How statement and proof provenance work
The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.
- Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
- AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
- AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.
These labels describe origin, not correctness: citations and verification chips remain separate evidence.
Comb Structure in co-E-Free Graphs
1 · Prerequisites
- Absolute and Conditional Convergence; Rearrangement; Products
- Binary Operations, Monoids, Groups and Subgroups
- Blockades, Combs and Pattern Graphs
- Cographs, Perfect Patterns and Pure Pairs
- Compactness in Metric Spaces
- Completeness, Completion, and Uniform Continuity
- Construction of the Natural Numbers
- Construction of the Real Numbers via Cauchy Sequences
- Construction of the Real Numbers via Dedekind Cuts
- Continuity, IVT, EVT, and Uniform Continuity
- Countability and Uncountability
- Finite Counting, Factorials and Binomial Coefficients
- Finite Probability Spaces and Random Variables
- Foundations of the Real Numbers for Analysis
- Graph Colouring
- Graphs, Walks and Connectivity
- Inclusion–Exclusion, the Pigeonhole Principle and Double Counting
- Induced Subgraphs and Hereditary Graph Classes
- Iterative Sparsification and the Five-Vertex Path
- Limits of Real Functions
- limsup, liminf, and Subsequential Limits
- Metric Spaces
- Modules, Substitution and Prime Graphs
- Monotone Functions, Discontinuities, and Continuity Sets
- Monotone Sequences, Bolzano-Weierstrass, and Cauchy Completeness
- Order, Zorn's Lemma, and the Axiom of Choice
- Polynomial Rödl, Virality and Erdős–Hajnal Equivalence
- Power Series and Real-Analytic Functions
- Properties of the Integral and the Working FTC
- Property (*) and Comb Outcomes
- Quotient Blockades and Mixing Relations
- Ramsey Theory
- Regular Pairs and Induced Counting
- Relations, Functions, and Quotients
- Rings, Subrings, Integral Domains and Fields
- Roots, Rational Powers, and Classical Inequalities
- Sequences and Limits
- Sequences and Series of Functions; Uniform Convergence
- Series: Convergence and the Nonnegative Tests
- Small-Graph Erdős-Hajnal Consequences
- Sparse Restricted Subgraphs and the Rödl–Nikiforov Theorems
- Suprema and Infima
- The Derivative and the Mean Value Theorems
- The Erdős–Hajnal Property and Homogeneous Sets
- The Exponential Function
- The Five-Cycle and the Erdős-Hajnal Property
- The Logarithm and General Powers
- The Riemann Integral: Definition and Integrability
- The Structural Criterion for Property (*)
- The ZFC Axioms and the Basic Set Constructions
- Topology of ℝ
- Vector Spaces, Linear Subspaces, Span and Direct Sums
2 · Summary
This page develops the overlap-quotient proof of the special-vertex co- comb partition and the resulting local route to property for .
3 · Logical flowchart
4 · Definitions, theorems and proofs
The family consisting of and co- has the Erdős–Hajnal property
Statement
The finite forbidden family has the Erdős–Hajnal property.
Facts & Assumptions
Given: The graphs and the graph co-.
The graph has the Erdős–Hajnal property (The graph has the Erdős-Hajnal property).
The graph has the Erdős–Hajnal property (The five-vertex path and its complement have the Erdős-Hajnal property).
The Erdős–Hajnal property passes to a hereditary subclass (The Erdős–Hajnal property and each of its constants pass to hereditary subclasses).
Huang--Ju--Zhou, Corollary 1.8, states the following leaf/co-leaf transfer. Let be a finite family, let have a leaf , and let have a co-leaf . If both families obtained from by replacing, respectively, by and by have the Erdős--Hajnal property, then has the Erdős--Hajnal property.
Proof
The class of -free graphs is a hereditary subclass of the class of -free graphs, and the class of -free graphs is a hereditary subclass of the class of -free graphs. Thus [F1]--[F3] give the Erdős–Hajnal property for both families, for every .
Fix and suppose that has the property. In , deleting the leaf of gives . The vertex is a leaf of , so it is a co-leaf of co-, and deleting it from co- leaves . Hence the two modified families in [F4] are exactly and .
Step 1.2, the induction hypothesis, and the second base family from step 1.1 let [F4] yield the property for .
Starting with and repeating step 2.1 through proves the property for .
The special-vertex-local structural-partition criterion implies property (*)
Statement
Let have a common Erdős–Hajnal constant . Suppose that, in every -free graph, every special-vertex comb occurring in the definition of property has a partition satisfying clauses (1), (2.1)--(2.3) of the structural comb partition. Then has property .
Facts & Assumptions
Given: The finite graph families and common constant in the Statement, and the supplied partition for each special-vertex comb in the property- trigger. For that comb, write . The local clauses mean that is -free, partitions into nonempty blocks forming a pure blockade with -free pattern, and each vertex in another is pure to each . These are the partition clauses of The structural comb-partition hypothesis; its universal assertion about all combs is not assumed.
A common Erdős–Hajnal constant supplies a clique or stable set of size at least in each nonempty -vertex -free or -free graph (The Erdős–Hajnal property and an Erdős–Hajnal constant for a hereditary graph class). Induced subgraphs of a family-free graph remain family-free (-free and -free graphs under the induced-subgraph convention).
A pure blockade has pairwise complete or anticomplete blocks; its pattern records precisely the complete pairs (Complete, anticomplete, pure, weakly sparse, and -sparse blockades, The pattern graph of a pure blockade). Blockades have disjoint nonempty blocks and the stated lower bounds on length and width (Blockades, their length, their width, and their support).
Integral geometric layers use the cutoff and consecutive blocks through the first cutoff attaining (Integral geometric layers of a decreasing block partition).
Positive real powers satisfy the product and iterated-power laws (The exponent, product, quotient, and iterated-power laws for positive real bases and real exponents); monotonicity follows from their exponential-logarithm definition (Real powers for positive bases, with the zero-base positive-exponent convention, Order, continuity, range, and the product, quotient, and reciprocal laws for the natural logarithm, The exponential function is strictly increasing).
The geometric series with ratio has sum (For , , and for the series diverges).
Proof
Set and . Fix an arbitrary -free finite graph and a special-vertex -comb from Property (*) for a finite graph family, with integral and real . Use its supplied local partition. Suppose that all three property- outcomes with these constants fail.
If for some , then is nonempty and [F1] supplies a clique or stable set of size at least , since . This contradicts the first failure. Hence every , and implies .
If every partition has a block of size at least , choose one for each of the finitely many indices . Fix distinct . Each vertex of is complete or anticomplete to by the local external-purity clause. Two vertices of with opposite relations would make any vertex of the nonempty mixed on , contrary to the same clause with reversed. Thus are pure. The disjoint sequence is consequently a pure blockade of width at least , contradicting the third failure.
By step 2.2 there is an index such that every , hence is at most this bound. Put , , and reorder these blocks as in nonincreasing size. Reordering preserves purity and changes the pattern only by relabelling. Since , we have . The reordered pattern is still -free.
Form the cutoffs of [F3]. They reach : for example, for the positive integer , the latter elementary inequality following by induction. Let be the first index with . Since , we have . For , the integer is at most and at most , so . Thus all layers are nonempty and partition the blocks in order.
For , put . Then , so , yielding . Also and each contains at most blocks, including when .
Suppose a preterminal layer , , has every block of size at least . The first blocks all have at least that size by their nonincreasing order. Their induced pattern is nonempty and -free, so [F1] gives a pattern clique or stable set of integral cardinality . By [F2], the blocks indexed by form a complete or anticomplete blockade of length and width at least .
Since and , this blockade has width at least and satisfies the second property- outcome. That contradicts step 1.1. Therefore every preterminal contains a block of size strictly less than .
The first layer contributes at most vertices. For , every block in follows the small block in and has size less than . Hence contributes less than .
Because and , the sum of the latter bounds is at most . All layers have been counted, so , contradicting step 2.1.
Thus one of the three outcomes holds for every special-vertex comb required by Property (*) for a finite graph family, with constants independent of and the comb. This proves that has property .
Relative to a complete nonedge pair in a co--free graph, a one-sided vertex mixed on an induced path avoids two consecutive nonneighbours and three consecutive neighbours
Statement
Let be co--free, let be an induced path, and let distinct vertices be nonadjacent and complete to . If is mixed on , then has neither two consecutive nonneighbours nor three consecutive neighbours on .
Facts & Assumptions
Given: as in the Statement.
In co-, adjacency is the complement of the five-path-with-middle-leaf edge set defining (The -graph and co-).
A path has distinct vertices and its listed consecutive edges, while an induced copy preserves both adjacency and nonadjacency. Hence an induced path has precisely its consecutive path edges among its own vertices (Walks, closed walks, trails, paths and cycles, with length equal to the number of traversed edges, Induced embeddings and induced copies of a graph).
Proof
Suppose two consecutive vertices of are nonneighbours of . Travelling from a neighbour of on to the first such consecutive pair and taking the first change gives an induced subpath with an edge and nonedges.
If instead has three consecutive neighbours, reverse if needed and take the last such run before an adjacency change. There is an induced subpath with edges and a nonedge.
On the nonedges are exactly the -edges under : they are . Thus this induced subgraph is co-, contrary to [F1].
On the nonedges are exactly the -edges under : they are . This is an induced co-, again a contradiction.
Both assumed runs are impossible, proving the two assertions.
Relative to a complete nonedge pair in a co--free graph, every one-sided vertex is pure to an induced
Statement
Let be co--free. If nonadjacent are complete to an induced copy of and , then is complete or anticomplete to that copy of .
Facts & Assumptions
Given: and a labeled induced as in the Statement.
The labeled has rim , hub complete to the rim, and leaves adjacent only to (The graphs ).
On any induced path to which is mixed and whose exterior vertices are , the preceding path-run lemma forbids two consecutive nonneighbours and three consecutive neighbours (Relative to a complete nonedge pair in a co--free graph, a one-sided vertex mixed on an induced path avoids two consecutive nonneighbours and three consecutive neighbours).
Proof
Suppose is mixed on the . If is complete to the rim and some is a nonedge, then is mixed on the induced path and has three consecutive neighbours there, contrary to [F2]. Thus is adjacent to every . If failed, induces co-; hence holds and is complete to , a contradiction.
If is anticomplete to the rim and some is an edge, then is mixed on with two consecutive nonneighbours, contrary to [F2]. Thus every is a nonedge. If were an edge, then would be mixed on with two consecutive nonneighbours, again contrary to [F2]. Hence is a nonedge, so is anticomplete to , also a contradiction.
It remains that is mixed on the rim. Any cyclic run of two rim nonneighbours or three rim neighbours, together with a vertex of the opposite adjacency supplied by mixedness, lies in an induced rim subpath to which [F2] applies. Thus the two run restrictions force, up to cyclic relabeling, . If were a nonedge, then would be mixed on with two consecutive nonneighbours; if were a nonedge, then would be mixed on with three consecutive neighbours. Hence [F2] gives ; then induces co-, impossible.
Every possible rim relation contradicts mixedness, so is pure to the induced .
The -overlap-chain relation in one comb block
Definition
Fix a comb block . Let be the set of vertices of that lie in an induced . For , write when there is a finite sequence in such that each consecutive pair lies in one induced copy of contained in .
This is the -overlap-chain relation. Its equivalence classes are the -overlap classes of . The relation is reflexive (the length-one sequence), symmetric (reverse a chain), and transitive (concatenate chains).
Every -overlap class is connected
Statement
Every -overlap class induces a connected graph.
Facts & Assumptions
Given: An -overlap class in one comb block.
Two vertices in are joined by a finite chain of induced copies with successive copies sharing a vertex (The -overlap-chain relation in one comb block).
The graph is connected (The graphs ).
Proof
Let . By [F1], choose an overlap chain from a copy containing to a copy containing . Within each copy, [F2] gives paths from its entering vertex to its shared vertex and then to its exiting vertex.
Concatenating these paths at the shared vertices is a walk in from to , and deleting repetitions gives a path. Thus every two vertices of are connected.
Hence is connected.
Purity on every induced propagates along an -overlap class
Statement
Let be an -overlap class and let . If is pure to every induced contained in , then is pure to .
Facts & Assumptions
Given: as in the Statement.
A chain of induced copies links the copies meeting any two vertices of (The -overlap-chain relation in one comb block).
A vertex pure to a nonempty set is either complete or anticomplete to it (Edges between disjoint vertex sets; complete, anticomplete, pure and mixed pairs).
Proof
Two consecutive copies in an overlap chain share a vertex. By [F2], cannot be complete to one and anticomplete to the other, since their shared vertex would then be both adjacent and nonadjacent to .
Thus the complete/anticomplete label is constant along every overlap chain. By [F1], every vertex of lies in a copy reached from any fixed copy, so all vertices of receive one label.
Hence is complete or anticomplete to , as claimed.
The -overlap blockade and its iterated mixed quotients
Definition
For a comb block , suppose and let be the ordered blockade whose blocks are the -overlap classes in , ordered by their least vertex in a fixed ordering of . Having defined , put where is its mixed-block reachability relation. These are the iterated mixed quotients of the -overlap blockade.
Iterated mixed quotients of an -overlap blockade terminate at a pure blockade
Statement
For a nonempty -overlap blockade, some iterated mixed quotient is a pure blockade.
Facts & Assumptions
Given: A nonempty initial overlap blockade .
Its blocks form a finite nonempty sequence of nonempty sets (Blockades, their length, their width, and their support).
The quotient blocks are the equivalence-class unions of mixed-block reachability (The quotient blockade obtained from mixed-block reachability).
Proof
If is not pure, two distinct blocks are mixed. They lie in one mixed-reachability class, so [F2] merges at least two blocks and strictly decreases the positive integer number of blocks.
Suppose no iterate were pure. Step 1.1 would give an infinite strictly decreasing sequence of positive integers, the successive numbers of blocks.
The set of values of that sequence has a least element by well-ordering, but its successor in the sequence is smaller, a contradiction. Therefore a first pure iterate exists.
A vertex mixed on a connected set has opposite adjacency on some edge of that set
Statement
If induces a connected graph and is mixed on , then some edge of has exactly one endpoint adjacent to .
Facts & Assumptions
Given: A connected set and a vertex mixed on it.
Mixedness supplies a neighbour and a nonneighbour of in (Edges between disjoint vertex sets; complete, anticomplete, pure and mixed pairs).
Connected vertices are joined by a path in the induced graph (Connected graphs and connected components defined by the existence of vertex paths).
Proof
Choose with an edge and a nonedge by [F1], and choose an -- path in by [F2].
Along this finite path, the adjacency indicator to begins at and ends at , so it first changes across one consecutive pair. That pair is an edge of with opposite adjacencies to .
This is the required mixed edge.
In a special-vertex comb of a co--free graph, vertices in other comb blocks remain pure to every -overlap quotient block
Statement
Let be co--free and let be a comb with an outside vertex complete to all and anticomplete to all . Fix , form the nonempty -overlap blockade in , and form its iterated mixed quotients. Every vertex of is pure to every block of every iterate.
Facts & Assumptions
Given: The special-vertex comb, an index , and its iterated overlap quotients.
Relative to any nonadjacent pair complete to an induced in a co--free graph, every one-sided vertex is pure to that ; in particular, with , every external comb-block vertex is pure to every induced in (Relative to a complete nonedge pair in a co--free graph, every one-sided vertex is pure to an induced ).
Purity on every propagates to its overlap class (Purity on every induced propagates along an -overlap class).
For a blockade of connected blocks, suppose distinct mixed quotient blocks have outside vertices with a nonedge, complete to , and complete to and anticomplete to . If no vertex of is mixed on , there are mixed member blocks inside with an outside triple satisfying the same adjacency conditions (A quotient-level mixed-block witness descends to two mixed member blocks).
If an outside vertex is mixed on a connected set, it has opposite adjacency to the endpoints of some edge of that set (A vertex mixed on a connected set has opposite adjacency on some edge of that set).
In a co--free graph, if nonadjacent outside vertices are complete to an induced path , a vertex mixed on cannot have two consecutive nonneighbours on (Relative to a complete nonedge pair in a co--free graph, a one-sided vertex mixed on an induced path avoids two consecutive nonneighbours and three consecutive neighbours).
Initial overlap classes are connected, and taking a mixed quotient preserves connectedness of blocks (Every -overlap class is connected, A quotient block of connected or anticonnected blocks is again connected or anticonnected).
Each next iterate replaces mixed-reachability classes of blocks by their unions (The -overlap blockade and its iterated mixed quotients).
Proof
Write for iterate . For any external comb-block vertex , the comb and special-vertex hypotheses give , with nonadjacent and complete to . Thus [F1] makes pure to every induced in , and [F2] makes it pure to each block of .
All blocks of every are connected: start with the initial classes and repeatedly apply connectedness preservation in [F6].
Fix and assume the assertion for . Suppose an external vertex is mixed on a block of . By the induction hypothesis, each member block of inside is complete or anticomplete to , and both labels occur. By [F7], a mixed chain inside joins blocks of opposite labels; at a change of label, consecutive mixed blocks have complete to and anticomplete to .
Consider any mixed blocks at level with an outside triple satisfying: is a nonedge, are complete to , and is complete to and anticomplete to . No vertex is mixed on . Indeed, if one were, connectedness and [F4] give an edge in with an edge and a nonedge. Then is induced, are outside and complete to it, and is mixed on it with consecutive nonneighbours , contrary to [F5].
The pair in step 1.3 has the required triple at level . Whenever its current level exceeds one, apply [F3] to : step 1.2 supplies connected member blocks and step 2.1 supplies the directional no-mixed-vertex hypothesis. The resulting mixed blocks at level have an outside triple with all the same adjacency conditions. Repeating this finite descent reaches mixed initial classes and an outside triple with a nonedge, complete to , and complete to and anticomplete to . If , the initial pair already has these properties.
Apply step 2.1 to . Every vertex of is pure to . Since the pair is mixed, some vertex is complete to and some vertex is anticomplete to ; otherwise all vertices have the same label and the pair is pure. Therefore any is adjacent to and nonadjacent to , and is mixed on . Such a vertex exists because a blockade block is nonempty.
The vertices are outside , nonadjacent, and both complete to . Also . Apply [F1] with to every induced contained in , then [F2] to the overlap class . It follows that is pure to , contradicting step 4.1. Thus no external vertex is mixed on any block at level . Together with the base case this proves the assertion for every iterate.
The pattern of the terminal -overlap quotient is -free
Statement
In a co--free graph, the pattern graph of a terminal pure iterated -overlap quotient is -free.
Facts & Assumptions
Given: A terminal pure quotient blockade in a co--free graph.
A pattern edge means its two nonempty blocks are complete; a pattern nonedge means they are anticomplete (The pattern graph of a pure blockade).
Every initial induced lies wholly in one initial overlap class, and quotienting only merges blocks (The -overlap blockade and its iterated mixed quotients).
Proof
Suppose the pattern contains an induced , and choose one vertex from each of its eleven corresponding nonempty blocks. By [F1], the selected vertices induce in the ambient graph.
If the pattern contains an induced co-, selecting one vertex from each of its six blocks and using [F1] similarly induces co- in the ambient graph, contradicting co--freeness.
The eleven selected vertices lie in distinct terminal blocks. But [F2] says the vertices of every induced must already lie in one initial overlap class and hence in one terminal block, a contradiction.
Neither forbidden induced graph occurs in the pattern.
A special-vertex comb in a co--free graph admits the structural partition
Statement
Let be co--free and let it contain an -comb and an outside vertex complete to all blocks and anticomplete to all teeth. For every , there is a partition such that is -free, and has a nonempty-block pure blockade partition whose pattern is -free and whose every block is pure to every vertex of .
Facts & Assumptions
Given: The co--free special-vertex comb of the Statement.
The terminal overlap quotient is pure and has -free pattern (Iterated mixed quotients of an -overlap blockade terminate at a pure blockade, The pattern of the terminal -overlap quotient is -free).
Every external comb-block vertex is pure to every terminal overlap quotient block (In a special-vertex comb of a co--free graph, vertices in other comb blocks remain pure to every -overlap quotient block).
A singleton sequence is a pure blockade with one-vertex pattern, and induced subgraphs of a co--free graph are co--free (Edges between disjoint vertex sets; complete, anticomplete, pure and mixed pairs, -free and -free graphs under the induced-subgraph convention).
Proof
Fix . Let be the vertices of contained in an induced , and put . Then is -free by definition and co--free as an induced subgraph, hence it is -free.
Assume-case nonempty: if , set and take its terminal overlap quotient as the partition. Its pure-blockade and pattern clauses are [F1], and its cross-block purity clause is [F2].
Assume-case empty: if , choose , set and . The singleton blockade on is pure, its pattern has one vertex and is forbidden-family-free, and every outside vertex is pure to it; is -free because was.
The two cases produce the required partition for this arbitrary , and therefore for every comb block.
The singleton family has property (*)
Statement
The singleton finite family has property .
Facts & Assumptions
Given: An arbitrary co--free graph and a special-vertex comb required by property .
has the Erdős–Hajnal property (The family consisting of and co- has the Erdős–Hajnal property).
The special-vertex comb has the required partition (A special-vertex comb in a co--free graph admits the structural partition).
The local criterion converts those two facts into property (The special-vertex-local structural-partition criterion implies property (*)).
Proof
For , its complement family is . Thus the given graph is in the setting of [F2].
Take . Fact [F1] supplies their common Erdős–Hajnal constant, and [F2] supplies the local partition for every special-vertex comb in the graph of step 1.1.
Applying [F3] now proves that has property .
5 · Examples, counterexamples and false statements
None yet.
Sources
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Lemma 6.3
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Corollary 1.8
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, proof of Lemma 5.1
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Claim 6.4.1
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Claim 6.4.2
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, proof of Lemma 6.4
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, proof of Claim 6.4.3
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Claim 6.4.3
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Lemma 6.4(2.2)
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Lemma 6.4
- Huang, Ju, and Zhou, Erdős-Hajnal beyond the five-vertex path, Sections 5--6.1