Upper and Lower Sums

The ancient problem of measuring objects with curved boundaries, famously tackled by Archimedes, lies at the heart of one of mathematics' most powerful ideas: the integral. While we intuitively grasp the concept of "area under a curve," how can we define this quantity with logical rigor, especially for complex and irregularly behaved functions? This question reveals a foundational gap between geometric intuition and mathematical certainty. This article addresses this gap by delving into the mechanics of upper and lower sums, the fundamental tools used to construct the definite integral.

We will begin by exploring the "Principles and Mechanisms," where we construct these sums using rectangles to create underestimates and overestimates of area, mimicking Archimedes' ancient strategy. You will learn the precise criterion that determines whether a function is "integrable"—that is, whether this squeezing process can successfully pinpoint a single, unambiguous value for the area. We will then move to "Applications and Interdisciplinary Connections," discovering the rich algebraic structure of integrable functions and identifying which classes of functions, from smooth to discontinuous, pass the integrability test. This journey will not only solidify your understanding of integral calculus but also reveal its deep connections to fields like computer science and physics.

How do we measure something with a curved edge? This is an ancient question, one that puzzled the greatest minds of antiquity. Think of finding the area of a circle. Archimedes had a brilliant idea: trap the circle between two polygons. He drew one polygon inside the circle and another outside it. The area of the circle, whatever it was, had to be greater than the area of the inner polygon and less than the area of the outer one. By using polygons with more and more sides, he could "squeeze" the true area with ever-increasing precision.

This beautiful, intuitive idea of trapping an unknown quantity between an underestimate and an overestimate lies at the very heart of integral calculus. It's the central mechanism we're going to explore. Instead of polygons, we'll use a collection of simple rectangles, but the spirit of the game is identical. We will construct two approximations: a "lower sum" of rectangles that fits entirely under our curve, and an "upper sum" of rectangles that completely covers it. The magic happens when we see how these two sums behave as we make our rectangles ever narrower.

Let's begin our journey with the simplest possible scenario. Imagine not a curve, but a perfectly flat, horizontal line defined by the function $f(x) = c$ over some interval $[a, b]$. The "area under the curve" here is just a simple rectangle with width $(b-a)$ and height $c$. We know from elementary geometry that its area is exactly $c(b-a)$. Can our new method confirm this?

First, we do what Archimedes did: we divide our domain. We slice the interval $[a, b]$ into smaller pieces by choosing a set of points called a ​​partition​​, $P = \{x_0, x_1, \dots, x_n\}$, where $a=x_0$ and $b=x_n$. For each small subinterval $[x_{i-1}, x_i]$, we'll build two rectangles. The first, our "underestimate" rectangle, will have a height equal to the minimum value the function takes on that subinterval. The second, our "overestimate" rectangle, will have a height equal to the maximum value.

The total area of the short, underestimating rectangles is called the ​​lower Darboux sum​​, denoted $L(P,f)$. The total area of the tall, overestimating rectangles is the ​​upper Darboux sum​​, $U(P,f)$.

For our constant function $f(x) = c$, this process is almost laughably simple. On any subinterval, what's the minimum value of $f(x)$? It's $c$. What's the maximum value? It's also $c$. There is no variation at all! This means for every single piece of our partition, the short rectangle and the tall rectangle are identical.

When we sum their areas, we find that the lower sum and the upper sum are exactly the same. Both are equal to $c(b-a)$, and this is true for any partition $P$ we could possibly dream up. The area is perfectly "boxed in" from the very start. There is no gap between our overestimate and our underestimate. This provides a crucial baseline: for the simplest case, the method works perfectly and unambiguously.

What happens if the function is not constant? Let's take a simple, increasing function like $f(x) = x^2$ on the interval $[0, 2]$. Because the function is always rising, on any subinterval $[x_{i-1}, x_i]$, the minimum value will be at the left endpoint, $f(x_{i-1})$, and the maximum will be at the right endpoint, $f(x_{i})$.

Let's start with a very crude partition, simply dividing the interval in two: $P = \{0, 1, 2\}$.

• For the first subinterval $[0, 1]$, the minimum height is $f(0)=0$ and the maximum height is $f(1)=1$.
• For the second subinterval $[1, 2]$, the minimum height is $f(1)=1$ and the maximum height is $f(2)=4$.

Calculating the sums, we get a lower sum $L(P,f) = (0 \times 1) + (1 \times 1) = 1$ and an upper sum $U(P,f) = (1 \times 1) + (4 \times 1) = 5$. Our estimate for the true area is trapped somewhere between 1 and 5. That's a pretty big gap! How can we do better?

The answer, again, comes from Archimedes: use more sides, or in our case, a finer partition. Let's add just one more point to our partition, say at $x=0.5$, creating a new, "refined" partition $P' = \{0, 0.5, 1, 2\}$. By splitting the subinterval $[0,1]$, we trim some of the excess area from our upper sum and fill in some of the empty space in our lower sum. If you perform the calculation, you'll find the new upper sum is smaller than before, and the new lower sum is larger.

This reveals a fundamental and beautiful principle. If a partition $P'$ is a ​​refinement​​ of $P$ (meaning $P'$ contains all the points of $P$ and at least one more), then: $L(P, f) \leq L(P', f) \quad \text{and} \quad U(P', f) \leq U(P, f)$ Adding more points to our partition can never make our estimate worse; it can only squeeze the gap from both sides, bringing our lower and upper bounds closer together.

This leads us to the crucial question: can we always make the gap between the upper and lower sums as small as we want, just by making the partition fine enough?

Let's investigate with some well-behaved functions. Consider a linear function $f(x) = kx + d$ on $[0, L]$, partitioned into $n$ equal subintervals. The gap between the upper and lower sums, $U(f, P_n) - L(f, P_n)$, turns out to be $\frac{k L^{2}}{n}$. For the function $f(x)=1/x$ on $[1,2]$, the gap is $\frac{1}{2n}$. Notice the pattern? In both cases, the gap is inversely proportional to $n$, the number of subdivisions. As we increase $n$, making our partition finer and finer, the gap shrinks towards zero!

This is the moment of triumph. When we can make the difference $U(f, P) - L(f, P)$ arbitrarily close to zero, we've succeeded. We say the function is ​​Darboux integrable​​ (or ​​Riemann integrable​​, the concepts are equivalent). This condition is the litmus test for whether a function has a well-defined area underneath it. The common value that both the upper and lower sums are converging towards is the one, true value of the ​​definite integral​​, written as $\int_a^b f(x) \, dx$.

We can be even more precise about what causes the gap. On any given subinterval, the difference between the maximum and minimum value of the function, $M_k - m_k$, is called the ​​oscillation​​ of the function on that piece. The total gap between the upper and lower sums is just the sum of each subinterval's oscillation multiplied by its width. So, a function is integrable if, by choosing narrow enough rectangles, we can make the total effect of these oscillations vanish. For a continuous function, small intervals mean small oscillations, which is why they are so beautifully integrable.

Does this squeeze play always work? Is every function integrable? It would be a simpler world if it were so, but not nearly as interesting. Nature has cooked up some functions that refuse to be pinned down.

Consider a truly bizarre creature, a variation of the ​​Dirichlet function​​. Let's define a function that is $c_1$ if $x$ is a rational number (a fraction) and $c_2$ if $x$ is an irrational number (like $\pi$ or $\sqrt{2}$), with $c_1 > c_2$. Try to visualize this function. It's not a line or a curve; it's an incomprehensible dust of two different values, interwoven so tightly that you can't separate them.

What happens when we apply our method here? Let's take any subinterval from our partition, no matter how microscopically small. A profound property of the real numbers is that every interval contains both rational and irrational numbers. This means that on every single subinterval, the function takes on both the value $c_1$ and the value $c_2$.

So, for every rectangle in our partition:

• The maximum height, $M_i$, is always $c_1$.
• The minimum height, $m_i$, is always $c_2$.

When we calculate the sums, the upper sum is always $c_1(b-a)$, and the lower sum is always $c_2(b-a)$, regardless of the partition. Refining the partition does absolutely nothing. The gap is permanent. It's stuck at $(c_1 - c_2)(b-a)$. The squeeze fails completely. Such a function is ​​not integrable​​. There is no single, unambiguous number that represents the "area" under this chaotic cloud of points. The upper and lower integrals remain stubbornly apart. Even for similar functions where there is some "structure", like $f(x)=x$ for rationals and $0$ for irrationals, the lower sum remains stuck at zero while the upper sum, though it may decrease with refinement, never gets close enough to meet it.

This exploration, from simple lines to chaotic dust, reveals the power and the limits of one of mathematics' most fundamental ideas. The mechanism of upper and lower sums provides a rigorous way to define area, but it also serves as a diagnostic tool, separating the "well-behaved" functions from the "pathological" ones. The success of the method gives us the integral, a cornerstone of modern science. Its failure teaches us about the surprising and intricate structure of the mathematical world itself.

Now that we’ve painstakingly built our machinery of upper and lower sums, you might be tempted to think of it as a rather formal, abstract device—a theoretical curiosity for establishing the existence of an integral. But that would be like admiring a master watchmaker’s tools without ever asking to see the watch! The real beauty of this framework, the true genius of Riemann's approach, is not just in defining the integral, but in allowing us to reason about it. It’s a powerful engine for discovery, letting us understand which functions can be integrated and how integrals behave when we manipulate these functions. It gives us a solid foundation upon which we can build a true "calculus of integrals," explore a veritable zoo of functions, and even build sturdy bridges to the world of computer science.

Let’s start by playing. What happens if we take an integrable function and… do things to it? Suppose we have a function $f(x)$ whose integral on an interval $[a,b]$ we understand. What if we create a new function, $g(x) = f(x) + C$, by simply lifting the entire graph of $f(x)$ by a constant amount $C$? Your intuition screams that the new area should just be the old area plus the area of the added rectangle, which is $C \times (b-a)$. The machinery of Darboux sums confirms this intuition with mathematical certainty. When you shift the function, the supremum and infimum on every little subinterval also shift by exactly $C$, and this constant can be factored out of the sum, leading directly to the expected result.

This is a simple start, but it's the first hint of a deeper structure. The space of integrable functions is remarkably well-behaved under common algebraic operations. Think of it as a club with certain membership rules. If two functions, $f$ and $g$, are members (i.e., they are integrable), who else gets to join?

It turns out the club is quite accommodating. If $f$ is integrable, is its absolute value, $|f|$, also integrable? Taking the absolute value can introduce sharp corners and complexities, so the answer isn't immediately obvious. Yet, by using a fundamental property of numbers—that for any two values $u$ and $v$, the difference between their absolute values, $\big| |u| - |v| \big|$, can never be more than the absolute difference $|u-v|$—we can show that the oscillation of $|f|$ on any small interval is less than or equal to the oscillation of $f$. This elegant little trick guarantees that if the difference between the upper and lower sums for $f$ can be made arbitrarily small, the same must be true for $|f|$.

The club's rules are even more generous. If you have two integrable members, $f$ and $g$, you can take their pointwise maximum, $h(x) = \max(f(x), g(x))$, and the resulting function $h$ is also guaranteed to be integrable. Even more impressively, their product, $(fg)(x) = f(x)g(x)$, is also integrable. And if a function $f$ is integrable and safely bounded away from zero (say, $f(x) \ge \delta > 0$), its reciprocal $1/f$ is also welcome in the club.

This is profound. It means we can take functions we know are integrable—like polynomials or sine waves—and add, subtract, multiply, and (carefully) divide them, and the result remains integrable. We are not just defining an integral; we are building a whole, robust mathematical system.

So, who are the members of this club to begin with? What kinds of functions pass the test of their upper and lower integrals squeezing together?

One of the most well-behaved classes of functions we encounter in nature are monotone functions—functions that are always increasing or always decreasing. Think of the distance you’ve traveled from home as you walk in one direction, or the temperature of a cup of coffee as it cools. It’s deeply satisfying to discover that every monotone function is Riemann integrable. The proof is a thing of beauty. For a uniform partition with $n$ tiny subintervals, the difference between the upper and lower sums telescopes down to a simple expression: $\frac{(b-a)(f(b)-f(a))}{n}$. As we make the partition finer ($n \to \infty$), this difference vanishes, and the trap closes perfectly.

But what about functions that aren't so well-behaved? What about functions with "jumps," or discontinuities? Here, the Riemann integral shows its forgiving nature. Consider a function that is zero everywhere except for a finite number of points, where it spikes to some value. What is its integral? Our intuition might be confused, but the Darboux sums give a clear answer. The lower sum is always zero, because every subinterval, no matter how small, contains points where the function is zero. The upper sum can be made as small as we please by "trapping" each of the few spike points within incredibly narrow intervals. The total contribution of these spikes to the upper sum can be squeezed down to nothing. The upper and lower integrals both converge to zero!

This is a marvelous insight. The Riemann integral doesn't "see" a finite number of points. The area under a single point, or a thousand points, is zero. This idea extends to simple jump discontinuities as well. If a function has a single jump, the only place where the upper and lower bounds of a subinterval will differ is in the one tiny interval that contains the jump. As the partition becomes finer, the width of this single problematic interval shrinks to zero, and its contribution to the overall difference $U(f,P) - L(f,P)$ vanishes. The integral exists and is perfectly well-defined. This is the mathematical crystallization of the idea that "small" sets of misbehavior don't ruin the big picture.

The theory of upper and lower sums doesn't just sit in its own beautiful, isolated world. It forms the bedrock for more advanced concepts and finds surprising echoes in the practical world of computation.

One of the deepest results in analysis tells us that the property of being integrable is preserved under uniform convergence. Imagine you have a very complicated function $f$, perhaps the solution to a differential equation in physics. A common strategy is to approximate it with a sequence of simpler, integrable functions $f_n$ (like polynomials or trigonometric series). But how can we be sure that the integral of our approximation, $\int f_n$, actually gets close to the true integral, $\int f$? If the sequence $f_n$ "settles down" towards $f$ in a sufficiently nice way (this is what uniform convergence means), we can use the upper and lower sum framework to prove that the limit function $f$ must also be integrable. Furthermore, the integral of the limit is the limit of the integrals. This result is a cornerstone that ensures the mathematical methods used throughout science and engineering are on solid ground.

And what about the world inside our computers? A computer cannot perform the infinite limiting process of the Riemann integral. It must approximate. One of the most fundamental methods is the trapezoidal rule, which approximates the area under a curve by a series of trapezoids. Where does this rule come from? It's no mere computational hack. For a given partition, the trapezoidal rule approximation is exactly the arithmetic average of the left-hand and right-hand Riemann sums. And because we know from our theory that for any integrable function, both the left and right Riemann sums converge to the true integral, their average must too. The abstract theory of Riemann sums directly blesses the numerical algorithms that power everything from financial modeling to weather prediction, bridging the gap between the continuous world of ideas and the discrete world of computation.

Finally, understanding a tool means knowing its limits. The entire construction of the Riemann integral is based on partitioning a closed and bounded interval $[a, b]$. If we want to ask about the area under a curve over an infinite range, like $[0, \infty)$, the standard definition simply doesn't apply. You cannot create a finite partition that ends at "infinity". This is not a failure of the theory, but a clarification of its purpose. It tells us we need a new idea—the improper integral—which builds upon the Riemann integral by taking a further limit. It also hints at the existence of even more powerful theories of integration, like the Lebesgue integral, designed to handle even "wilder" functions and more complex sets. Our journey with upper and lower sums, then, is not an end, but a gateway to an even grander mathematical landscape.