Section 4.5

The Existence of non-Lebesgue Sets

[7/0%]
Show
LaTeX
Problem 4.5.1

Let E∉LkE \notin \mathscr {L}^{k} denote the set constructed in this section, and let A⊆EA \subseteq E be such that A∈LkA \in \mathscr {L}^{k}. Then λˉk(A)=0\bar{\lambda }_{k}(A)=0.

?
Remark.
?

The construction of this section: for each x\boldsymbol {x} and y\boldsymbol {y} in Rk\mathbb {R}^{k}, write x∼y\boldsymbol {x} \sim \boldsymbol {y} iff there is r∈Qk\boldsymbol {r} \in \mathbb {Q}^{k} such that x+r=y\boldsymbol {x}+\boldsymbol {r}=\boldsymbol {y}; then ∼\sim is an equivalence relation on Rk\mathbb {R}^{k}, determining an infinite number of equivalence classes, each itself infinite. Invoking the Axiom of Choice, form a set E⊆RkE \subseteq \mathbb {R}^{k} consisting of exactly one element from each equivalence class, and consider the collection {E⊕r:r∈Qk}\left\{ E \oplus \boldsymbol {r}: \boldsymbol {r} \in \mathbb {Q}^{k}\right\} of the translates of EE by the elements of Qk\mathbb {Q}^{k}. Claim 2: if r1,r2∈Qk\boldsymbol {r}_{1}, \boldsymbol {r}_{2} \in \mathbb {Q}^{k} are distinct, then E⊕r1E \oplus \boldsymbol {r}_{1} and E⊕r2E \oplus \boldsymbol {r}_{2} are disjoint. Claim 3: Rk=⋃{(E⊕r):r∈Qk}\mathbb {R}^{k}=\bigcup \left\{ (E \oplus \boldsymbol {r}): \boldsymbol {r} \in \mathbb {Q}^{k}\right\}, the union being disjoint. Claim 4: λk∗(E)>0\lambda_{k}^{*}(E)>0. Claim 5: sup⁡{λk(K):K compact,K⊆E}=0\sup \left\{ \lambda_{k}(K): K \text{ compact}, K \subseteq E\right\} =0. Claim 6: the set EE is not in Lk\mathscr {L}^{k}. This is a variant of the classical construction given by Vitali in 1905.

Problem 4.5.2

There exists no translation invariant measure μ\mu defined for all subsets of Rk\mathbb {R}^{k} such that μ((a,b])=∏i=1k(bi−ai)\mu ((\boldsymbol {a}, \boldsymbol {b}])=\prod_{i=1}^{k}(b_{i}-a_{i}) for all bounded kk-dimensional rsc rectangles (a,b](\boldsymbol {a}, \boldsymbol {b}].

?
Problem 4.5.3

Here is an alternative but quite similar construction of a set outside L\mathscr {L}. We shall consider (0,1](0,1] as our space. Let +˙\dot{+} denote addition modulo 1: for x,y∈(0,1]x, y \in (0,1],

x+˙y={x+y,if x+y⩽1,x+y−1,if x+y>1. x \mathbin {\dot{+}} y= \begin{cases} x+y, & \text{if } x+y \leqslant 1, \\ x+y-1, & \text{if } x+y>1. \end{cases}

For each A⊆(0,1]A \subseteq (0,1] and x∈(0,1]x \in (0,1], we shall [for this exercise only] write A⊕xA \oplus x as translation modulo 1: A⊕x={a+˙x:a∈A}A \oplus x=\left\{ a \mathbin {\dot{+}} x: a \in A\right\}. Finally, [for this exercise only] we shall let Q\mathbb {Q} denote the rational numbers of (0,1](0,1].

?
(a)

For x,y∈(0,1]x, y \in (0,1] write x∼yx \sim y iff there is r∈Qr \in \mathbb {Q} such that x+˙r=yx \mathbin {\dot{+}} r=y. Then ∼\sim is an equivalence relation.

(b)

Let FF denote the set formed by choosing exactly one element from each equivalence class [use (AC)]. Show that (F⊕r1)∩(F⊕r2)(F \oplus r_{1}) \cap (F \oplus r_{2}) is empty if r1r_{1} and r2r_{2} are distinct elements of Q\mathbb {Q}.

(c)

(0,1]=⋃r∈Q(F⊕r)(0,1]=\bigcup_{r \in \mathbb {Q}}(F \oplus r), where the union is disjoint.

(d)

F∉LF \notin \mathscr {L} and 0<λ∗(F)⩽10<\lambda^{*}(F) \leqslant 1.

(e)

If there exists a translation invariant measure μ\mu defined on a σ\sigma-field on (0,1](0,1] that includes FF and is closed under translations, then μ((0,1])\mu ((0,1]) is either 00 or +∞+\infty.

(f)

Show that there can exist no translation invariant measure defined on all subsets of (0,1](0,1] that agrees with λˉ\bar{\lambda } on the Lebesgue sets of (0,1](0,1].

Remark.
?

Argue by contradiction for (d): if F∈LF \in \mathscr {L}, then

1=λˉ((0,1])=∑r∈Qλˉ(F⊕r)=∑r∈Qλˉ(F). 1=\bar{\lambda }((0,1])=\sum _{r \in \mathbb {Q}} \bar{\lambda }(F \oplus r)=\sum _{r \in \mathbb {Q}} \bar{\lambda }(F).

If λˉ(F)=0\bar{\lambda }(F)=0, we obtain 1=01=0. If λˉ(F)>0\bar{\lambda }(F)>0, then we obtain 1=+∞1=+\infty. Conclude that λˉ(F)\bar{\lambda }(F) cannot exist, which means that F∉LF \notin \mathscr {L}. For the other part, since F⊆(0,1]F \subseteq (0,1], clearly λ∗(F)⩽1\lambda^{*}(F) \leqslant 1. To show that λ∗(F)>0\lambda^{*}(F)>0, write

1=λ∗((0,1])=λ∗(⋃r∈Q(F⊕r))⩽∑r∈Qλ∗(F⊕r)=∑r∈Qλ∗(F). 1=\lambda ^{*}((0,1])=\lambda ^{*}\left(\bigcup _{r \in \mathbb {Q}}(F \oplus r)\right) \leqslant \sum _{r \in \mathbb {Q}} \lambda ^{*}(F \oplus r)=\sum _{r \in \mathbb {Q}} \lambda ^{*}(F).
Problem 4.5.4

Prove the following assertions.

?
(a)

There exists a sequence {En}n=1∞\left\{ E_{n}\right\}_{n=1}^{\infty } of disjoint sets [not necessarily in L\mathscr {L}] such that λ∗(⋃n=1∞En)<∑n=1∞λ∗(En)\lambda^{*}\left(\bigcup_{n=1}^{\infty } E_{n}\right)<\sum_{n=1}^{\infty } \lambda^{*}(E_{n}).

(b)

There exists a nonincreasing sequence {En}n=1∞\left\{ E_{n}\right\}_{n=1}^{\infty } of sets [not necessarily in L\mathscr {L}] with λ∗(En)<+∞\lambda^{*}(E_{n})<+\infty for each n∈Nn \in \mathbb {N} such that λ∗(⋂n=1∞En)<lim⁡nλ∗(En)\lambda^{*}\left(\bigcap_{n=1}^{\infty } E_{n}\right)<\lim_{n} \lambda^{*}(E_{n}).

Remark.
?

For (a): first, show that λ∗\lambda^{*} is translation invariant. Let {rn}n=1∞\left\{ r_{n}\right\}_{n=1}^{\infty } denote the rationals in (0,1](0,1], and let En=F⊕rnE_{n}=F \oplus r_{n}, where FF is the set from Exercise 4.5.3 [the set formed by choosing exactly one element from each equivalence class of ∼\sim on (0,1](0,1], where x∼yx \sim y iff x+˙r=yx \mathbin {\dot{+}} r=y for some rational rr] and ⊕\oplus denotes translation modulo 1.

Problem 4.5.5

Let A∈LkA \in \mathscr {L}^{k} be such that λˉk(A)>0\bar{\lambda }_{k}(A)>0. There exist x,y∈A\boldsymbol {x}, \boldsymbol {y} \in A so that x−y∈Qk\boldsymbol {x}-\boldsymbol {y} \in \mathbb {Q}^{k}.

?
Problem 4.5.6

Let T:R2→RT: \mathbb {R}^{2} \rightarrow \mathbb {R} be such that Tx=x1T \boldsymbol {x}=x_{1} for all x∈R2\boldsymbol {x} \in \mathbb {R}^{2}. That is, TT projects points [perpendicularly] onto the x1x_{1}-axis.

?
(a)

Such a mapping is linear and singular.

(b)

There exists A∈B(R2)A \in \mathscr {B}\left(\mathbb {R}^{2}\right) such that (i) TATA, viewed as a subset of R\mathbb {R}, is not in B\mathscr {B}, while (ii) TATA, viewed as a subset of R2\mathbb {R}^{2}, is in B(R2)\mathscr {B}\left(\mathbb {R}^{2}\right).

Remark.
?

Consider the set E×{0}E \times \left\{ 0\right\}, where EE is the one-dimensional form of the set constructed in this section.

Problem 4.5.7

For any linear mapping TT [singular or nonsingular] defined on Rk\mathbb {R}^{k}, A∈LkA \in \mathscr {L}^{k} forces TA∈LkTA \in \mathscr {L}^{k}.

?
Remark.
?

The result has been established for nonsingular TT (Claim 11(ii) of Section 4.4: if TT is a nonsingular transformation, then for every A∈LkA \in \mathscr {L}^{k} we have TA∈LkTA \in \mathscr {L}^{k} and λˉk(TA)=∣det⁡(T)∣λˉk(A)\bar{\lambda }_{k}(TA)=\left|\operatorname {det}\left(T\right)\right| \bar{\lambda }_{k}(A)). For singular TT, observe that TA⊆TRkTA \subseteq T\mathbb {R}^{k}, which is a hyperplane, hence a Borel set of λk\lambda_{k}-measure zero.