Inner Measure

How do we determine the "size" of an object, especially one with a bizarre or infinitely complex boundary? While our intuition serves us well for simple shapes, the mathematical world contains sets so strange that they challenge our very notion of measurement. This leads to a fundamental problem: the potential for ambiguity in assigning a single, definitive value for a set's size. The theory of inner measure provides a powerful solution by working in tandem with its counterpart, the outer measure. Together, they offer a rigorous framework for not only measuring well-behaved sets but also for precisely diagnosing and quantifying the "un-measurability" of more pathological ones.

This article will guide you through this fascinating concept. First, in "Principles and Mechanisms," we will explore the intuitive idea behind inner measure, formalize its definition, and see how its relationship with outer measure provides the key to understanding measurability. Following that, in "Applications and Interdisciplinary Connections," we will uncover how this seemingly abstract idea has profound consequences, revealing the limits of calculus, shaping the axioms of probability theory, and connecting to diverse fields like dynamical systems and combinatorics.

Imagine you're a surveyor tasked with measuring a bizarre, ethereal plot of land. You can't walk inside it freely, but you can try two things. First, you can lay a tarp over it, trying to find the smallest possible rectangular tarp that covers the entire property. This gives you an upper bound on its size—its ​​outer measure​​. But this method is generous; it includes all the nooks and crannies and might overestimate. So, you try a second strategy: from the boundary, you carefully lay down solid, perfectly measurable paving stones inside the plot, trying to cover as much ground as possible without going over the edge. The total area of the largest patch of stones you can lay is a lower bound on its size. This is the spirit of the ​​inner measure​​.

In mathematics, we formalize this wonderfully intuitive idea. After developing the concept of an ​​outer measure​​, $\mu^*(A)$, which gives us an upper estimate for the size of any set $A$, we naturally ask for the corresponding lower estimate. How do we define it? We look for the largest, most substantial, "well-behaved" piece that fits entirely inside our set $A$.

A ​​measurable set​​ is our version of a solid paving stone—a set whose size we can determine unambiguously. The ​​inner measure​​, denoted $\mu_*(A)$, is defined as the supremum, or the least upper bound, of the measures of all the measurable sets we can possibly tuck inside $A$. In essence, it's the measure of the largest possible measurable "core" of the set $A$.

So for any given set, we have two numbers:

1. ​​Outer Measure $\mu^*(A)$​​: The infimum (greatest lower bound) of the measures of all measurable sets that contain $A$. Think of it as the tightest possible "shrink-wrap" around the set.
2. ​​Inner Measure $\mu_*(A)$​​: The supremum (least upper bound) of the measures of all measurable sets that are contained within $A$. Think of it as the largest solid "kernel" within the set.

By definition, for any set $A$, it's always true that $\mu_*(A) \le \mu^*(A)$. The inside measurement can't be bigger than the outside measurement. The really interesting physics, as it were, lies in what happens when these two values are equal—and more tantalizingly, when they are not.

When does a set have a well-defined, unambiguous size? When our two surveyors—the one laying tarps from the outside and the one laying stones from the inside—agree on the number. This is the heart of the modern definition of measurability. A set $A$ is declared ​​Lebesgue measurable​​ if and only if its inner measure equals its outer measure: $\mu_*(A) = \mu^*(A)$ When this equality holds, we drop the asterisks and simply call this common value the ​​measure​​ of $A$, denoted $\mu(A)$. It's a conceptual handshake that confirms the set is not some fuzzy, ill-defined entity.

But what if the handshake fails? Consider a hypothetical non-measurable set $N_0$ on the interval $[0,1)$, which is known to have an inner measure of 0 and an outer measure of 1. Let's build a new set $E$ by taking a scaled-down version of $N_0$ and attaching a simple interval to it: $E = \{x/2 \mid x \in N_0\} \cup [1/2, 1]$. The original set's properties tell us its scaled-down version, let's call it $N$, has $\mu_*(N)=0$ and $\mu^*(N)=1/2$. Now, let's measure $E$.

When we approximate from the inside, the largest measurable piece we can fit is the interval $[1/2, 1]$, which has a measure of $1/2$. The non-measurable part $N$ contributes nothing to the solid core. So, we find $\mu_*(E) = 1/2$. When we approximate from the outside, we must cover both the interval and the "fuzzy" set $N$. The measure of the interval is $1/2$ and the outer measure of $N$ is $1/2$, and it turns out they add up nicely, giving an outer measure of $\mu^*(E) = 1$.

So for our set $E$, we have $\mu_*(E) = \frac{1}{2}$ and $\mu^*(E) = 1$. The inner and outer measures do not agree! This gap, $1 - 1/2 = 1/2$, is a direct measurement of the "non-measurability" of the set. The set $E$ is not well-behaved; it possesses an intrinsic ambiguity, a "fuzz," that our measurement tools have detected.

The most famous example of a non-measurable set is the ​​Vitali set​​, $V$. Using our new tools, we can understand its bizarre nature with stunning clarity. Let's try to find its inner measure. We take any measurable set $K$ that fits inside $V$. Now, here's the magic. Because of how the Vitali set is constructed (by picking one number from each class of reals that differ by a rational), if we take our set $K$ and start translating it by different rational numbers, all the translated copies, $K+q$, remain disjoint from each other.

Imagine you have a shape $K$ with some positive area, $\mu(K) > 0$. If you could make an infinite number of non-overlapping copies of it and squeeze them all into a finite region (say, the interval $[-1, 2]$), you'd have a paradox. The total area would be an infinite sum of a positive number, which must be infinite, yet it's confined to a finite area. The only way to resolve this contradiction is if your original shape had zero area to begin with.

This is exactly what happens with any measurable subset of a Vitali set. The rigorous proof shows that for any measurable set $K \subseteq V$, its measure must be zero. This means the largest measurable core we can fit inside a Vitali set has a measure of zero. $\mu_*(V) = 0$ Yet, it can be shown that the outer measure of this very same set is $\mu^*(V) = 1$. So for the Vitali set, we have the maximum possible discrepancy: $\mu_*(V)=0$ and $\mu^*(V)=1$. It is a set with no measurable "guts" at all, yet it's too substantial to be ignored from the outside. It's like a phantom, all skin and no substance. This "skin" is precisely the difference $\mu^*(V) - \mu_*(V) = 1$. When we encounter a composite object containing a Vitali set, its measurable part is precisely its inner measure, and what's left over is the pure, non-measurable "fuzz" of the Vitali set.

This brings us to a deep and beautiful point about the nature of reality, or at least our mathematical description of it. If a set is non-measurable, is it just a matter of finding a better ruler? Could we, perhaps, design a sequence of increasingly clever measurable sets $M_k$ that "approximate" our non-measurable set $N$ so well that the error—the size of the bits that don't match up, $\lambda^*(N \Delta M_k)$—goes to zero?

The answer is a resounding no. If such a sequence of approximations existed, the set $N$ would be forced to be measurable itself. The existence of a non-zero gap between inner and outer measure is not just a nuisance; it's a fundamental barrier. It implies that the set has an irreducible "fuzziness." It cannot be caged or perfectly described by any sequence of well-behaved sets. A non-measurable set is fundamentally elusive, forever slipping through the grasp of our instruments. It's a permanent and quantifiable ambiguity woven into the fabric of the number line.

You might be tempted to think this is just a strange quirk of the real numbers. But the principle is universal. We can invent strange new universes with different rules of measurement and find the same dilemma.

Consider the universe of natural numbers, $\mathbb{N}=\{1, 2, 3, \ldots\}$. Let's define a peculiar "probability" measure: a set is "large" and has measure 1 if it contains all but a finite number of elements (co-finite), and it's "small" and has measure 0 if it contains only a finite number of elements. What, then, is the measure of the set of even numbers, $E_{even}$?

It's not finite, so its measure can't be 0. But its complement, the odd numbers, is also not finite, so $E_{even}$ can't be co-finite, meaning its measure can't be 1. It falls through the cracks of our system. Let's apply our inner and outer measure tools.

• ​​Outer Measure​​: To cover $E_{even}$ with a "large" set, we must use a co-finite set. The smallest such set is the whole space $\mathbb{N}$. So, the outer measure is 1.
• ​​Inner Measure​​: To fill $E_{even}$ from within with "small" sets, we can only use finite sets. The largest finite subset of the evens is... still just a finite set. The supremum of their measures is 0.

So, even in this discrete world, we find $P_*(E_{even}) = 0$ and $P^*(E_{even}) = 1$. The gap appears again! The concept of an inner/outer measure disagreement as the signature of non-measurability is a deep and universal principle of measurement, not just a geometric curiosity.

Let's return to our familiar geometric space. There's a final, elegant symmetry to this story. Our "inside" approach, finding the inner measure, is often framed as filling our set with ​​compact​​ sets (which in $\mathbb{R}$ are simply closed and bounded sets). Our "outside" approach is equivalent to approximating the set with ​​open​​ sets.

It turns out that for any reasonable (or "regular") finite measure, these two approaches are perfectly dual. The ability to perfectly approximate any set from the inside using compact sets (a property called ​​inner regularity​​) is logically equivalent to the ability to perfectly approximate any set from the outside using open sets (​​outer regularity​​).

To see why, think about a set $A$ and its complement $A^c$. Approximating $A$ from the outside with an open set $U$ is the same as saying that $A^c$ is being approximated from the inside by the closed set $U^c$. In a finite measure space, knowing the measure of the best outer approximation for $A$ tells you the measure of the best inner approximation for $A^c$. This beautiful duality reveals that the concepts of inner and outer measure are not just two independent ideas we came up with. They are two faces of the same coin, a perfectly matched pair that together unlock a deep understanding of the very nature of size and space.

After our journey through the precise mechanics of inner and outer measure, you might be tempted to ask, "So what?" Are these concepts, and the strange non-measurable sets they reveal, just a peculiar treasure for the pure mathematician, a curiosity locked away in an ivory tower? The answer is a resounding "No!"

The gap between a set's inner and outer measure is not a bug; it's a feature. It is a diagnostic tool of incredible power. It tells us, with surgical precision, the exact point where our intuitive notions of length, area, and probability begin to fray. By understanding where the fabric of measurement tears, we learn about the deep and often surprising structure of that fabric itself. Let's see how this single idea—the tension between the "inside" and the "outside" of a set—illuminates a breathtaking range of scientific thought.

Our first stop is the world of calculus. You are familiar with the Riemann integral, the brilliant idea of approximating the area under a curve with a collection of thin rectangles. For well-behaved, continuous functions, this works beautifully. But what happens when we confront it with something truly wild?

Consider the indicator function of our friend, the Vitali set $V$, which we'll call $\chi_V$. This function is $1$ for points inside $V$ and $0$ for points outside it. To find its integral over the interval $[0, 1]$ using Riemann's method is to try to find the "length" of $V$. But a strange sickness afflicts our tools. Since both the Vitali set and its complement are dense, every single sliver of a subinterval, no matter how microscopically thin, contains points where $\chi_V$ is $1$ and points where $\chi_V$ is $0$.

What does this mean for our Riemann sums? The optimist, constructing the upper sum, always takes the maximum value, $1$, for the height of each rectangle, so the total area is stubbornly $1$. The pessimist, building the lower sum, always takes the minimum value, $0$, and their total area is just as stubbornly $0$. The gap between the upper and lower sums never closes. The Riemann integral simply fails to exist; it throws its hands up in despair.

This is where Lebesgue's view, armed with inner and outer measure, rides to the rescue. It doesn't just say the integral fails; it tells us why. The upper Lebesgue integral of $\chi_V$ is precisely the outer measure of $V$, $\mu^*(V)$, and the lower integral is its inner measure, $\mu_*(V)$. For a Vitali set on $[0,1]$, we established that $\mu_*(V) = 0$ and we can construct it such that $\mu^*(V) = 1$. The failure of the integral is a direct reflection of the non-measurability of the set!

This isn't just about simple indicator functions. We can build far more complex functions whose behavior is "poisoned" by an underlying non-measurable set. For instance, we could define a function that behaves like $3x^2$ on the Vitali set and like $-3x^2$ off of it. When we try to calculate its upper and lower Lebesgue integrals, we find they are dramatically different, say $1$ and $-1$, respectively. The gap, the ambiguity, persists. The inner measure acts as a fundamental test: if it doesn't match the outer measure, the bedrock of integration itself becomes unstable.

Probability theory, at its core, is measure theory in a disguise. The sample space is our universe of outcomes, and the "probability" of an event is simply the measure of the set of outcomes corresponding to that event. A core axiom is that the probability of the entire sample space is $1$. But for this logical framework to hold, the "events" we speak of must be measurable sets.

So, what is the probability of an outcome falling within a Vitali set $V$? The question itself is ill-posed. If we try to pin it down from the inside, by looking at measurable events (subsets) contained within $V$, we run into a curious fact: any measurable subset of a Vitali set must have measure zero. This means that the inner measure of $V$ is zero. From this "internal" point of view, the event is impossible. Yet its outer measure can be 1, suggesting from an "external" point of view that it is almost certain. The axioms of probability cannot assign a single, meaningful number. There exist "events" so pathologically constructed that the very notion of chance breaks down for them.

Faced with this, probabilists developed sophisticated tools. Instead of asking for the expectation of a random variable defined on such a set (like our friend $\chi_V$), they ask for its outer expectation. This is found by looking at all the "well-behaved" (i.e., measurable) random variables that are larger than our pathological one and finding the lowest possible expectation among them. For the indicator function of a Vitali set with outer measure 1, this outer expectation turns out to be exactly 1. This doesn't erase the ambiguity—the corresponding inner expectation would be 0—but it provides a rigorous way to bound our uncertainty. The gap between inner and outer measure becomes the fundamental measure of our ignorance.

The story doesn't end on the real number line. These ideas reverberate through many other fields, revealing deep connections.

Think about higher dimensions. If we take our one-dimensional, non-measurable Vitali set $V$ and extend it into a cylinder, creating the set $V \times [0,1]$ in the plane, what is its two-dimensional area, or measure? One might guess the problem becomes hopelessly complicated. But a beautiful theorem from geometric measure theory gives a clear answer: the outer measure of this product is simply the outer measure of $V$ multiplied by the length of the interval $[0,1]$. If $\mu^*(V)=1$, the 2D outer measure of the cylinder is 1. This demonstrates how our one-dimensional analysis provides the building blocks for understanding the geometry of much more complex, higher-dimensional objects. Even when combining a "good" measurable set with a "bad" non-measurable one, the concept of inner and outer measure allows us to make precise statements.

The construction of the Vitali set itself is not an isolated trick. It relies on partitioning the real numbers using the rational numbers. But what if we used a different rule? In the field of dynamical systems, one might study the behavior of repeatedly rotating a point on a circle by an irrational angle, say by $\sqrt{2}$ radians. The set of all points visited from a single starting point is called an "orbit." These orbits also partition the circle. If we once again use the Axiom of Choice to pick exactly one point from each distinct orbit, we create a new kind of non-measurable set. And guess what? It, too, has an inner measure of zero and an outer measure of one. This reveals a grand, unifying principle: non-measurable sets are not a quirk of the rational numbers but a deep consequence of group actions on spaces, a concept central to modern physics and geometry.

Finally, what happens when we start combining these strange sets? Let's take our Vitali set $V$ and add it to the famous Cantor set $C$—that infinitely porous "dust" of points that remains after repeatedly removing the middle third of intervals. The Cantor set is bizarre, but it is measurable and has a total length of zero. What is the measure of the sum $V+C$? This question belongs to the fascinating field of additive combinatorics. The answer is astonishing: the resulting set, $V+C$, is still non-measurable, and its inner measure is still zero. The proof of this surprising fact relies critically on analyzing the inner measure.

From the failure of simple integrals to the limits of probability and the structure of chaotic systems, the concept of inner measure proves its worth. It is the careful, internal auditor of our mathematical world. By listening to what it tells us—by paying attention to the gap between it and the outer measure—we gain a far richer, more honest, and ultimately more beautiful understanding of the universe of numbers and shapes we inhabit.