{"id":"403f0636-91db-4abf-886b-783701085b3c","arxiv_id":"1908.05037","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper completes termination of the foliated minimal model program on threefolds and proves that terminal singularities admit first integrals and log canonical singularities admit separatrices.","lead":"This math paper studies foliations, which are continuous ways of slicing a three-dimensional space into one-dimensional curves. It proves that the main simplification algorithm for foliated spaces terminates, and that certain singular foliations always contain an invariant curve, tools that will be used in future classification and dynamics results.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The non-canonical case of Theorem 6.1 rests on Lemma 6.8, which invokes the foliated MMP over a local analytic klt germ without supplying the needed local version of [CS21]'s quasi-projective MMP.","rationale":"The reader identified Lemma 6.8 and the [CS21] machinery as the weakest assumption, and I agree that this is the load-bearing point of the central claim. My stress-test sharpens it: the MMP invoked in Lemma 6.8 is stated over Q-factorial quasi-projective threefolds, whereas Theorem 6.1 concerns local analytic germs that are not assumed Q-factorial. The proof of Lemma 6.8 also implicitly relies on the fact that a non-canonical log canonical foliation singularity has a transverse exceptional divisor of discrepancy exactly −ε; this is plausible from [CS21, Theorem 1.16] but is not spelled out in the text. I found no internal contradiction in the main argument, and the external dependencies are to published or soon-to-be-published machinery, so I do not think the verdict should change. However, the paper would be more robust if it either stated a local analytic version of the F-dlt modification theorem or explicitly reduced the germ case to the quasi-projective setting. The reader's ACCEPT with MODERATE confidence remains appropriate.","tokens_in":50844,"tokens_out":20740,"duration_ms":238398,"concrete_test":"Write out, for the category of isolated klt threefold germs used in Section 6, the local analytic analogue of [CS21, Theorem 8.1] and Corollary 2.3, or give an explicit reduction of Lemma 6.8 to the quasi-projective MMP. A minimal test is to take an isolated klt threefold germ that is not globally Q-factorial and reproduce the steps of Lemma 6.8: construct the F-dlt modification and the subsequent δ-MMP; if the construction needs a compact quasi-projective model that the germ does not admit, the lemma lacks a stated hypothesis, and the non-canonical case of Theorem 6.1 is conditional on an unproved extension.","verdict_should_be":"UNCHANGED","load_bearing_attack":"To prove Theorem 6.1 for log canonical but not canonical F, the proof must enter the dicritical regime. The engine is Lemma 6.8: it asserts a birational morphism π: Y → X and a transverse π-exceptional divisor E0 with π^{-1}(P) ⊂ E0, constructed by taking an F-dlt modification of (X, F) and then running a (K_F + Σ ε(E'_i)E'_i − δE'_0)-MMP over X. Both Theorem 2.4 and Corollary 2.3 are stated for Q-factorial quasi-projective threefolds, while P ∈ X in Theorem 6.1 is an analytic germ and X is not assumed Q-factorial. No algebraization, compactification, or local analytic version of the foliated MMP is provided. This is load-bearing because Lemma 6.9 and the curve case of Theorem 6.1 use π^{-1}(P) ⊂ E0 to produce an invariant divisor whose pushforward is the separatrix. Without a local or localized MMP, the argument only covers foliation germs that are restrictions of quasi-projective models. This is not an internal contradiction: if [CS21] contains the needed local statement, the gap is expository. But the paper does not identify where that statement appears, and the reader's verdict should record the dependency.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops local and global applications of the recently developed minimal model program for co-rank one foliations on threefolds. Locally, it proves a singular analogue of Malgrange's theorem (Theorem 5.1), a classification of terminal foliation singularities as cyclic quotients of smoothings of Du Val singularities (Theorem 5.20), and the existence of separatrices for log canonical foliation singularities on isolated klt bases (Theorem 6.1). Globally, it proves termination of flips (Theorem 2.1), a non-vanishing theorem (Theorem 2.6), connectedness of non-klt loci (Theorem 3.1), inversion of adjunction (Theorem 3.12), a foliated relative vanishing theorem (Theorem 4.3), and a foliated Mori hyperbolicity statement (Theorem 7.1). The proofs systematically use F-dlt modifications and the foliated MMP, and the paper explicitly relies on [CS21] and [Spi20] for existence of flips, contractions, special termination, and F-dlt modifications.","tokens_in":51096,"tokens_out":4621,"duration_ms":49006,"significance":"If the results hold, this is a substantial advance: it completes the termination part of the foliated MMP on threefolds and gives a general separatrix theorem for log canonical foliation singularities, converting a discrepancy condition into dynamical information. The paper is carefully structured, with the main reductions clearly identified: foliation flips are reduced to usual threefold log terminal flips in Section 2.2, and lc separatrices are constructed through F-dlt modifications in Section 6. The statements are sharp, as Example 6.3 shows that the isolated klt hypothesis on the base is necessary for the stated level of generality, and Jouanolou's example shows that log canonicity is close to optimal. The proofs are detailed and I found no post-hoc selection or circular reasoning in the central arguments; the main concern is an unresolved gap between the stated quasi-projective MMP statements and the local analytic setting used in one load-bearing lemma.","major_comments":[{"comment":"The proof of Lemma 6.8 invokes Corollary 2.3 to run a (K_F + Σ ε(E_i')E_i' − δE_0')-MMP over X, where X is a germ of a normal threefold and not assumed to be quasi-projective. Corollary 2.3 is stated only for Q-factorial quasi-projective threefolds equipped with a projective morphism to an algebraic base; no algebraization of the analytic germ or local analytic version of the foliated MMP is supplied. This is load-bearing: Lemma 6.8 produces the transverse exceptional divisor E0 with π^{-1}(P) ⊂ E0, and Lemma 6.9 and the curve case of Theorem 6.1 use this divisor to construct the separatrix in the log canonical but not canonical case. If the needed local statement is already contained in [CS21], the paper should give a precise citation; otherwise the proof of Theorem 6.1 currently covers only germs that are restrictions of quasi-projective models.","section":"§6, Lemma 6.8"}],"minor_comments":[{"comment":"The abstract contains the typo 'non-vanshing'; it should read 'non-vanishing'.","section":"Abstract"},{"comment":"The phrase 'constrcuted' should be 'constructed'.","section":"§2, Corollary 2.3"},{"comment":"There is a typo 'Lemm 4.1' in the sentence referring to Lemma 4.1; it should read 'Lemma 4.1'.","section":"§4, proof of Lemma 4.2"},{"comment":"Several displayed diagrams and formulas contain OCR artifacts such as '/d47/d47', '/d31/d31', and 'Tr eor em` e'; these should be cleaned in the final version so that the arrows and foreign-language citations are readable.","section":"Throughout"},{"comment":"The references to 'Theor´ em` e A' and 'factorisation' contain encoding errors and should be typeset correctly.","section":"§5.4"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this is a substantial paper, and the main theorems are new and largely believable. It proves termination of flips for F-dlt co-rank one foliations on threefolds, gives a singular Malgrange theorem and a classification of terminal foliation singularities, and proves existence of separatrices for log canonical singularities. The proofs are detailed and the overall architecture is coherent: the reduction of foliation flips to usual threefold log terminal flips in Section 2 is clean, and the connectedness and hyperbolicity results look solid. The authors are honest about what is new versus what comes from [CS21], [Spi20], [CLN08], and [MS19]; the citation pattern is not self-serving.\n\nThe main soft spot is in the proof of Theorem 6.1. The non-canonical case rests on Lemma 6.8, which runs a foliated MMP over a local analytic klt germ. But Corollary 2.3, the relative MMP, is stated for Q-factorial quasi-projective threefolds, and Theorem 2.4 (F-dlt modifications) is also quasi-projective. No algebraization, compactification, or local analytic version of the foliated MMP is provided. This is load-bearing: Lemma 6.9 and the curve case use the exceptional divisor E0 containing π^{-1}(P) to produce the separatrix. If the local MMP is not available, the argument only covers germs that are restrictions of quasi-projective models. This is not an internal contradiction, and I suspect the gap is patchable—one can probably localize the MMP or pass to a neighborhood—but the paper doesn't say how. A referee should ask for this.\n\nSmaller issues: Lemma 4.1 leaves a proof to the reader, which is fine; the heavy reliance on [CS21] as a black box is fair in context, though it makes the paper less self-contained. There is also a typo in Lemma 6.8 where the modification is written with the same symbol on source and target.\n\nWho is this for? People working in birational geometry of foliations, moduli, or hyperbolicity. It will be a standard reference. I'd certainly cite it and would send it to a serious referee. The likely correct outcome is accept after the local MMP gap is addressed, not desk reject.","headline":"Completes the foliated MMP on threefolds with a local analytic gap in the separatrix theorem that a referee should press.","tokens_in":720,"tokens_out":1619,"would_cite":true,"duration_ms":44309,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14E30","37F75","32S65"],"pacs":[],"model":"deepseek-v4-flash","headline":"The minimal model program for threefold foliations converts discrepancy classes into local integrability: canonical singularities have first integrals, terminal singularities are explicit quotients, and log canonical singularities always…","keywords":["foliations","minimal model program","co-rank one foliations","threefolds","log canonical singularities","separatrices","first integrals","hyperbolicity"],"falsifier":"Construct a log canonical co-rank one foliation germ on an isolated klt threefold that has no formal or analytic separatrix; a concrete route is to take a separatrix-free Jouanolou-type 1-form on $\\mathbb{C}^3$, perturb it to make the singular locus isolated, and compute foliated discrepancies to see whether any such germ is log canonical while remaining separatrix-free.","tokens_in":50624,"feed_emoji":"🌀","tokens_out":10368,"duration_ms":98978,"temperature":0.7,"pith_summary":"This paper uses the minimal model program for co-rank one foliations on threefolds to show that the foliated discrepancy, a numerical invariant measuring how the canonical class changes under blow-ups, carries dynamical meaning. Locally, it proves that an isolated canonical foliation singularity on a Q-factorial threefold has a holomorphic first integral, that terminal foliation singularities are classified as quotients of smoothings of Du Val surface singularities, and that every log canonical foliation singularity on an isolated klt threefold admits a separatrix, i.e. an invariant analytic curve through the singular point. Globally, the same machinery yields termination of flips for foliated dlt pairs, a connectedness theorem for non-klt loci, a non-vanishing theorem, and a hyperbolicity criterion. If these results are right, the discrepancy decides when a foliation singularity is locally integrable and when it has local solutions.","feed_headline":"Log canonical foliation singularities always have a separatrix","feed_subtitle":"Even dicritical log canonical germs on isolated klt threefold bases must contain an invariant analytic curve.","key_machinery":"The central object is the F-dlt modification, a foliated analogue of a dlt modification: a birational morphism that extracts only divisors of foliation discrepancy equal to the transversality defect $-\\epsilon(E)$, arranged so that the transformed pair is foliated dlt. Its work is to turn a local singularity into a global model: in the dicritical case it extracts a transverse exceptional divisor $E_0$, and the existence of a separatrix is reduced, via foliated adjunction, to finding an invariant algebraic curve on $E_0$ for a restricted foliation with trivial first Chern class, which is then guaranteed by the classification of rank-one foliations with trivial canonical class on surfaces. A secondary mechanism, used for the Malgrange-type results, is the holomorphic Godbillon-Vey sequence attached to the foliation's defining 1-form, whose existence follows from control of the ambient singularity and which produces the holomorphic first integral.","core_discovery":"The central claim, stated as Theorem 6.1, is that a germ of a log canonical co-rank one foliation singularity on an isolated klt threefold singularity always admits a separatrix: a formal or convergent invariant hypersurface through the singular point. Because log canonical singularities are generally dicritical, the classical non-dicritical separatrix theorem does not apply; the paper supplies a new proof by passing to an F-dlt modification and finding an invariant algebraic curve on a transverse exceptional divisor. The paper also proves a singular Malgrange theorem (Theorem 5.1): an isolated canonical foliation singularity on an isolated analytically Q-factorial threefold has a holomorphic first integral, and for terminal singularities (Theorem 5.20) no assumption on the ambient germ is needed and the germs are classified, up to a $\\mathbb{Z}/n \\times \\mathbb{Z}/m$-cover, as smoothings of Du Val surface singularities. The global half of the paper establishes termination of flips for F-dlt pairs, connectedness of the non-klt locus, a non-vanishing theorem, and a foliated Mori hyperbolicity statement.","pith_inferences":["If Theorem 6.1 extends beyond dimension three, it would settle the log canonical case of the local separatrix problem for foliations in all dimensions; the paper itself raises exactly this as an open question.","The F-dlt modification construction suggests a practical separatrix search: extract a transverse exceptional divisor, compute the restricted foliation, and test for invariant algebraic curves; this could be implemented on explicit 1-forms to detect separatrices computationally.","The terminal classification should constrain deformation theory and moduli of threefold foliations, since every terminal germ must sit inside one of the six Du Val smoothing families before a finite quotient is taken.","The connectedness and hyperbolicity results point toward a log canonical version of the foliated MMP; the paper explicitly asks whether log canonical flips exist, and its separatrix theorem removes a known obstruction to that program."],"forward_implications":["Every log canonical co-rank one foliation singularity on an isolated klt threefold has at least one invariant analytic curve, so the local dynamical system defined by the foliation has a genuine solution curve through the singular point.","An isolated canonical foliation singularity on a Q-factorial threefold is locally a pullback of a smooth foliation: it admits a holomorphic first integral, and consequently a separatrix.","Terminal foliation singularities on threefolds have an explicit description: up to a $\\mathbb{Z}/n \\times \\mathbb{Z}/m$-cover they are smoothings of Du Val surface singularities, and the underlying threefold is terminal.","The foliated MMP terminates for F-dlt co-rank one pairs on Q-factorial threefolds, completing the program started in the companion paper; pseudo-effective $K_F + \\Delta$ then has a nonzero section, and non-klt loci satisfy a connectedness principle.","A foliated hyperbolicity criterion holds: under potential kltness and the absence of non-constant $\\mathbb{A}^1$-curves tangent to the foliation, both in the complement of the non-klt locus and in its strata, $K_F + \\Delta$ must be nef."],"supporting_citations":[{"why":"Supplies the foliated MMP on threefolds—F-dlt modifications, flips, divisorial contractions, and special termination—on which the local and global proofs are built.","marker":"[CS21]"},{"why":"Malgrange's Frobenius theorem with singularities is the statement that Theorem 5.1 extends, and its proof via Godbillon-Vey sequences is adapted.","marker":"[Mal76]"},{"why":"Provides desingularization of non-dicritical foliations and the classical existence theorem for separatrices used in the non-dicritical cases.","marker":"[CC92]"},{"why":"Supplies the classification of rank-one foliations with trivial first Chern class on surfaces, the key input for finding the invariant curve on the transverse exceptional divisor.","marker":"[McQ08]"},{"why":"Gives reduction of codimension-one foliation singularities to simple singularities in dimension three, used to construct foliated log resolutions.","marker":"[Can04]"},{"why":"Standard toolkit for discrepancy, adjunction, termination of log terminal flips, and the basepoint-free theorem used throughout.","marker":"[KM98]"},{"why":"Earlier foliated Mori theory results on negativity, adjunction, and extension of separatrices that the present proofs rely on.","marker":"[Spi20]"},{"why":"Proves Frobenius-type theorems and Godbillon-Vey sequences on singular varieties, directly used in the proof of Theorem 5.1.","marker":"[CLN08]"}],"fun_headline_variants":["Every lc foliation germ on a klt threefold has a separatrix","Dicritical lc foliation germs always contain invariant curves","Lc foliation singularities on klt threefolds have separatrices","Separatrix theorem for all log canonical foliation germs","Foliation threefolds: lc singularities never lack a separatrix"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the foliated minimal model program for co-rank one foliations on threefolds works as established in [CS21]; in particular, the separatrix proof for the dicritical log canonical case needs an F-dlt modification that extracts a transverse exceptional divisor, so a failure of that existence theorem would remove the support from Theorem 6.1.","fun_headline_variants_meta":{"raw":{"variants":["Every lc foliation germ on a klt threefold has a separatrix","Dicritical lc foliation germs always contain invariant curves","Lc foliation singularities on klt threefolds have separatrices","Separatrix theorem for all log canonical foliation germs","Foliation threefolds: lc singularities never lack a separatrix"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001085,"raw_usage":{"total_tokens":4486,"prompt_tokens":847,"completion_tokens":3639,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":463,"completion_tokens_details":{"reasoning_tokens":3546}},"tokens_in":463,"tokens_out":3639,"duration_ms":25538,"temperature":1.0,"reasoning_tokens":3546,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:26:48.048385+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a log canonical co-rank one foliation germ on an isolated klt threefold that has no formal or analytic separatrix; a concrete route is to take a separatrix-free Jouanolou-type 1-form on $\\mathbb{C}^3$, perturb it to make the singular locus isolated, and compute foliated discrepancies to see whether any such germ is log canonical while remaining separatrix-free.","supporting_citations":[],"review_version":1}