Wiki
Wiki

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

Updated

Erdős 699: an effective row-zero reduction at every fixed index

P699 research · The effective S-part theorem


Proposition. Fix i≥4i\ge4 and K≥1K\ge1. There is an effectively computable bound for every bad triple with [n]{p:p<i}≤K[n]_{\{p:p<i\}}\le K.

In particular, above an effective bound depending only on ii, at least one small prime has a strictly larger valuation in nn than in every other numerator factor n−1,…,n−i+1n-1,\ldots,n-i+1.

This is a reduction of possible position patterns, not effective finiteness for an entire remaining index.

Write S={p:p<i}S=\{p:p<i\} and let H=iH=i when ii is prime, H=1H=1 otherwise. Let RtR_t be the part of n−tn-t supported on primes at least ii. Badness and Lucas's theorem give

R0gcd⁡(R0,H)∣j.\frac{R_0}{\gcd(R_0,H)}\mid j.

Consequently, if j/n=a/qj/n=a/q is reduced, then

2≤q≤H[n]S≤Q:=HK.2\le q\le H[n]_S\le Q:=HK.

For every 0≤t<i0\le t<i, the same localization gives

Rt∣HFt(a,q),Ft(a,q)=∏r=0t(at−qr).(1)R_t\mid H F_t(a,q),\qquad F_t(a,q)=\prod_{r=0}^t(at-qr).\tag{1}

To see this prime by prime, an active prime power dividing n−tn-t forces j≡rj\equiv r modulo that power for some 0≤r≤t0\le r\le t. The identity qj=anqj=an then gives at−qr≡0at-qr\equiv0. At the single possible denominator prime p=ip=i, an inactive first power is covered by HH. No inversion of qq modulo a prime is needed.

Since gcd⁡(a,q)=1\gcd(a,q)=1, the polynomial Ft(a,q)F_t(a,q) is zero exactly when q∣tq\mid t. For q=2q=2 choose t=1,3t=1,3; for q≥3q\ge3 choose t=1,2t=1,2. Both selected values in (1) are nonzero. As 0<a/q≤1/20<a/q\le1/2, their product is bounded by 81Q681Q^6. Thus, for one of the two fixed polynomials

f2(X)=(X−1)(X−2),f3(X)=(X−1)(X−3),f_2(X)=(X-1)(X-2),\qquad f_3(X)=(X-1)(X-3),

we have

[ft(n)]S≥ft(n)81H2Q6.(2)[f_t(n)]_S\ge\frac{f_t(n)}{81H^2Q^6}.\tag{2}

BEG's effective Theorem 2.2, already used in the effectivity note, supplies computable C≥1C\ge1 and κ>0\kappa>0 such that, for both polynomials,

[ft(n)]S≤Cft(n)1−κ.[f_t(n)]_S\le C f_t(n)^{1-\kappa}.

Combining this with (2), and using ft(n)≥n2/4f_t(n)\ge n^2/4 for n≥6n\ge6, gives the effective bound

n≤2(81CH2Q6)1/(2κ).(3)\boxed{n\le2(81CH^2Q^6)^{1/(2\kappa)}.}\tag{3}

For the last assertion take K=lcm⁡(1,…,i−1)K=\operatorname{lcm}(1,\ldots,i-1). If [n]S>K[n]_S>K, some p<ip<i has vp(n)>⌊log⁡p(i−1)⌋v_p(n)>\lfloor\log_p(i-1)\rfloor; otherwise [n]S[n]_S divides KK. For every 1≤t<i1\le t<i, this forces vp(n−t)=vp(t)<vp(n)v_p(n-t)=v_p(t)<v_p(n). So its unique selected maximum is at position zero.

The stronger index-six and index-seven result excludes the complementary position patterns at every admissible row, using exact fixed-slope divisibilities. The present argument supplies the effective reduction uniformly for every fixed i≥4i\ge4.