Riemann Integrable Function

The concept of finding the area under a curve is a cornerstone of calculus, intuitively understood by slicing the area into thin rectangles and summing them up. The Riemann integral, formalized by Bernhard Riemann, provides the rigorous mathematical framework for this idea. However, this simple process encounters profound challenges when dealing with functions that are not smooth and continuous. This raises a critical question: what are the precise conditions under which a function can be integrated, and what happens when those conditions are not met?

This article embarks on a journey to answer that question. We will first delve into the core ​​Principles and Mechanisms​​ of Riemann integration, dissecting how it tames functions with various types of discontinuities and culminating in the elegant and powerful Lebesgue's Criterion. Following this, under ​​Applications and Interdisciplinary Connections​​, we will explore how this theory serves as a powerful lens in analysis and a bridge to other fields like number theory, while also probing its critical failures, particularly its inability to reliably handle limits, which ultimately reveals the necessity for a more robust theory of integration.

Imagine you want to find the area of a shape with a curvy top. A beautifully simple idea, dating back to the ancient Greeks but perfected by Bernhard Riemann in the 19th century, is to slice the area into a collection of thin vertical rectangles, calculate the area of each, and add them all up. If your function is a nice, smooth curve, you can imagine that as you make your rectangular slices thinner and thinner, their total area will get closer and closer to the "true" area under the curve.

This is the heart of the ​​Riemann integral​​. We try to trap the area between two approximations. For each thin slice, we can draw a rectangle whose height is the lowest value the function takes in that slice (this gives us the ​​lower sum​​), and another rectangle whose height is the highest value (giving the ​​upper sum​​). The true area, if it exists, must be somewhere between these two sums. A function is called ​​Riemann integrable​​ if, as we make our slices infinitesimally thin, the lower sum and the upper sum squeeze together and converge to the very same value. For a vast landscape of functions that we encounter in physics and engineering—continuous functions, for instance—this process works perfectly.

But the world of mathematics is filled with more exotic creatures than just smooth, continuous curves. The really interesting question, the one that opens the door to a deeper understanding, is not when this method works, but when it fails, and why. This is where our journey of discovery begins.

A function that jumps around is a challenge for our slicing method. If a function has a "jump" discontinuity, what height should we pick for the rectangle in the slice containing the jump? The beauty of Riemann's method is that for some functions, it doesn't matter!

Consider a simple step function, one that is constant for a while and then suddenly jumps to a new value. It has a finite number of these jumps. On almost all of our thin slices, the function is constant, and the upper and lower rectangles are identical. Only in the few slices where the jumps occur is there a difference. But as we make all slices thinner, the contribution of these few problematic slices to the total area becomes negligible. They are like single threads in a giant tapestry; their influence vanishes in the grand scheme of things.

But what if we have infinitely many discontinuities? Surely that must break everything. Let's look at a curious function: let $f(x) = 1$ if $x$ is the reciprocal of a positive integer (like $1, 1/2, 1/3, \dots$), and $f(x) = 0$ for all other points in the interval $[0, 1]$. This function has an infinite number of spikes. Yet, it is perfectly Riemann integrable, and its integral is zero! How can this be? The trick is that these discontinuities, while infinite in number, "bunch up" near the point $x=0$. For any given level of precision, we can isolate the vast majority of these spikes within a tiny region around zero, whose contribution to the area can be made as small as we please. The remaining spikes are finite in number and scattered, and just like with the step function, their influence vanishes as our slices get thinner.

This tells us something profound: the number of discontinuities isn't the whole story. Their arrangement matters. Even a function that oscillates infinitely, like $f(x) = \sin(1/x)$ near the origin, can be integrable. While its graph goes berserk as $x$ approaches zero, it is technically only discontinuous at the single point $x=0$. A single point has no "width," so it contributes nothing to the area, and the integral is well-behaved.

This leads to a powerful generalization. Any function on a closed interval that is ​​monotonic​​—meaning it only ever goes up or only ever goes down—is always Riemann integrable. Such a function can have jumps, but it can't oscillate wildly. It turns out that the set of its jump points must be at most "countable," meaning we can list them out one by one, just like we did for the points $1/n$. And as we've seen, such well-behaved infinite sets of discontinuities can be tamed.

We've seen that finite and even some infinite sets of discontinuities are permissible. So, what is the master rule that separates the integrable sheep from the non-integrable goats? The answer was provided by Henri Lebesgue, and it is one of the most elegant and powerful ideas in all of analysis.

​​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​​.

What on earth does "measure zero" mean? Intuitively, a set has measure zero if it is so thin and sparse that you can cover all of its points with a collection of tiny intervals whose total length can be made as small as you desire—arbitrarily close to zero. Think of it as a fine "dust" of points. A single point has measure zero. A finite collection of points has measure zero. Even a countable collection of points, like the set of all rational numbers ($\mathbb{Q}$), has measure zero! We can place the first rational in an interval of length $\epsilon/2$, the second in an interval of length $\epsilon/4$, the third in one of length $\epsilon/8$, and so on. The total length of all these intervals is $\epsilon$, which we can make as small as we like.

This criterion is a magic key that unlocks the mystery of our previous examples. The functions with a finite or countable number of discontinuities are integrable because those sets of points have measure zero. Now, let's test this key on a truly bizarre lock: Thomae's function, which is sometimes called the "popcorn function." Define a function that is $1/q$ if $x=p/q$ is a rational number in lowest terms, and $0$ if $x$ is irrational. This function has the mind-bending property of being discontinuous at every single rational number and continuous at every single irrational number. Since the rationals are dense, this function "jumps" everywhere! And yet, because the set of all rational numbers is countable and has measure zero, Thomae's function is beautifully Riemann integrable.

To drive the point home, we can go even further. Consider the infamous Cantor set, a fractal constructed by repeatedly removing the middle third of intervals. This process leaves behind a set of points that is ​​uncountable​​—it contains more points than the set of all integers or even all rational numbers—yet, miraculously, its total length, its "measure," is zero. If we define a function to be $7$ on the Cantor set and $2$ everywhere else, this function is discontinuous on an uncountable set of points. But since that set has measure zero, the function is still Riemann integrable! This is the ultimate testament to the power of Lebesgue's idea: it is not about counting the discontinuities, but about measuring them.

With such a powerful criterion, one might think we have conquered the concept of integration. But the world of Riemann integrable functions, for all its breadth, has sharp and surprising boundaries. It is a delicate and, in some sense, incomplete world.

First, there's a hard wall you cannot pass: a function must be ​​bounded​​ to be Riemann integrable. If a function shoots off to infinity, say like $f(x) = 1/\sqrt{x}$ near $x=0$, then in any slice of the area that includes the origin, the "highest point" is infinity. The upper sum will always be infinite, and the method fails from the start.

Second, what happens if the set of discontinuities does not have measure zero? Consider the ultimate troublemaker, the ​​Dirichlet function​​, defined as $D(x) = 1$ if $x$ is rational and $D(x) = 0$ if $x$ is irrational. In any slice of the interval, no matter how microscopically thin, there will always be both rational and irrational numbers. The lowest point is always $0$ and the highest point is always $1$. The lower sum for the entire interval will always be $0$, and the upper sum will always be $1$. They never get closer. They are stuck. The function is discontinuous at every point, and this set—the entire interval—does not have measure zero. The Dirichlet function is not Riemann integrable.

The final boundary is the most subtle and profound. The space of Riemann integrable functions is structurally fragile.

• ​​Fragile under Composition:​​ If you take two Riemann integrable functions, their sum and product are also Riemann integrable. But what about composing them, $f(g(x))$? One might expect this to work too. It doesn't. We can take the perfectly integrable Thomae's function and compose it with a simple integrable step function, and the result is the monstrous, non-integrable Dirichlet function. It's as if by mixing two harmless chemicals, you create a powerful explosive.

• ​​Fragile under Limits:​​ This is perhaps the most damning flaw. Imagine a sequence of functions, $f_1, f_2, f_3, \dots$, where each one is simple and easily Riemann integrable. What if this sequence converges, point by point, to a limit function $f$? Surely $f$ must also be integrable? The answer, shockingly, is no. We can construct a sequence of simple step functions, each with only a finite number of jumps, that converge pointwise to the non-integrable Dirichlet function. Each step on our ladder is safely in the "Riemann world," but the limit they are approaching is outside of it.

This means the space of Riemann integrable functions is not ​​complete​​. It's like the rational numbers, which are full of "holes"—for example, you can have a sequence of rational numbers that gets closer and closer to $\sqrt{2}$, but $\sqrt{2}$ itself is not a rational number. Similarly, we can have a sequence of Riemann integrable functions that gets closer and closer (in a specific sense of "distance" between functions) to a limit, but that limit function is not itself Riemann integrable.

These are not just mathematical parlor tricks. The inability to reliably take limits of functions and guarantee the limit is also integrable is a serious handicap for physics, probability theory, and advanced engineering. The world of Riemann, for all its intuitive beauty, is just not robust enough. It pointed the way, but a new, more powerful theory was needed—a theory that could handle these wilder functions and patch the holes in the fabric of integration. This necessity was the mother of invention for the Lebesgue integral, a story for our next chapter.

So, we have dutifully assembled the machinery of the Riemann integral. We've defined our partitions, our upper and lower sums, and we've squeezed functions between them to capture that single, unique number we call the integral. It's an elegant construction, to be sure. But the natural question to ask, the one a physicist or an engineer or simply a curious person should always ask, is: "So what?" What grand games can we play with this new tool? Where does it take us?

You might expect that the story of applications is one of calculating areas, volumes, work, and other physical quantities. And you would be right—that is the bread and butter of introductory calculus, the vital first step. But the true power of this idea, its deeper beauty, lies not just in getting answers, but in providing a new way to think. The Riemann integral is not just a calculator; it's a microscope for examining the intricate structure of functions, a bridge connecting disparate fields of mathematics, and, perhaps most profoundly, a stepping stone that reveals its own limitations and points the way toward an even more powerful theory.

One of the first deep insights the theory of integration gives us is a surprisingly sharp criterion for what makes a function "well-behaved" enough to be integrated. A bounded function on a closed interval is Riemann integrable if and only if its "bad spots"—its points of discontinuity—are negligible. And what does "negligible" mean? It means that the set of all these discontinuities has a total "length," or measure, of zero.

This single idea, the Lebesgue Criterion for Riemann Integrability, is a remarkable lens. It tells us that our intuition about what makes a function "jagged" or "pathological" can sometimes be misleading. Consider a function built upon a strange, dusty set like the Smith-Volterra-Cantor set—a set constructed by repeatedly removing middle portions of intervals, but in such a stingy way that the final set, while being nowhere dense, still has a positive length. We can define a function that is continuous everywhere except at the countably infinite endpoints of the removed intervals. To our eye, a graph of this function might look hopelessly complex, tied to this bizarre fractal dust. And yet, because the set of its discontinuities is merely a countable collection of points, its total measure is zero. Our microscope tells us it's clean! The function is Riemann integrable.

Now contrast this with a different kind of monster. Imagine we take every rational number in the interval $[0, 1]$ and surround it with a tiny open interval. We can do this cleverly so that the total length of all these little intervals adds up to some small number, say, $\epsilon = 0.01$. We then define a function that is $1$ inside this union of intervals and $0$ elsewhere. This set of "on" points is dense—it appears everywhere—but it's mostly empty space. The function is $0$ on a set of measure $1 - \epsilon = 0.99$. It seems like this function is "mostly zero." Yet, where is it discontinuous? It jumps from $0$ to $1$ (or $1$ to $0$) at every boundary point of our collection of intervals. The set of these boundary points turns out to be precisely the places where the function is zero! This set of discontinuities has a measure of $0.99$. It is anything but negligible. Our microscope reveals that this function, despite its simple on/off definition, is profoundly ill-behaved, and it is not Riemann integrable. The lesson is subtle and beautiful: for integrability, what matters is not how complicated a function's definition is, but the "size" of its set of imperfections.

This fine-grained understanding of function structure has consequences in surprising places. Let's take a stroll over to number theory. Consider a sequence of numbers, say the fractional parts of the multiples of an irrational number like $\sqrt{2}$: $\{ \sqrt{2} \}, \{ 2\sqrt{2} \}, \{ 3\sqrt{2} \}, \dots$. If you plot these points on the interval $[0, 1)$, you'll find they never repeat and seem to fill up the interval without any discernible pattern or clumping. We have a name for this behavior: we say the sequence is uniformly distributed. This means that for any subinterval $[a, b)$, the proportion of points from the sequence that fall into it will, in the long run, equal the length of the interval, $b-a$.

How on earth would you prove such a thing? The definition requires checking every possible interval. The trick, it turns out, is to use integration. The condition is equivalent to saying that for any "test" function $f$, the average value of the function at our sequence points converges to the integral of the function over the interval:

But what class of "test" functions do we need to check? Do we need to check all of them? The theory of Riemann integration gives us a spectacular shortcut. If we can prove this equality holds for the simplest functions—indicator functions of intervals—then the very structure of the Riemann integral guarantees it holds for all Riemann integrable functions! Why? Because any Riemann integrable function can be "squeezed" between two step functions (which are just sums of indicator functions). If the averages work for the simple step functions, they must work for the more complex function they are squeezing. The machinery of Darboux sums, that process of approximation by rectangles, provides the logical bridge that allows a result about simple intervals to be generalized into a powerful statement about a vast class of functions, revealing deep truths about the distribution of numbers.

We've seen the power and subtlety of the Riemann integral. It feels robust. It feels right. Now, let's break it.

In the physical sciences, we are constantly dealing with approximations and limits. We model a complex process as the limit of a sequence of simpler ones. A fundamental piece of faith is that if our approximations $f_n$ get closer and closer to the "real" function $f$, then the integral of $f_n$ should get closer and closer to the integral of $f$. We expect to be able to swap limits and integrals: $\lim \int = \int \lim$.

Let's put this faith to the test. Consider a sequence of functions. For $f_1$, we put a spike of height 1 at the first rational number, and it's 0 everywhere else. For $f_2$, we put spikes at the first two rational numbers. For $f_n$, we put spikes at the first $n$ rational numbers. Each of these functions, $f_n$, is discontinuous at only a finite number of points. They are perfectly Riemann integrable, and because the spikes have no width, their integral is always exactly 0. So, the limit of the integrals is clear:

Now, what about the function these $f_n$ are approaching? As $n \to \infty$, we are placing a spike at every rational number. The limit function, let's call it $f(x)$, is the famous Dirichlet function: it's $1$ if $x$ is rational and $0$ if $x$ is irrational. What is the integral of this function? The Riemann integral simply throws up its hands in defeat. In any tiny interval, no matter how small, there are both rational and irrational numbers. The function oscillates so wildly everywhere that the upper sums are always 1 and the lower sums are always 0. The integral does not exist. Our faith is shattered. We have a perfectly reasonable sequence of integrable functions whose integrals converge to 0, but the limit function itself is not even integrable.

This catastrophic failure is not just a mathematician's parlor trick. It is a profound signal that our tool, the Riemann integral, is too fragile. It cannot withstand the kinds of limiting processes that are ubiquitous in modern science. The problem lies in our notion of "closeness." The sequence $f_n$ converged to the Dirichlet function pointwise, but this is a very weak form of convergence.

To fix this, we must ascend to a higher viewpoint, the viewpoint of functional analysis. Here, we think of functions themselves as points in an abstract space. The distance between two functions $f$ and $g$ can be defined in different ways. If we define the distance as the integral of their difference, $d(f, g) = \int_0^1 |f(x) - g(x)| \, dx$, we find that our sequence of spiky functions is a Cauchy sequence. This means the functions in the sequence are getting closer and closer to each other. In a "complete" space, like the real number line, every Cauchy sequence is guaranteed to converge to a point within that space. But our sequence's limit, the Dirichlet function, is not in the space of Riemann integrable functions. It's as if a sequence of rational numbers were converging to $\sqrt{2}$, but we were forbidden from acknowledging that $\sqrt{2}$ exists. The space of Riemann integrable functions is full of "holes".

Is all hope lost? Not quite. If we use a much stricter definition of distance—the uniform distance, $\sup_x |f(x) - g(x)|$—then the space of Riemann integrable functions is complete. A uniform limit of Riemann integrable functions is always Riemann integrable. But this is too restrictive; many important limiting processes in physics and probability are not uniform.

The true resolution lies in forging a new, more powerful tool: the ​​Lebesgue integral​​. The space of Lebesgue integrable functions, $L^1$, is, in essence, the completion of the space of Riemann integrable functions. It fills in the holes. In this larger space, the Dirichlet function is a perfectly valid citizen. Its Lebesgue integral is 0, exactly matching the limit of the integrals of the spiky functions that approached it. The harmony between limits and integrals is restored. This robustness is not a mere mathematical nicety; it is the foundation upon which modern probability theory, quantum mechanics, and signal processing are built.

And so, the story of the Riemann integral finds its beautiful and humbling conclusion. It is a powerful and intuitive tool that serves us well for a vast range of problems. But its greatest application, in a sense, is to show us its own boundaries and to force upon us the discovery of a deeper and more profound way to understand the concepts of measure and integration. It is a classic tale in science: a trusted theory is pushed to its breaking point, and in its failure, it illuminates the path to a grander successor.