $v_p(a_p)>\\lfloor (k-1)/p \\rfloor$, and it equals $\\overline{V}_{k,0}$.","The constancy range improves from $\\lfloor (k-2)/(p-1) \\rfloor$ to $\\lfloor (k-1)/p \\rfloor$, enlarging the region in which the mod $p$ reduction is known explicitly.","In weights $k\\geq 2p+1$ the proof produces an explicit integral Kisin module whose Frobenius matrix is $\\begin{pmatrix}P&-1\\\\E^{k-1}&0\\end{pmatrix}$ with $P\\in m_F[u]$ of degree at most $k-1$ and $P(0)=a_p$; reducing this matrix modulo $p$ gives $\\begin{pmatrix}0&-1\\\\u^{k-1}&0\\end{pmatrix}$.","The reduction of $V_{k,a_p}$ modulo $p$ is therefore the same as that of $V_{k,0}$, which is the induction $\\operatorname{Ind}_{G_{\\mathbb{Q}_{p^2}}}^{G_{\\mathbb{Q}_p}}(\\omega_2^{k-1}\\chi)$.","Because the construction uses Kisin modules rather than $p$-adic local Langlands, the same descent algorithm is available for semi-stable, non-crystalline inputs and for representations beyond $\\mathrm{GL}_2(\\mathbb{Q}_p)$, as the paper notes."],"supporting_citations":[{"why":"Supplies the Kisin-module theory used throughout: the functor D to filtered (φ,N)-modules and the equivalence that lets an integral Kisin module determine the associated Galois representation.","marker":"[23]"},{"why":"Gives the previous constancy bound floor((k-2)/(p-1)) via a different integral p-adic Hodge theory construction; Theorem 1.1.1 improves this range.","marker":"[6]"},{"why":"Supplies the small-weight theorem used to cover h<2p, so Corollary 5.2.3 holds for every k≥2 rather than only k≥2p+1.","marker":"[4]"},{"why":"Gives the explicit description of V_{h+1,0} as an induction of ω_2^h χ, identifying the constant reduction in Corollary 5.2.3.","marker":"[10]"},{"why":"Its monodromy criterion underlies Corollary 2.2.5, which lets the paper verify the monodromy condition on one p-adic disc instead of all completions.","marker":"[25]"}],"fun_headline_variants":["Slope beyond floor bound forces constant mod p reduction","Mod p reduction stops varying past a slope threshold","Reduction independent of a_p for slope above floor bound","For high slopes, mod p reduction is fixed and explicit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim rests on the estimate that every allowed row operation raises the numerical error of a $\\gamma$-allowable matrix by at least $\\min\\{\\gamma,p-1\\}$; if that estimate fails for even one matrix, the descent produces no explicit Kisin module and the calculation of the reduction collapses.","fun_headline_variants_meta":{"raw":{"variants":["Slope beyond floor bound forces constant mod p reduction","Mod p reduction stops varying past a slope threshold","Reduction independent of a_p for slope above floor bound","For high slopes, mod p reduction is fixed and explicit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00046,"raw_usage":{"total_tokens":2326,"prompt_tokens":987,"completion_tokens":1339,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":603,"completion_tokens_details":{"reasoning_tokens":1275}},"tokens_in":603,"tokens_out":1339,"duration_ms":9575,"temperature":1.0,"reasoning_tokens":1275,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:24:50.563998+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a prime $p$ and a weight $k\\geq 2p+1$, choose $a_p$ with $v_p(a_p)$ just above $\\lfloor (k-1)/p \\rfloor$, and compute the semisimplification of $V_{k,a_p}$ modulo $p$ by an independent method (for instance, the construction of [6]); if the result is not isomorphic to $\\operatorname{Ind}_{G_{\\mathbb{Q}_{p^2}}}^{G_{\\mathbb{Q}_p}}(\\omega_2^{k-1}\\chi)$, the theorem is false.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Kisin-module theory used throughout: the functor D to filtered (φ,N)-modules and the equivalence that lets an integral Kisin module determine the associated Galois representation."},{"cited_title":"Berger, H","cited_arxiv_id":null,"evidence_quote":"Gives the previous constancy bound floor((k-2)/(p-1)) via a different integral p-adic Hodge theory construction; Theorem 1.1.1 improves this range."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the small-weight theorem used to cover h<2p, so Corollary 5.2.3 holds for every k≥2 rather than only k≥2p+1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the explicit description of V_{h+1,0} as an induction of ω_2^h χ, identifying the constant reduction in Corollary 5.2.3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Its monodromy criterion underlies Corollary 2.2.5, which lets the paper verify the monodromy condition on one p-adic disc instead of all completions."}],"review_version":1}