{"id":"55a64c65-72c3-403a-b39f-1d758b3330a4","arxiv_id":"1908.01416","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For excellent strictly Henselian Gorenstein rational surface singularities of mixed characteristic (0,p>5), there is a split finite cover by a regular scheme, based on a new classification of the defining equations.","lead":"This paper classifies the exact equations that define rational surface singularities in mixed characteristic and proves each can be covered by a smooth (regular) ring in a way that splits. The result matters because it carries a classification known in equal characteristic into the mixed characteristic setting and settles a question about which singularities are direct summands of regular rings.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"E8 descent: Claim 4.12 does not justify that the reduction modulo (p−Q) preserves degree 120; target regularity is asserted rather than proven.","rationale":"The paper's main theorem is a substantial classification and construction result, and most of the proof—the An, Dn, E6, E7 cases via cyclic covers, the characteristic-free classification arguments, and the descent from complete to Henselian via Lipman and Elkik—appears internally consistent. The single most delicate step is the E8 construction, which is the only case not handled by a cyclic cover. There, the split cover is produced via an explicit degree-120 map and a trace-splitting argument. The correctness of this step hinges entirely on Claim 4.12, whose proof conflates the rank of a localized finite free extension with the generic degree of the quotient rings after reduction. The claim as stated is false if the quotient B/pB is non-reduced: for example, A=k[[t]], B=k[[s]], t=s^2, p=(t) gives A_p→B_p of rank 2 but generic degree 1. In the application, the target A = W(k)[[u,v]]/(p−Q(f1,f2,f3)) is in fact regular, because the f_i have no linear terms and hence p−Q(f1,f2,f3) is nonzero modulo (p,u,v)^2; this would supply the missing domain hypothesis. However, the paper never explicitly proves this, and the étaleness check is relegated to a single sentence. Since the trace-splitting argument requires both finiteness and a degree-120 generic fiber of the reduced extension, a careful reader cannot verify Theorem A from the text alone. The concern is real and load-bearing, but it is local and likely fixable by adding a short justification, so the appropriate verdict is CONDITIONAL rather than REJECT. The suggested computational check would settle whether the descent actually preserves degree 120.","tokens_in":23401,"tokens_out":41167,"duration_ms":439350,"concrete_test":"Use a Gröbner basis computation over Z[1/30] to verify two facts for the explicit f1,f2,f3 in (4.11): (i) p−Q(f1,f2,f3) has a nonzero image in (p,u,v)/(p,u,v)^2, so A is a 2-dimensional regular local ring (hence a domain); (ii) the discriminant ideal of the finite map W(k)[[x,y,z]]/(x^2+y^3+z^5) → W(k)[[u,v]] is not contained in the ideal (p−Q), so the reduction is generically étale of degree 120. If both hold, the E8 descent is sound; if either fails, the trace splitting in Proposition 4.11 needs repair.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem A's E8 case (Proposition 4.11) relies on Claim 4.12 to conclude that after reducing the degree-120 cover W(k)[[x,y,z]]/(x^2+y^3+z^5) → W(k)[[u,v]] modulo the prime p = (p−Q), the resulting map R → A is a finite extension of degree 120. The proof of Claim 4.12 shows that A_p → B_p is finite free of rank d, then states 'of course' that this rank is the residual degree, which is the generic degree of A/p → B/pB. This identification is valid only when B/pB is a domain (or at least reduced); otherwise the length of B_p/pB_p can exceed the degree of the reduced generic fiber. In the application, A = W(k)[[u,v]]/(p−Q(f1,f2,f3)) is asserted to be regular—hence a domain—but no proof is given at that stage, and the étaleness on the punctured spectrum is only mentioned afterwards. Without a proof that A is a domain, the field trace argument does not go through, and the split cover is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper classifies, in mixed characteristic (0,p>5), the possible defining equations of a 2-dimensional rational double point inside a 3-dimensional regular local ring, proving the normal forms listed in Theorem B (An, Dn, E6, E7, E8). It then uses this classification to prove Theorem A: every excellent strictly Henselian 2-dimensional Gorenstein rational singularity is a rational double point and admits a finite cover by a regular scheme that splits as a map of O_X-modules. The final section applies Theorem A to show that 2-dimensional BCM-regular singularities of mixed characteristic (0,p>5) are finite direct summands of regular rings.","tokens_in":23603,"tokens_out":16478,"duration_ms":165038,"significance":"The classification and the split-cover theorem are a substantial advance: they extend Artin's equicharacteristic classification and Lipman's E8 treatment to mixed characteristic, with explicit cyclic covers for types A, D, E6, E7 and an explicit degree-120 cover for E8. The proof is organized by singularity type and includes detailed coordinate-change arguments. If the gaps in the E8 descent are repaired, the results would be a significant step for the theory of rational singularities and for mixed-characteristic BCM-regularity.","major_comments":[{"comment":"Claim 4.12 is false as stated. Take A = k[t], B = k[t,s]/(s^2 - t), and p = (t). Then A -> B is a finite extension of domains of generic degree 2, B is Cohen-Macaulay, and pd_A(A/p) is finite, but B/pB = k[s]/(s^2) is not reduced; the degree of the reduced generic fiber of A/p -> B/pB is 1, not 2. The proof of the claim identifies the rank of the finite free localization B_p over A_p with the residual degree, but when B/pB has nilpotents the length of the generic fiber exceeds the sum of the degrees of its reduced components. In the application, A = B0/pB0 is asserted to be regular and hence a domain, but that assertion is not part of Claim 4.12 and is not proved at the point where it is used. Therefore the trace-map argument in Proposition 4.11 does not currently establish that R -> A is a finite extension of degree 120.","section":"§4.11, Claim 4.12"},{"comment":"The finiteness of W(k)[[x,y,z]]/(x^2+y^3+z^5) -> W(k)[[u,v]] over W(k) is justified only by inverting p and invoking the equicharacteristic zero case. Localization at the element p is not faithfully flat because p lies in the maximal ideal, so finiteness of A0[1/p] -> B0[1/p] does not imply finiteness of A0 -> B0. The authors need a direct proof of integral finiteness over W(k), for example by showing that the f_i generate the invariant subring for a finite group action that is defined over W(k), or by proving that u and v are integral over the image of the hypersurface ring.","section":"§4.11, before Claim 4.12"},{"comment":"The ring A := W(k)[[u,v]]/(p - Q(f1,f2,f3)) is called a regular local ring without proof. This is true because p - Q(f1,f2,f3) has class p in m/m^2 (as Q(f1,f2,f3) is in (u,v)), so it is a regular parameter in the regular ring W(k)[[u,v]]; but the argument should be included, since the claim that A is a domain is needed to identify the generic degree of the reduced fiber with the localization rank. As written, the proof of the E8 case is incomplete at this point.","section":"§4.11, definition of A"}],"minor_comments":[{"comment":"The section title 'Prelimaries' should be spelled 'Preliminaries'.","section":"Table of contents / Section 2"},{"comment":"The phrase 'Assume noe p | n' should read 'Assume now p | n'.","section":"Lemma 5.1"},{"comment":"The phrase 'one can check easily, with for instance Macaulay2' appears for a finite polynomial identity used in the E7 reduction. The authors should either display the expanded identity or state that it is a direct finite expansion, since the surrounding proof is otherwise computational.","section":"Proposition 3.8"},{"comment":"The sentence 'The classification of such functions are used' should use the singular verb 'is used'.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The E8 descent issue is the only substantive obstacle I see. I believe it is repairable with a direct proof that the reduction is a domain and that the extension is finite over W(k), so I recommend major revision rather than rejection. The rest of the paper is careful and the classification arguments are convincing."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing you should know: this paper delivers the mixed characteristic (0,p>5) analogue of Artin's classification of rational double point equations, and uses it to prove that every excellent strictly Henselian Gorenstein rational surface singularity of dimension 2 admits a split finite cover by a regular scheme. That is a real result, and the paper earns it with explicit case-by-case arguments.\n\nThe classification in Theorem B is genuinely new for mixed characteristic. The proofs in Section 3 follow Artin's method but adapt it carefully to W(k)-coefficients, with Cauchy-sequence arguments to justify the terminating coordinate changes. The E8 cover is backed by explicit polynomials, credit to Klein and to Lipman, and the descent step is the right idea.\n\nI also want to flag: the stress-test concern about Claim 4.12 does not survive a close reading. In the application, the prime p is the principal ideal (p−Q), so in the quotient B/pB that ideal becomes zero. The localization B_p/pB_p is therefore the total ring of fractions of B/pB, and its dimension over the residue field of A_p is exactly the generic degree of the reduced fiber. The 'of course' in the claim is terse but correct. The target ring A is asserted to be regular; that is also fine, since the f_i have no linear terms, so p−Q(f_i) is congruent to p modulo the square of the maximal ideal.\n\nWhat is genuinely soft is a different point in Proposition 4.11. The paper states that the map W(k)[[x,y,z]]/(x^2+y^3+z^5) → W(k)[[u,v]] is a finite extension of degree 120, and justifies it by saying this follows after inverting p. That only proves generic finiteness. To conclude finiteness over W(k), you need the special fiber to be finite, which is true here—the images of f_1,f_2,f_3 in k[[u,v]] have positive degrees and cut out a finite-length quotient—but the paper does not say that. This is a presentation gap, not a fatal flaw; a referee should ask for a one-paragraph proof.\n\nMinor quibbles: some of the Macaulay2-backed computations in the E7 case are left to the reader, and the proof of Lemma 5.1 has a typo ('noe'). None of this affects the main claims.\n\nWho is this for? Anyone working on mixed characteristic singularities, splinters, or BCM-regularity. It deserves a serious referee; I would accept it for peer review and request the finiteness clarification and a slightly expanded Claim 4.12. I would cite it if I worked in this area.","headline":"Solid mixed-characteristic classification paper; the E8 descent worry dissolves on close reading, but the finiteness step before reduction deserves a clearer proof.","tokens_in":24125,"tokens_out":16210,"would_cite":true,"duration_ms":165871,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14J17","14B05","13H10"],"pacs":[],"model":"deepseek-v4-flash","headline":"In mixed characteristic $(0,p>5)$, every Gorenstein rational surface singularity is a rational double point with a split finite regular cover.","keywords":["rational surface singularities","rational double points","mixed characteristic","split finite covers","cyclic covers","Dynkin diagrams","BCM-regular singularities","classification of singularities"],"falsifier":"Take $p=7$ and compute the module structure of $W(k)[[u,v]]$ over $W(k)[[x,y,z]]/(x^2+y^3+z^5)$ under the map sending $x,y,z$ to the three explicit polynomials $f_1,f_2,f_3$; concretely, test whether some power of the maximal ideal annihilates the cokernel. If the extension is not finite, the trace-splitting proof of Proposition 4.11 fails.","tokens_in":2064,"feed_emoji":"📐","tokens_out":4425,"duration_ms":117162,"temperature":0.7,"pith_summary":"This paper extends the classical story of rational surface singularities to mixed characteristic $(0,p>5)$. Its main theorem asserts that every excellent, strictly Henselian, 2-dimensional Gorenstein rational singularity is a rational double point—one of the hypersurface types $A_n$, $D_n$, $E_6$, $E_7$, $E_8$—and that each such singularity admits a finite cover by a regular local ring for which the structure map splits as a map of modules. Splitting means the singular ring is a direct summand of a regular ring, the same behavior already known in equicharacteristic $0$ and in prime characteristic $p>5$. The authors first classify all defining equations (Theorem B), then build the covers type by type: cyclic covers for $A_n$, $D_n$, $E_6$, and $E_7$, and an explicit degree-120 cover for $E_8$. They apply the result to show that 2-dimensional BCM-regular singularities in this mixed-characteristic setting are finite direct summands of regular rings.","feed_headline":"Mixed-characteristic rational double points admit split covers","feed_subtitle":"Every listed singularity type gets an explicit cover: cyclic covers for A, D, E6, E7 and a degree-120 cover for E8.","key_machinery":"The load-bearing objects are the rational double point equations and the cyclic cover construction attached to the divisor class group. For each type except $E_8$, the paper identifies a pure height-one prime ideal whose class has finite order in the divisor class group; the cyclic cover of that order is a domain by the Tomari--Watanabe cyclic cover theorem, and the authors verify directly that it is regular and that the structure map splits. For $E_8$, whose divisor class group is trivial, the cover is instead the explicit subring $W(k)[[u,v]]$ mapped by the icosahedral polynomial triple $f_1,f_2,f_3$; the trace map of this degree-120 finite extension, divided by the degree, gives the splitting. A general descent claim (Claim 4.12) is used to pass the finiteness and degree of this extension from characteristic zero down to the mixed-characteristic Witt vector setting.","core_discovery":"The paper's central claim is that rational double points in mixed characteristic $(0,p>5)$ behave like their equicharacteristic counterparts not only in classification but also in covering: they are covered by regular schemes in a split way. Theorem B states that if $(S,\\mathfrak{n},k)$ is a 3-dimensional complete regular local ring of mixed characteristic with separably closed residue field and $f \\in \\mathfrak{n}^2-\\mathfrak{n}^3$ defines a rational double point, then after choosing minimal generators and multiplying by a unit, $f$ must be exactly one of $x^2+y^2+z^{n+1}$, $x^2+y^2z+z^{n-1}$, $x^2+y^3+z^4$, $x^2+y^3+yz^3$, or $x^2+y^3+z^5$. Theorem A then asserts that a strictly Henselian excellent Gorenstein rational singularity of dimension 2 is such a hypersurface and that there is a finite cover $Y \\to \\operatorname{Spec}(R)$ with $Y$ regular such that $\\mathcal{O}_R \\to \\pi_*\\mathcal{O}_Y$ splits. The covers are constructed explicitly, and in the $E_8$ case the construction exhibits three polynomials $f_1,f_2,f_3$ in two variables satisfying $f_1^2+f_2^3+f_3^5=0$ and generating a finite split extension of degree 120.","pith_inferences":["The explicit $E_8$ cover polynomials are written with rational coefficients, so it is worth testing whether the same triple yields a split regular cover in equicharacteristic $p>5$ and possibly at boundary primes where the current hypotheses stop applying.","The chain of cyclic covers $A \\leftarrow D \\leftarrow E_6 \\leftarrow E_7$, capped by the degree-120 $E_8$ cover, suggests that the split covers might assemble into a tower whose Galois group is related to the Weyl group or icosahedral group; the paper does not pursue this structural statement.","Because the classification is of equations up to unit multiplication rather than of rings up to isomorphism, the same Dynkin type can hide genuinely different mixed-characteristic singularities; the explicit covers may depend on the chosen $p$-adic parameter, so tracking that dependence is a natural next step."],"forward_implications":["Every $A_n$ singularity in this setting has a cyclic cover of index $n+1$ that is regular, since the divisor class of $(x+iy,z)$ has order $n+1$.","Every $D_n$ singularity has a degree-2 cyclic cover of type $A_{2n-5}$; $E_6$ has a degree-3 cyclic cover of type $D_4$; and $E_7$ has a degree-2 cyclic cover of type $E_6$, so the covers chain down to a regular ring.","For $E_8$, an explicit finite split regular cover is given by the polynomials $f_1,f_2,f_3$, and the cover is etale on the punctured spectrum whenever the residual characteristic is $p>5$.","As an application, every 2-dimensional BCM-regular local singularity of mixed characteristic $(0,p>5)$ is a finite direct summand of a regular ring (Theorem 5.2)."],"supporting_citations":[{"why":"Supplies the foundational machinery: rational surface singularities are resolved by quadratic transforms, the dual graphs are the A-D-E Dynkin diagrams, and the E8 equation has a purely mixed-characteristic treatment.","marker":"[Lip69]"},{"why":"Provides Artin's equicharacteristic classification and coverings of rational double points that this paper extends to mixed characteristic.","marker":"[Art77]"},{"why":"Gives the cyclic cover theorem used to show the index-$n$ covers built from torsion Weil divisors are domains.","marker":"[TW92]"},{"why":"Contains the icosahedral polynomial expressions $f_1,f_2,f_3$ satisfying $f_1^2+f_2^3+f_3^5=0$, which form the explicit $E_8$ cover.","marker":"[Kle56]"},{"why":"Supplies the theorem on minimal reductions used to prove that a 2-dimensional Gorenstein rational singularity has multiplicity 2 and is a hypersurface.","marker":"[LT81]"},{"why":"Provides the Henselian descent result that lets the completed $E_8$ extension descend to a strictly Henselian local ring without losing regularity.","marker":"[Elk73]"},{"why":"Defines BCM-regularity and supplies the purity criteria used in the final application to 2-dimensional BCM-regular singularities.","marker":"[MS18]"},{"why":"Establishes that direct summands of regular rings behave like splinters in mixed characteristic, which is used in the counterexample discussion for small primes.","marker":"[And18]"}],"fun_headline_variants":["Explicit split covers for all rational double points in mixed char","E8 rational double point: degree-120 split cover in mixed char","Mixed char rational surface singularities admit split regular covers","Classification and split covers for rational double points in mixed char","Split regular covers for Gorenstein rational singularities in mixed char"],"cache_read_input_tokens":26368,"weakest_assumption_plain":"The heart of the proof is a descent step: a degree-$120$ cover that is finite over the characteristic-zero ring is asserted to remain finite after reducing modulo the mixed-characteristic prime; if the reduction loses finiteness, the trace map that splits the $E_8$ cover may not exist.","fun_headline_variants_meta":{"raw":{"variants":["Explicit split covers for all rational double points in mixed char","E8 rational double point: degree-120 split cover in mixed char","Mixed char rational surface singularities admit split regular covers","Classification and split covers for rational double points in mixed char","Split regular covers for Gorenstein rational singularities in mixed char"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000461,"raw_usage":{"total_tokens":2329,"prompt_tokens":990,"completion_tokens":1339,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":606,"completion_tokens_details":{"reasoning_tokens":1263}},"tokens_in":606,"tokens_out":1339,"duration_ms":12645,"temperature":1.0,"reasoning_tokens":1263,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:14:07.257207+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $p=7$ and compute the module structure of $W(k)[[u,v]]$ over $W(k)[[x,y,z]]/(x^2+y^3+z^5)$ under the map sending $x,y,z$ to the three explicit polynomials $f_1,f_2,f_3$; concretely, test whether some power of the maximal ideal annihilates the cokernel. If the extension is not finite, the trace-splitting proof of Proposition 4.11 fails.","supporting_citations":[],"review_version":1}