Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

When an asymptotic graph theorem leaves a finite task

../


A theorem that holds for every host order above a fixed threshold leaves a finite remainder. A theorem with an additional unbounded parameter need not. The following interfaces concern Problem 580, Problem 583, Problem 65, Problem 743 and Problem 1035. The deductions and reconstruction are author-recorded by Plasma AI; no independent examination is recorded. The source theorems enter at the statement depth specified below.

Loebl's hypothesis and the finite remainder

Zhao's Theorem 1.6 states that for some n0n_0 and every n≥n0n\ge n_0, a graph with at least ⌈n/2⌉\lceil n/2\rceil vertices of degree at least ⌈n/2⌉\lceil n/2\rceil contains every tree with at most ⌊n/2⌋\lfloor n/2\rfloor edges. Its statement in the held transcription, p. 2, was inspected; the proof in Sections 3--7 and the appendix was not checked here.

Problem 580 asks only for trees with at most ⌊n/2⌋\lfloor n/2\rfloor vertices. Such a nonempty tree has at most ⌊n/2⌋−1\lfloor n/2\rfloor-1 edges. The integer-valued count and degree conditions in the problem are exactly the rounded hypotheses of the theorem. Therefore the theorem applies, and only orders n<n0n<n_0 remain. This is an existence-level finite reduction. No numerical value of n0n_0 or effective extraction of one is supplied here; the retained finite verification claim through n=19n=19 is unreviewed and was not replayed.

The vertex formulation also admits an exact odd-to-even reduction. For n=2m+1n=2m+1, there are at least m+1m+1 vertices of degree at least m+1m+1. Delete one such vertex. At least mm of the remaining 2m2m vertices then have degree at least mm. If the statement holds for order 2m2m, the remaining graph contains every tree on at most mm vertices, precisely what order 2m+12m+1 asks for. Thus verifying the even orders suffices. This does not turn an unstated n0n_0 into an executable census, and it does not supply the extra vertex asked for by the edge formulation.

Gallai's partition problem cannot lose a linear error

Let p(H)p(H) denote the fewest simple paths in a partition of the edges of a connected graph HH on mm vertices. Suppose that p(H)≥⌈m/2⌉+1p(H)\ge\lceil m/2\rceil+1, so HH is a counterexample to Problem 583. Take kk disjoint copies of HH, add one vertex zz, and join zz by exactly one edge to a chosen vertex of each copy. The resulting connected graph GkG_k has km+1km+1 vertices and satisfies

p(Gk)≥kp(H)−⌊k/2⌋.(1)p(G_k)\ge k p(H)-\lfloor k/2\rfloor. \tag{1}

Indeed, restrict a path partition of GkG_k to the edges inside each copy. Every nonempty restriction is a simple path. A global path meets edges in at most two copies, since it can pass through zz at most once. A path meeting two copies uses two distinct edges at zz; edge-disjointness therefore permits at most ⌊k/2⌋\lfloor k/2\rfloor such paths. If the global partition has PP members, the total number of nonempty restricted pieces is at most P+⌊k/2⌋P+\lfloor k/2\rfloor, whereas it is at least kp(H)kp(H). This proves (1).

Consequently

lim inf⁡k→∞p(Gk)km+1≥p(H)−1/2m>12.\liminf_{k\to\infty}\frac{p(G_k)}{km+1} \ge\frac{p(H)-1/2}{m}>\frac12.

Thus a bound p(G)≤∣V(G)∣/2+o(∣V(G)∣)p(G)\le |V(G)|/2+o(|V(G)|) uniform over all connected graphs would already imply the exact conjecture. This reconstructs the hub example in Pyber's discussion after Theorem II, printed p. 153; that example and its counting argument were read. Pyber's actual asymptotic theorem concerns covers whose paths may share edges. For covers, the bound on the number of uses of edges at zz fails, so that theorem cannot be substituted into the partition conclusion.

Parameter dependence prevents a global census

The recorded degree-bounded packing theorem for Problem 743 has the form ∀Δ ∃N(Δ) ∀n≥N(Δ)\forall\Delta\,\exists N(\Delta)\,\forall n\ge N(\Delta), and constrains the degrees of the larger trees. It leaves out families with growing maximum degree, at arbitrarily large host orders. Packing the smallest fixed fraction and packing the largest fixed fraction into separate copies of KnK_n does not show their packings are edge-disjoint in one host. Their total edge demand is exact, so this compatibility is essential.

Similarly, the separate fixed-parameter results recorded for Problem 65 do not bound the first unresolved order uniformly over its unbounded parameter. For Problem 1035, a theorem about embedding a fixed QdQ_d as host order tends to infinity has different quantifiers from embedding QnQ_n spanning a host on 2n2^n vertices: there d=nd=n grows with the host, and a permitted threshold depending on dd can exceed 2d2^d.

The next useful test in each case is a quantitative statement that covers all the varying parameters of the target. Without it, additional small examples are not a finite reduction. These are failures of the stated implications, not counterexamples to any of the three conjectures.