Preimage of a Set

In mathematics, a function is often seen as a one-way street: an input yields an output. But what if we were to travel in reverse? Given a specific result, how could we identify every possible starting point that might have led there? This "detective's question" highlights a knowledge gap that a simple inverse function often cannot fill, especially when multiple inputs lead to the same output. The answer lies in the powerful concept of the ​​preimage of a set​​, which shifts the focus from where you are going to where you could have come from. This article explores this fundamental idea, starting with its core "Principles and Mechanisms," where we deconstruct the preimage and its beautifully predictable behavior with set operations. We then proceed to "Applications and Interdisciplinary Connections," revealing how this single concept becomes a master key, unlocking profound insights and providing a unified language for fields like topology, measure theory, and abstract algebra.

Imagine a fantastic machine. You put something in—an "input"—and it spits something out—an "output." A function in mathematics is just like that. You give it a number, say $x=3$, it performs its operation, like $f(x)=x^2$, and out comes the number $9$. Simple enough. But what if we play the game in reverse? What if we have the output and want to know what the input could have been? If I show you the output $9$, you might say the input was $3$. But you'd only be half right! The input could also have been $-3$, because $(-3)^2$ is also $9$.

This reverse question—"given an output, what are all the possible inputs that could have produced it?"—is the central idea of a ​​preimage​​. It's fundamentally a detective's question. We have the evidence, and we're looking for all the possible culprits. Because there can be more than one culprit, we don't ask for the preimage of a single output value, but for the preimage of a set of output values. The answer, in turn, is always a set of inputs.

Let's make this concrete. Suppose you are a network administrator for a global company. You have servers scattered across the globe, and a function, let's call it $f$, that maps each server to its data center location. Now, your boss tells you, "We need to perform maintenance on all servers in London and Sydney." Your task is to find all the servers that need to be shut down. You're not evaluating the function; you're going backward. You have the set of outputs, $B = \{\text{London, Sydney}\}$, and you need to find all the inputs—the servers—that map into this set. This list of servers is precisely the preimage of the set $B$, which we write as $f^{-1}(B)$.

Notice the notation $f^{-1}$. It might look like an inverse function, but it's much more general. An inverse function only exists if every output comes from exactly one input. But our server function is not like that; there might be ten servers in London. The preimage concept doesn't care. It happily collects all inputs that land in the target set, whether it's one, ten, or a billion.

This idea of multiple inputs leading to the same output is crucial. Consider a function that assigns a "complexity score" to a string of code based on the symbols it contains. For instance, let's say the score is $2$ times the number of 'P's plus $5$ times the number of 'Q's. If we ask for all strings that have a score of $10$, we are asking for the preimage of the set $\{10\}$. A little investigation reveals that a string made of five 'P's and no 'Q's works ($2 \times 5 + 5 \times 0 = 10$). But so does a string of two 'Q's and no 'P's ($2 \times 0 + 5 \times 2 = 10$). So the preimage $f^{-1}(\{10\})$ is the set $\{PPPPP, QQ\}$. Two very different-looking inputs produce the exact same output. The preimage reveals this hidden connection.

The real fun begins when we move from discrete items like servers and strings to the continuous realm of real numbers. Imagine our function maps the entire real number line to another real number line. What does a preimage look like now?

Let's take the function $f(x) = \frac{1}{x^2+4}$. This function takes any real number $x$, squares it, adds 4, and takes the reciprocal. The outputs are always positive and never larger than $\frac{1}{4}$. Now suppose we ask: which input numbers $x$ produce an output that falls within the interval $[\frac{1}{20}, \frac{1}{8}]$? We are looking for the preimage $f^{-1}([\frac{1}{20}, \frac{1}{8}])$. To solve this, we work backward through the function's definition, carefully "un-doing" each step with inequalities. What we find is that the inputs that satisfy this condition are not in one continuous piece. Instead, they form two distinct intervals: $[-4, -2]$ and $[2, 4]$. A single, connected interval in the output space corresponds to a fragmented, symmetric set in the input space. The preimage acts like an X-ray, revealing the function's symmetry ($f(x)=f(-x)$) and its stretching and compressing of the number line.

The surprises don't stop there. A function can map a continuous stretch of inputs to a single discrete output. Consider the floor function $f(x) = \lfloor x \rfloor$, which takes a real number and gives the greatest integer less than or equal to it. The preimage of the set $\{3\}$ is not a single point; it's the entire interval $[3, 4)$. All numbers from $3$ up to, but not including, $4$ get squashed down to the integer $3$. We can do this with more complex functions, too. For a function like $f(x) = \lfloor x^2 - x \rfloor$, the question "Which inputs map to the set of integers $\{-1, 0, 1\}$?" leads to a beautiful puzzle. Solving it reveals that the preimage is the entire continuous open interval $(-1, 2)$. A discrete target set in the output can correspond to a continuous chunk of the input line!

This idea readily expands to higher dimensions. Imagine a function on the Cartesian plane, $f(x, y) = |x| + |y|$, which measures the "taxicab distance" of a point from the origin. If we ask for the preimage of the interval $[1, 2]$, we are asking: "What is the set of all points $(x,y)$ whose taxicab distance from the origin is between 1 and 2, inclusive?" The answer is not an annulus (a ring between two circles), which you might get with the standard distance $x^2+y^2$. Instead, it's a stunning geometric shape: the closed region between two concentric squares, tilted at 45 degrees. The preimage paints a picture of the function's structure. In a completely different context, for a bizarre function that maps rational numbers to $1$ and irrational numbers to $-1$, the preimage of the interval $[0, 2]$ is simply the set of all rational numbers, $\mathbb{Q}$. The preimage can be a familiar geometric shape, or it can be a "dust" of points infinitely sprinkled along the number line.

At this point, you might think that these preimages are wild and unpredictable. But here is the most profound and beautiful part of the story. Mathematicians discovered that, in a deep sense, the preimage operation is extraordinarily simple and well-behaved. It follows a few elegant and unbreakable laws. While its cousin, the image operation (going forward from inputs to outputs), can be messy, the preimage is the "good child" of set theory.

What are these laws? Let's say you have two sets of outputs, $B_1$ and $B_2$. The preimage operation plays beautifully with the basic set operations of union, intersection, and complement:

1. ​​Preimages and Unions:​​ The preimage of the union of two sets is the union of their preimages. $f^{-1}(B_1 \cup B_2) = f^{-1}(B_1) \cup f^{-1}(B_2)$. In plain English: The set of inputs that land in either $B_1$ or $B_2$ is, of course, the set of inputs that land in $B_1$ OR the set of inputs that land in $B_2$.

2. ​​Preimages and Intersections:​​ The preimage of the intersection of two sets is the intersection of their preimages. $f^{-1}(B_1 \cap B_2) = f^{-1}(B_1) \cap f^{-1}(B_2)$. In plain English: An input lands in both $B_1$ and $B_2$ if and only if it lands in $B_1$ AND it lands in $B_2$. This property holds not just for two sets, but for any collection of sets, no matter how many.

3. ​​Preimages and Complements:​​ The preimage of the complement of a set is the complement of its preimage. $f^{-1}(B^c) = (f^{-1}(B))^c$. In plain English: The set of inputs that land outside of $B$ is exactly the set of inputs that do not land inside of $B$.

These properties might seem almost trivial, but their power is immense. The corresponding statements for forward images are often false! For example, the image of an intersection is not necessarily the intersection of the images, as we can have two separate input sets whose outputs "collide". The preimage operation doesn't have this "collision" problem; it simply sorts the inputs based on where they land. This reliable, predictable behavior is precisely why preimages are a cornerstone of higher mathematics, especially in the field of topology, where the very definition of a continuous function is elegantly stated using preimages.

Finally, preimages also behave predictably with compositions of functions. If you apply one function $f$ and then another function $g$, taking the preimage of the final result is the same as taking the preimages through $g$ and then $f$, in reverse order: $(g \circ f)^{-1}(U) = f^{-1}(g^{-1}(U))$. It's like taking off your shoes and socks: you must reverse the order in which you put them on.

From tracking down servers to painting geometric masterpieces and underpinning the definition of continuity, the simple act of "looking backward" turns out to be one of the most powerful and unifying ideas in all of mathematics.

You've now seen the machinery of the preimage, the formal definition of "looking backward." You might be thinking, "Alright, a neat bit of set theory. What's the big deal?" The answer, which I hope you'll find as delightful as I do, is that this is not just a definition. It is a key. It is a special lens that, once you learn how to use it, reveals a hidden and profound unity across vast landscapes of science and mathematics. By simply asking, "Where could I have come from to land here?", we unlock a startlingly powerful way of thinking.

Let's embark on a journey through a few of these landscapes and see what the preimage lens reveals.

Our first stop is topology, the mathematical study of shape and space. You've likely encountered the concept of a continuous function, perhaps with the somewhat clunky "epsilon-delta" definition from calculus. It's a perfectly good definition, rigorously nailing down the idea that "nearby inputs give nearby outputs." But it's a bit like describing a beautiful sculpture by listing the coordinates of every point on its surface. It's correct, but it misses the holistic elegance of the form.

The concept of a preimage gives us a much more profound and elegant way to talk about continuity. The idea is this: ​​A function is continuous if and only if the preimage of every open set is an open set.​​

Why is this so powerful? An "open set" is, intuitively, a region that doesn't include its own boundary—like the interval $(0, 1)$ but not $[0, 1]$. The preimage definition says that a function is continuous if, no matter what boundary-less region you pick in the output space, the entire collection of input points that map into it also forms a boundary-less region.

Consider a function that is famously not continuous: the floor function, $f(x) = \lfloor x \rfloor$, which maps any real number to the greatest integer less than or equal to it. Let's look at the preimage of the set $\{3, 4\}$ in the codomain of integers. Because every set of integers is considered "open" in the discrete topology, we are pulling back an open set. The set of all real numbers $x$ such that $\lfloor x \rfloor$ is either 3 or 4 is the interval $[3, 5)$. Is this set open in the real numbers? No! It contains the point $x=3$, but any open interval around 3, like $(3-\varepsilon, 3+\varepsilon)$, will always contain numbers less than 3, which are not in our set. The preimage is not open. Our test has failed, and it has failed precisely at a point of discontinuity.

You see the same phenomenon with a simple step function, say one that jumps from a value of 10 to 20 at $x=7$. The set of inputs that map to the single point $\{10\}$ is $(-\infty, 7)$. This set is open. But the set of inputs that map to the closed set $\{10\}$ is not a closed set, because its limit point, 7, is not included. A continuous function would have preserved this "closedness" on the way back. The preimage acts as a perfect litmus test for continuity, revealing flaws in the fabric of the function with beautiful precision.

So, preimages tell us about the shape of sets. But often in science, from physics to finance, we want to know about their size. What is the volume of this region of space? What is the probability of this event? This is the world of measure theory. To even begin to answer these questions, we need our functions to be "well-behaved" in a specific way—they must be measurable.

And how do we define measurability? You guessed it: with preimages. A function is measurable if the preimage of any "well-behaved" (i.e., measurable) set of outputs is also a "well-behaved" set of inputs. For the real numbers, the most important measurable sets are the Borel sets, which include all intervals and any sets you can make from them through unions, intersections, and complements.

A simple linear function, $f(x) = mx + b$, is of course measurable. If we ask, "What set of inputs $x$ results in an output greater than some value $a$?", we are asking for the preimage of the interval $(a, \infty)$. A quick calculation shows this is another open interval, $(\frac{a-b}{m}, \infty)$, which is certainly a measurable set. This seems simple, but it's the foundation.

Now for a more spectacular example. Consider the space of all $2 \times 2$ matrices. This is a four-dimensional space, one for each entry of the matrix. Within this vast space, there is a special, intricate surface corresponding to all the singular matrices—those with a determinant of zero. Is this ghostly surface a "well-behaved" measurable set? Answering this directly would be a nightmare. But with preimages, the argument is breathtakingly simple. The set of singular matrices is just the preimage of the set $\{0\}$ under the determinant function. The determinant, $\det(A) = ad-bc$, is a simple polynomial of the matrix entries, making it a continuous function. The set $\{0\}$ is a closed set, and therefore a Borel set. Since the preimage of a Borel set under a continuous function is always a Borel set, the set of all singular matrices is a perfectly well-behaved, measurable set. What seemed impossibly complex becomes almost trivial.

This tool allows for even more subtle reasoning. Imagine a function where the set of inputs mapping to any single output point has a measure of zero. What can we say about the set of all inputs that map to a rational number? The rational numbers, $\mathbb{Q}$, are a "dust" of points, infinitely many but countably so. It seems like the set of inputs mapping to this dust might be complicated. But the preimage of $\mathbb{Q}$ is just the union of the preimages of each individual rational number. Since there are only countably many rational numbers, and the measure of each individual preimage is zero, the measure of the whole collection must also be zero, by a fundamental property of measures called countable subadditivity. We have determined the "size" of a very complex set with an argument of pure elegance.

(As a tantalizing peek into deeper waters, this powerful connection between continuity and measurability has its limits. While the preimage of any Borel set under a continuous function is always Borel, there exist strange continuous functions and pathological Lebesgue measurable sets where the preimage is, shockingly, not Lebesgue measurable!

Our journey now takes a turn, from the geometric world of topology and measure theory to the structural world of abstract algebra. Here, preimages do more than describe shapes and sizes; they reveal the fundamental skeleton of mathematical structures.

In group theory, a homomorphism is a function between two groups that respects their structure. One of the most fundamental results is that ​​the preimage of a subgroup under a homomorphism is always a subgroup​​. This is a profound statement about the preservation of structure. The most famous example is the kernel of a homomorphism, which is simply the preimage of the identity element.

Let's see this in action. Consider a group of upper-triangular matrices, and a homomorphism $\phi$ that maps a matrix $\begin{pmatrix} a & b \\ 0 & c \end{pmatrix}$ to the pair of its diagonal entries, $(a,c)$. Now, consider a special subgroup in the target space: the set of all pairs $(x,y)$ such that their product is 1. What is the preimage of this subgroup? Which matrices in our original group map to these special pairs? The condition is that the product of the diagonal entries, $a \cdot c$, must be 1. But for a $2 \times 2$ matrix, $ac$ is nothing but its determinant! So, the preimage is the set of all matrices in our group whose determinant is 1. We have discovered a fundamentally important object, a piece of the special linear group, simply by looking backward from a subgroup.

This idea extends far beyond groups. In functional analysis, we study spaces whose "points" are themselves functions. Consider the space of all continuous functions on the interval $[0,1]$. We can define an "evaluation map," $E_p$, which takes a function $f$ and returns its value at a specific point $p$. What is the preimage of $\{0\}$ under this map? By definition, it's the set of all continuous functions $f$ for which $f(p)=0$. This is not just a random collection. It is a vast, infinite-dimensional plane (a hyperplane) running through the origin of this function space. It is a core structural and geometric object, and we found it just by asking: "Which functions give a value of 0 at point $p$?"

Finally, let's look at applications where the preimage takes on a dynamic or geometric life of its own.

In topology, we often study complicated spaces by finding simpler spaces that "map onto" them. A classic example is the exponential map $p(z) = e^z$, which takes a complex number $z=x+iy$ and maps it to another complex number. Let's take a simple, nice region in the output space: an annulus, or a ring, of all points whose distance from the origin is between 1 and 2. What is the preimage of this beautiful, finite ring? What does the world look like before we apply this map? The answer is not another ring, but an infinite vertical strip in the complex plane, defined by $0 \operatorname{Re}(z) \ln(2)$. The map takes this infinite strip and essentially wraps it around the origin, over and over again, to create the annulus. The preimage has "unrolled" the space for us, revealing its hidden periodic structure. This is the central idea behind the beautiful theory of covering spaces.

Perhaps the most exciting use of preimages is in the study of dynamical systems and chaos theory. Here, we apply a function over and over, and we want to understand the long-term behavior. Looking at preimages is like running the movie backward. Consider the famous logistic map, $f(x) = 4x(1-x)$, a simple formula known to produce chaotic behavior. What are the points that land on $0$ after one iteration? The equation $f(x)=0$ gives us two points: $x=0$ and $x=1$. Now, what about the points that land on $0$ after two iterations? That means $f(f(x))=0$, so $f(x)$ must be either $0$ or $1$. We're looking for the preimage of the set $\{0, 1\}$. By tracing these preimages backward step by step, we can find all the points that will eventually land on a specific spot. For the logistic map, the number of points that land on 0 after $n$ steps is not random at all, but a beautifully ordered sequence: $2^{n-1}+1$. By looking backward, we find a stunningly predictable order hidden within a system that is the very emblem of chaos.

From defining the very notion of continuity, to measuring the size of sets, to uncovering deep algebraic structures and decoding the history of chaotic systems, the simple act of looking backward through a function is a thread that weaves together the fabric of modern mathematics. The preimage is not just a definition; it is a way of seeing.