Lebesgue's Criterion for Riemann Integrability

The Riemann integral, a cornerstone of calculus, offers a straightforward way to find the area under a curve by summing the areas of vertical rectangles. While this method works perfectly for smooth, well-behaved functions, it encounters profound difficulties with functions that exhibit erratic jumps or chaotic oscillations. This raises a fundamental question: where, precisely, is the dividing line between functions that the Riemann integral can handle and those it cannot? When does the simple idea of summing rectangles break down?

This article addresses this knowledge gap by exploring the landscape of function discontinuities, from simple finite jumps to the "everywhere-discontinuous" chaos of the Dirichlet function. It culminates in the elegant solution provided by Henri Lebesgue. Over the following sections, you will discover the brilliant concept of "measure zero" and how it forms the basis of Lebesgue's masterstroke. The first chapter, "Principles and Mechanisms," will unpack the theory, explaining why counting discontinuities is insufficient and how measuring their collective "size" provides the final answer. Following that, "Applications and Interdisciplinary Connections" will demonstrate the criterion's surprising power in analyzing signals, function transformations, and even bizarre mathematical objects like space-filling curves, revealing it as an indispensable tool across science and engineering.

Imagine you’re trying to find the area under a curve. The method taught by Bernhard Riemann, which you likely learned in calculus, is beautifully simple: slice the area into a forest of thin vertical rectangles and sum their areas. If the function is well-behaved, like a smooth parabola, you can make the rectangles thinner and thinner, and their total area will converge to a single, definite number. This is the ​​Riemann integral​​. It feels robust, universal. But what happens when the curve is not so well-behaved? What if it jumps, wiggles, or vibrates with such ferocity that the tops of our rectangles refuse to settle down? This is where our journey begins—into the wilder territories of functions, seeking a master rule that tells us precisely when this simple idea of summing rectangles works and when it breaks down.

Let’s start with a simple puzzle. If a function is perfectly smooth everywhere except for a single "jump" discontinuity—like a step in a staircase—can we still find its area? Of course! You can trap that single jump within one of our rectangles. As you make that rectangle infinitesimally thin, its contribution to the total area vanishes to nothing. The rest of the curve is well-behaved, so the integral exists.

What about two jumps? Or a hundred? The same logic holds. As long as you have a ​​finite number of discontinuities​​, you can isolate each one in a vanishingly small box, and the Riemann integral remains perfectly well-defined.

Now, you might be tempted to draw a line in the sand right here. Perhaps the rule is that the integral exists only if there are a finite number of "bad" points. But nature is far more subtle. Consider a function that has discontinuities at every point of the form $\frac{1}{n}$ for all positive integers $n=1, 2, 3, \dots$. This is a ​​countably infinite​​ set of jumps! These points pile up closer and closer as they approach zero. And yet, this function is still Riemann integrable. Why? Even though there are infinitely many discontinuities, they are, in a sense, "sparse." They are isolated points that can still be contained within a collection of tiny intervals whose total width can be made as small as we please.

This tells us something profound: simply counting the number of discontinuities—finite versus infinite—is not the right way to think about the problem. A function can survive an infinite number of discontinuities if they are arranged in the "right" way. However, this fragile peace is about to be shattered.

Let’s meet the true villain of this story, the function that breaks the Riemann integral in the most spectacular way possible. It’s called the ​​Dirichlet function​​, $f(x)$, defined on the interval $[0, 1]$:

Try to picture this function. It’s impossible! Between any two rational numbers, there is an irrational one. Between any two irrationals, there's a rational. This means the graph is a chaotic cloud of points, oscillating infinitely between 0 and 1 in every conceivable interval, no matter how small.

When we try to apply Riemann's method here, we hit a wall. To calculate the "upper sum," we take the highest point in each rectangular slice. Since every slice contains a rational number, this highest value is always 1. The total upper sum is always the full area of a $1 \times 1$ box, which is 1. To calculate the "lower sum," we take the lowest point. Since every slice contains an irrational number, this lowest value is always 0. The total lower sum is 0. The upper and lower sums are stubbornly stuck at 1 and 0, and they will never meet, no matter how thin we make the slices. The Riemann integral simply fails to exist. The function is too "discontinuous everywhere" for the method to handle.

So, we have a puzzle. Finite and even some countably infinite sets of discontinuities are fine, but the "everywhere-discontinuous" Dirichlet function is not. The brilliant insight came from the French mathematician Henri Lebesgue. He proposed that the right question isn’t "How many points of discontinuity are there?" but rather, "How much space do the points of discontinuity take up?"

Lebesgue developed a new way to "measure" the size of a set of points, now called the ​​Lebesgue measure​​. For an interval like $[0, 0.5]$, its measure is just its length, $0.5$. But for a scattered set of points, the idea is more clever. Imagine you want to measure the "size" of the rational numbers. You can cover each rational with a tiny interval. Because you can enumerate them (first, second, third, ...), you can make the interval covering the first rational very small, the one covering the second even smaller, and so on. In fact, you can make the total length of all these covering intervals as small as you like—smaller than $0.1$, smaller than $0.001$, smaller than any tiny number $\epsilon > 0$. If the total length can be made arbitrarily small, Lebesgue declared that the set has ​​measure zero​​.

This is a revolutionary idea. A set can be infinite—even densely packed like the rational numbers—and still have a total "length" or measure of zero. It’s like a dust cloud of infinitely many particles, which nevertheless occupies zero volume.

Armed with this powerful new tool, Lebesgue delivered his masterstroke, a single, elegant criterion that solves our puzzle completely.

​​Lebesgue's Criterion for Riemann Integrability:​​ A bounded function on a closed interval is Riemann integrable if and only if the set of its points of discontinuity has Lebesgue measure zero.

This simple statement beautifully explains everything we've observed.

• A function with a finite number of discontinuities? That finite set of points has measure zero. So, the function is integrable.
• A function with a countably infinite set of discontinuities (like at $\{\frac{1}{n}\}$ or all rationals)? That countable set has measure zero. As long as the function is bounded, it is integrable. A classic case is any ​​monotonic function​​ (one that is always non-decreasing or non-increasing); it can be shown that its set of discontinuities is always countable at most, and therefore it is always Riemann integrable.
• The Dirichlet function? It's discontinuous at every point in the interval $[0, 1]$. The set of discontinuities is the entire interval, which has a measure of 1, not zero. Therefore, it is not Riemann integrable.

Lebesgue's criterion is a perfect example of mathematical elegance. It replaces a confusing menagerie of special cases with one unifying principle. If you know that a function is continuous almost everywhere—meaning the "bad" set of discontinuities is negligibly small (measure zero)—then you know it is Riemann integrable.

The world opened up by measure theory is filled with fascinating objects that challenge our intuition. One of the most famous is the ​​Cantor set​​. You build it by starting with the interval $[0, 1]$, removing the open middle third $(\frac{1}{3}, \frac{2}{3})$, then removing the middle thirds of the two remaining pieces, and so on, forever.

What’s left is a strange, dusty set of points. The total length of all the pieces you removed is $\frac{1}{3} + \frac{2}{9} + \frac{4}{27} + \dots$, which is a geometric series that sums to exactly 1. This means the Cantor set that remains, despite containing an uncountably infinite number of points (as many as the entire interval $[0,1]$!), must have a Lebesgue measure of zero.

What does this imply for integrals? It means you can have a function that is discontinuous on the uncountable Cantor set, and it will still be Riemann integrable, provided it's bounded. This definitively proves that "uncountable" does not mean "large" in the sense of measure. The size of a set's infinity (its cardinality) and its "spatial size" (its measure) are two completely different things.

Lebesgue's criterion also provides us with a toolkit for reasoning about more complex functions. But we must be careful, as our intuition can sometimes lead us astray. Let's test our understanding with a few scenarios.

• ​​If $f$ is integrable, is $|f|$ integrable?​​ Yes. The absolute value function is continuous. If $f$ is continuous at a point, then $|f|$ will be too. So, the set of discontinuities of $|f|$ can only be a subset of the discontinuities of $f$. If the latter has measure zero, so does the former.

• ​​If $|f|$ is integrable, is $f$ integrable?​​ No! This is a crucial trap. Consider a function that is $1$ for rational numbers and $-1$ for irrational numbers. This function, much like the Dirichlet function, is discontinuous everywhere and thus not Riemann integrable. However, its absolute value, $|f|$, is the constant function $1$, which is continuous everywhere and perfectly integrable.

• ​​If $f^2$ is integrable, is $f$ integrable?​​ Again, no. The same counterexample works. If $f$ is $1$ on rationals and $-1$ on irrationals, then $f^2$ is the constant function $1$, which is integrable. But $f$ itself is not. An operation like squaring can "heal" the discontinuities in a way that hides the wild behavior of the original function.

• ​​If $f$ and $g$ are not integrable, is it possible for $f+g$ to be integrable?​​ Yes! Let $f$ be the Dirichlet function (1 on rationals, 0 on irrationals) and let $g$ be its opposite (0 on rationals, 1 on irrationals). Neither is integrable. But their sum, $(f+g)(x)$, is always equal to 1. This is a constant function and is perfectly integrable. The "badness" of two functions can perfectly cancel out.

These examples teach us a final, vital lesson. The criterion is a sharp, precise tool. It applies to a function based on the geometric properties of its own unique set of discontinuities. The integrability of a related function, like its square or absolute value, doesn't automatically tell you about the original. You must always return to the fundamental question: what is the measure of the set where this specific function fails to be continuous? In that simple question lies the key to the entire theory of Riemann integration.

Having grappled with the gears and levers of the Lebesgue criterion, we might feel we've been wrestling with angels in the dark. We have this fantastically precise tool, but what is it for? Is it merely a way for mathematicians to create ever-more peculiar functions, like a craftsman building intricate but useless puzzle boxes? Not at all! In science and engineering, we are constantly faced with functions that are not the smooth, well-behaved creatures of a first-year calculus course. They are signals full of static, data with missing points, or physical phenomena that change abruptly. Lebesgue's criterion is our guide, telling us when our old friend, the Riemann integral, can be trusted to give a meaningful answer, and when we must tread more carefully. It’s the art of knowing how much "messiness" we can ignore.

Let’s begin our journey with something simple, almost trivial. Imagine a function that jumps. A light switch is either on or off. The price of a stock ticks up or down in discrete steps. Consider a function as simple as the "ceiling" function, $f(x) = \lceil x \rceil$, which rounds any number up to the nearest integer. As you trace its graph, you glide along flat plateaus, and then, snap, you jump up to the next level every time you cross an integer. Over an interval like $[-2.5, 2.5]$, these jumps happen only at a handful of specific points: $-2, -1, 0, 1, 2$. These are the discontinuities. Our criterion asks: what is the "size"—the measure—of this set of misbehaving points? Well, it's just a finite collection of five points. You can't even draw them with a thick enough pencil to give them any length. Their total length, their measure, is zero. They are like a few specks of dust on a long glass rod. You can still see right through it. The Riemann integral, therefore, has no trouble at all; the function is perfectly integrable.

This seems easy. But what if the dust is infinite? What if we have not five, but infinitely many discontinuities? Surely that must jam the machinery? Let's construct a function that's a bit more mischievous. Imagine a function on $[0,1]$ that is zero everywhere, except that it has a little "blip" at every point of the form $\frac{1}{n}$ for $n=1, 2, 3, \dots$. The set of discontinuities is now the infinite collection $\{1, \frac{1}{2}, \frac{1}{3}, \dots\}$. It has infinitely many points! But look at them. They get closer and closer, piling up near zero, like a trail of breadcrumbs leading to a single spot. Is this infinite collection "large" or "small"? In the world of measure, it's still infinitesimally small. It’s a countable set, and any countable collection of points is like a countable collection of dimensionless dust specks—their total measure is zero. So, as long as our function doesn't shoot off to infinity (a condition we call boundedness), the Riemann integral can handle it just fine. This is a profound idea: infinity comes in different sizes, and for the Riemann integral, a countable infinity of discontinuities is a kind of "tame" infinity it can gracefully ignore.

So, when does the machine finally break? We need to find a type of "dust" that isn't just a collection of discrete points, but something more pervasive. Let's enter a true gallery of rogues. Our first star is the infamous Dirichlet function, $\chi_\mathbb{Q}$, which is $1$ for rational numbers and $0$ for irrational numbers. Try to imagine its graph. Between any two points, no matter how close, you have an infinity of rationals and an infinity of irrationals. The function doesn't just jump; it vibrates with an impossible frenzy between $0$ and $1$ at every single location. It is discontinuous everywhere. The set of its discontinuities is the entire interval $[0,1]$. This is no longer dust; it's a thick, opaque fog. The measure of this set is the length of the interval, which is $1$. This is not zero. And so, as our criterion declares, this function is emphatically not Riemann integrable.

This leads us to a more subtle distinction. What if we have a set of discontinuities that is "full of holes" but is still uncountably infinite? Enter the Cantor set. We start with the interval $[0,1]$ and remove the middle third. Then we remove the middle third of the two remaining pieces. We repeat this process forever. What's left is a strange "dust" of points. It contains no intervals, yet it has as many points as the original line. It's an uncountable infinity of discontinuities for its characteristic function. But what is its measure? At each step, we remove length. The total length removed is $\frac{1}{3} + \frac{2}{9} + \frac{4}{27} + \dots = 1$. The Cantor set, this uncountable infinity of points, has a total length of zero! It is an uncountable set of measure zero. Our criterion, with its beautiful clarity, says that the characteristic function of the Cantor set is Riemann integrable, and its integral is zero.

But wait, the story gets even stranger. We can construct a "fat" Cantor set. We do the same thing—start with $[0,1]$, and iteratively remove middle pieces—but at each step, we remove a smaller and smaller fraction. For example, we might remove an interval of length $\frac{1}{4^k}$ from each of the $2^{k-1}$ pieces at step $k$. If you add up the total length removed, you'll find it's less than one. What remains is a set that, like the standard Cantor set, is nowhere dense and uncountable. But this one has a positive length! It's a "fat dust" that has real bulk. Now, if we consider its characteristic function, the set of discontinuities is the fat Cantor set itself. Since it has positive measure, the Lebesgue criterion delivers its verdict: this function is not Riemann integrable [@problem_to_be_cited]. This is the heart of the matter: the question of Riemann integrability is not about countability, or density, or topology alone. It is, purely and simply, a question of measure.

The criterion also shows its power in the world of function transformations, a world essential to physics and engineering. Suppose you have a signal $f(t)$ that you know is well-behaved enough to be integrated (meaning its discontinuities form a set of measure zero). What happens if you look at the signal on a warped timescale, say by studying the function $g(x) = f(x^k)$ for some integer $k \geq 2$? The transformation $x \mapsto x^k$ squeezes and stretches the domain. Does this create new, problematic discontinuities? The answer is no. A continuous stretching or squeezing of the axis might move the "dust" of discontinuities around, but it cannot turn a set of zero measure into a set of positive measure. If $f$ is Riemann integrable, then $g(x) = f(x^k)$ is too. Integrability is robust under such smooth changes of variables.

Perhaps the most surprising application comes when we compose functions. Can we take a hopelessly non-integrable function and "tame" it? Let's return to the chaotic Dirichlet function, $g_D(x)$, which is $1$ on rationals and $0$ on irrationals. It's the very definition of non-integrable. Now, let's pass its output through a second function, say $f(y) = (2y-1)^2$. What happens? For every rational input $x$, $g_D(x)=1$, and our composite function is $h(x) = f(g_D(x)) = f(1) = (2(1)-1)^2 = 1$. For every irrational input $x$, $g_D(x)=0$, and our composite function is $h(x) = f(g_D(x)) = f(0) = (2(0)-1)^2 = 1$. Look what happened! The output is now $1$ for every input $x$. Our new function, $h(x)$, is a constant function! It is perfectly continuous and thus gloriously Riemann integrable. We fed pure chaos into our machine, but because the second stage of the machine mapped the two chaotic states to the same output, the chaos was neutralized. This is more than a mathematical trick; it's a deep principle. It shows how one system can be insensitive to the noise generated by another, a concept that echoes in signal processing, control theory, and even computing, where errors in an underlying state can be rendered harmless by a higher-level process.

Finally, we push the boundaries into truly modern mathematics, where our intuitions about space are challenged. Imagine a "space-filling curve." This is a monster of a function: a continuous, unbroken line, drawn over a finite time interval, that manages to visit every single point in a two-dimensional square. Now, consider a simple shape in that square, say, two vertical bars, $A = ([0, \frac{1}{3}] \cup [\frac{2}{3}, 1]) \times [0,1]$. Its characteristic function $\chi_A$ is certainly integrable over the square. But what happens if we define a one-dimensional function $f(t)$ by looking at where the space-filling curve $g(t)$ is at time $t$? That is, $f(t) = \chi_A(g(t))$. The function $f(t)$ will be $1$ if the curve is inside the bars and $0$ if it is outside. Is this function $f$ Riemann integrable? The boundary of our shape $A$ consists of a few line segments. In two dimensions, their area is zero. But a peculiar property of some space-filling curves is that the set of times $t$ for which the curve $g(t)$ is on this boundary can have a positive length! The curve spends a non-zero amount of time traversing a shape of zero area. At all these times, the function $f(t)$ is discontinuous. Since the set of discontinuities has positive measure, Lebesgue's criterion tells us that $f(t)$ is not Riemann integrable. This is a stunning result. A composition of a continuous function and a simple integrable function can fail to be integrable. It shows that the concept of "size" or "measure" is subtle and can behave in strange ways when we move between dimensions. It is in these wild territories, where our geometric intuition falters, that the cold, hard logic of Lebesgue's criterion for integrability becomes an indispensable beacon.