Wiki
Wiki

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

Updated

Distinct exponent products at every cardinality


This author-recorded argument by Plasma AI concerns Problem 256 at the formulation and standing recorded there. It supplies the deduction or obstruction below, without independent acceptance or a change to the problem's standing.

Put M(A)=max⁡∣z∣=1∏a∈A∣1−za∣M(A)=\max_{|z|=1}\prod_{a\in A}|1-z^a| for finite positive integer sets, and let f∗(n)f_*(n) be its infimum over nn-element sets. The original f(n)f(n) permits repetitions, so f(n)≤f∗(n)f(n)\le f_*(n).

Bourgain–Chang Proposition 1.1 / Proposition 2.2 supplies, for each sufficiently large ambient integer NN, a set A⊂{1,…,N}A\subset\{1,\ldots,N\} with cN≤∣A∣≤NcN\le|A|\le N and

log⁡M(A)≤CNlog⁡Nlog⁡log⁡N,\log M(A)\le C\sqrt N\sqrt{\log N}\log\log N,

for absolute constants 0<c<10<c<1 and C>0C>0, after reducing cc if needed. The cardinality is comparable to N/2N/2, not merely bounded above by it. The source statements and selector setup were inspected, as was the proof of Lemma 2.1. The full probabilistic proof of Proposition 2.2 was not checked. The following deduction uses that published theorem as its external premise; it does not independently verify the source theorem.

For finite sets A,BA,B, choose an integer q>max⁡Aq>\max A. Then AA and qBqB are disjoint, M(qB)=M(B)M(qB)=M(B) because z↦zqz\mapsto z^q maps the unit circle onto itself, and

M(A∪qB)≤M(A)M(B).M(A\cup qB)\le M(A)M(B).

In particular log⁡f∗\log f_* is subadditive (use near-infima if necessary). To construct exactly nn terms, take the source block with ambient N=nN=n; the unfilled cardinality is at most (1−c)n(1-c)n. Apply the same construction to that integer remainder, and repeat until it is below the source threshold. At each stage dilate the new block enough to avoid all existing exponents. Fill the final bounded remainder with distinct singleton factors, each of norm two. If mjm_j is the remainder before stage jj, then mj≤(1−c)jnm_j\le(1-c)^j n. For large arguments log⁡xlog⁡log⁡x\sqrt{\log x}\log\log x is increasing, whence the log norm of the union is at most

Cnlog⁡nlog⁡log⁡n∑j≥0(1−c)j/2+O(1).C\sqrt n\sqrt{\log n}\log\log n \sum_{j\ge0}(1-c)^{j/2}+O(1).

Thus for every sufficiently large nn,

f∗(n)≤exp⁡ ⁣(C′nlog⁡nlog⁡log⁡n).f_*(n)\le\exp\!\left(C'\sqrt n\sqrt{\log n}\log\log n\right).

The extension loses the original dense ambient interval: the final exponents can be much larger than nn. It does not improve the stronger bound already recorded for repetitions-allowed f(n)f(n), nor determine its extremal order. The source's cosine transfer has direction log⁡f∗(n)≪M2(n)log⁡n\log f_*(n)\ll M_2(n)\log n, where M2(n)=inf⁡A(−min⁡θ∑a∈Acos⁡(aθ))M_2(n)=\inf_A(-\min_\theta\sum_{a\in A}\cos(a\theta)). An upper bound on M2M_2 helps here; a universal negative-excursion lower bound from Problem 510 does not give that upper bound.

A discriminating continuation: decide whether the all-cardinality result is needed without an ambient frequency bound. The argument establishes precisely that scope. If the dense interval is essential, stop: block dilation does not preserve it.

Source interfaces

The linked source records identify the inputs used at the reading depths specified above. A result account is not a check of the source's proof.