{"id":"79ab1f4e-07e7-4383-9bc2-39249178e56e","arxiv_id":"2508.21134","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A systematic study of subtoposes yields explicit formulas for generating Grothendieck topologies and new preservation theorems for pullbacks of subtoposes.","lead":"This mathematics monograph by two topos theorists develops systematic methods for computing with subtoposes: it gives explicit formulas for generating Grothendieck topologies and proves new theorems about how subtoposes behave under pullback. A generalist might care because it connects logical provability to geometric topology generation, a bridge that could eventually feed into computational or machine-learning frameworks.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Finite-union preservation for arbitrary pullbacks rests on a deferred factorization (embedding followed by locally connected morphism) that is not standard and not proved in the visible text; if that factorization fails, Corollary I.3.20 collapses.","rationale":"The reader's weakest assumption is precisely the factorization theorem used to deduce finite-union preservation for all geometric morphisms. My stress-test pass confirms this is the single most load-bearing point: the visible proof of Corollary I.3.20 is only an 'indication' that refers to Chapter IV, and the factorization itself is not a standard result I can verify from the excerpt. I did not find an actual counterexample, and the finite-union result might be true by other means, so the appropriate verdict remains CONDITIONAL as the reader gave. I agree with the reader's identification of the weakest point and do not see a reason to move the verdict up or down.","tokens_in":77695,"tokens_out":28158,"duration_ms":301356,"concrete_test":"Obtain the full text of Chapitre IV, §2c and verify the factorization theorem. As a targeted stress test, apply the factorization to a concrete non-locally-connected morphism whose codomain has nontrivial subobject classifier, e.g., f: Set^N -> Sh(Sierpinski) where N is the discrete category on two objects and Sh(Sierpinski) is the topos of sheaves on the two-point Sierpinski space. Exhibit the intermediate topos E'', check that E'' -> Sh(Sierpinski) is locally connected by verifying the Beck-Chevalley condition (Definition I.3.18(2)) for a non-identity morphism of Sh(Sierpinski), and check that Set^N -> E'' is an embedding. If the manuscript's construction fails this example, or if the claimed factorization is not proved anywhere, the proof of Corollary I.3.20 is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline new result is Corollary I.3.20/Chapter IV: for any geometric morphism f: E' -> E, the pullback operation f^{-1} on subtoposes preserves finite unions. The proof, announced in the preface and in the indication following Corollary I.3.20, is not self-contained: it depends on a factorization theorem asserted in Chapter IV, section 2c, namely that every geometric morphism factors as an embedding E' -> E'' followed by a locally connected morphism E'' -> E. For embeddings, f^{-1} is the meet operation, which preserves finite unions because the lattice of subtoposes is distributive. For locally connected morphisms, f^{-1} has a further left adjoint f_!, so it preserves all unions. Thus the finite-union statement follows from the stated factorization. The load-bearing assumption is that this factorization is true in full generality. It is not a standard factorization in the topos-theory literature (the standard surjection/embedding factorization does not make the first factor locally connected), and the visible excerpt does not include Chapter IV section 2c, so no proof is available. If the factorization is false, the proof of finite-union preservation does not go through by the announced route. I have no counterexample from the visible text, so the concern is about proof completeness and the availability of a nonstandard theorem, not an evident internal inconsistency.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a systematic account of subtoposes of Grothendieck toposes through Galois connections between sieves and presheaves, monomorphisms and objects, and sieves and subpresheaf embeddings. It claims to derive, as fixed points of these connections, the equivalences between Grothendieck topologies and subtoposes and between topologies and theories, and to obtain explicit formulas for the topology generated by a family of sieves or covering families. It further studies Heyting operations on subtoposes and direct/inverse image operations, and states two preservation theorems: inverse image of subtoposes preserves finite unions along any geometric morphism, and arbitrary unions along locally connected morphisms. The visible text contains the preface, Chapter I and part of Chapter II; the announced Chapter III and Chapter IV are largely not present, and several key statements are only backed by sketches or deferred proofs.","tokens_in":77994,"tokens_out":3561,"duration_ms":43370,"significance":"If the full proofs are supplied, the paper would make a useful contribution: the Galois-connection framework is elegant and likely to clarify structural relationships among topologies, subpresheaves, and subtoposes; the closed formula for generated topologies and the bridge between provability and topology generation have potential applications to logical and geometric computations. The claimed preservation of finite unions by arbitrary inverse image functors is a natural and potentially useful new result. The paper draws on substantial prior work ([TST], [Local fibrations], [Denseness]); the novelty lies mostly in the uniform framework and in the new generation and preservation formulas. However, the present text is not self-contained enough for the main claims to be certified from what is supplied.","major_comments":[{"comment":"The finite-union preservation claim is proved only by deferral: \"On rappellera en effet au chapitre IV que tout morphisme de topos peut s'écrire comme le composé d'un plongement et d'un morphisme localement connexe.\" This factorization is load-bearing and is not proved in the visible text, nor is a reference given. The standard surjection/embedding factorization does not generally make the first factor locally connected, so this cannot be treated as folklore. The proof of Corollary I.3.20 must either include a complete proof of the factorization or cite a precise theorem with proof in the literature.","section":"Chapter I, Corollary I.3.20"},{"comment":"The existence of an extra left adjoint f_! for inverse image of subtoposes along a locally connected morphism, and hence preservation of arbitrary unions, is announced as a Chapter IV result but is not proved in the supplied text. Since this theorem is used for the finite-union claim, it is essential that the manuscript contain a full proof, not merely a sketch or a forward reference.","section":"Chapter I, §3.f, Theorem I.3.16"},{"comment":"The paper’s central new contributions are the closed generation formula for topologies (Chapter III, §1), the tree/multi-covering formula (Chapter III, §2), and the detailed treatment of inverse images along locally connected morphisms (Chapter IV). None of these chapters appears in the text supplied for review. The abstract and preface cannot substitute for the actual proofs. The manuscript must be re-submitted with the full chapters, or, if this is a partial submission, the missing material must be added or a complete version declared.","section":"Chapters III and IV (not included in the supplied text)"},{"comment":"The bridge between quotient theories and topologies, used in the translation of provability into topology generation, is presented with only sketches (\"Esquisse de démonstration\") for points (i)–(iv). Some of this is based on [TST], but if the presentation is meant to be new, the inference rules of geometric logic must be explicitly matched with the axioms of a topology; a sketch is not enough for a central logical equivalence used in the paper.","section":"Chapter I, §2.c, Proposition I.2.13"}],"minor_comments":[{"comment":"Several displayed formulas contain typographical artifacts, especially in Definition I.1.1 and the surrounding braces; these should be cleaned up in the final version.","section":"General"},{"comment":"The order convention on ST(E) is stated only implicitly; since union and intersection are order-reversing with respect to the topology inclusion, the direction of the order should be explicitly fixed to avoid confusion.","section":"Chapter I, §3.a"},{"comment":"The phrase \"dualité des cribles et des préfaisceaux\" is helpful, but the notation R, F_R, G_R becomes heavy; a small table of notation would improve readability.","section":"Chapter II, §2"},{"comment":"The preface promises applications to \"apprentissage profond topossique\" but the visible text does not return to this; if these applications are not developed, the promise should be toned down or the applications deferred explicitly.","section":"Preface"},{"comment":"The paper cites [Local fibrations], [Denseness], and [TST] for crucial background. Full bibliographic details should be listed and, for the factorization theorem used in Ch. IV, a precise statement with proof should be supplied.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The submitted text is incomplete: it stops in Chapter II, while the advertised main theorems (Chapters III and IV) are absent. Before a full review can be completed, the authors must provide the complete manuscript. The load-bearing factorization theorem in Chapter IV is nonstandard and needs a proof, not a forward reference. The paper is potentially valuable, but the current version cannot be evaluated as a finished journal submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about arXiv:2508.21134. First, it is not a routine survey: the explicit closed-form generation formula in Chapter III and the tree-based multi-cover formula are genuinely useful, and the second one directly addresses Johnstone's transfinite-induction worry, which is a real selling point. Second, the most advertised new result—that pullback along any geometric morphism preserves finite unions of subtoposes—is not fully proved in what I can see. It relies on a factorization of every geometric morphism as an embedding followed by a locally connected morphism, asserted in Chapter IV section 2c. That is not a standard factorization in the literature, and the excerpt we have stops before the proof. The stress-test note is right to flag this as load-bearing. I have no counterexample, but the proof is incomplete as written.\n\nWhat the paper does well: the Galois-connection framework in Chapter II is clean and does real work, not just packaging. It makes the generation formula almost formal, and the intersection/union computations for subtoposes become natural consequences. The logical translation of provability into topology generation is explained with enough concreteness that a reader could implement it for simple theories. The authors are also honest that many arguments are only sketched ('Esquisse de demonstration'), but that honesty cuts both ways: for a refereed monograph, a sketch is not enough for a theorem this central.\n\nThe soft spots are concentrated in Chapter IV. The locally connected case for arbitrary unions is fine if the definition of local connectedness is as they state, because the left adjoint to pullback gives preservation of all joins. The problem is the reduction from arbitrary morphisms to locally connected ones. I also note that the paper leans heavily on [TST] for the bridge equivalences, but those are independently established, so the circularity burden is low. The product-of-spaces application looks plausible but is not where I would focus referee energy.\n\nBottom line: this is a paper for topos theorists and categorical logicians who work with subtopos lattices and generation of topologies. It deserves a serious referee, but the referee should demand a complete proof of the factorization theorem or a rewrite of Corollary I.3.20 as conditional on it. I would use the generation formulas in my own work; I would not yet cite the finite-union preservation as a theorem.\n\nFor peer review: send it out, but with a clear request to fill in Chapter IV section 2c or move the finite-union result to an appendix with a full proof.","headline":"A serious systematic monograph with real new formulas, but the headline finite-union preservation theorem depends on a deferred factorization proof that isn't in the visible text.","tokens_in":78467,"tokens_out":1451,"would_cite":true,"duration_ms":19972,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["18B25","18F10","03G30"],"pacs":[],"model":"deepseek-v4-flash","headline":"Pulling back subtoposes along any geometric morphism preserves finite unions, and along locally connected morphisms arbitrary unions; the paper proves this with explicit formulas for generated Grothendieck topologies.","keywords":["Grothendieck topologies","subtoposes","generation of topologies","geometric logic","provability","locally connected morphisms","Galois connections","classifying toposes"],"falsifier":"Test the identity on a small explicit site: take a functor between finite categories (for the locally connected case, a fibration with a Giraud topology) and two topologies on the target; compute the pullback of the intersection of the topologies and the intersection of the pullbacks using the Chapter IV formulas. If they differ, finite-union preservation is false; likewise for an infinite family along a locally connected morphism. Alternatively, test the closed generation formula: for a pullback-stable family of sieves on a small category, a sieve belongs to the generated topology exactly whe","tokens_in":77579,"feed_emoji":"🔗","tokens_out":21067,"duration_ms":193532,"temperature":0.7,"pith_summary":"This monograph builds a systematic calculus for subtoposes, the topos-level analogue of subspaces. The paper's central move is to treat a subtopos as equivalently a Grothendieck topology on any presenting site or a quotient of any presenting geometric theory, and to derive that duality — not assume it — as the fixed points of a Galois connection between sieves and presheaves. This turns provability in first-order geometric logic into a problem of generating topologies, for which the paper gives two explicit formulas. The central new claim is that pulling subtoposes back along any geometric morphism preserves finite unions, and that pullback along locally connected morphisms preserves arbitrary unions, making the lattice of subtoposes behave predictably under all geometric operations. If correct, these results make subtopos computations effective: joins, intersections and differences of subtoposes can be read off from formulas rather than built step by step.","feed_headline":"Pulling back subtoposes preserves finite unions along every topos map","feed_subtitle":"For locally connected maps the pullback also preserves arbitrary unions, sharpening the logic–geometry dictionary.","key_machinery":"Three pieces carry the argument. A Galois connection between sieves and presheaves — with an intrinsic version between monomorphisms and objects of a topos — has fixed points that are exactly the Grothendieck topologies and the subtoposes, and generating a topology is the composite of its two adjoints. Two generation formulas compute generated topologies: a closed sieve-closure formula for pullback-stable families, and a multi-cover tree formula whose pullback-stability removes the need for transfinite iteration. For the union theorems: the adjunction f^{-1} ⊣ f_* between subtopos lattices induced by a geometric morphism, the left adjoint f_! ('extraordinary direct image') existing precisely","core_discovery":"A subtopos is equivalently a Grothendieck topology on any presenting site or a quotient of any presenting geometric theory; the paper derives this duality, rather than postulating it, as the fixed points of a Galois connection between sieves and presheaves. Two explicit formulas compute generated topologies: a closed form for pullback-stable families, and a tree-based multi-cover formula avoiding transfinite iteration. The inverse-image operation on subtoposes preserves all intersections, preserves finite unions for every geometric morphism, and preserves arbitrary unions when the morphism is locally connected, via a left adjoint 'extraordinary direct image' and a factorization of any morphi","pith_inferences":["The factorization template — prove a subtopos identity for locally connected morphisms and for inclusions, then extend to all morphisms — could promote further identities, for instance the behaviour of the co-Heyting difference operation under pullback, from the locally connected case to arbitrary geometric morphisms; the paper only applies this template to unions.","The paper notes that the closed formula suits sieve presentations (typical of logical contexts such as the dense or De Morgan topologies) while the multi-cover formula suits pre-cover presentations (typical of geometric contexts such as Zariski or étale); a natural testable project is an automated provability checker that switches between the two depending on the site.","The preface announces that oriented and fibred products of toposes will be treated in a future version using these methods; if the same generation formulas control those products, the union-preservation theorems should transfer by the same adjunction argument.","If data elements are represented as subtoposes rather than vectors, as the preface contemplates, the union-preservation theorems imply that the operation combining data-points commutes with any morphism that changes representation — the topos-level analogue of a linear map commuting with vector addition, and a condition one could test in a concrete representation-learning setting."],"forward_implications":["Provability in geometric logic becomes a closure computation: a sequent is provable from a family of axioms exactly when the associated sieve families are covering for the topology those axioms generate, so the closed generation formula makes the pullback-stable cases explicit and finite.","Subtopos joins commute with inverse images: finite joins along every geometric morphism, arbitrary joins along locally connected ones; consequently every action of a topos-theoretic correspondence or chain on subtoposes, being a composite of direct and inverse images, preserves finite unions.","The multi-cover tree formula generates topologies without transfinite iteration, settling a point on which the standard treatise had left transfinite induction seemingly unavoidable.","Topologies presenting a finite product of sheaf toposes are exactly those generated by the pullbacks of the presenting topologies along the canonical projections; the generation formula therefore computes finite products of toposes, and when the factors are sheaf toposes of locally compact spaces (all but possibly one) the sheaf topos of the product space is the product of the sheaf toposes.","Along locally connected morphisms the inverse image of subtoposes acquires a left adjoint, the extraordinary direct image, which preserves arbitrary intersections of subtoposes and adds a new adjoint layer to the calculus of subtopos operations."],"supporting_citations":[{"why":"supplies the original bridge between subtoposes, topologies and quotient theories, and the first version of the generation formula (proposition 4.1.1) that this paper refines.","marker":"[TST]"},{"why":"provides the standard topos-theory framework and the remark that computing generated topologies seems to require transfinite induction, which the Chapter III multi-cover formula avoids.","marker":"[Elephant]"},{"why":"proves that locally connected morphisms are exactly essential morphisms whose essential image is a fibration, the characterization on which the Chapter IV adjunction results rest.","marker":"[Local fibrations]"},{"why":"proves that fibrations equipped with Giraud topologies induce locally connected geometric morphisms, connecting Giraud topologies to the pullback formulas of Chapter IV.","marker":"[Denseness]"},{"why":"supplies the general method of Galois connections by which Chapter II derives the topologies–subtoposes and topologies–closure-properties dualities as fixed points.","marker":"[Galois]"},{"why":"gives the classifying-topos theorem and the syntactic-category constructions for geometric theories that ground the logical side of the provability translation.","marker":"[Categorical Logic]"},{"why":"introduces the 'toposes as bridges' technique that frames the paper's translation of logical provability into topology generation.","marker":"[Mémoire]"}],"fun_headline_variants":["Pullback of subtoposes preserves finite unions for all maps","Locally connected maps: pullback preserves all unions of subtoposes","Explicit formulas for generated Grothendieck topologies","New duality: Grothendieck topologies as subtoposes","From logic to geometry: provability as topology generation"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The whole argument for arbitrary morphisms depends on one factorization theorem: every geometric morphism can be written as an inclusion followed by a locally connected morphism, where the locally connected part itself rests on the stability (Beck–Chevalley) condition built into its definition; without that factorization, finite-union preservation is only proven for the two special cases separately.","fun_headline_variants_meta":{"raw":{"variants":["Pullback of subtoposes preserves finite unions for all maps","Locally connected maps: pullback preserves all unions of subtoposes","Explicit formulas for generated Grothendieck topologies","New duality: Grothendieck topologies as subtoposes","From logic to geometry: provability as topology generation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001137,"raw_usage":{"total_tokens":4594,"prompt_tokens":815,"completion_tokens":3779,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":3693}},"tokens_in":559,"tokens_out":3779,"duration_ms":29401,"temperature":1.0,"reasoning_tokens":3693,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T14:33:28.750048+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test the identity on a small explicit site: take a functor between finite categories (for the locally connected case, a fibration with a Giraud topology) and two topologies on the target; compute the pullback of the intersection of the topologies and the intersection of the pullbacks using the Chapter IV formulas. If they differ, finite-union preservation is false; likewise for an infinite family along a locally connected morphism. Alternatively, test the closed generation formula: for a pullback-stable family of sieves on a small category, a sieve belongs to the generated topology exactly whe","supporting_citations":[],"review_version":1}