{"id":"d8045f9b-c793-42c7-9f65-883d370c1127","arxiv_id":"2507.07922","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Generalised ultracategories yield a proof that every topos with enough points is equivalent to the category of left ultrafunctors from its generalised ultracategory of points to Set.","lead":"This paper introduces a new categorical structure called generalised ultracategories, which extend Lurie's ultracategories, and uses it to prove that any topos with enough points can be reconstructed from its category of points. The result extends Makkai and Lurie's conceptual completeness theorems and provides a unified framework where topological spaces and toposes are both examples of the same structure.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 7.3 depends on an unproved extension of Lurie's sheaf correspondence: Lult(X,Set)≃Sh(X) is only sketched for arbitrary topological spaces.","rationale":"The reader's weakest-assumption analysis is correct and points to the most exposed link in the proof. I reviewed all three equivalences used to prove Theorem 7.1. Theorem 7.2 also has gaps: the spaces I_μ and I_{μ,(λ_i)} are not fully verified as topological spaces, and the reconstruction of a left ultrafunctor from a cloven Cartesian functor is only sketched. Theorem 7.5 similarly compresses the full faithfulness argument into 'follows from the fact that E is the colimit.' However, Theorem 5.7 is the cleanest load-bearing concern because the paper explicitly leaves the functoriality and inverse equivalence to the reader and delegates part of the proof to a separate preprint. It is also used inside Theorem 7.3, which is indispensable to the main theorem. The paper does not appear internally inconsistent, and the claimed theorem may well be true, but the evidence supplied is insufficient to move the verdict beyond conditional; the reader's CONDITIONAL verdict is therefore the right one.","tokens_in":41496,"tokens_out":15139,"duration_ms":172318,"concrete_test":"Work out Theorem 5.7 explicitly for the Sierpinski space: list all left ultrafunctors F:S→Set, construct the proposed total space E=⊔F(x), check Wyler's axioms UQ1 and UQ4 so that E is a topological space, and prove the projection E→S is étale. Then define the inverse sheaf-to-ultrafunctor construction and verify the two compositions are naturally isomorphic. If the Sierpinski case goes through, repeat for a non-sober space such as the cofinite topology on N; if either step fails or requires an additional hypothesis (compactness, sobriety, or the unpublished [9, Lemma 3.4]), Theorem 7.3 and hence Theorem 7.1 are not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central chain in Theorem 7.1 passes through Theorem 7.3, whose fibre equivalence (displayed in §7.2) uses Lult(X,Set)≃Sh(X) for every topological space X. This is a genuine extension of Lurie's compact-Hausdorff result, but Theorem 5.7 is not proved in this paper. The proof sketches only one direction, constructing a topology on E=∐F(x) by ultrafilter convergence, and then states: 'We leave to the reader showing that these two processes are functorial and are inverses of each other (up to equivalence).' It also invokes [9, Lemma 3.4] from an earlier preprint and does not verify that the topology on E is well-defined, independent of choices, or that the inverse construction from étale bundles satisfies the full generalised-ultracategory axioms of §2. If this equivalence fails for some non-compact or non-Hausdorff space, the identification Top//M_E≃Top//E in Theorem 7.3 breaks, and with it the equivalence Lult(M_E,M_E')≃Geom(E,E'). This is a proof gap rather than a demonstrated falsehood, but it is load-bearing: without Theorem 5.7, the main theorem is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces 'generalised ultracategories', a relational enrichment of Lurie's ultracategories in which ultraproducts are replaced by formal generalised Hom-sets. It shows that topological spaces, viewed with their specialisation preorders, form generalised ultracategories via ultrafilter convergence, and that the category of points of a topos carries a canonical generalised ultrastructure. The main theorem (Theorem 7.1) asserts that for toposes E and E' with enough points, the category of left ultrafunctors between their generalised ultracategories of points is equivalent to the category of geometric morphisms Geom(E,E'). The proof is structured as three equivalences: Lult(M_E,M_E') ≃ ClovenCart(Top//M_E, Top//M_E'); Top//M_E ≃ Top//E as 2-fibrations over Top; and ClovenCart(Top//E, Top//E') ≃ Geom(E,E'), the last using the Butz-Moerdijk representation of toposes with enough points as colimits of topological groupoids.","tokens_in":41673,"tokens_out":6520,"duration_ms":76248,"significance":"If the main theorem is correct, it is a substantial extension of Makkai's and Lurie's conceptual completeness theorems from coherent toposes to all toposes with enough points, and it provides a pleasing reconstruction principle: a topos with enough points is recovered from its generalised ultracategory of points. The paper also contributes a useful framework of ultrapreorders and shows that topological spaces embed fully faithfully into generalised ultracategories. The author is explicit about the intended architecture, and the appendices contain substantial verification work. On the other hand, the paper currently rests on several results that are stated with proofs left to the reader or delegated to the author's previous preprints; the central chain is therefore not yet fully supported. I found no circularity: the reliance on [9] and [10] is dependence on prior work, not use of the main theorem to prove itself.","major_comments":[{"comment":"Theorem 5.7, the equivalence Lult(X,Set) ≃ Sh(X) for every topological space X, is load-bearing for Theorem 7.3 and hence for Theorem 7.1, but it is not proved. The proof constructs a topology on E=∐F(x) from a left ultrafunctor F and then states: 'We leave to the reader showing that these two processes are functorial and are inverses of each other (up to equivalence).' It also invokes [9, Lemma 3.4] from an earlier preprint without verifying its hypotheses, and it does not check that the inverse construction from an étale bundle satisfies all the generalised ultracategory axioms of §2. Since the paper uses this equivalence for arbitrary (not necessarily compact Hausdorff) spaces, this is a genuine gap in the proof of the main theorem.","section":"§5.1, Theorem 5.7"},{"comment":"The proof of Theorem 7.2 depends on topological spaces I_μ and I_{μ,(λ_i)} whose existence is asserted rather than established. The text says 'The proof that this construction satisfies 5.3 is left to the reader' for I_μ, and for I_{μ,(λ_i)} it says 'To show that this is a topological space, we have of course to use [25] (by checking lots of cases).' Appendix D provides a case analysis, but many steps are conclusions announced without derivation, and the analysis does not fully verify the closure conditions of Theorem 5.3. In particular, the claimed continuity of the map I_{R_I ι_i λ_i dμ} → I_{μ,(λ_i)} is only stated, yet this map is used to prove the composition axiom for the reconstructed left ultrafunctor f.","section":"§7.1, Theorem 7.2 and Appendix D"},{"comment":"Theorem 7.3 claims an equivalence of 2-fibrations over Top between Top//M_E and Top//E, but the proof establishes only that the fibres are equivalent. The passage from fibre-wise equivalence to an equivalence of discrete 2-fibrations is dismissed with 'the proof can be deduced by inspecting the following diagram'. This is insufficient: one must prove naturality with respect to continuous maps, compatibility with cartesian lifts, and coherence for 2-cells. Because Theorem 7.1 passes through this fibred equivalence, the gap is load-bearing.","section":"§7.2, Theorem 7.3"},{"comment":"The key exchange equivalence used in the proof of Theorem 7.3 is asserted rather than proved. The text says 'It is easily verifiable that these maps σ'_μ indeed satisfy the compatibility axioms' and 'one can easily verify that this gives F a left ultrastructure', but no verification is supplied. The restriction to J-continuous lex functors is also handled by a stalkwise epimorphism argument that relies on the unproved equivalence Lult(X,Set) ≃ Sh(X) and on Sh(X) having enough points. This equivalence is essential for identifying the fibres of Top//M_E and Top//E.","section":"§7.2, fibre equivalence Lult(X,Fun(C,Set)) ≃ Fun(C,Lult(X,Set))"},{"comment":"The proof that the forgetful functor from Top/E to Topos has colimit E is not complete. Lemma 7.8, which supplies the comparison maps between the pullbacks of a sheaf along different points, ends with 'one can easily notice that the basic open sets in both coincide'; this is the key topological step in proving that the constructed cocone is a colimit. Without a full proof of Lemma 7.8, the equivalence ClovenCart(Top//E,Top//E') ≃ Geom(E,E') in Theorem 7.5 is not established, and Theorem 7.1 remains unsupported.","section":"§7.3, Lemma 7.6 and Lemma 7.8"}],"minor_comments":[{"comment":"The statement of Theorem 5.8 says there is a canonical generalised ultrastructure 'on A', but the construction defines Hom(A,∫_I M_i dμ) as Hom_B(A,∫_I M_i dμ), which only makes sense when the objects A and M_i lie in the full subcategory B. The intended statement is presumably a structure on B; this should be corrected to avoid confusion.","section":"§5.2, Theorem 5.8"},{"comment":"The notation for fibred functors is inconsistent: the text alternately uses 'ClovCart' and 'ClovenCart', and the definition of a cloven Cartesian functor is stated informally. Please standardise the terminology and the notation.","section":"§6"},{"comment":"In the proof of Theorem 7.2, the sentence 'Now we claim that for any topological space X and any left ultrafunctor h, we have a natural isomorphism of left ultrafunctors between f∘h and F(h)' is not justified by the surrounding text; the diagram involving X_μ only shows equality of underlying functors, not equality of ultrastructure. This point needs a proper argument even if the overall strategy is correct.","section":"§7.1"},{"comment":"The verification that the underlying category is a category uses many diagrams that are labelled 'commutes by ...' without giving the diagram in full. Since this is an appendix intended to supply the deferred proof, more explicit justifications are needed for the reader to follow the composition axiom.","section":"Appendix A"},{"comment":"There are numerous typographical and formatting issues: 'genralised', 'E F E', inconsistent use of 'Hom' vs. 'Hom', and a diagram in §7.3 whose typesetting is garbled (e.g., the line 'HomGroupoids(X,G •)≃ ∗'). These do not affect the mathematics but should be cleaned up.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper's central idea is attractive and the three-step strategy is coherent, but the current version leaves too many load-bearing verifications to earlier preprints or to the reader. In particular, Theorem 5.7 and the topological constructions of §7.1 are foundational for the main theorem and must be proved in full, or explicitly stated as assumptions, before the paper can be accepted. I would also ask the editor to request that the author clarify the relation to Saadia's virtual ultracategories [24]; the note in §1 acknowledges the overlap but does not say whether the two notions agree."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it introduces a genuinely new notion, generalised ultracategories, and uses it to state a natural extension of Lurie's conceptual completeness: for toposes with enough points, left ultrafunctors between categories of points are equivalent to geometric morphisms. Second, the proof as written has a load-bearing gap: the key sheaf correspondence Lult(X,Set) ≃ Sh(X) for arbitrary topological spaces (Theorem 5.7) is only sketched and delegated to a prior preprint, and the chain from the main theorem passes through it. If that equivalence fails for non-compact or non-Hausdorff spaces, the main theorem is unsupported.\n\nWhat the paper does well: the definition of generalised ultracategories is carefully laid out, with axioms that generalise Lurie's ultracategories to a relational setting. The examples are compelling—topological spaces become ultrapreorders, and points of toposes carry a canonical generalised ultrastructure. The overall proof strategy is coherent and uses known results (Moerdijk-Butz representation) in a sensible way. The author also honestly notes an independent similar result by Saadia [24].\n\nThe soft spots are real. Theorem 5.7 is load-bearing but not proved; the proof sketches one direction and leaves the inverse to the reader. This is not a minor omission: the equivalence is a genuine extension of Lurie's compact-Hausdorff result, and the paper does not verify that the topology on the etale bundle is independent of choices or that the inverse construction satisfies all generalised-ultracategory axioms. Similarly, the spaces I_μ and I_{μ,(λ_i)} in Theorem 7.2 are defined by convergence relations, but the proof that they are topological spaces is left to the reader (only one case is handled in Appendix D). These gaps are potentially fixable, but they are not cosmetic.\n\nWho this is for: people working in categorical logic and topos theory, especially those interested in conceptual completeness and ultracategories. It deserves a serious referee; the main theorem is significant and the gaps are identifiable and likely repairable. I would not cite it as a proof of the main theorem until the gaps are filled, but it is worth engaging with.","headline":"A genuinely new generalisation of ultracategories with a plausible main theorem, but the proof leans on an unproved sheaf correspondence; worth reviewing, not yet citable as a proof.","tokens_in":42278,"tokens_out":2318,"would_cite":true,"duration_ms":24891,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["18B25","03G30"],"pacs":[],"model":"deepseek-v4-flash","headline":"A topos with enough points can be reconstructed from its generalised ultracategory of points: left ultrafunctors between point categories are exactly geometric morphisms between the toposes.","keywords":["generalised ultracategories","toposes with enough points","conceptual completeness","geometric logic","left ultrafunctors","ultrafilter convergence","ultrapreorders","topological groupoids"],"falsifier":"Take a non-compact Hausdorff space $X$ and compute both sides of Theorem 5.7 explicitly: the etale space over $X$ built from a left ultrafunctor $F$, and the left ultrafunctor built from an etale space by taking stalks. If the two processes fail to be inverse up to equivalence, or if the topology defined on the total space fails the convergence axioms, then the equivalence $\\mathrm{Top}//M_E \\simeq \\mathrm{Top}//E$ is not established, because its proof uses this identification fibrewise for arbitrary $X$.","tokens_in":41204,"feed_emoji":"🔄","tokens_out":12092,"duration_ms":116227,"temperature":0.7,"pith_summary":"This paper introduces generalised ultracategories, a relational variant of the established ultracategory framework in which formal Hom-sets $\\mathrm{Hom}(A,\\int_I M_i\\,d\\mu)$ represent maps into an ultraproduct even when genuine ultraproducts do not exist. It argues that any topos with enough points can be reconstructed from its points: the main theorem gives an equivalence of categories between left ultrafunctors $\\mathrm{Lult}(M_E,M_{E'})$ between the generalised ultracategories of points of two such toposes and geometric morphisms $\\mathrm{Geom}(E,E')$ between the toposes themselves. This extends the known conceptual completeness results, previously available for coherent toposes, to the larger class of toposes with enough points. The guiding observation is that topological spaces, viewed as ultrapreorders, encode ultrafilter convergence through these formal Hom-sets, and the category of points of a topos carries the same kind of convergence structure.","feed_headline":"A topos with enough points is rebuilt from its points","feed_subtitle":"Left ultrafunctors equal geometric morphisms: completeness beyond coherent toposes.","key_machinery":"The central object is the generalised ultracategory: a category equipped with formal Hom-sets $\\mathrm{Hom}(A,\\int_I M_i\\,d\\mu)$ for every object $A$, family of objects $(M_i)_{i\\in I}$, and ultrafilter $\\mu$ on $I$, together with change-of-base maps $\\Xi$ and composition maps $\\beta$ satisfying unit, associativity, and compatibility axioms. These formal Hom-sets encode ultraproduct-like structure without requiring an ultraproduct operation to exist. The load-bearing construction is that topological spaces become ultrapreorders under this structure, with ultrafilter convergence captured by which formal Hom-sets are non-empty, and that the category of points of a topos inherits a generalised ultrastructure by embedding into the functor category over a site of definition. The bridge between the two worlds is the equivalence $\\mathrm{Sh}(X) \\simeq \\mathrm{Lult}(X,\\mathrm{Set})$ for every topological space $X$ (Theorem 5.7), which lets the author identify the lax-slice categories $\\mathrm{Top}//M_E$ and $\\mathrm{Top}//E$. A known representation of toposes with enough points as colimits of topological groupoids then supplies the final comparison between geometric morphisms and cloven Cartesian functors between lax slices.","core_discovery":"On the paper's own terms, the central result is Theorem 7.1: for toposes $E$ and $E'$ with enough points, with $M_E$ and $M_{E'}$ their respective generalised ultracategories of points, there is an equivalence of categories between $\\mathrm{Lult}(M_E,M_{E'})$ and $\\mathrm{Geom}(E,E')$. Replacing $E'$ by the classifying topos of the theory of objects $S[O]$ yields $\\mathrm{Lult}(M_E,\\mathrm{Set}) \\simeq E$, so a topos with enough points is reconstructed from its generalised ultracategory of points. The equivalence is proved in three stages: left ultrafunctors between point categories correspond to cloven Cartesian functors between lax-slice categories over the 2-category of topological spaces (Theorem 7.2); the lax slice over $M_E$ is equivalent, fibrewise over spaces, to the lax slice over $E$ (Theorem 7.3); and geometric morphisms between toposes are equivalent to cloven Cartesian functors between their lax slices (Theorem 7.5).","pith_inferences":["Because the proof of Theorem 7.2 uses topological spaces to realise generalised morphisms, the author's technique suggests that the 2-category of generalised ultracategories is densely generated by topological spaces; if made precise, this would let questions about arbitrary generalised ultracategories be reduced to questions about spaces and their sheaves.","The paper leaves open whether its generalised ultracategories coincide with the virtual ultracategories of an independent recent approach; if they do, the two frameworks would mutually support a reconstruction theorem for all toposes with enough points.","The reconstruction is routed through the colimit-of-topological-groupoids presentation of a topos, so the equivalence $\\mathrm{Lult}(M_E,\\mathrm{Set}) \\simeq E$ is in principle constructive; a careful reading of the proof could yield an explicit inverse functor for any site whose groupoid presentation is tractable."],"forward_implications":["Every topos with enough points $E$ is equivalent to $\\mathrm{Lult}(M_E,\\mathrm{Set})$, so the generalised ultracategory of points determines the topos up to equivalence.","Two toposes with enough points are Morita equivalent exactly when their generalised ultracategories of points are equivalent, making the generalised ultrastructure a complete invariant for such toposes.","The coherent case is recovered as a special instance, since coherent toposes have enough points and, for them, the new notion of left ultrafunctor agrees with the earlier one.","Geometric morphisms between toposes with enough points can be computed as cloven Cartesian functors between lax-slice categories over the 2-category of topological spaces, offering a new route to questions about geometric logic."],"supporting_citations":[{"why":"Supplies the ultracategory framework, the notion of left ultrafunctor, and the coherent-topos reconstruction theorem this paper extends.","marker":"[16]"},{"why":"Gives the representation of toposes with enough points as colimits of topological groupoids, used in the final step of the main theorem.","marker":"[4]"},{"why":"Provides the logical topological groupoids and their colimit computations used in Section 7.3.","marker":"[5]"},{"why":"Establishes topological spaces as relational algebras for the ultrafilter monad, the fact used to encode ultrafilter convergence as generalised ultrastructure.","marker":"[1]"},{"why":"The author's previous account of ultracategories as colax algebras for a pseudo-monad supplies the colax associator and coherence machinery behind the axioms.","marker":"[10]"},{"why":"The author's earlier bundle-of-metric-spaces work contains the lemma used to show that the etale space construction in Theorem 5.7 is well defined.","marker":"[9]"},{"why":"Fixes the theory of fibred 2-categories in which the equivalences of lax-slice categories over Top are stated and compared.","marker":"[3]"},{"why":"Supplies the sheaf-theoretic and geometric-logic background, including the comparison lemma used to prove independence from the choice of site.","marker":"[13]"}],"fun_headline_variants":["Toposes with enough points are rebuilt from their points","Left ultrafunctors reconstruct toposes from their points","Conceptual completeness for toposes with enough points","Topos reconstruction from points via generalised ultracategories","Generalised ultracategories: from points to toposes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The chain of equivalences rests on Theorem 5.7, which claims that for every topological space $X$, sheaves on $X$ are the same as left ultrafunctors from $X$ to sets; the paper sketches this identification but leaves the verification that the two constructions are inverse, and compatible along continuous maps, to the reader, so any gap there would propagate through the whole reconstruction.","fun_headline_variants_meta":{"raw":{"variants":["Toposes with enough points are rebuilt from their points","Left ultrafunctors reconstruct toposes from their points","Conceptual completeness for toposes with enough points","Topos reconstruction from points via generalised ultracategories","Generalised ultracategories: from points to toposes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001166,"raw_usage":{"total_tokens":4802,"prompt_tokens":898,"completion_tokens":3904,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":514,"completion_tokens_details":{"reasoning_tokens":3827}},"tokens_in":514,"tokens_out":3904,"duration_ms":29431,"temperature":1.0,"reasoning_tokens":3827,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:29:27.815155+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a non-compact Hausdorff space $X$ and compute both sides of Theorem 5.7 explicitly: the etale space over $X$ built from a left ultrafunctor $F$, and the left ultrafunctor built from an etale space by taking stalks. If the two processes fail to be inverse up to equivalence, or if the topology defined on the total space fails the convergence axioms, then the equivalence $\\mathrm{Top}//M_E \\simeq \\mathrm{Top}//E$ is not established, because its proof uses this identification fibrewise for arbitrary $X$.","supporting_citations":[{"cited_title":"Ultracategories.Preprint available athttps: // www","cited_arxiv_id":null,"evidence_quote":"Supplies the ultracategory framework, the notion of left ultrafunctor, and the coherent-topos reconstruction theorem this paper extends."},{"cited_title":"Representation of topoi by topological spaces.Journal of Pure and Applied algebra 130 223-235, 1998","cited_arxiv_id":null,"evidence_quote":"Gives the representation of toposes with enough points as colimits of topological groupoids, used in the final step of the main theorem."},{"cited_title":"Topological representation of sheaf cohomology of sites","cited_arxiv_id":null,"evidence_quote":"Provides the logical topological groupoids and their colimit computations used in Section 7.3."},{"cited_title":"Relational algebras","cited_arxiv_id":null,"evidence_quote":"Establishes topological spaces as relational algebras for the ultrafilter monad, the fact used to encode ultrafilter convergence as generalised ultrastructure."},{"cited_title":"Ultracategories as colax algebras for a pseudo-monad on cat.https: // arxiv","cited_arxiv_id":null,"evidence_quote":"The author's previous account of ultracategories as colax algebras for a pseudo-monad supplies the colax associator and coherence machinery behind the axioms."},{"cited_title":"Bundles of metric structures as left ultrafunctors.arxiv preprinthttps: // arxiv","cited_arxiv_id":null,"evidence_quote":"The author's earlier bundle-of-metric-spaces work contains the lemma used to show that the etale space construction in Theorem 5.7 is well defined."},{"cited_title":"Fibred 2-categories and bicategories.Journal of pure and applied algebra, 218(6):1034–1074, 2014","cited_arxiv_id":null,"evidence_quote":"Fixes the theory of fibred 2-categories in which the equivalences of lax-slice categories over Top are stated and compared."},{"cited_title":"Oxford University Press, 2002","cited_arxiv_id":null,"evidence_quote":"Supplies the sheaf-theoretic and geometric-logic background, including the comparison lemma used to prove independence from the choice of site."}],"review_version":1}