Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Statement
Let be a finite simple rooted graph with nonempty internal set , root set , and positive integers . Suppose
There are an integer and constants such that
This is a lower-bound theorem. No upper bound for any rooted power is assumed. The roots may be adjacent, and need not be bipartite.
Generic obstruction
Put , , and choose , so and . Apply the [[extremal_graph_theory/adamczewski_2026_erdos571/finite_coefficient_obstruction|finite coefficient obstruction]] over and its algebraic closure . This supplies and a nonzero coefficient polynomial over . If , the polynomial bipartite graph over with coefficients contains no .
Every finite extension of embeds in . Evaluation of polynomials commutes with that embedding, and applying the embedding coordinatewise gives an injective edge-preserving map from the graph over into the graph over . Consequently, the same nonvanishing condition excludes over every such . The image of remains a nonzero polynomial and has the same total degree .
A dense specialization
Let . Choose every coefficient of polynomials of total degree at most in variables independently and uniformly from . Their common zero condition defines a bipartite graph on two copies of . There are potential edges.
By interpolation, one potential edge occurs with probability . Two distinct edges have distinct left-right coordinate tuples, even when they share an endpoint. Since , the two evaluations are independent and their joint edge probability is . If counts edges, this gives
Therefore
The finite-field Schwartz–Zippel inequality says that a nonzero polynomial
of total degree , evaluated at independent uniform field elements,
vanishes with probability at most . This is the external polynomial
zero-count bound used here; the formal source invokes Mathlib's
schwartz_zippel_totalDegree.
Choose . Since , we have , and hence . Thus the two bad probabilities above have sum less than one:
Some coefficient tuple therefore satisfies and . Its graph is -free by the obstruction. The same works for every sufficiently large power .
Transfer to every large order
Let be sufficiently large and choose a power of two with
large enough for the previous construction. It gives an -free graph on vertices with at least edges. A uniformly chosen induced subgraph on vertices has expected edge count
Every such subgraph remains -free. For , some choice has at least
edges. This proves the assertion for all sufficiently large . Subsampling, rather than adding isolated vertices, also handles forbidden graphs that have isolated roots.
Source and dependencies
This is Proposition 2.1, p. 2 of the preliminary
exposition, with its essential
omitted deductions supplied by the linked same-source lemmas. The pinned
formal source gives FiniteVariance.exists_dense_outside, lines
3930–3956; PolynomialEdgeVariance.edge_moments, lines 3990–4016;
DenseGenericSamples.exists_dense_avoiding, lines 4120–4165;
GenericFiniteFieldLower.dense_specializations, lines 4242–4289;
TransferLowerBound and DyadicLowerTransfer, lines 4294–4396; and
UnconditionalRootedLower.exists_power_lower, lines 4407–4437.
The finite-field and algebraic-closure existence facts, the elementary finite expectation identities, and the stated Schwartz–Zippel inequality are the external background inputs. The generic fibers, compactness and obstruction arguments are proved on the linked result pages. This route uses no Lang–Weil estimate; its relation to earlier random-algebraic lower-bound proofs is discussed in the source record.
Bears on. #571.