{"id":"f4162985-66dc-491f-984d-bfb51086f039","arxiv_id":"2506.08733","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A subfield of the reals is constructed in which the set of squares, defined by a quantifier-free formula, is not Borel, and the field has the independence property.","lead":"The paper builds a subfield of the real numbers that is wild in a model-theoretic sense: it can define, with one simple formula, a set that is not Borel. The construction is a stepping stone toward understanding whether NIP, a tameness property, forces definable sets to be measurable.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the two proof details flagged by the reader are fillable and do not threaten Theorem 3.3.","rationale":"The reader's verdict is CONDITIONAL because of two proof gaps: the WLOG step in Proposition 2.3 and the unproved Lemma 3.2. I examined both and found them fillable. The WLOG step works by scaling each element of a size-c algebraically independent set by a small rational; multiplication by nonzero elements of K preserves algebraic independence, and the scaled set retains cardinality c. The cardinality of U_{σ,α} follows from T_{σ,α}⊆U_{σ,α}−U_{σ,α}, which is immediate from the definitions, so the potential worry about |K|=c is not a real obstruction. Lemma 3.2(ii) follows from the preservation of transcendence degree under algebraic extensions, and Lemma 3.2(i) follows from unique factorization in a rational function field. The central measure-theoretic construction in Theorem 3.3 is sound: A∩[0,1]⊆D and A′∩[0,1]⊆K\\D, together with the inner-measure-zero conditions from Proposition 2.3, force the Borel set B to have measure 1 on both B∩[0,1] and [0,1]\\B, a contradiction. I do not find a load-bearing flaw. The paper would benefit from adding the T⊆U−U cardinality observation and a one-line proof of Lemma 3.2, but these are editorial improvements, not correctness risks. Hence I do not move the verdict; I keep the reader's CONDITIONAL as a fair assessment of the manuscript's presentation.","tokens_in":12361,"tokens_out":51444,"duration_ms":599442,"concrete_test":"Verify the cardinality argument in Proposition 2.3 by explicitly checking that T_{σ,α}⊆U_{σ,α}−U_{σ,α}; if this inclusion holds, then |U_{σ,α}|≥|T_{σ,α}|=c, which settles the WLOG step without any dependence on |K|. Also confirm R. Robinson's Fact 2.7 applies to arbitrary subfields of R (formally real with archimedean ordering) by consulting the cited reference; if it only holds for real closed fields, the IP claim in Theorem 3.3 would need a more restrictive hypothesis.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I find no load-bearing concern for the central claim. The two spots the reader flagged are not fatal. (1) The 'without loss of generality' step in Proposition 2.3 is justified: since Q⊆K, each element of a size-c algebraically independent set T can be multiplied by a sufficiently small nonzero rational to land in the Steinhaus neighborhood S=(R\\C_σ)−(R\\C_σ); scaling each element by a nonzero element of K preserves algebraic independence over K, and the scaled set retains cardinality c. The cardinality of U_{σ,α} can be proven directly from T_{σ,α}⊆U_{σ,α}−U_{σ,α}, so |U_{σ,α}|≥|T_{σ,α}|=c, without relying on the relative algebraic closure cardinality argument. (2) Lemma 3.2 is true: Q(√C∪C′) is algebraic over Q(C∪C′), so the transcendence degree is preserved, making √C∪C′ algebraically independent; and in the rational function field Q(√C∪C′), no element of C′ is a square by unique factorization. The measure-theoretic contradiction in Theorem 3.3 is sound. The only minor issue is that Proposition 2.3's proof should explicitly state the T⊆U−U argument for readers concerned about the case |K|=c, but the theorem itself uses only K=Q, where the original argument also works.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the relation between model-theoretic tameness and Borel measurability for subfields of R in the language of rings. Its main result (Theorem 3.3) constructs a subfield K of R of cardinality c that has the independence property and ∅-defines the set D of its squares, which is not Borel with respect to the order topology on K. The proof combines a transfinite construction (Proposition 2.3) of continuum many pairwise disjoint, measure-theoretically wild, algebraically independent subsets of R with R. Robinson's theorem that Z is definable in purely transcendental extensions of real fields, and a measure argument showing that D cannot equal B∩K for any Borel B⊆R. The paper also records that K is not o-minimal, is undecidable, and admits maximal families of archimedean and non-archimedean orderings, and it situates the result relative to Shelah's Conjecture on NIP fields.","tokens_in":12545,"tokens_out":14654,"duration_ms":183988,"significance":"If correct, the result provides a striking example that first-order definability in the pure ring language on a subfield of R does not imply Borel measurability, and it clarifies that the independence property can coexist with non-Borel definable sets. The construction is elementary and largely self-contained, using only standard facts such as Steinhaus's Theorem and R. Robinson's theorem, and the measure-theoretic contradiction in Theorem 3.3 is elegant. The paper does not answer Question 1.1 for NIP fields, but it supplies a natural test case and connects the question to Shelah's Conjecture. The main proof is mathematically sound modulo two local justifications discussed below, neither of which affects the central claim.","major_comments":[],"minor_comments":[{"comment":"In the choice of a_{σ,α}, the inference 'Since |D_{σ,α}|<|T_{σ,α}|=c, this yields |U_{σ,α}|=c' is only justified when the field generated by K, D_{σ,α}, and U_{σ,α} has cardinality <c, which holds for |K|<c but not in general for the stated hypotheses. The conclusion is nevertheless correct, because for every t∈T_{σ,α} the witnesses u,v with t=u−v lie in U_{σ,α}, so T_{σ,α}⊆U_{σ,α}−U_{σ,α}, and |U_{σ,α}−U_{σ,α}|≤|U_{σ,α}| for infinite U_{σ,α}; please either replace the argument with this observation or restrict the proposition to |K|<c, the case actually used in Theorem 3.3.","section":"§2.1, Proposition 2.3"},{"comment":"The 'without loss of generality' step that scales T_{σ,α} into the Steinhaus difference (R\\C_σ)−(R\\C_σ) should be spelled out: multiplying each t by a sufficiently small nonzero rational q∈Q⊆K preserves algebraic independence over K and cardinality, and Steinhaus's Theorem supplies an interval around 0 contained in the difference.","section":"§2.1, Proposition 2.3"},{"comment":"Lemma 3.2 is stated without proof and is used essentially in Theorem 3.3; since it is described as straightforward, please include a short proof (for instance, via the rational function field Q(C∪C′) and unique factorization) or a precise reference.","section":"§3, Lemma 3.2"},{"comment":"The notation A_{≥0} is used without definition; please define A_{≥0}=A∩[0,∞), and note explicitly that the equalities µ_*(A∩[0,1])=µ_*(A′∩[0,1])=1 follow from Lemma 2.5 together with µ_*(R\\A)=µ_*(R\\A′)=0.","section":"§3, Theorem 3.3"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. First, this is a correct and genuinely new construction: a subfield K of R that has the independence property and ∅-defines a non-Borel set (the set of squares) in the ring language. The technical core, Proposition 2.3, is a transfinite construction of continuum many pairwise disjoint algebraically independent sets, each with inner measure zero and outer measure full. I checked the induction and the measure argument; they hold. Second, the main open question motivating the paper—whether NIP ordered fields have Borel-definable sets—remains open, and the paper is honest about that.\n\nWhat is new: the construction itself. The family of algebraically independent non-Borel sets with these measure properties does not appear in the cited literature, and the application to the set of squares is a clean counterexample showing that ring-language definability in a subfield of R does not imply Borel measurability. The proof of Theorem 3.3 combines Steinhaus's theorem, a cardinality trick to keep the sets non-Borel, and Robinson's definability of Z in F(t). Remark 3.4(a) also gives a neat cardinality argument for 2^c many subfields with distinct sets of squares, though it cannot reach non-Borelness directly.\n\nSoft spots, in proportion. Lemma 3.2 is stated without proof; it is straightforward algebra (Q(√C∪C′) is algebraic over Q(C∪C′), and no element of C′ is a square by unique factorization), so I do not see a gap. The 'without loss of generality' in Proposition 2.3, moving T into the Steinhaus difference, is terse: you need to scale by a sufficiently small rational, not just any nonzero element of K. That works because Q⊆K and the difference is a neighborhood of 0, and scaling preserves both cardinality and algebraic independence. Worth making explicit for the reader, but it is not a flaw.\n\nWho this is for: model theorists and people working in tame geometry or definability versus measurability. The paper is well-organized and clearly written. I would send it to a serious referee; the conditions are minor exposition fixes, not substantive corrections.","headline":"A correct and genuinely new construction of a subfield of R that is not NIP and ∅-defines a non-Borel set of squares; the main proof is sound, with only two terse spots that are fillable.","tokens_in":13161,"tokens_out":1632,"would_cite":true,"duration_ms":19225,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["03C64","28A05","12L12","12J15","03C40","03C45","54H05"],"pacs":[],"model":"deepseek-v4-flash","headline":"A subfield of the real numbers of cardinality continuum defines a non-Borel set of squares in the language of rings, and has the independence property.","keywords":["Borel","first-order definable","NIP","independence property","non-measurable","tame geometry","subfield of the reals","transfinite induction"],"falsifier":"Check the asserted “without loss of generality” in Proposition 2.3: for $K=\\mathbb Q$ and $C=[0,1]$, the Minkowski difference is a neighborhood of 0; one would need to confirm that every such neighborhood contains a $\\mathbb Q$-algebraically independent set of cardinality $\\mathfrak c$. If some small $\\delta>0$ yields only smaller algebraically independent sets, the induction step and hence the main theorem fail.","tokens_in":12072,"feed_emoji":"🧮","tokens_out":13676,"duration_ms":157626,"temperature":0.7,"pith_summary":"Tame geometry usually lets model-theoretic tameness conditions imply good topological behavior, such as definable sets being Borel. This paper constructs a subfield $K$ of the real numbers, of cardinality the continuum, in which a very simple first-order formula in the language of rings—“$x$ is a square”—defines a set that is not Borel in the order topology. The same field has the independence property, is undecidable, and admits both archimedean and non-archimedean orderings. The construction is a transfinite induction producing continuum many pairwise disjoint, algebraically independent subsets of $\nobreak R$ that are invisible to Borel measure from both inside and outside. If correct, ring-language definability in a subfield of the reals does not imply Borel measurability, and the field is a counterexample to several tameness/measurability bridges, though not to the open NIP question because the constructed field has the independence property.","feed_headline":"A ring formula in a subfield of R can define a non-Borel set","feed_subtitle":"The new field shows ring-theoretic definability does not force Borel measurability, a key bridge in tame geometry.","key_machinery":"The load-bearing construction is Proposition 2.3, a transfinite induction that produces a family $(A_\\alpha)_{\\alpha<\\mathfrak c}$ of pairwise disjoint subsets of $\\mathbb R$, algebraically independent over the base field $K$, such that each $A_\\alpha$ has inner measure $0$ and the complement of $A_\\alpha$ also has inner measure $0$; such sets cannot be Borel. At each induction step, a classical result on Minkowski differences of positive-measure sets is used to place the next real number outside a prescribed Borel set while preserving algebraic independence. The definable non-Borel set is $D=\\{y^2:y\\in K\\}$, defined by the formula $\\exists y\\, (x=y^2)$. The independence property follows from a classical result that $\\mathbb Z$ is definable without parameters in $F(t)$ whenever $F$ is a real field with an archimedean ordering and $t$ is transcendental over $F$.","core_discovery":"The paper's main theorem (Theorem 3.3) asserts the existence of a subfield $K\\subseteq \\mathbb R$ of cardinality $\\mathfrak c$ such that $K$, viewed as a structure in the language of rings $\\{+,-,\\cdot,0,1\\}$, has the independence property and $\\emptyset$-defines a set $D\\subseteq K$ that is not a Borel set with respect to the order topology on $K$. The set $D$ is the set of squares in $K$, defined by the formula $\\exists y\\, (x=y^2)$. The field is built as $K=\\mathbb Q(\\sqrt{A_{\\ge 0}}\\cup A')$ for two disjoint algebraically independent sets $A,A'\\subseteq \\mathbb R$ that are non-Borel in a strong sense: each has inner measure $0$ and its complement also has inner measure $0$. The proof shows that if $D$ were Borel, it would have to be approximated by Borel sets in $[0,1]$ with measure both $1$ and $0$ simultaneously, a contradiction. Because $\\mathbb Z$ is $\\emptyset$-definable in $K$, the independence property follows; because $\\mathbb Z$ is definable, $K$ is undecidable.","pith_inferences":["The same method would produce a non-Borel set of squares in any subfield $K=\\mathbb Q(S)$ where $S$ is a set of continuum many algebraically independent reals arranged so that some of their square roots lie in $K$ and some do not; the transfinite construction is one way to get such $S$, but the definability-to-measurability failure may be much more common.","If the unproved “without loss of generality” step in Proposition 2.3 is repaired, the construction yields a whole family of $\\mathfrak c$ many pairwise disjoint non-Borel algebraically independent sets; this could be used to build many non-isomorphic wild subfields, possibly including some that avoid the independence property if the $\\mathbb Z$-definability step were replaced.","A natural next test is whether an NIP subfield of $\\mathbb R$ can define a non-Borel set; the present construction suggests that the obstacle is not the orderings or the Borel $\\sigma$-algebra but the algebraic independence of the defining parameters, so an NIP example would need a different definability mechanism."],"forward_implications":["The set of squares in $K$ is not only non-Borel but also not of the form $M\\cap K$ for any Lebesgue measurable $M\\subseteq \\mathbb R$, since the proof's measure argument goes through with Lebesgue measure.","$K$ cannot be o-minimal, since every definable set in an o-minimal ordered field is Borel in the order topology; in particular $K$ is not real closed and not almost real closed.","$\\mathbb Z$ is $\\emptyset$-definable in $K$, so $K$ is undecidable and has the independence property; hence $K$ is a wild field from the tame-geometry perspective.","$K$ admits $2^{\\mathfrak c}$ pairwise non-isomorphic archimedean orderings and $2^{\\mathfrak c}$ pairwise non-isomorphic non-archimedean orderings, so the failure of tameness does not depend on choosing one ordering.","The motivating question of whether every NIP ordered field has Borel definable sets remains open: the constructed $K$ has the independence property, so it does not settle that question."],"supporting_citations":[{"why":"Supplies the theorem that a positive-measure Lebesgue-measurable set has a Minkowski difference that is a neighborhood of 0, used in Proposition 2.3 to place the next real number outside a given Borel set.","marker":"[26]"},{"why":"Supplies the result that $\\mathbb Z$ is $\\emptyset$-definable in $F(t)$ for real $F$ with an archimedean ordering and $t$ transcendental over $F$, which yields the independence property in Theorem 3.3.","marker":"[22]"},{"why":"The motivating result connecting tame geometry to statistical learning; the paper's Question 1.1 asks whether its o-minimal measurability conclusions extend to NIP fields, and Theorem 3.3 answers only the non-NIP side.","marker":"[17]"},{"why":"Provides the measure-theoretic setup: Borel measure, Lebesgue completion, and inner and outer measure facts used throughout Proposition 2.3 and Lemma 2.5.","marker":"[1]"},{"why":"Gives the descriptive-set-theoretic fact that every positive-measure Borel set has cardinality $\\mathfrak c$, used in the choice of $w_\\sigma$ during the induction.","marker":"[13]"},{"why":"Records that definable sets in o-minimal ordered fields are Borel, the benchmark against which the non-Borel definable set is contrasted.","marker":"[11]"},{"why":"Supplies the facts about Borel $\\sigma$-algebras in topological spaces used to identify the trace and order-topology $\\sigma$-algebras on $K$ in Lemma 3.1.","marker":"[2]"}],"fun_headline_variants":["Subfield of R defines non-Borel set from squares","Ring-definable non-Borel set exists in a real subfield","Independence property and non-Borel definable set in R-subfield","A subfield of reals with non-Borel definable squaring"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction's transfinite induction assumes, without proof, that at every stage a continuum-sized algebraically independent set can be placed inside the Minkowski difference of the complement of any positive-measure Borel set; if that “without loss of generality” step fails, the non-Borel sets $A_\\alpha$ cannot be produced.","fun_headline_variants_meta":{"raw":{"variants":["Subfield of R defines non-Borel set from squares","Ring-definable non-Borel set exists in a real subfield","Independence property and non-Borel definable set in R-subfield","A subfield of reals with non-Borel definable squaring"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000309,"raw_usage":{"total_tokens":1732,"prompt_tokens":883,"completion_tokens":849,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":785}},"tokens_in":499,"tokens_out":849,"duration_ms":8975,"temperature":1.0,"reasoning_tokens":785,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:08:21.559361+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the asserted “without loss of generality” in Proposition 2.3: for $K=\\mathbb Q$ and $C=[0,1]$, the Minkowski difference is a neighborhood of 0; one would need to confirm that every such neighborhood contains a $\\mathbb Q$-algebraically independent set of cardinality $\\mathfrak c$. If some small $\\delta>0$ yields only smaller algebraically independent sets, the induction step and hence the main theorem fail.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the result that $\\mathbb Z$ is $\\emptyset$-definable in $F(t)$ for real $F$ with an archimedean ordering and $t$ transcendental over $F$, which yields the independence property in Theorem 3.3."},{"cited_title":"Kaiser , ‘First order tameness of measures’, Ann","cited_arxiv_id":null,"evidence_quote":"Records that definable sets in o-minimal ordered fields are Borel, the benchmark against which the non-Borel definable set is contrasted."}],"review_version":1}