{"id":"7a859977-e0d4-414a-a864-9edbe188154f","arxiv_id":"2509.06339","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The author proves, modulo two of her own unpublished preprints, that Teichmüller space has a well-rounded, equivariant deformation retraction onto a complex of dimension 4g-5.","lead":"This paper defines a \"well-rounded\" deformation retraction of Teichmüller space, analogous to the classic well-rounded retract for SL(n,Z), and claims one exists for every surface genus at least 2. If correct, it strengthens the known minimal-dimensional spine of Teichmüller space with a natural symmetry-compatible property.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central lemma's stretch path is not well-defined: the weighted sum m(x) can contain intersecting curves from different blocks, so m(x) is not a multicurve and no stretch path exists.","rationale":"The paper's central claim is Theorem 1.1 (Theorem 4.5), which is derived from Lemma 4.3 and the author's unpublished [13]. The reader's conditional verdict is well founded. In a good-faith reading, the main new ingredient is Lemma 4.3, which aims to construct an equivariant homotopy of a dual D into the thin part when the horizon labels do not span H_1. The proof's core mechanism is a flow along 'stretch paths' generated by a weighted sum m(x) of block labels. I examined this construction carefully. The weighted sum is taken over an open cover where each U_i is assigned a block label m_i. The blocks are equivalence classes of x under equality of the multicurve m_[α](x), which is constructed from the local geometry at x and therefore can vary from point to point. Consequently, two blocks meeting near a stratum boundary can have labels containing curves that intersect. The sum m(x) then has positive weights on intersecting curves. By the definition given in the proof, a multicurve must have pairwise disjoint representatives; m(x) is therefore not a multicurve. The paper gives no definition of a stretch path for an arbitrary weighted set of intersecting curves — the standard notion (Thurston's stretch maps, earthquakes) applies to measured laminations or at most pairwise disjoint weighted multicurves. Thus the asserted existence of a stretch path γ_{m(x)} is unsupported. This is a more basic obstruction than the smoothness issue the reader noted; even continuity or well-definedness is not established. The same concern appears in Theorem 4.5's iteration, which relies on Lemma 1.2. Hence the central claim remains unproven as written. The concern is not about external consensus; it is internal to the proof's definitions. A concrete check is to see if two block labels in a common U_i can intersect. If they cannot, the proof needs to supply a lemma proving such compatibility; if they can, the stretch-path flow does not exist. I do not see that the paper provides such a lemma. Therefore I recommend keeping the reader's CONDITIONAL verdict; the gap is real but potentially repairable, so rejection is not forced.","tokens_in":12028,"tokens_out":7890,"duration_ms":66494,"concrete_test":"To settle the concern, check whether the support of the partition of unity at a point x in D^δ can include two block labels m_i and m_j with i(γ_i, γ_j)>0 for some curves γ_i∈m_i and γ_j∈m_j. If such a configuration occurs, then m(x) is not a multicurve and the stretch path is undefined. A concrete place to look is the genus-5 example in [2, Remark 4.4], where the systoles fill but fail to span H_1; compute m_[α](x) for two nearby points in adjacent blocks and test whether the resulting multicurves intersect. Alternatively, prove that all block labels appearing in a common U_i are sub-multicurves of a single multicurve; if that proof is impossible, the lemma's proof fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Lemma 4.3, the partition of unity yields m(x) := Σ χ(x)ϕ_i(x) m_i, where each m_i labels a block of the equivalence relation x~y ⇔ m_[α](x) = m_[α](y). The proof then states that 'the weighted multicurve m(x) determines a stretch path γ_{m(x)}'. However, by the paper's own definition earlier in the proof, a multicurve is a set of weighted curves with pairwise disjoint representatives. The block labels m_i are multicurves for different values of m_[α](x); as x moves across a block boundary, m_[α](x) changes, and the supports of the ϕ_i can include two blocks whose labels contain curves with positive geometric intersection. Then m(x) has positive weights on intersecting curves and is not a multicurve. Since stretch paths are defined for measured laminations (or at most pairwise disjoint weighted multicurves) and no definition is given for arbitrary weighted sets of intersecting curves, the homotopy ψ_t is not constructed. Even if one interprets m(x) as a measured lamination, the transition can change the lamination's transverse measure only continuously, not smoothly, and the asserted smooth dependence is not established. Thus the proof of Lemma 4.3, and hence Theorem 1.1, has an unresolved gap at the construction of the stretch-path flow.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces a notion of 'well-rounded deformation retraction' for the mapping class group action on Teichmüller space, modelled on Ash's well-rounded retract for SL(n,Z). Definition 4.1 requires every locally top-dimensional cell of the spine to have a dual labelled by a set of curves spanning H_1(S_g;Q). The main theorem (Theorem 1.1/4.5) asserts that for every g≥2 such a retraction exists onto a CW complex of dimension 4g−5. The proof rests on Lemma 1.2/4.3, which states that if the curves labelling a dual are homologically deficient, then the horizon map carries the dual to a boundary in the barycentric subdivision of Harvey's curve complex; the proof constructs a homotopy of the dual into the thin part using cyclic covers and stretch paths. The paper also discusses duality, the horizon map, sets of minima, and the relation to the author's earlier construction [13].","tokens_in":12286,"tokens_out":6905,"duration_ms":62540,"significance":"If correct, Theorem 1.1 gives an optimal-dimensional equivariant spine with a natural filling/homology-spanning property, strengthening the analogy between mapping class groups, Out(F_n), and GL(n,Z). The definition of well-roundedness via duals labelled by homology-spanning curve sets is a useful conceptual contribution, and the paper is unusually explicit about its limitations, including the open converse to Lemma 1.2 and the possible lack of a genuine cell decomposition. The main caveats are that Lemma 4.3 is not proved at the claimed level of rigor and that Theorem 4.5 is inherited from the author's unpublished [13] rather than proved here.","major_comments":[{"comment":"The weighted object m(x) := Σ χ(x)ϕ_i(x)m_i need not be a multicurve: the supports of the partition functions can meet blocks whose labels contain curves with positive geometric intersection, and the preceding claim only shows that the labels share a common submulticurve, not that their union is pairwise disjoint. Since stretch paths are introduced for multicurves and no stretch path is defined for an arbitrary weighted set of intersecting curves, the homotopy ψ_t is not actually constructed. This is a load-bearing gap for Lemma 4.3 and hence for Theorem 4.5.","section":"Section 4, proof of Lemma 4.3"},{"comment":"Even if m(x) is reinterpreted as a measured lamination, the assertion 'these stretch paths vary smoothly with x' is unsupported. A partition-of-unity interpolation across a block boundary can change the weights continuously, but there is no argument that the resulting lamination, or its stretch path, varies smoothly, nor that the combinatorial type of m(x) changes in a controlled way at the boundary. The proof needs a precise statement of the regularity of the map x ↦ γ_{m(x)}.","section":"Section 4, proof of Lemma 4.3"},{"comment":"Theorem 4.5 is stated as a consequence of Lemma 1.2 and the construction in [13]. Since [13] is an unpublished preprint and the author's own earlier work, the paper should either state the relevant theorem from [13] as an explicit assumption or include enough detail to verify the induction. In particular, the claims that each iteration replaces an orbit of cells by cells of smaller dimension and that only finitely many iterations are possible are not justified in the present paper.","section":"Section 4, proof of Theorem 4.5"},{"comment":"The step using Harer's theorem on ∂T^{ε_M}_g needs more argument: a subcomplex of dimension less than 2g−2 is null-homologous in a wedge of (2g−2)-spheres, but the inference that a dual can be homotoped relative to its boundary out of T^{ε_M}_g requires a null-homotopy of the inclusion and control over the collar, not just a homology statement.","section":"Section 4, proof of Theorem 4.5"}],"minor_comments":[{"comment":"The phrase 'a homotopy of D fixing the points D∩T^δ_g and taking D into T^δ_g' appears to have the thick and thin parts reversed; the later text says the homotopy lands in the δ′-thin part.","section":"Section 4, proof of Lemma 4.3"},{"comment":"The proof contains typos 'an homotopy' and 'dimension dimension'.","section":"Section 4, proof of Theorem 4.5"},{"comment":"The notation 'h(D)_v' is often typeset with a missing space ('the seth(D)_v'); please fix the spacing.","section":"Throughout"},{"comment":"It would help to add a remark explaining in what category the family of stretch paths is smooth: smooth in the point x with respect to a fixed cell decomposition, or continuous on the whole complex.","section":"Section 4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is heavily self-citational, with Theorem 4.5 depending on the unpublished preprint [13] and with the main technical machinery coming from [17]. The editor may wish to verify the status of those preprints and the amount of overlap with the present paper. The Introduction's discussion of the 'failed analogy' in [6] is somewhat polemical and could be trimmed without affecting the mathematics."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the genuinely new idea: defining well-rounded deformation retractions for Teichmüller space via duals labelled by curve sets that span H_1(S_g;Q) is a real addition to the spine literature, and Lemma 1.2's necessary condition for cycles in the curve complex is an interesting statement in its own right. If Theorem 1.1 were solid, it would be a nice strengthening of the 4g-5 spine results, though it does not resolve a long-open problem.\n\nBut the proof of Theorem 1.1 turns on Lemma 4.3, and that proof has a gap that looks real. The stress-test note gets it right: the partition-of-unity construction produces m(x) as a weighted sum of block labels m_i, and nothing ensures that curves from different block labels are pairwise disjoint. The paper earlier defines a multicurve as a set of pairwise disjoint curves. So m(x) may not be a multicurve, and stretch paths are only defined for measured laminations. The line 'As these stretch paths vary smoothly with x' is asserted, not proved, and the 'sufficiently large M' and 'arbitrarily close to parallel' arguments are heuristic. If m(x) is not a well-defined lamination, the homotopy ψ_t is not constructed and Lemma 4.3 fails.\n\nThe second soft spot is the reliance on the author's own preprints [13] and [17]. The main theorem is essentially 'take the retraction from [13] and apply Lemma 1.2', so a referee cannot certify Theorem 1.1 without verifying two unpublished papers. That is heavy, even if the papers are available online. Finally, the paragraph attacking Fortier Bourque's paper is unprofessional and should be removed.\n\nOn the positive side, the duality framework is presented clearly enough to follow, and the connection between homology spanning and collapsibility is plausible. The conjecture about Out(F_n) is a reasonable speculation.\n\nBottom line: This is a research announcement with an interesting framework and a theorem whose proof is not yet trustworthy. I would send it to a serious referee, but the report should ask for a rigorous proof of Lemma 4.3 and for [13] to be verified or folded into the paper. If those gaps can be filled, it will be a useful contribution. As it stands, I would not base my own work on it.","headline":"A genuinely new definition and a plausible theorem, but the key stretch-path construction in Lemma 4.3 has a real gap, and the main result depends on unpublished work.","tokens_in":12835,"tokens_out":3712,"would_cite":false,"duration_ms":33023,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["57K20","32G15"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that for every genus $g \\ge 2$, Teichmüller space admits an equivariant deformation retraction onto a $4g-5$-dimensional well-rounded CW complex.","keywords":["Teichmüller space","mapping class group","deformation retraction","systole-filling spine","curve complex","well-rounded retract","duality","virtual cohomological dimension"],"falsifier":"Compute the horizon image $h(D)$ for an explicit dual $D$ whose label set fails to span $H_1(S_g;\\mathbb{Q})$ — for example the genus-5 example cited in the paper — and check whether $h(D)$ is a boundary in the curve complex, and simultaneously check whether the partition-of-unity multicurve $m(x)$ is single-valued and continuous across the block boundaries of $D$; a nonzero homology class in $h(D)$ or a discontinuity in $m(x)$ would disprove Lemma 4.3 and Theorem 1.1.","tokens_in":11798,"feed_emoji":"📐","tokens_out":14566,"duration_ms":119091,"temperature":0.7,"pith_summary":"This paper proves that for every closed oriented surface of genus $g\\ge 2$, Teichmüller space $T_g$ carries a mapping-class-group-invariant deformation retraction onto a CW complex of dimension $4g-5$ whose locally top-dimensional cells are 'well-rounded': each such cell has a dual labelled by a set of curves that spans $H_1(S_g;\\mathbb{Q})$. This is the correct analogue of the well-rounded retractions of $\\mathrm{SL}(n,\\mathbb{Z})$, with the matching notion of duality supplied by sets of minima and the horizon map into the barycentric subdivision of the curve complex. The key lemma states that if the curves labelling a dual fail to span rational homology, the horizon map sends that dual to a boundary, so the corresponding cell can be removed. Iterating this removal terminates in a well-rounded complex of the minimal possible dimension, because $4g-5$ is the virtual cohomological dimension of the mapping class group. A corollary is an elementary necessary condition for a cycle in the geometric realization of the curve complex to represent nonzero homology.","feed_headline":"Teichmüller space retracts to a 4g-5 well-rounded spine","feed_subtitle":"The collapse is mapping-class-group invariant and every top cell gains a dual whose curves span the surface's homology.","key_machinery":"The mechanical core of the paper is a duality between cells of the spine and sets of minima, connected to the curve complex by the horizon map $h$. A set of minima $\\mathrm{Min}(C)$ is the set of hyperbolic structures where a positively weighted sum of geodesic lengths over a filling curve set $C$ is minimized; the horizon map sends $\\mathrm{Min}(C)$ to the subcomplex of the barycentric subdivision of the curve complex spanned by multicurves that can be made arbitrarily short on $\\mathrm{Min}(C)$. Duals are unions of such sets, and the set $h(D)_v$ of curves labelling a dual $D$ records these short multicurves. The load-bearing lemma, Lemma 1.2 and its later restatement as Lemma 4.3, says that when $h(D)_v$ does not span $H_1(S_g;\\mathbb{Q})$, the image $h(D)$ is a boundary in the curve complex. To prove it, the paper builds an equivariant homotopy from weighted multicurves $m(x)$ assembled from a partition of unity, where the weights come from distance-level multicurves in a cyclic cover determined by a homology class missing from the label set; these weights determine stretch paths that move the dual into the thin part of Teichmüller space.","core_discovery":"The central claim, Theorem 1.1 (restated as Theorem 4.5), is that for every genus $g\\ge 2$ there is a well-rounded deformation retraction of Teichmüller space $T_g$ onto a CW complex of dimension $4g-5$. 'Well-rounded' is defined through duality: every locally top-dimensional cell of the complex has a dual labelled by a set of curves whose rational homology classes span $H_1(S_g;\\mathbb{Q})$, in direct analogy with the well-rounded retractions for $\\mathrm{SL}(n,\\mathbb{Z})$. The proof takes the image of an earlier equivariant deformation retraction onto the spine and examines its locally top-dimensional cells one by one. Whenever the curves labelling a dual do not span homology, Lemma 1.2 produces a homotopy of that dual into the thin part of Teichmüller space, showing the complex has nonempty boundary and allowing an equivariant retraction that removes the offending cell. Because each iteration drops dimension and there are only finitely many cell orbits, the process terminates, and the dimension cannot fall below $4g-5$ since that is the virtual cohomological dimension of the mapping class group.","pith_inferences":["If the converse to Lemma 1.2 is true, as the paper suspects, then well-rounded retracts would be minimal against any further equivariant collapse, not just minimal in dimension, because new duals would inherit homology-spanning labels.","The same dual-labelling criterion could be transplanted to other group actions with a curve-complex-like boundary, such as $\\mathrm{Out}(F_n)$ on Outer space with the free factor complex playing the role of the curve complex; the necessity of spanning homology would be the algebraic obstruction that makes the analogy work.","The cyclic-cover mechanism in Lemma 4.3 suggests looking for explicit high-genus examples where a filling systole set has homology-defective labels; the paper's cited genus-5 case is a natural test bed, and finding such examples would make the lemma's hypothesis directly checkable."],"forward_implications":["For every genus $g\\ge 2$, the mapping class group acts on a CW complex of dimension $4g-5$ that is both an equivariant spine for Teichmüller space and well-rounded in the paper's dual-labelling sense.","The iterative collapse always terminates: each step removes an orbit of cells and lowers dimension, and only finitely many cell orbits exist.","Any further equivariant retraction of a well-rounded complex is again well-rounded, because the duals of newly created cells contain the duals of the cells they came from.","A cycle in the barycentric subdivision of the curve complex whose vertices are labelled by a set of curves that does not span $H_1(S_g;\\mathbb{Q})$ cannot represent a nontrivial homology class.","The complex has minimal possible dimension: the lower bound $4g-5$ equals the virtual cohomological dimension of the mapping class group."],"supporting_citations":[{"why":"Supplies the earlier equivariant deformation retraction of $T_g$ onto a $4g-5$-dimensional subcomplex of the spine; Theorem 4.5 iterates its cell-removal construction.","marker":"[13]"},{"why":"Defines the horizon map and the equivariant duality between cells and sets of minima that the paper's notion of well-roundedness depends on.","marker":"[17]"},{"why":"Introduces sets of minima and proves they are nonempty exactly when the curve set fills; these sets are the building blocks of all duals.","marker":"[25]"},{"why":"Establishes the spine $P_g$ and the equivariant deformation retraction of $T_g$ onto its thick part, used to push duals into the thin part in Lemma 4.3.","marker":"[29]"},{"why":"Proves the barycentric subdivision of the curve complex is equivariantly homotopy equivalent to the boundary of the thick part, allowing the horizon image to be read as a cycle there.","marker":"[18]"},{"why":"Proves the virtual cohomological dimension of the mapping class group is $4g-5$, giving the lower bound that makes the resulting spine minimal in dimension.","marker":"[11]"}],"fun_headline_variants":["Well-rounded spine of Teichmüller space in dimension 4g-5","Teichmüller retraction mirrors SL(n,Z) via curve duality","4g-5 well-rounded spine: Teichmüller space analogy","New proof: Teichmüller space has a 4g-5 spine","Equivariant collapse of Teichmüller space to 4g-5 spine"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the weighted collection of curves assembled from the partition of unity varies continuously as the point moves across the piecewise blocks of a dual cell, so the stretch paths genuinely form a homotopy, and that the earlier unpublished construction of the $4g-5$-dimensional equivariant spine is correct; if either fails, Lemma 4.3 and hence Theorem 1.1 collapse.","fun_headline_variants_meta":{"raw":{"variants":["Well-rounded spine of Teichmüller space in dimension 4g-5","Teichmüller retraction mirrors SL(n,Z) via curve duality","4g-5 well-rounded spine: Teichmüller space analogy","New proof: Teichmüller space has a 4g-5 spine","Equivariant collapse of Teichmüller space to 4g-5 spine"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000213,"raw_usage":{"total_tokens":1405,"prompt_tokens":915,"completion_tokens":490,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":531,"completion_tokens_details":{"reasoning_tokens":387}},"tokens_in":531,"tokens_out":490,"duration_ms":4085,"temperature":1.0,"reasoning_tokens":387,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:17:26.230117+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the horizon image $h(D)$ for an explicit dual $D$ whose label set fails to span $H_1(S_g;\\mathbb{Q})$ — for example the genus-5 example cited in the paper — and check whether $h(D)$ is a boundary in the curve complex, and simultaneously check whether the partition-of-unity multicurve $m(x)$ is single-valued and continuous across the block boundaries of $D$; a nonzero homology class in $h(D)$ or a discontinuity in $m(x)$ would disprove Lemma 4.3 and Theorem 1.1.","supporting_citations":[{"cited_title":"Schmutz-Thurston Duality","cited_arxiv_id":"2508.04587","evidence_quote":"Defines the horizon map and the equivariant duality between cells and sets of minima that the paper's notion of well-roundedness depends on."},{"cited_title":"Schmutz Schaller","cited_arxiv_id":null,"evidence_quote":"Introduces sets of minima and proves they are nonempty exactly when the curve set fills; these sets are the building blocks of all duals."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proves the barycentric subdivision of the curve complex is equivariantly homotopy equivalent to the boundary of the thick part, allowing the horizon image to be read as a cycle there."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proves the virtual cohomological dimension of the mapping class group is $4g-5$, giving the lower bound that makes the resulting spine minimal in dimension."}],"review_version":2}