← all posts

Vitali Covering Theorem

2025-04-04·5 min read·

A proof of the Vitali covering theorem via Zorn's lemma and a 5B covering argument.

Lemma (Vitali's Covering Lemma)

Let (X,d)(X,d) be a separable metric space. Suppose F\mathcal{F} is any collection of closed balls in XX such that

R=sup{rad(B)BF}<.R = \sup\{\mathrm{rad}(B) \mid B \in \mathcal{F}\} < \infty.

Then there exists a countable subcollection {Bn(xn,rn)}F\{B_n(x_n, r_n)\} \subseteq \mathcal{F} such that

BFBn=1Bn(xn,5rn).\bigcup_{B \in \mathcal{F}} B \subseteq \bigcup_{n=1}^\infty B_n(x_n, 5r_n).

Proof

We use Zorn's Lemma to construct a countably maximal subcollection of F\mathcal{F}.

Theorem (Zorn's Lemma)

Let SS be an ordered set in which every chain has an upper bound. Then SS has at least one maximal element.

Divide F=n=0Fn\mathcal{F} = \bigcup_{n=0}^\infty \mathcal{F}_n, where

Fn={BF2n1R<rad(B)2nR}.\mathcal{F}_n = \{B \in \mathcal{F} \mid 2^{-n-1}R < \mathrm{rad}(B) \leq 2^{-n}R\}.

Let (Pn,)(P_n, \leq) be the ordered set defined by

Pn={PFnAB=, A,BP},P_n = \{\mathcal{P} \subseteq \mathcal{F}_n \mid A \cap B = \varnothing,\ \forall A, B \in \mathcal{P}\},

with the order relation P1P2    P1P2\mathcal{P}_1 \leq \mathcal{P}_2 \iff \mathcal{P}_1 \subseteq \mathcal{P}_2.

Proof.

Step 1: PnP_n has at least one maximal element.

Let {Ti}iI\{\mathcal{T}_i\}_{i \in I} be any chain in PnP_n, and set T=iITi\mathcal{T} = \bigcup_{i \in I} \mathcal{T}_i. Clearly T\mathcal{T} is an upper bound of {Ti}iI\{\mathcal{T}_i\}_{i \in I}. We show that no two elements of T\mathcal{T} intersect, so that TPn\mathcal{T} \in P_n.

Suppose for contradiction that there exist A,BTA, B \in \mathcal{T} with ABA \cap B \neq \varnothing. If AA and BB both lie in a single Ti\mathcal{T}_i, then TiPn\mathcal{T}_i \notin P_n, a contradiction. Otherwise, suppose Ta,Tb{Ti}iI\mathcal{T}_a, \mathcal{T}_b \in \{\mathcal{T}_i\}_{i \in I} contain AA and BB respectively. Since {Ti}\{\mathcal{T}_i\} is a chain, either TaTb\mathcal{T}_a \leq \mathcal{T}_b or TbTa\mathcal{T}_b \leq \mathcal{T}_a, so AA and BB both belong to one of them — contradicting that collection being in PnP_n. Therefore T\mathcal{T} has no two elements intersecting.

By Zorn's Lemma, PnP_n has at least one maximal element, say KnPnK_n \in P_n, for each nNn \in \mathbb{N}.

Step 2: Construction of GG.

For n=0n = 0, let G0=K0G_0 = K_0. For n=1n = 1, let

G1=K1{BFBC=, CG0}.G_1 = K_1 \cap \{B \in \mathcal{F} \mid B \cap C = \varnothing,\ C \in G_0\}.

For n>1n > 1, let

Gn=Kn{BFBC=, CG0Gn1}.G_n = K_n \cap \{B \in \mathcal{F} \mid B \cap C = \varnothing,\ C \in G_0 \cup \cdots \cup G_{n-1}\}.

Set G=n=0GnG = \bigcup_{n=0}^\infty G_n. Then GG is a collection of disjoint closed balls in F\mathcal{F}.

Step 3: GG is countable.

Fix nNn \in \mathbb{N}. Since (X,d)(X,d) is separable, there exists a countable dense set SXS \subseteq X. Let ε=2n1R\varepsilon = 2^{-n-1}R, and let InXI_n \subseteq X be the set of centers of balls in GnG_n.

For each xInx \in I_n with B(x,rx)GnB(x, r_x) \in G_n, we have ε<rx\varepsilon < r_x. Since SS is dense in XX, there exists sxSs_x \in S such that sxB(x,ε)B(x,rx)s_x \in B(x, \varepsilon) \subseteq B(x, r_x).

We claim sxsys_x \neq s_y for all xyx \neq y in InI_n. Indeed, if sx=sys_x = s_y, then B(x,rx)B(y,ry)B(x, r_x) \cap B(y, r_y) \neq \varnothing, contradicting disjointness of GnG_n. So the map xsxx \mapsto s_x is injective from InI_n into SS, hence InI_n is countable, and therefore GnG_n is countable. Since G=n=0GnG = \bigcup_{n=0}^\infty G_n is a countable union of countable sets, GG is countable.

Step 4: Covering estimate.

Let B(x,r)FB(x,r) \in \mathcal{F}. Then there is nNn \in \mathbb{N} such that B(x,r)FnB(x,r) \in \mathcal{F}_n. We show B(x,r)B(c,p)GB(c,5p)B(x,r) \subseteq \bigcup_{B(c,p) \in G} B(c, 5p).

Case 1: B(x,r)KnB(x,r) \notin K_n. Then B(x,r)B(x,r) must intersect some ball in KnK_n, otherwise {B(x,r)}Kn\{B(x,r)\} \cup K_n would be a larger disjoint collection in PnP_n, contradicting maximality of KnK_n. Let B(y,r1)KnB(y, r_1) \in K_n intersect B(x,r)B(x,r).

If B(y,r1)GnB(y, r_1) \in G_n: since R2n1<r1R2nR2^{-n-1} < r_1 \leq R2^{-n}, we have r<2r1r < 2r_1. For all tB(x,r)t \in B(x,r),

d(t,y)d(t,x)+d(x,y)r+(r+r1)<2r1+2r1+r1=5r1.d(t, y) \leq d(t,x) + d(x,y) \leq r + (r + r_1) < 2r_1 + 2r_1 + r_1 = 5r_1.

Hence B(x,r)B(y,5r1)B(x,r) \subseteq B(y, 5r_1).

If B(y,r1)GnB(y, r_1) \notin G_n: there exists B(z,r2)GiB(z, r_2) \in G_i with B(y,r1)B(z,r2)B(y, r_1) \cap B(z, r_2) \neq \varnothing for some i<ni < n. Taking i=n1i = n-1, we have r<r2r < r_2 and r1<r2r_1 < r_2. Then

d(z,y)r1+r22r2,d(z, y) \leq r_1 + r_2 \leq 2r_2,

and for all tB(y,r1)t \in B(y, r_1): d(z,t)2r2+r13r2<5r2d(z,t) \leq 2r_2 + r_1 \leq 3r_2 < 5r_2, so B(y,r1)B(z,5r2)B(y, r_1) \subseteq B(z, 5r_2). Moreover,

d(z,x)d(z,y)+d(y,x)2r2+r1+r4r2,d(z, x) \leq d(z,y) + d(y,x) \leq 2r_2 + r_1 + r \leq 4r_2,

and for all tB(x,r)t \in B(x,r): d(z,t)4r2+r5r2d(z,t) \leq 4r_2 + r \leq 5r_2. Hence B(x,r)B(z,5r2)B(x,r) \subseteq B(z, 5r_2).

Case 2: B(x,r)KnB(x,r) \in K_n. If B(x,r)GnB(x,r) \in G_n we are done. Otherwise there exists B(y,r1)GiB(y, r_1) \in G_i intersecting B(x,r)B(x,r) for some i<ni < n. Taking i=n1i = n-1, we have r<r1r < r_1. For all tB(x,r)t \in B(x,r):

d(t,y)d(t,x)+d(x,y)r+r1+r<3r1<5r1.d(t, y) \leq d(t,x) + d(x,y) \leq r + r_1 + r < 3r_1 < 5r_1.

Therefore B(x,r)B(y,5r1)B(x,r) \subseteq B(y, 5r_1).

In all cases, every ball in F\mathcal{F} is contained in a 55-dilate of some ball in GG, which shows

BFBB(c,p)GB(c,5p).\bigcup_{B \in \mathcal{F}} B \subseteq \bigcup_{B(c,p) \in G} B(c, 5p).

References

[1] L. C. Evans and R. F. Gariepy, Measure Theory and Fine Properties of Functions, Revised Edition, CRC Press.

2025-04-04

comments

loading...

leave a note