../
Source. S. Lee, Relative independence of Erdős problem #501, second
version dated 2026-06-01, Lemma 2.1 (statement, physical p. 2; proof,
pp. 4--5), in the six-page PDF held by its library source card,
Lee (2026).
Its one input is the section inequality reconstructed in
Lemma 3.1.
Standing. This is an author-recorded reconstruction. It is not an
independent review and changes no status and assigns no tier.
Definitions
m, m∗ and the upper integral are as on the Lemma 3.1 page. Let
ν:P(R)→[0,∞] be a countably additive
measure extending Lebesgue measure. Two facts about ν are used.
- (1) ν(S)≤m∗(S) for every S⊆R (the source's
(1)): if S⊆⋃iJi with open intervals Ji, then
ν(S)≤∑iν(Ji)=∑im(Ji), and m∗(S) is the infimum of
the right side over all such covers. In particular m∗(S)<1 implies
ν(S)<1.
- The upper-integral identity
∫Ra1Sdm=am∗(S) for a≥0 and
S⊆R: a measurable T⊇S gives the majorant
a1T with integral am(T), and every measurable majorant h≥a1S
has T={h≥a}⊇S measurable and ∫h≥am(T)≥am∗(S)
when a>0; for a=0 both sides vanish.
For a family (Ay)y∈R of subsets of R and
x∈R, put Bx={y∈R:x∈Ay}, the set of indices
whose sets contain x.
Statement
Let ν be as above and let (Ay)y∈R satisfy m∗(Ay)<1
for every y∈R. If C⊆R and ν(C)=∞,
then there is an x∈C with ν(C∖Bx)=∞.
Proof
Suppose, toward a contradiction, that
ν(C∖Bx)<∞for every x∈C
(the source's (12)).
A window of measure above one. For N≥1 put CN=C∩[−N,N].
The sets CN increase to C, so continuity from below gives
ν(CN)→ν(C)=∞ (the source's (13)). Fix N with
D:=CN satisfying ν(D)>1. For k≥1 define
Dk={x∈D:ν(C∖Bx)≤k}
(the source's (14)). By the assumption (12), D=⋃k≥1Dk, and
the Dk increase, so continuity from below gives k≥1 with
ν(Dk)>1. By (1) and Dk⊆[−N,N],
1<ν(Dk)≤m∗(Dk)≤2N<∞
(the source's (15)).
A larger window. For every integer M>N,
ν(CM)≤ν([−M,M])=2M<∞.
By (13) and (15), choose an integer M>N with
ν(CM)>m∗(Dk)−1km∗(Dk)
(the source's (16)); the right side is a finite positive number by (15).
Since m∗(Dk)/(m∗(Dk)−1)>1, in particular ν(CM)>k.
Lower bound on the vertical sections. Let x∈Dk. Then
ν(C∖Bx)≤k; as CM has finite ν-measure and
CM∖Bx⊆C∖Bx,
ν(Bx∩CM)=ν(CM)−ν(CM∖Bx)≥ν(CM)−k
(the source's (17)). Define
H={(x,y)∈Dk×CM:x∈Ay}⊆R2.
For x∈Dk, Hx={y∈CM:x∈Ay}=Bx∩CM, and Hx=∅
for x∈/Dk. So ν(Hx)≥(ν(CM)−k)1Dk(x) for every
x∈R, and by monotonicity of the upper integral and the
identity for a1S with a=ν(CM)−k≥0,
∫Rν(Hx)dm(x)≥(ν(CM)−k)m∗(Dk)
(the source's (18)).
Upper bound on the horizontal sections. For y∈CM,
Hy=Ay∩Dk, and Hy=∅ for y∈/CM. Since
m∗(Ay)<1 for every y, m∗(Hy)≤1CM(y) for every
y∈R. The function y↦m∗(Hy) is ν-measurable,
ν being defined on all subsets, so
∫Rm∗(Hy)dν(y)≤∫R1CMdν=ν(CM)
(the source's (19)).
Contradiction. The section inequality (11) of Lemma 3.1 applied to
H, with (18) and (19), gives
(ν(CM)−k)m∗(Dk)≤ν(CM).
Dividing by ν(CM), which is positive and finite,
(1−ν(CM)k)m∗(Dk)≤1.
But (16) is equivalent, after multiplying by m∗(Dk)−1>0 and dividing
by ν(CM), to
(1−ν(CM)k)m∗(Dk)>1,
a contradiction. Hence (12) fails: some x∈C has
ν(C∖Bx)=∞.
Boundary. Only the values ν(C∖Bx), ν(CM) and the
outer measures m∗(Dk), m∗(Ay) enter; no measurability of the
sets Ay or Bx for Lebesgue measure is assumed. The lemma drives
the recursion in
Theorem 1.1.