Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Rank threshold families
Theorem: proved
Let m1,…,ms≥2 be pairwise coprime integers and fix one
residue bi(modmi) for each. Fix an integer r≥1. For every r-element subset
S⊆{1,…,s}, install the CRT-induced class modulo
qS=∏i∈Smi, activated at n≥qS. Let B be this
delayed union and β its full periodic density. Then
X≥1supX∣B∩[1,X]∣≪log(e/β)1.(1)
The implied constant is absolute, and the same bound holds for
HB(X)/H⌊X⌋. If r>s, both unions are empty.
All product moduli are distinct, since the blocks are pairwise coprime
and greater than one. Thus these are genuine singleton systems. The
blocks need not be prime powers. This theorem excludes another
small-full-footprint construction, but does not control arbitrary
subfamilies of products or incoherent residues.
Three elementary counting bounds
Put pi=1/mi, λ=∑ipi, and let
Y=∑iYi, with independent Yi∼Bernoulli(pi).
CRT gives β=Pr(Y≥r). Writing the supremum on the left of
(1) as ρ, we have
ρ≤2rβ,ρ≤λ/r,ρ≤(r−1−λ)2λ+3(r>λ+1).(2)
For the first bound, partition the deleted integers below N among
their first covering product classes, and write their normalized counts
as cS. For a block face J, activation gives
S⊇J∑cS≤∏j∈Jmj1.
On Haar space use Z=∑SqScS1CS. Expand intersections
using
Since Z is supported on the full union of measure β,
Cauchy--Schwarz proves the first assertion of (2).
Every actual deletion has at least r active block hits. Block i
has at most N/mi active hits up to N. Summing proves the second.
For the third, average over 1≤n≤N, and put
ZN(n)=mi≤N∑1n≡bi(modmi),λN=mi≤N∑1/mi.
An active certificate has at most one block larger than N,
so deletion implies ZN(n)≥r−1. If k blocks occur in this
sum, their single and pair CRT counts have normalized error at most
1/N. Thus
E(ZN−λN)2≤λN+Nk2+N2λNk≤λN+3,
using k≤N and λN≤k/2. Chebyshev's inequality,
with λN≤λ, proves (2).
An elementary independent-tail lower bound
When λ≥104 and r≤1.1λ,
β≥32exp[−150(1+λ(r−λ)+2)].(3)
To prove this, set a=(r−λ)+ and
d=4(a+2λ)/λ≤0.48. Tilt each Bernoulli odds by
1+d; the new probabilities are
pi′=1+dpi(1+d)pi.
Their mean M satisfies
M−λ=i∑1+dpidpi(1−pi)≥2+ddλ,M≤(1+d)λ≤1.5λ.
It follows that M−r≥3λ. The tilted variance is at
most 1.5λ, so Chebyshev gives tilted probability at least
2/3 to r≤Y≤M+3λ. On this event, using
log(1+x)≥x−x2/2,
For the last inequality, put u=a/λ. Its left side is at
most 20u2+92u+104≤150(1+u2). Changing measure proves (3).
Combining the bounds
Let L=log(1/β), with bounded L absorbed in the constant.
If r≤L/(2log2), (2) gives ρ≤e−L/2. Otherwise,
if λ≤L, it gives ρ≪L−1/2. If
λ>L and r−1−λ≥λ/20, its last bound
gives ρ≪1/λ≪L−1/2.
In the remaining case, r≤1.1λ for sufficiently large L,
and (3) implies (r−λ)+2/λ≫L. In particular
r>λ+1, and the last bound in (2) is O(1/L). This proves
(1). Partial summation of the uniform counting bound proves its harmonic
version.