Homepage Solution manuals Terence Tao An Introduction to Measure Theory Exercise 1.1.14 (Metric entropy formulation of Jordan measurability)

Exercise 1.1.14 (Metric entropy formulation of Jordan measurability)

Let E d be a bounded set. For each integer n , let E ( E , 2 n ) denote the number of dyadic cubes of sidelength 2 n that are contained in E , and let E ( E , 2 n ) be the numebr of dyadic cubes of sidelength 2 n that intersect E (i.e., metric entropy of E at scale 2 n ). Show that E is Jordan measurable if and only if

lim n 2 dn ( E ( E , 2 n ) E ( E , 2 n ) ) = 0

in which case one has

m ( E ) = lim n E ( E , 2 n ) = lim n E ( E , 2 n ) .

Answers

Note that for any collection C of dyadic cubes, by properties of elementary measure we have m( C) = E 2dn where E = #C. We now show both directions of equivalence.

  • Suppose that lim n2dn (E (E,2n) E (E,2n)) = 0. We use the criteria from Exercise 1.1.5 to prove the Jordan measurability. Pick an arbitrary 𝜖 > 0. By assumption, we can find n large enough so that

    2dn (E (E,2n) E (E,2n)) 𝜖

    We look closely at the definition of metric entropy. Notice that

    • em the collection of all dyadic cubes that intersect E form an elementary outer cover of E. (Pick an arbitrary x E. Then for x = (x1,,xd) we can find integers i1,,id such that i1 x1 2n < i1 + 1,,id xd 2n < id + 1, i.e., there exists a dyadic cube [ i1 2n, i1+1 2n ) × × [id 2n, id+1 2n ) which contains x E, i.e., intersects E.)
    • Similarly, the dyadic cubes that are contained in E by definition form an inner cover of E.

    Since each of the dyadic boxes has a Jordan measure of 2nd, the volume of the outer cover formed by this boxes is 2dn E (E,2n). Similarly, the volume of an inner cover formed by the contained dyadic boxes is 2dn E (E,2n). But the Jordan measure of their set difference is the difference in their Jordan measures; thus, by our choice of n𝜖 it cannot exceed 𝜖, as desired.

  • Now suppose that E is Jordan measurable. We use the following strategy to prove this assertion:

    1.
    item first, we demonstrate that for large enough scale n the measure of dyadic cubes intersecting a box B and a measure of dyadic cubes contained in B approximate B very good;
    2.
    this we extend without difficulty to elementary sets;
    3.
    if elementary sets can be approximated by dyadic cubes, and Jordan measurable sets can be approximated by elementary sets, then Jordan measurable sets can be approximated by dyadic cubes

    Thus, let B = [a(1),b(1)] × × [a(d),b(d)] be a box. We partition B into dyadic cubes along each dimension, i.e., we find integers between i1,,id and j1,,jd such that

    i1 a12n < i1 + 1j1 b12n < j1 + 1 id ad2n < id + 1jd bd2n < jd + 1

    Notices that the following collection of dyadic cubes naturally covers B:

    B¯ := kd,ikj+1 [ k1 2n, k1 + 1 2n ] × × [kd 2n, kd + 1 2n ]

    (Pick some x : ai xi bi from B. Then there exist integers ki xi2n < ki + 1 and by ii ai xi bi < ji + 1 follows ii ki < ki + 1 ji + 1.) The measure of this dyadic cube cover is by additivity:

    m (B¯) = kd,ikj+1 1 2nd = [ l=1d(j l + 1 il)] 1 2nd [ l=1d(b l2n + 1 a l2n + 1)] 1 2nd = l=1db l al + 2 2n

    which can be verified to approximate l=1dbl al arbitrarily close for sufficiently large n𝜖 . Also notice that the approximation error goes down with increasing n (thus, the boundary holds for all n > n𝜖). Similarly, taking tighter borders we obtain a dyadic inner cover of B:

    B̲ := kd,i+1kj [k1 2n, k1 + 1 2n ] × × [kd 2n, kd + 1 2n ]

    Which also approximates B arbitrarily good:

    m (B̲) = kd,i+1kj 1 2nd = [ l=1d(j l il 1)] 1 2nd [ l=1d(b l2n 1 a 2n 1)] 1 2nd = l=1db l al 2 2n

    This naturally extends to elementary sets, since we can partition F into N disjoint boxes Bl and then take 𝜖N precision for each box.
    Now consider the original Jordan measurable set E. Fix an arbitrary 𝜖 > 0 for our limit. By definition of Jordan measurability, we can find an elementary outer cover E¯ and an elementary inner cover E̲ such that both do not differ in measure from E by 𝜖. Now we cover both elementary covers by dyadic covers E¯¯ and E̲̲ of sidelength 2n𝜖, which do not differ from original covers by more that 𝜖2. From this we conclude

    m (E¯¯) m(E) m (E¯) + 𝜖 2 m(E) = 𝜖 2 + 𝜖 2 = 𝜖 m(E) m (E̲̲) 𝜖

    In other words, we have found n𝜖 large enough so that

    2dn𝜖 E(E,2n𝜖 ) m(E) 𝜖

    m(E) 2dn𝜖 E(E,2n𝜖 ) 𝜖

    More generally,

    n𝜖 : n n𝜖 : E(E,2n𝜖 ) m(E) < 𝜖

    n𝜖 : n n𝜖 : m(E) E(E,2n𝜖 ) < 𝜖

    as desired.

User profile picture
2020-03-30 00:00
Comments
  • Seems to me the cover should have a greater measure than the box and the inner cubes should have less.
    Shinkenjoe2024-07-10

If lim n2dn (E(A,n) E(A,n)) = 0 : Let S(A,n) be the set of dyadic cubes of scale 2n that are contained in A, and let S(A,n) be the set of dyadic cubes of scale 2n that intersect A. Then, S(A,n) A S(A,n), and these are elementary sets. Since dyadic cubes of the same scale are always disjoint, we get that

2dnE (A,n) = m ( S(A,n)) J(A) J(A) m ( S(A,n)) = 2dnn E(A,n).

Thus, taking n we get that J(A) = J(A). Moreover, we have that

m(A) = lim n2dnE(A,n) = lim n2dnE (A,n).

If A is Jordan measurable: Let B = I1 × × Id be a box where Ij is the interval with endpoints aj < bj, define - (B)n := [a1 2n,b1 + 2n] × × [ad 2n,bd + 2n], the enlargement of B by 2n in each coordinate direction. - (B)n := (a1 + 2n,b1 2n) × × (ad + 2n,bd 2n), shrinking B by 2n in each coordinate direction.

Under these definitions, if a dyadic cube Dn(z) of scale 2n intersects B then it is contained in (B)n; if Dn(z) is not contained in B then Dn(z) does not intersect (B)n. Finally note that

m ((B)n) m ((B) n) = j=1d (b j aj + 2 2n) j=1d (b j aj 2 2n) = S{1,,d} jS (bj aj) ( (2 2n) d|S| (2 2n) d|S|) 2d 4 2n m(B).

Fix 𝜀 > 0 and let E A F be elementary sets such that m(F) m(A) + 𝜀 and m(E) m(A) 𝜀. Suppose that E = k=1nEk and F = j=1mFj where En,Fj are boxes.

Let Dn(z) be a dyadic cube of scale 2n such that Dn(z) intersects A but is not contained in A. Then, there exists j such that Dn(z) intersects some Fj, so Dn(z) is contained in (Fj) n. Also, Dn(z) is not contained in Ek for all k, so for all k we get that Dn(z) does not intersect (Ek) n. Thus, if we define

(E)n := k=1n (E k) n and (F)n := j=1m (F j) n

then

m ((E)n) = k=1nm ( (E k) n) k=1nm (E k) (1 2d 22n) = m(E) (1 2d 22n) m ((F)n) j=1mm ( (F j) n) j=1mm (F j) (1 + 2d 22n) = m(F) (1 + 2d 22n)

and since for any Jordan measurable set J we have that 2dnE+(J) m(J) 2dnE(J),

2dn (E(A,n) E (A,n)) 2dn (E ((F)n,n) E ((E) n,n)) m ((F)n) m ((E) n) m(F) m(E) + 2d 22n (m(F) + m(E)) 𝜀 + 2d 22n (2m(A) + 𝜀)

Taking n , we have that for all 𝜀 > 0,

0 lim n2dn (E(A,n) E (A,n)) 𝜀

so this limit must be 0 .

User profile picture
2021-12-04 15:47
Comments