Hausdorff measure. Jordan Bell Department of Mathematics, University of Toronto. October 29, 2014

Transcription

1 Hausdorff measure Jordan Bell Department of Mathematics, University of Toronto October 29, Outer measures and metric outer measures Suppose that X is a set. A function ν : P(X) [0, ] is said to be an outer measure if (i) ν( ) = 0, (ii) ν(a) ν(b) when A B, and, (iii) for any countable collection {A j } P(X), ν ν(a j ). A j We say that a subset A of X is ν-measurable if ν(e) = ν(e A) + ν(e A c ), E P(X). (1) Here, instead of taking a σ-algebra as given and then defining a measure on this σ-algebra (namely, on the measurable sets), we take an outer measure as given and then define measurable sets using this outer measure. Carathéodory s theorem 1 states that the collection M of ν-measurable sets is a σ-algebra and that the restriction of ν to M is a complete measure. Suppose that (X, ρ) is a metric space. An outer measure ν on X is said to be a metric outer measure if implies that ρ(a, B) = inf{ρ(a, b) : a A, b B} > 0 ν(a B) = ν(a) + ν(b). We prove that the Borel sets are ν-measurable. 2 That is, we prove that the Borel σ-algebra is contained in the σ-algebra of ν-measurable sets. Theorem 1. If ν is a metric outer measure on a metric space (X, ρ), then every Borel set is ν-measurable. 1 Gerald B. Folland, Real Analysis, second ed., p. 29, Theorem Gerald B. Folland, Real Analysis, second ed., p. 349, Proposition

2 Proof. Because ν is an outer measure, by Carathéodory s theorem the collection M of ν-measurable sets is a σ-algebra, and hence to prove that M contains the Borel σ-algebra it suffices to prove that M contains all the closed sets. Let F be a closed set in X, and let E be a subset of X. Because ν is an outer measure, ν(e) = ν((e F ) (E F c )) ν(e F ) + ν(e F c ). In the case ν(e) =, certainly ν(e) ν(e F ) + ν(e F c ). In the case ν(e) <, for each n let E n = {x E \ F : ρ(x, F ) n 1 }, which satisfies ρ(e n, F ) n 1. Because ρ(e n, E F ) ρ(e n, F ) n 1, the fact that ν is a metric outer measure tells us that ν((e F ) E n ) = ν(e F ) + ν(e n ). (2) Because F is closed, for any x E \ F we have ρ(x, F ) > 0, and hence Therefore E \ F = E n. (3) n=1 E = (E F ) (E F c ) = (E F ) E n = ((E F ) E n ), hence for each n, using this and (2) we have n=1 n=1 ν(e) ν((e F ) E n ) = ν(e F ) + ν(e n ). To prove that ν(e) ν(e F ) + ν(e F c ), it now suffices to prove that lim ν(e n) = ν(e F c ). n Let D n = E n+1 \ E n. For x D n+1 and y X satisfying ρ(x, y) < ((n + 1)n) 1, we have ρ(y, F ) ρ(x, y) + ρ(x, F ) < which implies that y E n. Thus, ρ(d n+1, E n ) 1 n(n + 1) + 1 n + 1 = 1 n, 1 n(n + 1). (4) 2

3 For any n, using (4) and the fact that ν is a metric outer measure, and ν(e 2n+1 ) = ν(d 2n E 2n ) ν(d 2n E 2n 1 ) = ν(d 2n ) + ν(e 2n 1 ) = ν(d 2n ) + ν(d 2n 2 ) + + ν(d 2 ) + ν(e 1 ) n ν(d 2j ), ν(e 2n ) = ν(d 2n 1 E 2n 1 ) ν(d 2n 1 E 2n 2 ) = ν(d 2n 1 ) + ν(e 2n 2 ) = ν(d 2n 1 ) + ν(d 2n 3 ) + + ν(d 3 ) + ν(d 1 ) + ν(e 0 ) n = ν(d 2j 1 ). But E n E so ν(e n ) ν(e), and hence each of the series ν(d 2j) and ν(d 2j 1) converges to a value ν(e). Thus the series ν(d j) converges to a value 2ν(E). But for any n, ν(e \ F ) = ν E n ν(e n ) + ν(d j ). j=n D j Because the series ν(d j) converges, the sum on the right-hand side of the above tends to 0 as n, so j=n ν(e \ F ) lim inf ν(e n) lim sup ν(e n ) ν(e \ F ); n n the last inequality is due to (3), which tells us ν(e n ) ν(e \ F ). Therefore, which completes the proof. We shall use the following. 3 lim ν(e n) = ν(e \ F ) = ν(e F c ), n 3 Gerald B. Folland, Real Analysis, second ed., p. 29, Proposition

4 Lemma 2. Let (X, ρ) be a metric space. Suppose that E P(X) satisfies, X E and that d : E [0, ] satisfies d( ) = 0. Then the function ν : P(X) [0, ] defined by ν(a) = inf d(e j ) : E j E and A E j, A P(X) is an outer measure. We remark that if there is no covering of a set A by countably many elements of E then ν(a) is an infinimum of an empty set and is thus equal to. 2 Hausdorff measure Suppose that (X, ρ) is a metric space and let p 0, δ > 0. Let E be the collection of those subsets of X with diameter δ together with the set X, and define d(a) = (diam A) p. By Lemma 2, the function H p,δ : P(X) [0, ] defined by H p,δ (A) = inf d(e j ) : E j E and A E j, A P(X) is an outer measure. If δ 1 δ 2 then H p,δ1 (A) H p,δ2 (A), from which it follows that for each A P(X), as δ tends to 0, H p,δ (A) tends to some element of [0, ]. We define H p = lim H p,δ and show that this is a metric outer measure. 4 Theorem 3. Suppose that (X, ρ) is a metric space and let p 0. Then H p : P(X) [0, ] defined by H p (A) = lim H p,δ (A), A P(X). is a metric outer measure. Proof. First we establish that H p is an outer measure. It is apparent that H p ( ) = 0. If A B, then, using that H p,δ is a metric outer measure, H p (A) = lim H p,δ (A) lim H p,δ (B) = H p (B). 4 Gerald B. Folland, Real Analysis, second ed., p. 350, Proposition

5 If {A j } P(X) is countable then, using that H p,δ is a metric outer measure, = lim H p A j H p,δ lim = = A j H p,δ (A j ) lim H p,δ(a j ) H p (A j ). Hence H p is an outer measure. To obtain that H p is a metric outer measure, we must show that if ρ(a, B) > 0 then H p (A B) H p (A) + H p (B). Let 0 < δ < ρ(a, B) and let E be the collection of those subsets of X with diameter δ together with the set X. If there is no covering of A B by countably many elements of E, then H p (A B) H p,δ (A B) =. Otherwise, let {E j } E be a covering of A B. For each j, because diam E j δ < ρ(a, B), it follows that E j does not intersect both A and B. Write E = {E aj } {E bj }, where E aj B = and E bj A =. Then A E aj and B E bj, so (diam E j ) p = (diam E aj ) p + (diam E jb ) p H p,δ (A) + H p,δ (B). This is true for any covering of A B by countably many element of E, so H p,δ (A B) H p,δ (A) + H p,δ (B). The above inequality is true for any 0 < δ < ρ(a, B), and taking δ 0 yields completing the proof. H p (A B) H p (A) + H p (B), We call the metric outer measure H p : P(X) [0, ] in the above theorem the p-dimensional Hausdorff outer measure. From Theorem 1 it follows that the restriction of H p to the Borel σ-algebra B X of a metric space is a meausure. We call this restriction the p-dimensional Hausdorff measure, and denote it also by H p. It is straightforward to verify that if T : X X is an isometric isomorphism then H p T = H p. In particular, for X = R n, H p is invariant under translations. 5

6 We will use the following inequality when talking about Hausdorff measure on R n. 5 Lemma 4. Let Y be a set and (X, ρ) be a metric space. If f, g : Y X satisfy then for any A P(Y ), ρ(f(y), f(z)) Cρ(g(y), g(z)), y, z Y, H p (f(a)) C p H p (g(a)). Proof. Take δ > 0 and ɛ > 0. There are countably many sets E j that cover g(a) each with diameter C 1 δ and such that (diam Ej ) p H p (g(a)) + ɛ. Let a A. There is some j with g(a) E j, so a g 1 (E j ) and then f(a) f(g 1 (E j )). Therefore the sets f(g 1 (E j )) cover f(a). For u, v f(g 1 (E j )), there are y, z g 1 (E j ) with u = f(y), v = f(z). Because g(y), g(z) E j, hence ρ(u, v) = ρ(f(y), f(z)) Cρ(g(y), g(z)) Cdiam E j, diam f(g 1 (E j )) Cdiam E j. Since the sets f(g 1 (E j )) cover f(a) and each has diameter Cdiam E j δ, H p,δ (f(a)) (diam f(g 1 (E j ))) p C p (diam E j ) p C p (H p (g(a)) + ɛ). This is true for all δ > 0, so taking δ 0, H p (f(a)) C p (H p (g(a)) + ɛ). This is true for all ɛ > 0, so taking ɛ 0, H p (f(a)) C p H p (g(a)). 3 Hausdorff dimension Theorem 5. If H p (A) < then H q (A) = 0 for all q > p. 5 Gerald B. Folland, Real Analysis, second ed., p. 350, Proposition

7 Proof. Let δ > 0. Then H p,δ (A) H p (A) < Let {E j } be countably many sets each with diameter δ such that A E j and (diam Ej ) p H p,δ (A) + 1 H p (A) + 1. This gives us H q.δ (A) (diam E j ) q = (diam E j ) q p (diam E j ) p δ q p (diam E j ) p δ q p (H p (A) + 1). This is true for any δ > 0 and q p > 0, so taking δ 0 we obtain H q (A) = 0. For A P(X), we define the Hausdorff dimension of A to be inf{q 0 : H q (A) = 0}. If the set whose infimum we are taking is empty, then the Hausdorff dimension of A is. 4 Radon measures and Haar measures Before speaking about Hausdorff measure on R n, we remind ourselves of some material about Radon measures and Haar measures. Let X be a locally compact Hausdorff space. A Borel measure µ on X is said to be a Radon measure if (i) it is finite on each compact set, (ii) for any Borel set E, and (iii) for any open set E, µ(e) = inf{µ(u) : U open and E U}, µ(e) = sup{µ(k) : K compact and K E}. It is a fact that if X is a locally compact Hausdorff space in which every open set is σ-compact, then every Borel measure on X that is finite on compact sets is a Radon measure. 6 Suppose that G is a locally compact group. A Borel measure µ on G is said to be left-invariant if for all x G and E B G, µ(xe) = µ(e). A left Haar measure on G is a nonzero left-invariant Radon measure on G. It is a fact that if µ and ν are left Haar measures on G then there is some c > 0 such that µ = cν. 7 6 Gerald B. Folland, Real Analysis, second ed., p. 217, Theorem Gerald B. Folland, Real Analysis, second ed., p. 344, Theorem

8 5 Hausdorff measure in R n Let m n denote Lebesgue measure on R n. Lemma 6. If E is a Borel set in R n, then H n (E) 2 n m n (E). Proof. Let ɛ > 0 and let {E j } be countably many closed sets that cover E and such that (diam Ej ) n H n (E) + ɛ. The isodiametric inequality (which one proves using the Brunn-Minkowski inequality) states that if A is a Borel set in R n, then ( ) n diam A m n (A). 2 Using this gives 2 n m n (E j ) H n (E) + ɛ. But because the sets E j cover E we have m n (E) m n ( E j ) m n (E j ), so we get m n (E) H n(e) + ɛ 2 n. This expression does not involve the sets E j (which depend on ɛ), and since this expression is true for any ɛ > 0, taking ɛ 0 yields m n (E) H n(e) 2 n. Let Q = { x R n : x 1 1 2,..., x n 1 }. 2 Lemma 7. 0 < H n (Q) <. Proof. For any m 1, the cube Q can be covered by m n cubes q 1,..., q m n of side length 1 m. Let 0 < δ < 1 and let m > 1 δ. The distance from the center of q j to one of the vertices of q j is r = ( ) 2 ( ) n + + = 2m 2m 2m. 8

9 Inscribe q j in a closed ball b j with the same center as q j and radius r. These balls cover Q. Hence H p,δ (Q) (diam b j ) n = (2r) n = (2r) n m n = n n/2. m n Taking δ 0 gives H p (Q) n n/2 <. On the other hand, by Lemma 6, m n H n (Q) 2 n m n (Q) = 2 n > 0. Theorem 8. There is some constant c n > 0 such that H n = c n m n. Proof. R n is a locally compact Hausdorff space in which every open set in R n is σ-compact. Therefore, to show that H n is a Radon measure it suffices to show that H n is finite on every compact set. If K is a compact subset of R n, there is some r > 0 such that K rq. By Lemma 4 and Lemma 7 we get H n (rq) <, so H n (K) <. Therefore H n is a Radon measure. Because H n (Q) > 0, H n is not the zero measure. Any translation is an isometric isomorphism R n R n, so H n is invariant under translations. Thus H n is a left Haar measure on R n. But Lebesgue measure m n is also a left Haar measure on R n, so there is some c n > 0 such that proving the claim. H n = c n m n, 9