The Existence of Non-Measurable Sets

The intuitive concept of "size"—be it length, area, or volume—is something we take for granted. In mathematics, this idea is formalized through the theory of measure, which seeks to assign a numerical size to every possible set of points. For centuries, it was assumed that any set, no matter how complex, could be measured in a way that respected simple rules: a combined shape's measure is the sum of its non-overlapping parts, and moving a shape doesn't change its size. However, this foundational intuition turns out to be surprisingly flawed. A profound discovery in the early 20th century revealed the existence of "non-measurable sets," mathematical entities that defy our ability to measure them consistently.

This article explores the startling reality of these mathematical "monsters." It addresses the knowledge gap between our intuition about size and the logical consequences of modern set theory. The journey begins by establishing the principles of measure and following the precise recipe used to construct a non-measurable set, revealing why it breaks the rules. From there, we will investigate the far-reaching consequences of their existence, demonstrating how these paradoxical sets act as crucial guardians of mathematical theorems and reveal deep, unexpected connections between geometry, analysis, and abstract algebra.

Imagine you have a string. How long is it? You take out a ruler and measure it. Simple enough. Now, imagine you have a very complicated, wiggly shape drawn on a piece of paper. What's its area? A bit harder, perhaps. You might overlay it with graph paper and count the little squares inside, getting a good approximation. In mathematics, we want to make this idea of "size"—length, area, volume—precise. We call it ​​measure​​.

What properties would we demand of a good, reliable measure? First, the measure of a simple shape, like the length of an interval $[a, b]$, should be exactly what we expect: $b-a$. Second, if we take two shapes that don't overlap and put them together, the measure of the combined shape should be the sum of their individual measures. This is ​​additivity​​. Finally, if we pick up a shape and move it somewhere else without stretching or rotating it (a rigid motion or ​​isometry​​), its measure shouldn't change. This is ​​invariance​​. These rules seem not just reasonable, but completely obvious. They form the bedrock of our intuition about space and size.

For a very long time, mathematicians assumed that any set of points, no matter how bizarre or complicated, could be assigned a measure that obeys these simple rules. The shocking truth, discovered in the early 20th century, is that this is not the case. There exist "unmeasurable" sets—mathematical monsters that defy our intuition about size. Understanding how these creatures are born and what makes them so strange is a fantastic journey into the foundations of modern mathematics.

How do we distinguish a well-behaved, "measurable" set from a pathological, "non-measurable" one? The definitive test was formulated by the mathematician Constantin Carathéodory. The idea is wonderfully intuitive, even if the formula looks a bit formal.

Imagine you have a set $E$ that you want to test. Think of $E$ as a kind of filter or a cookie cutter. The test is to see how this cutter $E$ interacts with any other set, let's call it $A$. The set $E$ "cuts" $A$ into two pieces: the part of $A$ that is inside $E$ ($A \cap E$), and the part of $A$ that is outside $E$ ($A \cap E^c$).

A set $E$ is ​​Lebesgue measurable​​—our gold standard for "well-behaved"—if for every possible test set $A$, it makes a clean cut. That is, the measure of the original set $A$ is exactly the sum of the measures of its two pieces: $\lambda^*(A) = \lambda^*(A \cap E) + \lambda^*(A \cap E^c)$ Here, $\lambda^*$ stands for the ​​outer measure​​, which is our best attempt at assigning a size to any set, even the weird ones. For a measurable set, the outer measure is just its measure.

A set $E$ is ​​non-measurable​​ if it fails this test for at least one set $A$. For a non-measurable set, something bizarre happens: the set acts like a defective cutter that somehow "creates" extra measure out of thin air. For some test set $A$, the sum of the pieces is strictly greater than the whole: $\lambda^*(A) \lt \lambda^*(A \cap E) + \lambda^*(A \cap E^c)$ This inequality is the definitive signature of a non-measurable set. It’s a mathematical paradox in miniature. Imagine cutting a one-foot-long string into two pieces, and finding that the lengths of the pieces add up to, say, 1.75 feet! That’s precisely the kind of strangeness we’re talking about. For instance, one could encounter a hypothetical non-measurable set $V$ inside the interval $[0,1)$ which forces precisely this kind of result, where the measure of the interval (1) is less than the sum of the measures of its parts inside and outside $V$,.

"Alright," you might say, "I believe you that such a thing could exist, but I'd have to see one to really be convinced." Fair enough. Let's follow the recipe for constructing the most famous of these creatures: the ​​Vitali set​​. The recipe has two main steps, one of which involves a rather controversial ingredient.

Step 1: Sorting the Reals

First, we need to organize all the real numbers in the interval $[0, 1)$ into families, or ​​equivalence classes​​. We'll declare two numbers, $x$ and $y$, to be in the same family if their difference, $x-y$, is a ​​rational number​​ (a fraction of two integers). So, for example, $0.1$, $0.1 + 1/2 = 0.6$, and $0.1 - 1/5 = -0.1$ are all in the same family (if we "wrap around" the interval). On the other hand, $\pi-3$ and $1/2$ are in different families, because their difference is irrational.

This process carves up the entire interval $[0, 1)$ into a vast, uncountable number of disjoint families. Each family is a countable set of points, yet its members are sprinkled densely throughout the interval. Think of it like a deck of cards where each suit has infinitely many cards, and the cards of every suit are thoroughly shuffled throughout the entire deck.

Step 2: The Controversial Choice

Here comes the magic wand. We want to create a new set, let's call it $V$, by picking exactly one member from each of these families. The collection of families is uncountably infinite. How can we be sure it's possible to perform this infinite act of choosing, all at once?

We can't prove it's possible from the other basic axioms of mathematics. Instead, we must invoke a powerful and controversial principle: the ​​Axiom of Choice​​ (AC). This axiom simply declares that, given any collection of non-empty sets (our families), a new set can be formed by choosing exactly one element from each of them. It doesn't tell you how to choose; it just guarantees the existence of such a "choice set." It’s an axiom of pure existence. With this axiom, we can officially declare that our set $V$, the Vitali set, exists.

We’ve created our set $V$. Now we put it to the test. Can we assign it a Lebesgue measure, $\lambda(V)$? Let's assume we can and see where it leads. The argument is a beautiful proof by contradiction.

Let's consider what happens when we "translate" our set $V$ by rational numbers. Enumerate all the rational numbers between $-1$ and $1$ as a list $q_1, q_2, q_3, \dots$. Now form the translated sets $V_k = V + q_k = \{v+q_k \mid v \in V\}$.

Two amazing things are true about this collection of sets $\{V_k\}$:

1. They are all ​​disjoint​​. If two of them, say $V_k$ and $V_j$, shared a common point, it would mean $v_1 + q_k = v_2 + q_j$ for some $v_1, v_2$ in $V$. But this rearranges to $v_1 - v_2 = q_j - q_k$, which is a rational number. By the very way we built $V$, it can't contain two different numbers whose difference is rational. So we must have $v_1=v_2$, which means $q_k=q_j$. The translated sets don't overlap.
2. Their ​​union covers​​ the entire interval $[0, 1)$. Any number $x$ in $[0, 1)$ belongs to some family, and that family has a representative, $v$, in our set $V$. This means $x-v$ is some rational number $q$. So $x = v+q$, which means $x$ is in one of our translated sets.

Now, let's try to measure $V$. Suppose its measure is $\lambda(V) = \alpha$.

• ​​Case 1: The measure is positive ($\alpha \gt 0$).​​ Since all the translated sets $V_k$ are just moved-over copies of $V$, they must all have the same positive measure $\alpha$. We have a countably infinite number of these disjoint sets. By the rule of additivity, the measure of their union must be the sum of their individual measures: $\alpha + \alpha + \alpha + \dots = \infty$. But wait—the union of all these sets is contained within the interval $[-1, 2)$, which has a finite measure of $3$. We have a contradiction: the measure can't be both infinite and less than or equal to 3. So, $\alpha$ cannot be positive.
• ​​Case 2: The measure is zero ($\alpha = 0$).​​ If the measure of $V$ is zero, then the measure of every translate $V_k$ is also zero. The measure of their union would be the sum of their measures: $0 + 0 + 0 + \dots = 0$. But this union contains the interval $[0, 1)$, which has measure 1. Another contradiction: the measure can't be 0 if the set contains a region of measure 1.

Both possibilities lead to absurdity. The only escape is to conclude that our initial assumption was wrong. The Vitali set $V$ simply cannot be assigned a measure that is consistent with the rules of additivity and translation invariance. It is non-measurable.

Non-measurable sets are like mathematical ghosts. They exist, but they are ethereal and defy our attempts to pin them down with a single number for their "size". Digging a little deeper reveals even more of their strange nature.

A non-measurable set like $V$ is, in a sense, both big and small at the same time. Its ​​inner measure​​—the size of the largest, most "solid" measurable set you can fit inside it—is zero. This means it's like a fractal dust cloud, with no substantial chunks. Yet its ​​outer measure​​—the size of the smallest measurable set you can use to cover it—is positive. It's substantial enough to cause all the paradoxes we've seen.

It's natural to wonder if the rational numbers were somehow special in our construction. What if we had used integers instead? If we define families by $x \sim y \iff x-y \in \mathbb{Z}$, we could indeed choose a set of representatives. In fact, the interval $[0,1)$ is one such choice! It's perfectly measurable, and its integer translates tile the real line without any paradox. The trick to creating a non-measurable set lies in using a group (like $\mathbb{Q}$) that is ​​countable​​ but also ​​dense​​ in the real numbers. This recipe can be generalized to other groups as well, such as numbers of the form $a+b\sqrt{2}$ where $a,b$ are rational, leading to other, equally strange non-measurable sets.

The existence of these one-dimensional oddities is just the beginning. The same principles, when applied to rotations in three-dimensional space, lead to one of the most astonishing results in all of mathematics: the ​​Banach-Tarski paradox​​. This theorem states that it's possible to take a solid ball, decompose it into a small, finite number of non-measurable pieces, and then reassemble those same pieces, using only rotations and translations, to form two solid balls, each identical to the original!

This sounds like a violation of the conservation of mass, and it would be, if the "pieces" were things you could hold. But they are not. They are infinitely complex, non-measurable point sets, whose existence is guaranteed by the Axiom of Choice. The Banach-Tarski paradox is not a contradiction in mathematics; it is a profound proof that no volume measure can exist that satisfies our intuitive rules (additivity and invariance) and can measure every possible subset of 3D space. If we lived in a mathematical universe where the Axiom of Choice were false, it might be possible for every set to be measurable, and the ghost of Banach-Tarski would be laid to rest.

The discovery of non-measurable sets marks a turning point in our understanding of the infinite. It teaches us that our everyday intuition, forged in a finite world, can be a treacherous guide in the wilder realms of mathematics. These "monsters" force us to be more careful and precise, revealing a universe that is far stranger, and far more interesting, than we ever imagined.

Alright, we’ve been through a rather abstract and mind-bending construction. We’ve used the full force of the Axiom of Choice to conjure up a set that seems to defy reason—a set you can’t 'measure'. A fair question to ask at this point is, "So what?" Is this just a pathological curiosity, a monster lurking in a dark corner of mathematics that we can safely ignore in our day-to-day work? It’s a wonderful question, and the answer is a delightful and resounding no. These non-measurable sets are not just curiosities; they are probes. They are tools that allow us to test the very limits of our mathematical machinery, revealing hidden structures and deep connections we might never have suspected. By studying where our intuition breaks, we learn how our world is truly put together.

Let’s start close to home, in the world of functions and calculus. We like functions to be well-behaved. But what happens if we build a function using one of our non-measurable sets? Suppose we take our Vitali set $V$ on the interval $[0,1)$ and define a simple periodic function: it’s $1$ whenever a number’s fractional part falls into $V$, and $-1$ otherwise. Is this function measurable? Well, for it to be measurable, we'd need to be able to measure the set of points where the function takes a certain value, say, the value $1$. But that set is, by our very construction, a collection of shifted copies of the non-measurable set $V$! It inherits the non-measurability of its parent set. So, the existence of a non-measurable set immediately implies the existence of non-measurable functions. The pathology has spread.

You might think, 'Fine, some functions are badly behaved. I’ll just stick to the nice ones.' But the situation is more subtle. The world of measurable functions is a strange place. Consider a function $f$ that is $1$ on a non-measurable set $A$ and $-1$ everywhere else. As we’ve just seen, this function is non-measurable. Now, what about its square, $f^2$? Well, $(-1)^2$ is $1$, and $(1)^2$ is also $1$. So $f^2(x)$ is just the constant function $1$ for all $x$. A constant function is perfectly continuous and, therefore, perfectly measurable! This is a fascinating lesson: you can take a 'bad' non-measurable function, perform a simple algebraic operation, and end up with a 'good' measurable one. It tells us that the property of measurability is delicate. You can’t always work backwards. It’s like discovering that you can unscramble an egg—a clear sign that the rules of the game are not what you initially thought.

One of the most important roles these 'monsters' play is to stand guard over our most powerful theorems. They show us precisely why the 'fine print'—the hypotheses of a theorem—is not just legalistic nonsense, but the very foundation holding the entire structure up.

Take the celebrated theorem of Fubini, which tells us that for a 'nice enough' function of two variables, we can calculate its volume by integrating slices—it doesn't matter if we slice vertically then horizontally, or vice-versa. This is the workhorse of multivariable calculus and physics. But what does 'nice enough' mean? It means the function must be integrable, which requires it to be measurable. Is this just a technicality? Let's see. Using a non-measurable set, we can build a function $f(x,y)$ on a square that has a bizarre property: every vertical slice and every horizontal slice integrates to zero. So if you were to apply Fubini's theorem blindly, you’d calculate that both iterated integrals are zero. You might conclude the total 'volume' under the function is zero. And yet, the function itself is non-measurable! It's not integrable at all, and the very concept of its 'volume' or double integral is undefined. The non-measurable set allowed us to build a mirage. It showed us that without the guardrail of measurability, Fubini's theorem can lead you completely astray. These sets aren't breaking the theorem; they are illuminating its boundaries.

The same thing happens with the Monotone Convergence Theorem (MCT), another cornerstone of analysis that allows us to find the integral of a limit by taking the limit of integrals. The theorem requires a non-decreasing sequence of measurable functions. What if we drop that requirement? We can construct an increasing sequence of functions, where each function in the sequence is non-measurable. The sequence converges beautifully to a limit function. But guess what? The limit function is also non-measurable, so the conclusion of the MCT—that the limit function is measurable—fails completely. Once again, the non-measurable set acts as a perfect counterexample, demonstrating that the hypothesis of measurability is absolutely indispensable.

Beyond policing our theorems, non-measurable sets reveal profound and often shocking truths about the very nature of space and the concept of measure itself.

Consider the simple act of casting a shadow. If you take a well-defined object in the plane, say a shape with a measurable area, you would expect its shadow, or projection onto an axis, to be a well-defined line segment with a measurable length. Amazingly, this is not always true. It's possible to construct a set in the plane that is Lebesgue measurable—in fact, it can be a part of the $x$-axis, having zero area—but its projection onto that very axis is a non-measurable set!. How can this be? The answer lies in a subtle but crucial property of Lebesgue measure called 'completeness'. The line itself has zero area, and the Lebesgue measure is 'complete', which means any subset of a zero-measure set is declared measurable (with measure zero). Our strange set, $N \times \{0\}$, is a subset of the $x$-axis, so it gets a pass; it's declared measurable with area zero. But when we project it, we just get back the original non-measurable set $N$.

This tells us something deep about the fabric of our measurement system. The process of 'completing' a measure, which seems like a natural way to tidy things up, has strange topological consequences. The need for completion itself can be seen beautifully with another famous set, the Cantor set. This dusty fractal is a closed set, so it's a perfectly respectable 'Borel set', and it has measure zero. However, the collection of all Borel sets has the same cardinality as the real numbers, while the Cantor set has so many points that the collection of its subsets is vastly larger. This means there must be subsets of the Cantor set that are not Borel sets. So we find a non-Borel set hiding inside a Borel set of measure zero! This is why we need to 'complete' the measure, but as we saw, this completion can break our intuition about projections.

These sets also show a stubborn persistence. If you take a non-measurable Vitali set $V$ and add to it, point by point, all the points of the Cantor set $C$, what happens? The Cantor set is like a sparse dust of measure zero. You might think 'smearing' the Vitali set over it might regularize it somehow. But it doesn't. The resulting set $C+V$ is just as non-measurable as the original Vitali set. The property of non-measurability is a stubborn one.

We now arrive at the most spectacular consequence of all: the famous Banach-Tarski paradox. This is the theorem that says you can take a solid ball, cut it into a finite number of pieces, and then, using only rotations and shifts, reassemble those pieces to form two identical copies of the original ball. No stretching, no cheating.

How can this be? Where does the extra volume come from? The secret, of course, is that the 'pieces' in this cutting are non-measurable sets. Our intuition about volume (which is just a word for 'measure') simply does not apply to them. They are so intricately scattered and complex that they don't have a volume. The paradox isn't a contradiction; it's a brilliant demonstration that the idea of volume is not a property of all subsets of space.

But here’s an even deeper mystery. This trick works for a solid ball in three dimensions, but it fails for a flat disk in two dimensions. You cannot duplicate a disk this way. Why the difference? After all, we can construct non-measurable Vitali sets on a circle just as easily as on a line. The key is not just in the existence of the sets, but in the nature of the transformations we are allowed to use—the group of rigid motions.

This is where our story takes a turn into the beautiful world of abstract algebra. The group of rigid motions in the plane (rotations and translations) is what mathematicians call 'amenable'. You can think of this as meaning it's 'tame' or 'well-behaved'. It possesses a kind of averaging property that prevents the wild shuffling needed for the paradox. In stark contrast, the group of rotations in 3D space, $SO(3)$, is 'non-amenable'. It is 'wild'. It contains within it a mathematical structure called a 'free group', which acts like a perfect shuffler with no constraints. This wildness of the 3D rotation group is what allows it to take the non-measurable pieces and rearrange them in such a shocking way. The Banach-Tarski paradox, then, is not just a story about set theory. It is a profound statement about the fundamental difference in the geometric character of rotations in two and three dimensions, a difference captured by the algebraic properties of their corresponding groups.

So, these non-measurable sets, born from a seemingly abstract axiom, turn out to be anything but idle curiosities. They are the ultimate stress test for our mathematical frameworks. They draw the line where our bedrock theorems in analysis hold firm, and where they would otherwise crumble. They reveal unexpected fragility in intuitive ideas like projection, and they unveil deep, beautiful, and startling connections between the logic of sets, the structure of space, and the algebraic nature of symmetry. They teach us a lesson that is central to the scientific endeavor: our intuition is a powerful guide, but its limits are where the most profound discoveries are often made.