Ma · Mathematics

Why the Points at Distance 1 from a Compact Set Have Measure Zero

Published By the 9to5Science editorial team

At a glance

  1. 1. Define the distance

    Let K be a nonempty compact subset of Euclidean space and consider points x with distance from K exactly 1.

  2. 2. Use differentiability

    The distance function is Lipschitz, so it is differentiable almost everywhere.

  3. 3. Compare gradients

    At a density point of the level set, its gradient must be zero; outside K, a differentiable distance function has gradient norm 1.

  4. 4. Conclude measure zero

    The two gradient facts contradict each other if the level set has positive measure.

Title: Why the Points at Distance 1 from a Compact Set Have Measure Zero Category: mathematics

Let KK be a nonempty compact subset of Rn\mathbb{R}^n, and define its distance function by

f(x)=dist⁡(x,K)=inf⁡p∈K∣x−p∣.f(x)=\operatorname{dist}(x,K)=\inf_{p\in K}|x-p|.

The set in question is the level set

E={x∈Rn:f(x)=1}.E=\{x\in\mathbb{R}^n:f(x)=1\}.

We will show that EE has nn-dimensional Lebesgue measure zero. The argument uses two standard facts: distance functions are Lipschitz, and Lipschitz functions are differentiable almost everywhere.

The distance function is Lipschitz

For any x,y∈Rnx,y\in\mathbb{R}^n, the triangle inequality gives

f(x)≤∣x−y∣+f(y).f(x)\leq |x-y|+f(y).

Interchanging xx and yy shows that ∣f(x)−f(y)∣≤∣x−y∣|f(x)-f(y)|\leq |x-y|. Thus ff is 1-Lipschitz. By Rademacher’s theorem, it is differentiable at almost every point of Rn\mathbb{R}^n.

Compactness ensures that every point xx has a nearest point p∈Kp\in K satisfying ∣x−p∣=f(x)|x-p|=f(x). In fact, a nearest point also exists for any nonempty closed subset of finite-dimensional Euclidean space, but compactness is enough here.

The gradient has norm 1 where distance is positive

Suppose f(x)=r>0f(x)=r>0 and that ff is differentiable at xx. Choose a nearest point p∈Kp\in K, and set u=(x−p)/ru=(x-p)/r, a unit vector. For 0<t<r0<t<r, the point x−tux-tu lies on the segment from xx toward pp. Its distance from KK is at most its distance from pp:

f(x−tu)≤∣x−tu−p∣=r−t.f(x-tu)\leq |x-tu-p|=r-t.

The 1-Lipschitz property also gives f(x−tu)≥f(x)−t=r−tf(x-tu)\geq f(x)-t=r-t. Therefore equality holds. Taking the directional derivative in direction −u-u gives

∇f(x)⋅(−u)=−1.\nabla f(x)\cdot(-u)=-1.

So ∣∇f(x)∣≥1|\nabla f(x)|\geq 1. On the other hand, a differentiable 1-Lipschitz function has gradient norm at most 1. Hence

∣∇f(x)∣=1.|\nabla f(x)|=1.

In particular, this holds at every differentiability point of the level set EE, since its points have distance 1 from KK.

Why a positive-measure level set is impossible

Assume, for contradiction, that EE has positive Lebesgue measure. The Lebesgue density theorem says that almost every point of EE is a density point: near such a point, the proportion of a small ball occupied by EE tends to 1. Rademacher’s theorem says that ff is differentiable almost everywhere. Thus there is a point x∈Ex\in E that is both a density point of EE and a differentiability point of ff.

At a density point of a level set, the derivative of a differentiable function that is constant on that level set must be zero. To see the idea, take any direction vv. At small scale hh, a density-one set comes arbitrarily close to the point x+hvx+hv relative to the scale hh. Thus there are points yh∈Ey_h\in E with yh=x+hv+o(h)y_h=x+hv+o(h). Since f(yh)=f(x)=1f(y_h)=f(x)=1, differentiability implies ∇f(x)⋅v=0\nabla f(x)\cdot v=0. This holds for every vv, so ∇f(x)=0\nabla f(x)=0.

But the previous section showed that ∣∇f(x)∣=1|\nabla f(x)|=1 at every differentiability point of EE. This contradiction proves

Ln(E)=0.\boxed{\mathcal{L}^n(E)=0.}

The proof establishes the claim for every fixed positive distance, not just distance 1: for any r>0r>0, the level set {x:dist⁡(x,K)=r}\{x:\operatorname{dist}(x,K)=r\} has Lebesgue measure zero.

A concrete example

Let KK be the closed disk of radius 2 centered at the origin in R2\mathbb{R}^2. For a point outside the disk, its distance to KK is ∣x∣−2|x|-2. Therefore the points at distance exactly 1 form the circle

∣x∣=3.|x|=3.

A circle has two-dimensional Lebesgue measure zero, even though it has infinitely many points. The theorem shows that the same zero-area conclusion holds for much more irregular compact sets, including ones with no smooth boundary.

In one dimension, the result also has a simple description. The complement of a closed subset of R\mathbb{R} is a union of disjoint open intervals, and there are at most countably many such intervals. Within each bounded gap, the points at a fixed positive distance rr from the set are the two points at distance rr from the endpoints if the gap has length greater than 2r2r, the midpoint if its length is exactly 2r2r, and no points if its length is less than 2r2r. On an unbounded gap there is at most one such point. The level set is therefore countable and has measure zero.

Common mistakes

  • Confusing a level set with a neighborhood. The set of points at distance at most 1 can have positive measure; the result concerns points at distance exactly 1.
  • Assuming the level set is finite or countable in every dimension. It can contain a whole circle or other large lower-dimensional sets. The conclusion is zero nn-dimensional measure, not few points.
  • Using compactness without identifying its role. Compactness guarantees nearest points exist. The proof also works for any nonempty closed set in finite-dimensional Euclidean space, since nearest points exist there as well.

References

  1. Real Analysis: Modern Techniques and Their Applications, Gerald B. Folland
  2. Measure Theory and Fine Properties of Functions, Lawrence C. Evans and Ronald F. Gariepy

Spotted an error? Send a correction. Read how we write and check articles in our editorial policy.