{"id":"16998f04-8232-4472-9eee-397f3e39104c","arxiv_id":"2506.08742","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Proper faces of convex sets in arbitrary real vector spaces are characterized by generalized semispaces, compatible total preorders, and regular step-affine functions; all three descriptions are equivalent.","lead":"This mathematics paper proves three equivalent ways to describe the faces, or boundary pieces, of convex sets in infinite-dimensional vector spaces. It extends a result previously known only in finite dimensions and answers a question raised in a 2023 paper.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"As stated, The sufficiency directions of Theorems 3, 4, and 6 do not prove properness: the RHS can hold for F=Q or F=∅, which are not proper faces under the paper's own definition.","rationale":"I read the main proof chain in good faith. The generalized-semispace construction in Theorem 3 and its use in Theorems 4 and 6 are internally coherent for nonempty faces different from Q: Proposition 1 gives Q\\F convex and affF disjoint from Q\\F; Zorn's lemma provides a generalized semispace containing Q\\F; and the sufficiency argument proves the face property. The external dependence on [7] and [8] is a real but secondary verification risk. My stress-test found a more immediate internal problem: the sufficiency direction never proves properness. The examples above are not pathological infinite-dimensional constructions; they are one-dimensional and finite. Therefore the theorems' universal statements are false exactly in the boundary cases F=∅ and F=Q. This does not invalidate the intended characterization of nonempty, non-whole faces, but the paper as written cannot be accepted without a corrected statement. Hence conditional acceptance: fix the missing guards and check that the proofs cover them, or adjust the definition of proper face.","tokens_in":10048,"tokens_out":22725,"duration_ms":303793,"concrete_test":"Apply Theorem 3 to X=R, Q={0}, M={0}, S=(0,∞). The RHS is satisfied: M is a nonempty affine manifold, S is a generalized semispace generated by M, Q⊂M∪S, and F=M∩Q={0}. But F=Q is not a proper face under the paper's definition. Repeating with Theorem 6 and u(x)=x, or with Theorem 4 and the usual order, gives the same failure. This one-line instantiation settles that the current iff statements are false; the fix is to require F≠∅ and F≠Q on the RHS, or to redefine proper face consistently and still exclude F=Q.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The weakest point is not the quoted classification results but the sufficiency direction's failure to enforce that the constructed F is proper. Section 2 defines a proper face as a nonempty face different from Q, but Theorems 3, 4, and 6 only verify that a set with the stated representation is a face. In Theorem 3, take X=R, Q={0}, F=Q, M={0}, and S=(0,∞). S is a generalized semispace generated by M, Q⊂M∪S, and F=M∩Q, so the RHS holds while F is not a proper face. The same counterexample works for Theorem 6 with u(x)=x. For Theorem 4, take Q={0} and any compatible total preorder on R, e.g. the usual order: Min(Q)=Q, so the RHS holds while Q is not a proper face of itself. Conversely, a compatible total preorder with no minimum, such as lexicographic order on R^2, produces F=Min(X)=∅, which the paper explicitly does not count as proper. Thus the universal iff statements are false as written. The defect is localized: the necessity proofs already use nonemptiness of F, and the sufficiency proofs establish that F is a face. Adding the conditions F≠∅ and F≠Q to the RHS, or changing the definition of proper face, repairs the theorems without altering the construction.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims three equivalent, topology-free characterizations of proper faces of convex sets in arbitrary real vector spaces: Theorem 3 characterizes a proper face F of Q as an intersection F = M ∩ Q with a nonempty affine manifold M and a generalized semispace S generated by M such that Q ⊆ M ∪ S; Theorem 4 characterizes F as the minimum set Min(Q | ⪯) of a compatible total preorder; Theorem 6 characterizes F as the zero set of a regular step-affine function u with u ≥ 0 on Q. Theorem 7 declares the three statements equivalent, and Theorem 8 gives finite-rank and rank-one refinements that recover lexicographically exposed and exposed faces. Corollary 9 then derives Martinez-Legaz's finite-dimensional lexicographic characterization.","tokens_in":10303,"tokens_out":12921,"duration_ms":108517,"significance":"The proposed characterizations are natural and would give a complete, purely algebraic description of all faces without topological assumptions. The use of generalized semispaces, compatible preorders built from recession cones, and step-affine functions is elegant, and the necessity directions are largely well motivated. The paper is not fully self-contained: Proposition 5 imports the classification of generalized semispaces by regular step-affine functions from the author's earlier work, which is acceptable if that result is correct. The main contribution is the reduction of face theory to halfspace-type objects, but as written the central iff statements are false, so the paper needs a substantive correction before the claimed equivalence can be accepted.","major_comments":[{"comment":"The sufficiency directions prove only that the set F is a face, not that it is a proper face. Concretely, in any nontrivial real vector space X take Q = {x0}, M = {x0}, and let S be any generalized semispace generated by M (which exists by Zorn). Then Q ⊆ M ∪ S and F = M ∩ Q = Q, so the right-hand side of Theorem 3 holds even though F is not a proper face. The same defect is present in Theorem 6: for a nonzero linear functional l, the regular step-affine function u(x) = l(x - x0) satisfies u ≥ 0 on Q and F = {x ∈ Q : u(x) = 0} = Q. For Theorem 4, the compatible total preorder defined by l yields Min(Q | ⪯) = Q. Thus the universal iff statements are false as written. The fix is local: add the conditions F ≠ ∅ and F ≠ Q to the right-hand sides of Theorems 3, 4, 6 and Theorem 7(b)–(d), and in the sufficiency proofs explicitly observe that the already-proved face property plus these two conditions gives properness. The necessity proofs already assume properness and are unaffected; the internal constructions (semispace selection, preorder from recession cone, step-affine function from Proposition 5) remain valid.","section":"Theorems 3, 4, 6 and Theorem 7"},{"comment":"The improperness issue propagates to Theorem 8. In the sufficiency proof of part (a), the paper says 'by Theorem 6 F is a face of Q' and then proceeds to show F is lexicographically exposed. Since Theorem 6 as stated yields only a face, not a proper face, the proof would also apply to F = Q, for which the conclusion 'F is lexicographically exposed' is not meaningful under the paper's definition of a proper face. Once Theorem 6 is repaired as suggested, the proof of Theorem 8 works, but the dependence on a corrected Theorem 6 must be stated.","section":"Theorem 8 and Section 6"}],"minor_comments":[{"comment":"The phrase 'All three characterization are equivalent each other' should be 'All three characterizations are equivalent to one another'.","section":"Abstract"},{"comment":"In the statement, 'A subset F⊂Qia a proper face' should read 'F⊂Q is a proper face'; in the proof, the expression 'S u S Mu' should contain explicit union symbols, e.g. S_u ∪ M_u.","section":"Theorem 6"},{"comment":"The theorem states that u maps X to X, but the correct codomain is R, i.e. u : X → R, as used in Theorems 6 and 8.","section":"Theorem 7(d)"},{"comment":"The displayed definition of Min(M∪S | ⪯_S) quantifies over y ∈ Q, but for the argument it must quantify over y ∈ M∪S; otherwise the subsequent equality M = Min(M∪S | ⪯_S) does not follow. This is a typo in the quantifier, not a substantive gap.","section":"Theorem 4, necessity proof"},{"comment":"Reference [10] contains the typo 'Pronidence' for 'Providence'.","section":"References"},{"comment":"The manuscript uses the symbol F for disjoint union in expressions such as 'M F S', which is easily confused with the face F. A different symbol such as ⨆ or an explicit verbal clarification would improve readability.","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":"The central flaw is a missing properness condition in the sufficiency directions; it is easy to repair without changing the constructions. Given that the paper leans heavily on the author's earlier classification results, the editor may want to require a precise statement of Proposition 5 and its provenance in the revision. The manuscript is otherwise a coherent contribution to convex analysis, and I would be willing to look at a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper: it genuinely delivers new algebraic characterizations of faces in infinite-dimensional real vector spaces, and its main theorems, as written, are false. The flaw is localized and repairable, but it's real.\n\nWhat's new: the semispace characterization in Theorem 3, the compatible-preorder version in Theorem 4, and the step-affine function version in Theorem 6 are not in the finite-dimensional papers and answer a question posed in [4]. The mutual equivalence and the reduction to prior classification results are coherent. The necessity proofs are solid: Proposition 1 gives the geometric facts, and the Zorn argument producing a generalized semispace containing Q\\F is standard and correctly applied. The step-affine duality via Proposition 5 is elegant, and the finite-dimensional corollary recovering Martinez-Legaz's lexicographical theorem is a nice close.\n\nNow the soft spot. The sufficiency directions of Theorems 3, 4, and 6 show only that the given F is a face, not that it is proper. The paper defines a proper face as nonempty and not equal to Q, but the RHS of each theorem is satisfied by F=Q and F=∅. A concrete counterexample: take X=R, Q={0}, M={0}, and S=(0,∞). Then S is a generalized semispace generated by M, Q is contained in M∪S, and F=M∩Q=Q, which is not a proper face. The same construction kills Theorem 4 (any compatible total preorder makes the singleton its own minimum) and Theorem 6 (u(x)=x). The fix is trivial: add F≠∅ and F≠Q to the RHS, or replace \"proper face\" with \"face\" and state the properness condition separately. The necessity direction already guarantees those two conditions, so the paper's core mathematics survives intact.\n\nMinor issues: Theorem 7(d) has a type error, u:X→X should be u:X→R; Theorem 6 contains a typo (\"Qia\"); and the reliance on quoted classifications from the author's earlier papers [7,8] is a transparency concern more than a correctness one. A referee might ask for those statements to be restated, but they are published results.\n\nBottom line: this is a serious contribution to convex analysis in general vector spaces, and the central theorem holds up after a small amendment. I would send it to a referee and accept after minor revision, but I would not cite the current arXiv version without noting the needed fix.","headline":"New topology-free face characterizations, but the theorems as stated fail because the sufficiency directions don't enforce properness; a small added condition fixes them.","tokens_in":10848,"tokens_out":4698,"would_cite":false,"duration_ms":55449,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["52A05","52A99"],"pacs":[],"model":"deepseek-v4-flash","headline":"In any real vector space, every proper face of a convex set is the zero set of a nonnegative regular step-affine function, and equally the minimum set of a compatible total preorder.","keywords":["convex sets","faces of convex sets","generalized semispaces","compatible total preorders","step-affine functions","lexicographically exposed faces","infinite-dimensional vector spaces","conical halfspaces"],"falsifier":"Find one convex set $Q$ in an infinite-dimensional real vector space and one proper face $F$ for which no affine manifold $M$ satisfies $F = M \\cap Q$ with $Q \\subset M \\cup S$ for some generalized semispace $S$ generated by $M$; equivalently, find a proper face that is not the zero set of any nonnegative regular step-affine function. The standard example of a non-exposed face (cited in the paper from [1, p.179]) is a natural search area, since non-exposed faces are where the classical linear-function description breaks down.","tokens_in":9803,"feed_emoji":"📐","tokens_out":9363,"duration_ms":104700,"temperature":0.7,"pith_summary":"This paper claims that in every real vector space, with no topology or finite-dimension assumption, a nonempty subset $F$ of a convex set $Q$ is a proper face exactly when it can be described in any of three equivalent algebraic ways: as $Q \\cap M$ for an affine manifold $M$ and a generalized semispace $S$ with $Q \\subset M \\cup S$; as the minimum set of a compatible total preorder (a translation- and positive-scaling-invariant reflexive transitive relation in which any two points are comparable); and as the zero set of a nonnegative regular step-affine function. These three characterizations are proved equivalent, so a face is visible in purely geometric, purely order-theoretic, and purely functional terms. The point of the extension is that infinite-dimensional spaces contain faces that no single linear function exposes, and lexicographic minimization with finitely many linear functions also misses some of them; the paper's framework supplies descriptions that work for every face.","feed_headline":"Every face of a convex set is cut out by a step-affine function","feed_subtitle":"Three equivalent descriptions—halfspaces, total preorders, step-affine functions—cover all faces in infinite-dimensional spaces.","key_machinery":"The machinery is the generalized semispace: a maximal by inclusion convex subset of $X$ that is disjoint from a given affine manifold $M$; Proposition 2 shows it has the cone-like structure $S = \\{a + t(x - a) \\mid x \\in S,\\ t > 0\\}$ for any $a \\in M$. Theorem 3 is built on it: a face $F$ is cut out by taking $M = \\mathrm{aff}\\,F$ and any semispace $S$ generated by $M$ that contains $Q \\setminus F$, so $Q \\subset M \\cup S$ and $F = M \\cap Q$. From this single construction, the other two characterizations are derived. The compatible total preorder comes from the recession cone $0^{+}S$ of $S$, which is a conical halfspace and yields the order $x \\preceq y$ iff $y - x \\in 0^{+}S$. The step-affine function comes from Proposition 5, quoted from earlier work, which identifies $S$ as $\\{u > 0\\}$ and $M$ as $\\{u = 0\\}$ for a regular step-affine function $u$. Theorem 8 then splits the step-affine version into finite rank (lexicographically exposed faces) and rank one (exposed faces).","core_discovery":"The central result is a three-way equivalence, stated as Theorem 7, for any convex set $Q$ in a real vector space $X$ and any $F \\subset Q$: $F$ is a proper face of $Q$ iff (b) there exist a nonempty affine manifold $M$ and a generalized semispace $S$ generated by $M$ with $Q \\subset M \\cup S$ and $F = M \\cap Q$; iff (c) $F$ is the set $\\mathrm{Min}(Q\\,|\\preceq)$ of minimal elements for some compatible total preorder $\\preceq$ on $X$; iff (d) there is a regular step-affine function $u$ with $u(x) \\ge 0$ on $Q$ and $F = \\{x \\in Q \\mid u(x) = 0\\}$. Theorem 3 establishes (b), Theorem 4 establishes (c), and Theorem 6 establishes (d). The proof route passes through (b): Proposition 1 shows that for a proper face, $Q \\setminus F$ is convex and $\\mathrm{aff}\\,F$ is disjoint from $Q \\setminus F$, so a generalized semispace generated by $\\mathrm{aff}\\,F$ can contain $Q \\setminus F$. Then (c) follows by taking the recession cone of the semispace, and (d) follows by representing the semispace as the positive set of a regular step-affine function.","pith_inferences":["A natural next step the paper only hints at is to index faces by the order type of the defining cortege: finite rank gives lexicographic faces, while countable rank would give a new intermediate class in infinite-dimensional spaces.","The preorder formulation could be applied to vector optimization without topology: minimal elements of compatible total preorders give a face-based solution concept analogous to lexicographic scalarization.","One could test the sharpness of Proposition 5 by asking whether every generalized semispace in a given space can be generated by a cone over a linearly independent set of linear functionals; if not, the step-affine representation would need a wider function class."],"forward_implications":["In any real vector space, faces can be studied without any topology: a face is fully determined by the algebraic data of a maximal convex set disjoint from an affine manifold.","Every proper face of a convex set is the minimum set of some compatible total preorder, so face structure is fully expressible in order-theoretic terms.","Every proper face is the zero set of a nonnegative regular step-affine function; this generalizes the classical picture in which an exposed face is the zero set of a supporting affine function.","Faces that are lexicographically exposed are exactly those representable by a finite-rank step-affine function, and exposed faces by rank-one step-affine functions, so the finite-dimensional theorem becomes a corollary."],"supporting_citations":[{"why":"Supplies the standard definitions of convex sets, faces, exposed faces, and the classical example of a non-exposed face that motivates the need for a broader characterization.","marker":"[1]"},{"why":"Notes the convexity of $Q \\setminus F$ for a proper face and raises the question of generalizing lexicographic face characterizations to general vector spaces.","marker":"[4]"},{"why":"Establishes the finite-dimensional lexicographic characterization of faces that the present paper extends to infinite-dimensional spaces.","marker":"[5]"},{"why":"Supplies the recession-cone criteria and classification of halfspaces used in Theorem 4 to build a compatible total preorder from a generalized semispace.","marker":"[7]"},{"why":"Source of Proposition 5, the duality between generalized semispaces and regular step-affine functions that carries Theorem 6.","marker":"[8]"},{"why":"Provides the Kuratowski-Zorn lemma used to assert the existence of generalized semispaces containing a given convex set disjoint from an affine manifold.","marker":"[11]"},{"why":"Supplies the background on compatible preorders and their positive cones that underlies the order-theoretic characterization in Theorem 4.","marker":"[12]"}],"fun_headline_variants":["Faces in infinite dimensions: three equivalent cuts","Step-affine functions slice every convex face","Total preorders and semispaces reveal all faces","Infinite-dim faces: semispace, preorder, step-affine","Convex faces characterized by three tools in any dimension"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the imported duality that generalized semispaces are exactly the positive sets of regular step-affine functions, together with the recession-cone criterion that makes a compatible preorder total; if either of those classifications has a hidden restriction, the equivalence would fail for some face.","fun_headline_variants_meta":{"raw":{"variants":["Faces in infinite dimensions: three equivalent cuts","Step-affine functions slice every convex face","Total preorders and semispaces reveal all faces","Infinite-dim faces: semispace, preorder, step-affine","Convex faces characterized by three tools in any dimension"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000115,"raw_usage":{"total_tokens":1080,"prompt_tokens":960,"completion_tokens":120,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":576,"completion_tokens_details":{"reasoning_tokens":42}},"tokens_in":576,"tokens_out":120,"duration_ms":2293,"temperature":1.0,"reasoning_tokens":42,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:05:28.195019+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find one convex set $Q$ in an infinite-dimensional real vector space and one proper face $F$ for which no affine manifold $M$ satisfies $F = M \\cap Q$ with $Q \\subset M \\cup S$ for some generalized semispace $S$ generated by $M$; equivalently, find a proper face that is not the zero set of any nonnegative regular step-affine function. The standard example of a non-exposed face (cited in the paper from [1, p.179]) is a natural search area, since non-exposed faces are where the classical linear-function description breaks down.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the standard definitions of convex sets, faces, exposed faces, and the classical example of a non-exposed face that motivates the need for a broader characterization."},{"cited_title":"Set-Valued and Variational Analysis, Article 14 (2023) https://doi.org/10.1007/s11228-023-00671-6","cited_arxiv_id":null,"evidence_quote":"Notes the convexity of $Q \\setminus F$ for a proper face and raises the question of generalizing lexicographic face characterizations to general vector spaces."},{"cited_title":"Acta Mathematica Vietnamica 22(1), 207–211 (1997)","cited_arxiv_id":null,"evidence_quote":"Establishes the finite-dimensional lexicographic characterization of faces that the present paper extends to infinite-dimensional spaces."},{"cited_title":"Mathematical Notes 64(2), 164 – 169 (1998) https://doi.org/10.1007/BF02310300","cited_arxiv_id":null,"evidence_quote":"Supplies the recession-cone criteria and classification of halfspaces used in Theorem 4 to build a compatible total preorder from a generalized semispace."},{"cited_title":"In: Przeworska–Rolewich, D","cited_arxiv_id":null,"evidence_quote":"Source of Proposition 5, the duality between generalized semispaces and regular step-affine functions that carries Theorem 6."},{"cited_title":"Springer-Verlag (2006)","cited_arxiv_id":null,"evidence_quote":"Provides the Kuratowski-Zorn lemma used to assert the existence of generalized semispaces containing a given convex set disjoint from an affine manifold."},{"cited_title":"Harper and Row, (1967) 12","cited_arxiv_id":null,"evidence_quote":"Supplies the background on compatible preorders and their positive cones that underlies the order-theoretic characterization in Theorem 4."}],"review_version":1}