Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Coding refinement
This page proves a stronger sufficient affine-code criterion for the local seven-edge property than the sevenfold-XOR criterion in Certificates and obstructions. It also proves that the stronger criterion still cannot give an asymptotic counterexample in its stated range.
Fano classification extension extends the classification and sufficient threefold-XOR criterion to , using Füredi's fractional matching theorem. The polynomial obstruction below retains its own narrower entropy hypotheses.
Scope. These are obstructions to particular sufficient conditions. They do not prove the conjectured upper bound for arbitrary affine-code families satisfying the local property, or for arbitrary set families. No counterexample to Problem 644 is obtained.
Comparison with the independent input. Sections 7.2–7.5 of first independent record have been checked. Its sevenfold-XOR lemma is valid, but the exact obstruction in Certificates and obstructions already forces under that criterion, even for specially structured codes; the independent random-code entropy diagnostic is therefore weaker. Its conditional Fano-line XOR estimate is also correct. The missing classification is supplied below for , and the later polynomial argument excludes an asymptotic improvement by the refined distance criterion throughout that range. Neither conclusion extends to arbitrary coding criteria or asserts that the classification holds for larger deficiency.
The same comparison applies to the checked §§6.3–6.4 of second independent record. Its condition that the linear code contain a word of weight exactly is sufficient for the stated covering loss, but existence of that particular word is not needed by the forbidden-interval argument below. The later nonlinear forbidden-distance theorem removes affine closure from this obstruction as well. The record's robust-classification gap is therefore filled only in the displayed narrow-deficiency range, while its weaker sevenfold-XOR route is excluded outright.
Exact Fano structure sufficiently close to the threshold
Comparison with the third record. Its §9 correctly leaves the code construction unproved, but the present notes have progressed beyond its small-deficiency discussion. The checked fractional-matching argument classifies all degree-four spectra whenever , equivalently . The threefold-XOR bound then holds throughout that range. This does not extend the separate polynomial/entropy obstruction below beyond its stated range. The nonlinear forbidden-distance argument already rules out that sufficient distance criterion in the latter range, more strongly than the input's heuristic random-code entropy comparison. Distinct-seven XOR separately excludes a total-XOR avoidance criterion for . None rules out every code-defined family satisfying .
Let be -subsets of an -point set, with no two-point transversal. Suppose
The degree of a point is its number of containing sets, and its spectrum is the corresponding subset of . Every degree is at most four: a point of degree at least five, together with a point in the intersection of the at most two remaining sets, would be a transversal. Here guarantees that any two sets intersect.
Put
In particular . Call the degree-four points standard and let be the other points. Then
We prove that the spectra of the standard points are exactly the seven complements of the lines of a Fano plane on , with every one of these spectra occurring. This adapts the good-pair argument in Claims 2–5 of Section 2 of the supplied 1999 paper; all estimates needed here are proved below.
Write and . For every ,
Hence . Choose a maximum pair, say . The number of points outside its union is at least
so there is a standard point outside . Its missing spectrum has the form . Necessarily
because a point of this triple intersection would complete a two-point transversal.
If another standard point outside had missing triple , , then both and would avoid . Consequently
a contradiction. Thus all standard points outside the union miss . Together with (4), this gives
and therefore
Using (3) and (5), every pair satisfies
For any pair , its outside region consequently has size at least
It contains a standard point. If two standard points there had different missing triples and , then the same argument as (4) would give
where the strict inequality is exactly .
It follows that every pair of indices is in exactly one missing triple of a standard point. These triples form a Steiner triple system on seven points, hence the Fano plane. In particular all its seven lines occur as missing triples, as claimed.
A threefold-XOR criterion
Continue with the preceding bad seven-tuple, and let be its Fano plane on the row indices. A degree-three spectrum cannot be a line : the complementary cap occurs as the spectrum of a standard point, and these two points would cover all seven rows.
For each line , define
For a point of degree , the number of containing it is the number of Fano lines meeting its spectrum oddly. The possibilities are
| Spectrum | Degree | Number of odd line intersections |
|---|---|---|
| Empty | 0 | 0 |
| Singleton | 1 | 3 |
| Pair | 2 | 4 |
| Nonline triple | 3 | 3 |
| Fano cap | 4 | 0 |
The table follows from the facts that each Fano point is on three lines and each pair determines one line. For a nonline triple, its three pair-lines are distinct; they have even intersection, while three other lines have singleton intersection. The excluded line triple is the only additional degree-three possibility.
In every row of the table, the last entry is at most . Therefore
Now let be an affine subspace, and let
If a -uniform family has all its incidence vectors in , then the XOR of any three members also belongs to . Thus (7) proves the following sufficient condition:
If , the local property already follows from the degree argument, without any condition on .
Large transversal number forces a forbidden weight interval
The next observations require no local two-point property. Suppose is -uniform and , with an integer. Set
Every -point set contains an edge, so . More generally, every set of at least points contains an edge.
If has weight , choose supported inside the support of . Then and . Hence
Also, counting the -sets containing each edge gives
Polynomial bound for one forbidden weight
Suppose a binary linear space has no word of weight , where is prime and . Then
Here is a self-contained polynomial proof. Fix a weight layer
For distinct , one cannot have . Otherwise would be a positive multiple of , and would force it to equal , contrary to the hypothesis.
Over , associate to each the polynomial
On incidence vectors of , this polynomial is one at and zero at every other member. Multilinearizing does not change its values on binary vectors and leaves degree at most . These polynomials are linearly independent, so
Summing over the layers proves (11).
An asymptotic obstruction valid throughout the range of (8)
Let . Uniformly over affine spaces and -uniform families as above, if
then
To prove this, suppose along a sequence that for a fixed . By (9), forbids every weight from to . The prime number theorem supplies a prime
with for large . Thus weight is forbidden. Also , so .
Write
and use natural-log entropy
Here
By (10), Stirling's formula gives
By (11) and the usual binomial-sum entropy bound,
There is a uniform positive gap between the main terms of (13) and (14). Indeed, is increasing in , is increasing in , and is increasing separately in when . Consequently the gap is at least
For completeness, this is strictly positive without relying on a numerical calculation. Since ,
The first follows by bounding on ; the second follows from . Furthermore,
Thus (15) is at least
using and . This contradicts (13)–(14) and proves (12).
Finally, suppose a sequence of families satisfies the sufficient criterion (8), but has for some fixed . If , the universal bound already gives a contradiction for large . Otherwise take
Then , while (8) gives . Hence
contradicting (12). Therefore the stronger sufficient criterion also forces
This conclusion concerns the distance criterion (8), not arbitrary affine families with the local property. A construction using more information than the minimum weight of an affine coset remains outside the argument.