Rectifiability of sets and measures - University of …toro/Courses/15-16/...References P. Mattila....
Transcript of Rectifiability of sets and measures - University of …toro/Courses/15-16/...References P. Mattila....
Rectifiability of sets and measures
Tatiana Toro
University of Washington
IMPA
February 15, 2016
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 1 / 1
References
P. Mattila. Geometry of sets and measures in Euclidean spaces,Cambridge Stud. Adv. Math. 44, Cambridge Univ. Press, Cambridge,1995.
D. Preiss, Geometry of measures in Rn: distribution, rectifiability, anddensities, Ann. of Math. 125 (1987), 537-643.
D. Preiss, X. Tolsa and T. Toro, On the smoothness of Holderdoubling measures, Calculus of Variations and PDE’s 35 (2009),339-363.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 2 / 1
Geometry of Measures
In the 1920’s Besicovitch studied sets with locally finite “length” in theplane.What is the length of a set?
The basic question: do the infinitesimal properties of the “length” of a setE in the plane yield geometric information about E itself?What are the infinitesimal properties of the ‘length” of a set E? What sortof geometric information can one expect?
This simple question marks the beginning of the study of the geometry ofmeasures and the associated field known as Geometric Measure Theory!
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 3 / 1
Hs : s-dimensional Hausdorff measure in Rm
Let A ⊂ Rm, 0 ≤ s <∞, 0 < δ ≤ ∞
Hsδ(A) := inf
∞∑j=1
(diamCj)s |A ⊂
∞⋃j=1
Cj , diamCj ≤ δ
Hs(A) = lim
δ→0Hsδ(A) = sup
δ>0Hsδ(A).
Properties:
Hs is a Borel measure.
Hm = Lm (m-Lebesgue measure) in Rm.
Hs ≡ 0 for s > m.
Hs(λA) = λsHs(A) for λ > 0.
dimA := inf {0 ≤ s <∞|Hs(A) = 0}
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 4 / 1
Examples
Hausdorff dimension 1 Hausdorff dimensionlog 4log 3
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 5 / 1
Structure Theorem
Besicovitch (1929-1939) Suppose X ⊂ R2 and 0 < H1(X ) <∞. Then
X = Y ∪ Z
Y ⊂⋃∞
i=1 Γi , where Γi is a Lipschitz image of R.
H1(πL(Z )) = 0 for almost every line L ⊂ R2. Here πL denotes theorthogonal projection of R2 onto L.
Federer (1947) Suppose X ⊂ Rm, n ∈ N and 0 < Hn(X ) <∞. Then
X = Y ∪ Z
Y ⊂⋃∞
i=1 Σi , where Σi is a Lipschitz image of Rn
Hn(πL(Z )) = 0 for almost every n-dimensional plane L ⊂ Rm. HereπL denotes the orthogonal projection of Rm onto L.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 6 / 1
Rectifiable sets
A map f : Rn → Rm is Lipschitz if there exists M > 0 such that forx , y ∈ Rn
|f (x)− f (y)| ≤ M|x − y |
Σ = f (Rn) is a Lipschitz image of Rn.
E ⊂ Rm is n-rectifiable, n ∈ {1, · · · ,m} if there exists a family {Σi}iof Lipschitz images of Rn such that
Hn
(E\
∞⋃i=1
Σi
)= 0,
i.e.
E ⊂
( ∞⋃i=1
Σi
)∪ Σ0
with Hn(Σ0) = 0.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 7 / 1
Remarks
By Whitney’s extension theorem E ⊂ Rm is n-rectifiable iff thereexists a family {Σi}i n-dimensional smooth sub-manifolds and a setΣ0 such that
E ⊂
( ∞⋃i=1
Σi
)∪ Σ0
with Hn(Σ0) = 0.
In 1998 B. White showed using a “slicing” argument that Federer’sresult can be deduced from Besicovitch.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 8 / 1
Purely unrectifiable sets
Z ⊂ Rm is n-purely unrectifiable if 0 < Hn(Z ) <∞ andHn(πL(Z )) = 0 for almost every n-dimensional plane L ⊂ Rm.
Example: 4-corner Cantor set
I There exists C > 1 such that for each x ∈ E∞ and r ∈ (0,√
2)
C−1r ≤ H1(E∞ ∩ B(x , r)) ≤ Cr
I For almost every line L in R2, H1(πL(E∞)) = 0.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 9 / 1
Structure Theorem
Besicovitch (1929-1939) Suppose X ⊂ R2 and 0 < H1(X ) <∞. Then
X = Y ∪ Z
Y is 1-rectifiable.
Z is 1-purely unrectifiable.
Federer (1947) Suppose X ⊂ Rm, n ∈ N and 0 < Hn(X ) <∞. Then
X = Y ∪ Z
Y is n-rectifiable.
Z is n-purely unrectifiable.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 10 / 1
Rectifiable measures
A measure µ in Rm is Radon if it is Borel regular and µ(K ) <∞ foreach compact subset K ⊂ Rm.
Federer A Radon measure µ in Rm is n-rectifiable iff there exists afamily {Σi}i of Lipschitz images of Rn such that
µ(Rm\∞⋃i=1
Σi ) = 0
Mattila & Preiss A Radon measure µ in Rm is n-rectifiable iffµ� Hn and there exists a family {Σi}i of Lipschitz images of Rn
such that
µ(Rm\∞⋃i=1
Σi ) = 0
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 11 / 1
Remarks
If E ⊂ Rm Borel with 0 < Hn(E ) <∞, Hn E defined byHn E (A) = Hn(E ∩ A) is a Radon measure. The rectifiablity of Eis equivalent to the rectifiability of Hn E (in the sense of Mattila &Preiss)
How different are these two notions of rectifiability?
I Garnett-Killip-Schul: There exists a 1-rectifiable doubling measure µ(in the sense of Federer) supported in all of Rm. Hence H1 ⊥ µ.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 12 / 1
Garnett-Killip-Schul example in R2
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 13 / 1
First iterations
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 14 / 1
First iterations
Courtesy of M. Badger
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 15 / 1
Properties of the Garnett-Killip-Schul example
sptµ = {x ∈ Rm : µ(B(x , s)) > 0, ∀s > 0} = Rm
µ is doubling, i.e. there exists C > 0 such that for x ∈ sptµ = Rm
and s > 0µ(B(x , 2s)) ≤ Cµ(B(x , s))
If f : R→ Rm and Γ = {(x , f (x)) : x ∈ R} then µ(Γ) = 0.
There exists a family {Σi}i of Lipschitz images of R such that
µ(Rm\∞⋃i=1
Σi ) = 0.
What is the difference between a Lipschitz graph and a Lipschitzimage?
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 16 / 1
Besicovitch question
Let E ⊂ R2 such that 0 < H1(E ) <∞. Suppose that
limr→0
H1(E ∩ B(x , r))
2r= 1 H1 − a.e. x ∈ E . (1)
What can be said about the structure of E?
Examples of sets satisfying (1):
E ⊂ R× {0} Borel, 0 < H1(E ) <∞
E C 1 curve.
Examples of sets not satisfying (1):
4-corner Cantor set E∞.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 17 / 1
Densities
For a Radon measure µ in Rm we say that the n-density at x exists if
limr→0
µ(B(x , r))
rn= θn(µ, x) ∈ (0,∞)
What can be said about a Radon measure µ for which the n-densityexists µ-a.e. x ∈ Rm?
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 18 / 1
Radon measures with n-density in Rm are n-rectifiable
µ = Hn E µ
1928-1938 Besicovitchn = 1, m = 2
1944 Morse & Randolphn = 1, m = 2
E n-rectifiable µ n-rectifiable
1947 Federer: If Hn E Radon andE n-rectifiable then
θn(Hn E , x) = ωn, Hn − a.e. x ∈ E
ωn = measure of the unit ball in Rn
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 19 / 1
Radon measures with n-density in Rm are n-rectifiable
µ = Hn E µ
1928-1938 Besicovitchn = 1, m = 2
1944 Morse & Randolphn = 1, m = 2
1950 Mooren = 1, m ≥ 3
1961 Marstrandn = 2, m = 3
1975 Mattilan ≤ m, n,m ∈ N
1987 Preissn,m ∈ N
E n-rectifiable µ n-rectifiable
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 20 / 1
Theorem (Besicovitch) Let E ⊂ R2 such that 0 < H1(E ) <∞ and
limr→0
H1(E ∩ B(x , r))
2r= 1 H1 − a.e. x ∈ E ,
then E is 1-rectifiable.
Theorem (Preiss) Let µ be a Radon measure in Rm such that
limr→0
µ(B(x , r))
rn= θn(µ, x) ∈ (0,∞)
for µ-a.e. x ∈ Rm. Then µ is n-rectifiable.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 21 / 1
Questions
1 Are there conditions that ensure the Federer-rectifiability of ameasure? (Badger-Schul n = 1).
2 Are there conditions that ensure the Mattila-Preiss-rectifiability of ameasure?
3 What happens when we consider the s-density for s 6∈ N?
4 Does the density ratio µ(B(x ,r))rn encode further regularity for µ?
Questions about infinitesimal structure of a measure.How do we study infinitesimal structure of mathematical objects?
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 22 / 1
Weak convergence of Radon measures
Let µ, µi (i = 1, 2, · · · ) be Radon measures in Rm. The following areequivalent
limi→∞
ˆϕ dµi =
ˆϕ dµ for all ϕ ∈ Cc(Rm)
lim supi→∞
µi (K ) ≤ µ(K ) for each compact set K ⊂ Rm and
lim infi→∞
µi (U) ≤ µ(U) for each open set U ⊂ Rm.
limi→∞
µi (B) = µ(B) for each Borel set B ⊂ Rm with µ(∂B) = 0.
Notation: µi ⇀ µ.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 23 / 1
Compactness of Radon measures
Let {µi}i be a sequence of Radon measures in Rm such that for eachcompact set K ⊂⊂ Rm
supiµi (K ) <∞.
There exists a subsequence {µik}ik and a Radon measure µ such that
µik ⇀ µ.
Remark: The set of Radon measures in Rm can be metrized in sucha way that it becomes a complete metric space and the metricconvergence coincides with the weak convergence.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 24 / 1
Tools to study infinitesimal structures: dilations
The map Ta,r (x) = x−ar maps B(a, r) onto B(0, 1).
ar 1
0
The image of a Radon measure µ under Ta,r is the Radon measureTa,r [µ] defined by
Ta,r [µ](E ) = µ(rE + a)
where rE + a = {rx + a : x ∈ E}. Note
Ta,r [µ](B(0, 1)) = µ(B(a, r)).
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 25 / 1
Tangent measures
A non-zero Radon measure ν is a tangent measure to µ at a ∈ Rm if thereexist sequences {ri}i , ri → 0 and {ci}i , ci > 0 such that
ciTa,ri [µ] ⇀ ν as i →∞,
i.e. ∀ϕ ∈ Cc(Rm)
limi→∞
ci
ˆϕ(
x − a
ri) dµ(x) =
ˆϕ dν.
The set of all tangent measures to µ at a is denoted by
Tan(µ, a).
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 26 / 1
Examples
If A ⊂ Rm Borel measurable, µ = Hm A by Lebesgue densitytheorem for µ a.e. a ∈ Rm
Tan(µ, a) = {cHm : c > 0}.
If f ∈ L1loc(Rm), f > 0 and µ(E ) =´E f (x) dx then for µ a.e. a ∈ Rm
Tan(µ, a) = {cHm : c > 0}.
For ϕ ∈ Cc(Rm)
ci
ˆϕ(
x − a
ri)f (x) dx = ci
ˆϕ(y)f (riy + a)rmi dy
= ci rmi
ˆϕ(y)(f (riy + a)− f (a)) dy
+ci rmi f (a)
ˆϕ(y) dy
· · ·
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 27 / 1
Examples
If A ⊂ Rm Borel measurable, µ = Hm A by Lebesgue densitytheorem for µ a.e. a ∈ Rm
Tan(µ, a) = {cHm : c > 0}.
If f ∈ L1loc(Rm), f > 0 and µ(E ) =´E f (x) dx then for µ a.e. a ∈ Rm
Tan(µ, a) = {cHm : c > 0}.
If Σ ⊂ Rm a C 1 n-dimensional submanifold and µ = Hm Σ then
Tan(µ, a) = {cHn (TaΣ− a) : c > 0}
where TaΣ is the tangent plane to Σ at a.
Mattila constructed a family of examples where the tangents are notunique and none of them are flat either.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 28 / 1
Existence
Preiss (1987) For any Radon measure µ, Tan(µ, a) 6= ∅, µ- a.e.a ∈ Rm.
Let µ be a Radon measure in Rm, if for a ∈ Rm
ca = lim supr→0
µ(B(a, 2r))
µ(B(a, r))<∞ (µ asymptotically doubling at a)
then every sequence {ri}i , ri → 0 contains a subsequence {rik}ik suchthat
µa,rik =1
µ(B(a, rik ))Ta,rik
[µ] ⇀ ν 6= 0 and ν ∈ Tan(µ, a).
Moreover any tangent measure is obtained this way up to a constant.
Tatiana Toro (University of Washington) Historic recollection February 15, 2016 29 / 1