Source. S. Korsky, A resolution of the de Bruijn--Erdős
consecutive-gap problem, arXiv:2609.07196v2, Section 7: Theorem 7.1
(p. 13) and Lemma 7.2 (pp. 13--14) of the retained PDF, read in the
canonical conversion and checked against the text layer at the displayed
constants; held by its library card,
Korsky 2026, resolution.
Standing. Author-recorded reconstruction of Lemma 7.2; not an
independent review; changes no status and assigns no tier. Theorem 7.1 is
an external input, cited by the source to G. Halász, On Roth's method in
the theory of irregularities of point distributions, in Recent Progress
in Analytic Number Theory, vol. 2 (Academic Press, 1981), 79--94; that
paper is not held and the statement is not checked against it here.
Definitions
For a finite set P⊂[0,1]2 of M points put
DP(u,v)=#(P∩((0,u]×(0,v]))−Muv(0≤u,v≤1);
endpoint conventions do not affect the integrals below. Points, Pn and
Nn(⋅) are as on the
Lemma 2.1 page;
only integer times occur here.
The imported input (Theorem 7.1, Halász)
There is an absolute constant cH>0 such that every set P of
M≥2 points in [0,1]2 satisfies
∫01∫01DP(u,v)dudv≥cHlogM.
This is the unnormalized form of Halász's planar L1 discrepancy
theorem, as the source states it; the source gives no further derivation
and none is supplied here.
Statement (Lemma 7.2, p. 13)
There are absolute constants c4>0 and S0 with the following
property. Suppose S≥S0, B≥1, and, for all sufficiently large
integers n,
∫TNn((x,x+D/n])−Ddx≤B(0≤D≤S).(7.1)
Then B≥c4logS.
Proof
Put L=⌊S⌋, let n0 be a threshold for (7.1), choose an
integer N>max(L,n0), and put w=L/N. For a∈T let
Ma=NN((a,a+w]) be the number of the first N points in the arc of
length w=L/N starting at a. Applying (7.1) at n=N with D=L,
∫T∣Ma−L∣da≤B.(7.2)
The planar point sets. View the oriented arc (a,a+w] as a copy of
(0,1]. For each xi∈PN∩(a,a+w] form the point
(wxi−a,Ni)∈(0,1]2,
the first coordinate measured along the arc. Let Pa be the
resulting set of Ma points, and define
Ga(u,v)=N⌊Nv⌋((a,a+uw])−Luv.
A point xi of the arc lies in the box (0,u]×(0,v] exactly when
xi∈(a,a+uw] and i≤Nv, that is, i≤⌊Nv⌋; so, off
a null set of (u,v) caused by endpoints,
Lower bound from Halász. Since ∫01∫01uvdudv=1/4,
Theorem 7.1 gives, whenever Ma≥2,
∥Ga∥L1([0,1]2)≥cHlogMa−41∣Ma−L∣.(7.3)
If B≥L/4, then B≥c4logS for any fixed c4 once
S0 is large, because L≥S−1 grows faster than logS; so
assume B<L/4. By Markov's inequality and (7.2),
the set of a with ∣Ma−L∣>L/2 has measure at most (2/L)B<1/2, so
{a:L/2≤Ma≤3L/2}≥21.
On this set Ma≥L/2≥2 for L≥4, and
logMa≥log(L/2); integrating (7.3) over it and using (7.2)
for the subtracted term,
∫T∥Ga∥1da≥2cHlog(L/2)−4B.(7.4)
Upper bound from (7.1). Fix (u,v) and put n=⌊Nv⌋. If
n≥n0, set D=nuw; then D≤Nuw=Lu≤L≤S, the arc
(a,a+uw] has length uw=D/n, and
Ga(u,v)=(Nn((a,a+uw])−nuw)+Lu(Nn−v),
because nuw=Lu⋅n/N. The second term has absolute value at most
Lu/N≤w, since 0≤v−n/N<1/N; the integral over a of the absolute
value of the first term is at most B by (7.1). On the strip
0≤v<n0/N, where n<n0, trivially ∣Ga(u,v)∣≤n0+L. Integrating
over (u,v) and then a,
∫T∥Ga∥1da≤B+NL+Nn0(n0+L)=B+oN→∞(1),(7.5)
with L and n0 fixed.
Conclusion. Combining (7.4) and (7.5) and letting N→∞,
45B≥2cHlog(L/2).
Since L=⌊S⌋ and S≥S0 is large, log(L/2)≥21logS,
say, and B≥c4logS with an absolute c4>0 after adjusting the
constants; the case B≥L/4 treated above is covered by the same c4 once
S0 is large enough.
Role in the argument
Proposition 6.4 supplies (7.1) with B=C3A and S=Ar/Λ2;
the
Section 8 proof
takes A=clogr and compares C3A with
c4logS≥c4(logr)/3.