Independent examination of the E0836 edge-intersection argument

---

Subject and independence

Examination date: 2026-10-14 UTC. Subject freeze: 2026-10-14T11:26:16.490126+00:00. Reviewer: an independent adversarial mathematical reviewer, model gpt-6-astra. The reviewer neither authored nor collaborated on this argument and had no exposure to it before this examination. The context began fresh, with the charge to find a false statement, missing hypothesis, unsupported inference, quantifier error, or false consequence. Project attribution is to Plasma AI.

The mathematical subject consists of these two preserved extractions and their path/date/cut manifest:

• The frozen problem formulations preserve the supplied extraction of wiki/problems/graph_coloring/E0836/_index.md. Their retained path is wiki/research/erdos_836/evidence/assets/linear_intersection_subject/target.md. All 20 lines were read; lines 3–18 display the two formulations.
• The frozen argument preserves the supplied extraction of wiki/problems/graph_coloring/E0836/claims/2026_04_27_price.md. Its retained path is wiki/research/erdos_836/evidence/assets/linear_intersection_subject/argument.md. All 39 lines were read: the conclusion and argument at lines 3–25, the Covers paragraph at lines 29–32, and the dependency statement at lines 36–37.
• The subject manifest identifies the original paths, freeze time, and uniform [cut] marker. Its retained path is wiki/research/erdos_836/evidence/assets/linear_intersection_subject/manifest.json.

These are snapshots of the supplied extracts, not copies of the complete original cards. The original paths identify provenance and were not opened. The extraction's first, vertex-count question was visible but is outside this examination. All coverage here concerns the second, edge-intersection question. The mathematical subject is the recorded argument itself; fidelity to another version of an external document is not asserted.

The frozen packet was checked against its supplied identity manifest before mathematical reading and remained unchanged through the examination. The subscription table was checked and was empty. No incoming communication, author account, private plan, prior examination, sibling result, mathematical history, other source, or web search was consumed. No excluded exposure occurred. The review used its own continuation only. Agreement with a previous result or with an author's computation supplies no part of the reasoning below.

The allowed mathematical reading was exactly the two extracts and their manifest, and all of it was read. Operating reading was limited to the supplied root AGENTS instructions; README.md; the introduction, Claims, Evidence, Tiers, Naming and Project claims sections of docs/anatomy.md; all of docs/verification.md, docs/evidence.md, docs/math_authoring.md and docs/weaving.md; campaign/AGENTS.md; all of campaign/guides/verification.md; the Reading section of campaign/guides/research.md; and the verified commission, steps, configuration, and native operating instructions. Operating links were not followed to other mathematical material.

The independent derivations and exact witnesses are contained in this report. The review uses no mathematical inputs beyond the frozen packet and ordinary foundations. No computer calculation, executable mathematical checker, or kernel proof carries the verdict. Reproduction consists of reading the preserved extracts and checking the displayed derivations and finite sets below; there is no mathematical rerun command or cached-success premise.

Finding. The stated non-2-colorable theorem, including its specialization to chromatic number exactly 3 and uniform linear consequence, has verdict refutation-failed under the conventions below. The first formulation's parenthetical, if taken as the complete hypothesis of mere 3-colorability, gives a false extension: refuted-as-stated. The frozen argument explicitly acknowledges that distinction. These are separate scopes, not opposing verdicts on the same proposition.

Restatement and conventions

Let $r$ be any integer with $r\ge 2$. A hypergraph is a pair $G=(V,\mathcal E)$, where $V$ is a set and $\mathcal E$ is a family of distinct $r$-element subsets of $V$. The vertex set and edge family may be infinite. Intersecting means that every two distinct edges meet. A proper $q$-coloring is a map from $V$ to a set of $q$ colors with no monochromatic edge; it need not use every available color. The coloring is of vertices, and properness does not require every edge to use all $q$ colors. The equation $\chi(G)=3$ means that a proper 3-coloring exists and a proper 2-coloring does not exist.

The proposition being assessed is:

For every such intersecting hypergraph that admits no proper 2-coloring, there are distinct edges $E,F\in\mathcal E$ with $\lvert E\cap F\rvert\ge\lfloor r/2\rfloor$.

The finite and infinite cases are both included. Infinitary assertions are read in ordinary classical set theory with choice; the exact compactness use is proved below from Zorn's lemma. The finite argument requires no infinitary compactness. No choice-free claim is made.

The numerical consequence to check is a constant independent of both $r$ and $G$. The displayed bound gives the explicit all-order conclusion $\lvert E\cap F\rvert\ge r/3$, so it implies the asserted $\gg r$ statement. No extremal optimality claim for each value of $r$ is being added.

Isolated vertices are permitted by this proposition. Colors on isolated vertices do not affect any edge, and the proof below restricts a finite subhypergraph to the union of its edges. Thus the modified formulation's requirement that every vertex lie in an edge is harmless for this conclusion.

The simple-hypergraph convention is sufficient for the statement. If repeated edge occurrences are allowed instead, removing repeated copies leaves colorability unchanged. Distinct supports obtained from the simple family are also distinct occurrences; alternatively, two occurrences with the same support already have intersection $r$. No counterexample is introduced by that extension of convention.

The exact-chromatic-number condition implies the non-2-colorability hypothesis immediately. In fact every nonempty intersecting $r$-uniform hypergraph with $r\ge2$ has a proper 3-coloring: fix an edge $A$, give one vertex of $A$ color 2, the other vertices of $A$ color 3, and all vertices outside $A$ color 1. The edge $A$ uses colors 2 and 3. Every different edge meets $A$ and, by equal finite sizes and distinctness, also meets its complement, so it uses color 1 and another color. This is an elementary convention check, not a replacement for the intersection-bound argument. Mere 3-colorability omits the essential absence of a proper 2-coloring.

Weakest steps

1. Compactness, finite minimality, and the optimized coloring

The frozen argument invokes compactness but does not include a separate proof of the lemma whose proof it attributes to the note. The following is an explicit foundational expansion of that invoked step. It does not certify anything about an unseen document.

Suppose every finite edge subfamily of a hypergraph is properly 2-colorable. Let $\Omega=\{0,1\}^V$. For each edge $e$, let $C_e\subseteq\Omega$ consist of assignments nonconstant on $e$. Every finite intersection of sets $C_e$ is nonempty: a coloring of the corresponding finite subhypergraph can be extended by color 0 off its vertex set. These finite intersections generate a proper filter on $\Omega$.

Consider the proper filters extending the generated filter, ordered by inclusion. A nonempty chain has its union as a proper-filter upper bound, because each finite collection of members lies together in one filter in the chain; the generated filter itself is an upper bound for the empty chain. Zorn's lemma therefore gives a maximal proper filter $\mathcal U$. Maximality implies that for every $S\subseteq\Omega$, exactly one of $S$ and $\Omega\setminus S$ belongs to $\mathcal U$. Both cannot belong to a proper filter. If $S$ does not belong, adjoining it must make the generated filter improper, so some member of $\mathcal U$ is disjoint from $S$; upward closure then puts $\Omega\setminus S$ in $\mathcal U$.

For each vertex $v$, the two assignment sets specifying its two possible colors partition $\Omega$. Give $v$ the color whose assignment set belongs to $\mathcal U$. If an edge $e$ were monochromatic in this coloring, the intersection of $C_e$ with those same-color assignment sets for all vertices of $e$ would be empty. All these sets belong to $\mathcal U$, and there are only finitely many of them because $e$ has $r$ vertices. This contradicts properness of the filter. The resulting coloring is therefore proper on the entire hypergraph. Contrapositively, a non-2-colorable hypergraph has a finite non-2-colorable edge subfamily.

Apply this to $G$ and choose a non-2-colorable subfamily with the fewest edges inside the resulting finite family. Denote it by $H$, with vertex set $W=\bigcup_{e\in H}e$. Both $H$ and $W$ are finite, and $H$ remains intersecting and $r$-uniform. This is a finite minimum, not an infinite deletion procedure whose termination is assumed. The family has at least two distinct edges: an empty family is colorable, and a single edge of size at least two can be colored with both colors.

Fix $A\in H$. By minimality, $H\setminus\{A\}$ has a proper red-blue coloring. Such a coloring must make $A$ monochromatic, since otherwise it would color $H$ properly. Interchanging color names makes $A$ red. Among all proper colorings of $H\setminus\{A\}$ with $A$ red, choose one with the fewest blue vertices. This set of colorings is nonempty and finite. Write $B\subseteq W$ for its blue vertex set.

Define

$$t=\max\{\lvert E\cap F\rvert:E,F\in H,\ E\ne F\}.$$

The set being maximized is nonempty and finite, so the maximum is attained by distinct edges. Intersectingness and simplicity give $1\le t\le r-1$. The word distinct in this definition is necessary: allowing $E=F$ would instead give the useless value $r$. Reading the frozen paragraph's largest intersection in the convention of its stated distinct-edge conclusion is made explicit here.

2. The two recolorings and the incidence lower bound

Fix $x\in A$ and change just $x$ from red to blue. The edge $A$ is now bichromatic because $r\ge2$. No new red monochromatic edge can be created by this change. Since $H$ is not 2-colorable, some edge $F_x$ becomes monochromatic blue. It must contain $x$, and all its other vertices were already blue. Consequently

$$F_x\cap A=\{x\},\qquad C_x:=F_x\setminus\{x\}\subseteq B, \qquad \lvert C_x\rvert=r-1.$$

Choose one such $F_x$ for each $x\in A$. These choices are finite. Distinct values of $x$ give distinct edges, because their intersections with $A$ are different singletons. Also $F_x\ne A$.

Now fix any $b\in B$ and change just $b$ from blue to red. The edge $A$ stays red. Non-2-colorability of $H$ alone would say nothing useful about this change: $A$ is already monochromatic. The relevant reason is instead the minimum defining $B$. If the new coloring were proper on $H\setminus\{A\}$, it would be an admissible coloring with one fewer blue vertex. Therefore a new red monochromatic edge $G_b$ appears in $H\setminus\{A\}$. It contains $b$, and all its other vertices were originally red. In particular,

$$G_b\cap B=\{b\},\qquad G_b\ne A, \qquad \lvert G_b\cap A\rvert\le t.$$

For every $x\in A\setminus G_b$, the edges $F_x$ and $G_b$ are distinct and meet. Their common vertex cannot be $x$. Every other vertex of $F_x$ is originally blue, whereas $G_b$ has only the one originally blue vertex $b$. Thus their intersection is exactly $\{b\}$, and $b\in C_x$. It follows that the integer

$$d(b):=\lvert\{x\in A:b\in C_x\}\rvert \ge r-\lvert A\cap G_b\rvert\ge r-t.$$

This holds for every blue vertex. It is valid for any fixed choices of the $F_x$ and any allowed choice of each $G_b$; it does not require coordinated choices or a common $G_b$. The possible equality $F_x=G_b$ when $r=2$ causes no problem: the comparison just made only uses $x\notin G_b$, which already rules out that equality.

3. Double counting, integer rounding, and the linear consequence

Fix $x\in A$. For $y\ne x$, a common vertex of $F_x$ and $F_y$ must be originally blue: their only originally red vertices are the different vertices $x$ and $y$. Hence the following is an exact count of the pairs $(y,b)$ with $y\in A\setminus\{x\}$ and $b\in F_x\cap F_y$:

$$\sum_{y\in A\setminus\{x\}}\lvert F_x\cap F_y\rvert =\sum_{b\in C_x}(d(b)-1).$$

The subtraction of 1 removes the already counted membership of $b$ in $C_x$. Each summand on the left is at most $t$, since $F_x\ne F_y$. There are $r-1$ such summands. Each of the $r-1$ summands on the right is at least $r-t-1$ by the incidence lower bound. Therefore

$$(r-1)t\ \ge\ (r-1)(r-t-1).$$

Because $r-1>0$, cancellation is valid, and $2t\ge r-1$. Since $t$ is an integer,

$$t\ge\left\lceil\frac{r-1}{2}\right\rceil =\left\lfloor\frac r2\right\rfloor.$$

The distinct pair attaining $t$ belongs to $H\subseteq\mathcal E$, so it is the required pair in $G$. The argument neither substitutes a supremum for an attained value nor silently includes an edge paired with itself.

The all-order linear factor is checked in the ratio actually claimed. For even $r=2k$, the displayed lower bound divided by $r$ is $1/2$. For odd $r=2k+1\ge3$, it is $k/(2k+1)\ge1/3$, because $k\ge1$. Thus $\lfloor r/2\rfloor/r\ge1/3$ for every integer $r\ge2$, with equality in this numerical comparison at $r=3$. This proves the claimed linear scale with an absolute constant, not merely for large $r$. It does not turn the integer bound into $\lvert E\cap F\rvert\ge r/2$ when $r$ is odd.

Strongest attack and exact witnesses

The blue recoloring cannot follow from noncolorability alone

The strongest attempted proof attack targeted the existence of $G_b$. After changing a blue vertex to red, $A$ is still red, so a careless appeal to noncolorability cannot force any other bad edge. The following exact finite witness demonstrates this obstruction even without isolated vertices.

On $V=\{1,2,3,4,5,6,7\}$ take the seven triples

$$123,\quad145,\quad167,\quad246,\quad257,\quad347,\quad356,$$

where, for example, $145$ denotes $\{1,4,5\}$. Each vertex occurs in three triples, and every pair of vertices occurs in exactly one triple. These facts can be checked directly from the seven displayed sets. No two triples share two vertices. For a fixed triple, its three vertices have six incidences with the other six triples, at most one in each; therefore it meets every other triple in exactly one vertex. In particular $t=1$.

This hypergraph is not 2-colorable. In a proper red-blue coloring each triple would contribute exactly two pairs of differently colored vertices. Pair uniqueness would then require 14 such pairs. If there are $a$ red vertices, there are only $a(7-a)\le12$ such pairs, a contradiction. Deleting any edge does give a proper 2-coloring: color that edge red and its complement blue. Every remaining edge has one red vertex and two blue vertices. Thus this is also an edge-minimal example of the kind used by the frozen argument.

Choose $A=123$ and initially color exactly $\{4,5,6,7\}$ blue. The coloring is proper on $H\setminus\{A\}$, and $A$ is red. Changing vertex 4 to red keeps every other edge bichromatic. In the original coloring no edge has 4 as its only blue vertex. Thus the asserted $G_4$ does not exist for this admissible but nonminimal coloring.

This attacks an omitted-minimum shortcut, not the actual frozen hypothesis. The coloring with blue set $B=\{5,6,7\}$ is proper on $H\setminus\{A\}$ and has fewer blue vertices. It is minimal: at most two blue vertices meet at most five of the seven triples, since each vertex lies in three triples and any two vertices share exactly one triple. At least two all-red triples would remain, including one other than $A$.

For this minimizing coloring the recoloring witnesses can be chosen as

$$F_1=167,\quad F_2=257,\quad F_3=356, \qquad G_5=145,\quad G_6=246,\quad G_7=347.$$

Every $b\in B$ occurs in exactly two of the $C_x$, so $d(b)=2=r-t$. The double-count bounds are both equal to 2, and $2t=r-1=2$. The attempt to refute the recorded argument therefore fails precisely because its minimum-blue condition supplies the missing force. That condition is present in the frozen paragraph, not added by this review. The same witness also refutes an attempted strengthening to $2t\ge r$ or intersection at least $r/2$ for odd $r$.

Endpoint, distinctness, and infinite-domain attacks

At $r=2$, the three edges $12,13,23$ on three vertices form an intersecting non-2-colorable hypergraph with $t=1$. Noncolorability follows because two of three vertices must receive the same one of two colors, and their pair is an edge. Here the target bound is exactly 1. In the general derivation, $r-1=1$ remains a legal divisor, recoloring one vertex of $A$ really makes $A$ bichromatic, and the integer rounding still gives the correct bound. The seven-triple witness supplies the odd endpoint $r=3$ and the lost unit in $r-1$. These finite examples test boundary inferences; the universal statement rests on the symbolic derivation, not extrapolation from examples.

The distinctness attack was to allow a repeated $F_x$ or to use the self-pair in defining $t$. The singleton traces $F_x\cap A=\{x\}$ prohibit the first, and the explicit distinct-pair maximum prohibits the second. The only other intersection bounded by $t$ is $A\cap G_b$, and those edges are distinct because $G_b$ contains the blue vertex $b$. No estimate uses $t$ to bound an unverified distinct pair.

For infinite hypergraphs, minimizing the cardinality of an infinite blue set would not justify a strict improvement after changing one vertex: an infinite cardinality can stay unchanged. The frozen argument avoids that failure by reducing to a finite non-2-colorable subhypergraph first. The foundational derivation above supplies exactly that interface for arbitrary vertex sets. Thereafter every minimum and count is finite. Extra isolated vertices have no effect on the edge witnesses. No countable-enumeration assumption or finite experimental evidence is substituted for this compactness step.

The literal 3-colorability formulation has counterexamples

For each integer $r\ge2$, take a center $c$ and disjoint sets $X,Y$ of $r-1$ vertices each, neither containing $c$. Let the entire edge family be

$$E=\{c\}\cup X,\qquad F=\{c\}\cup Y.$$

This finite hypergraph is $r$-uniform, intersecting, and has every vertex in an edge. Its only distinct pair has intersection $\{c\}$. It has a proper 2-coloring, with $c$ red and all other vertices blue, and also a proper 3-coloring using all three colors: color $c$ with 1, $X$ with 2, and $Y$ with 3. Its chromatic number is 2, not 3.

At $r=4$, its intersection size 1 is less than $\lfloor r/2\rfloor=2$. For any proposed absolute constant $\gamma>0$, choosing an integer $r>\max\{2,1/\gamma\}$ gives $1<\gamma r$. Thus no positive uniform linear bound follows from mere 3-colorability, even when there are two distinct edges and no isolated vertices. This independently establishes the precise obstruction described in the frozen argument's star caveat.

There is also a separate distinctness obstruction: a hypergraph with one $r$-element edge is properly 3-colorable but has no pair of distinct edges at all. This is not needed for the quantitative counterexample, which has two edges, but prevents treating distinct-edge existence as automatic under the weakened hypothesis.

Correspondence to the displayed questions

The frozen Covers paragraph identifies the modified formulation and is consistent with the proof. Its affirmative consequence must not be extended to the literal weak parenthetical reading of the first formulation. The frozen paragraph already records this qualification. Nothing here decides the first, vertex-count question in either formulation.

Premises, reconstruction, and actual reading

Dependency reading at verdict filing, 2026-10-14 UTC: the complete supplied target and argument extracts, including their consequence and dependency sentences, were inspected for actual premise consumption. No native L-claim is cited or needed by this derivation. No live native dependency standing or ledger was consumed. There is no pending native premise grade or batch acceptance order on which this verdict rests.

The assumptions are finite integer $r\ge2$, uniformity, pairwise intersection, absence of a proper vertex 2-coloring, and ordinary classical foundations with choice for the infinitary compactness argument. The non-2-colorability hypothesis is assumed as part of the conditional theorem; this review does not decide colorability for arbitrary inputs.

The reconstruction is disclosed rather than attributed to the supplied paragraph. It expands the invoked foundational compactness step, the finite minimum, the two different reasons the recolorings fail, the incidence count, and the integer and constant calculations. It also makes the distinct-pair convention for $t$ explicit. These use the frozen argument's same sequence $H,A,F_x,G_b,t$ and its displayed double count. No substantive new hypergraph lemma or alternative proof route replaces a missing link. The strongest attack above explains why the already stated minimum-blue condition matters.

The extract's assertion that the note proves its compactness lemma is not independently verified as a statement about that note. Neither the note nor the problem page's Notes section was supplied or read. Likewise the historical attribution of an earlier $r/\log r$ bound is context, not a premise used to prove this result, and has not been checked against external literature. The linear consequence here is established directly. As a comparison of the written scales alone, $(r/3)/(r/\log r)=\log r/3$ grows without bound; this checks the claimed improvement in scale, not the historical theorem or its attribution. No source-fidelity or external-source proof-coverage conclusion is claimed.

Checklist

The following tables cover both the Erdos-specific checklist and every shared canonical failure mode and named pattern in docs/verification.md. They assess the precise non-2-colorable proposition and separately record the false weak-parenthetical extension.

Erdos-specific checklist

Shared canonical failure modes

Shared named patterns

The neighboring correlated-sample discipline is inapplicable because there is no statistical inference. Evidence coverage is as stated: a general finite derivation, its explicit foundational extension to infinite hypergraphs, and exact witnesses for the convention and inference attacks. The examination is neither a computational batch nor a second review cycle; no claim of multiple independent cycles is made.

Verdict and grading

Refutation-failed for the complete frozen non-2-colorable linear-intersection proposition, in ordinary classical foundations, and for its application to the modified second question and to the first formulation when chromatic number 3 has its standard exact meaning. The recorded proof's specific recoloring-and-counting route survives the attacks above, with the disclosed foundational and elementary inference expansions. The stated bound is $\lfloor r/2\rfloor$ and gives the uniform factor $1/3$ for all $r\ge2$.

Refuted-as-stated for the extension obtained by treating the first formulation's parenthetical, mere proper 3-colorability, as the entire hypothesis. The displayed two-edge family is a counterexample to a uniform linear bound; at $r=4$ it also contradicts the displayed floor bound. This does not refute the frozen theorem, whose non-2-colorability hypothesis excludes the witness, and agrees with the precise limitation the frozen argument records.

No verdict covers the vertex-count question, any unprovided external document or proof, historical attribution, a kernel formalization, or subsequent changes to the subject. No native dependency standing was consumed. No substantive missing mathematical interface remains for the scoped theorem under the foundations stated here.

This report does not grade itself, assert a tier, or change standing. A grader distinct from the author and reviewer must assess its independence, complete contract, reconstructed inferences, coverage, and verdict before any acceptance decision.