Wiki
Wiki

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

Updated


Source. Croot, published paper, p. 234, the construction preceding Theorem 1. This refines the Erdős–Szemerédi construction cited there by allowing prime-power factors.

Let f(x)f(x) be the maximum number of pairwise disjoint integer congruence classes with distinct moduli in [2,x][2,x]. Write T(x)=log⁡xlog⁡log⁡xT(x)=\sqrt{\log x\log\log x}.

Statement. For each η>0\eta>0 and all sufficiently large xx,

f(x)≥xexp⁡(−(2+η)T(x)).f(x)\ge x\exp(-(\sqrt2+\eta)T(x)).

Complete relative proof. Put c=1/2c=1/\sqrt2, and let pp be the largest prime at most ecT(x)e^{cT(x)}. Bertrand's postulate gives

log⁡p=cT(x)+O(1).\log p=cT(x)+O(1).

Take every integer m≤x/pm\le x/p all of whose prime-power divisors are strictly less than pp, and use the modulus q=pmq=pm. In particular p∤mp\nmid m. Write the maximal prime-power factors of mm in decreasing order as

dr>dr−1>⋯>d1,m=dr⋯d1.d_r>d_{r-1}>\cdots>d_1, \qquad m=d_r\cdots d_1.

The djd_j are powers of distinct primes, and all satisfy dj<pd_j<p. Define a residue am(modpm)a_m\pmod{pm} by

am≡dr(modp),am≡dj−1(moddj) (2≤j≤r),am≡0(modd1).a_m\equiv d_r\pmod p,\qquad a_m\equiv d_{j-1}\pmod{d_j}\ (2\le j\le r),\qquad a_m\equiv0\pmod{d_1}.

The Chinese remainder theorem applies because the listed moduli are pairwise coprime. When m=1m=1, define a1≡0(modp)a_1\equiv0\pmod p instead.

To prove disjointness, associate the descending list (p,dr,…,d1)(p,d_r,\ldots,d_1) to mm. For two distinct integers m,m′m,m', their lists have a longest common initial segment, which contains pp. Let DD be its last entry. The next entries of the two lists are different; if a list has ended, use 00 for its next entry. Both entries lie in [0,D)[0,D), and the construction makes them the respective residues of ama_m and am′a_{m'} modulo DD. Thus these residues differ modulo DD, although DD divides both moduli. An integer cannot belong to both congruence classes.

It remains to count the moduli. Their number is

ψ∗(x/p,p−1).\psi^*(x/p,p-1).

Set x′=x/px'=x/p. Since log⁡p=O(T(x))=o(log⁡x)\log p=O(T(x))=o(\log x),

T(x′)T(x)⟶1,log⁡(p−1)T(x′)⟶c.\frac{T(x')}{T(x)}\longrightarrow1,\qquad \frac{\log(p-1)}{T(x')}\longrightarrow c.

For every fixed 0<ε<c0<\varepsilon<c, the second relation eventually places p−1p-1 between L(c−ε,x′)L(c-\varepsilon,x') and L(c+ε,x′)L(c+\varepsilon,x'). Monotonicity in the smoothness cutoff and the complete prime-power smoothness deduction therefore imply, by letting ε\varepsilon tend to zero after taking limits,

ψ∗(x/p,p−1)=xpexp⁡(−(12c+o(1))T(x))=xexp⁡(−(c+12c+o(1))T(x))=xexp⁡(−(2+o(1))T(x)).\begin{aligned} \psi^*(x/p,p-1) &=\frac xp\exp\left(-\left(\frac1{2c}+o(1)\right)T(x)\right)\\ &=x\exp\left(-\left(c+\frac1{2c}+o(1)\right)T(x)\right)\\ &=x\exp(-(\sqrt2+o(1))T(x)). \end{aligned}

The constructed distinct moduli give the desired lower bound for f(x)f(x).

Source clarification. The phrase on p. 234 requiring both divisibility by pp and all prime-power factors to be less than pp must refer to the factors of q/pq/p. Taken literally for qq, it would exclude every modulus. The formula q=p dr⋯d1q=p\,d_r\cdots d_1 and the subsequent CRT conditions identify the intended meaning. The empty list and the disjointness argument are written out above.

Dependencies. The smooth-number theorem in Lemma 1 remains an external analytic input. Bertrand's postulate and the Chinese remainder theorem are classical inputs; the construction and all counting deductions specific to this paper are included.

Bears on. Problem 202: a lower bound for its maximum with coefficient 2\sqrt2 on the scale T(x)T(x), not the sharp coefficient.