Source. Scott D. Hughes, Sums of distinct divisors of factorials,
arXiv:2609.10902v1, Theorem 2 (quoted from Berend–Harmse), display (1) and
Corollary 3 with its proof, physical p. 2 of the five-page PDF held by
Hughes (2026);
the library records them on
the Theorem 2 page.
Read in the canonical conversion beside the PDF and checked against the page
image. Consumed by
the Theorem 1 reconstruction.
Standing. Author-recorded reconstruction; not an independent review; it
changes no status and assigns no tier. The Berend–Harmse estimate is an
imported theorem, quoted below in the form the source prints; the 1993 paper
is not held and was not read, so the import is second-hand.
Definitions
log is the natural logarithm and lgt=log2t. For real x≥216
put
εx=(x1)2lgx−lg(lgx).
Expanding lgx=logx/log2 and lg(lgx)=log(logx/log2)/log2,
The window of index j≥2 is the interval [(j−1)!,j!];
its logarithmic width is 21logj.
Imported theorem (Berend–Harmse, as quoted)
D. Berend and J. E. Harmse, Gaps between consecutive divisors of
factorials, Ann. Inst. Fourier (Grenoble) 43 (1993), no. 3, 569–583,
Theorem 2, in the form the source prints: for every integer n≥216 and
every real D with (n−1)!≤D≤n! there is a divisor x
of n! with
Only the outer inequality ∣x/D−1∣≤εn is consumed. The second
inequality between the two bounds is a numerical comparison, which was
checked here: the base-two logarithm of the ratio of the left bound to the
right one equals
Monotonicity. Write L=logx. The exponent in (1) is
(2lgx−lg(lgx))logx=log21(2L2−Lloglog2L),
whose derivative in L is log21(L−log(L/log2)−1).
At L=16log2 this is positive, since 16log2−log16−1>0, and it
increases with L. So log(1/εx) increases for x≥216,
and the sequence (εj)j≥216 is decreasing.
Size. At j=216 the exponent is 216−lg16=4, so
ε216=2−64, and therefore εj≤2−64<41
for every integer j≥216.
Statement
Let j≥216 and n≥j be integers, and let a<b be consecutive
divisors of n! with (j−1)!≤ab≤j!. Then
logab≤3εj.
Proof
Apply the imported theorem with j in place of n and D=ab, which
lies in the required range. It gives a divisor x of j! with
∣x/D−1∣≤εj. Since j≤n, j! divides n!, so x is a
divisor of n!; as a<b are consecutive divisors of n!, no divisor of
n! lies strictly between them, so x≤a or x≥b.
If x≤a, then 1−εj≤x/D≤a/D=a/b, so
b/a≤(1−εj)−2 and log(b/a)≤−2log(1−εj).
If x≥b, then b/a=b/D≤x/D≤1+εj, so
b/a≤(1+εj)2 and
log(b/a)≤2log(1+εj)≤2εj.
For 0≤u≤41,
−log(1−u)=log(1+1−uu)≤1−uu≤34u.
With u=εj<41 the first case gives
log(b/a)≤38εj. Both cases give
log(b/a)≤3εj.