Preimage

In mathematics, some concepts appear at first glance to be simple matters of definition, yet they hold the key to understanding vast and complex structures. The preimage is one such concept. While easily defined as a form of "reverse-lookup" for a function, this simple idea is one of the most powerful and unifying tools in modern mathematics. The common knowledge gap is not in knowing what a preimage is, but in appreciating what it does—how it sculpts shapes, defines continuity, and reveals hidden connections between disparate fields.

This article takes you on a journey to uncover that power. In the first section, ​​Principles and Mechanisms​​, we will build a deep, intuitive understanding of the preimage, using analogies of machines and round-trips to explore its fundamental properties and surprising behaviors. Following that, the ​​Applications and Interdisciplinary Connections​​ section will showcase how this single concept becomes a cornerstone of geometry, topology, abstract algebra, and analysis, demonstrating that the simple act of looking backward is a profound way of seeing the world.

Alright, let's get our hands dirty. We've talked about this idea of a "preimage," but what is it, really? Forget the dusty definitions for a moment. Think of a function, any function, as a machine. You put something in one end (an element from a set we call the ​​domain​​), and the machine spits something out at the other end (an element in a set we call the ​​codomain​​). The function $f(x) = x^2$ is a simple machine like this. You put in a $2$, you get out a $4$. You put in a $-3$, you get out a $9$.

The preimage is not about running the machine backwards. That would be an inverse function, and many machines can't be run in reverse (how would you "un-square" a $4$ to know if it came from a $2$ or a $-2$?) Instead, the preimage is a different kind of tool. It's a "reverse-lookup" device. You stand at the output end of the machine, point to a bin of results—say, the bin labeled '9'—and you ask the device: "Show me everything from the input side that ended up here." The device would light up both the '3' and the '-3' on the input side. You're not getting a single answer; you're getting a whole set of answers.

Let's make this more concrete. Imagine a fantastically simple machine: a projector. Our domain is the entire two-dimensional plane, $\mathbb{R}^2$, and our function, let's call it $\pi_1$, simply reads the first coordinate of any point $(x, y)$ and outputs that. So, $\pi_1(x, y) = x$. The point $(5, 10)$ goes in, the number $5$ comes out. The point $(5, -100)$ goes in, the number $5$ still comes out. The machine completely ignores the second coordinate.

Now, let's use our reverse-lookup device. We stand at the output, which is the number line $\mathbb{R}$, and we point to a single value, say $x_0 = 5$. We ask: what is the preimage of the set $\{5\}$? We are asking for all points $(x,y)$ in the plane such that $\pi_1(x,y) = 5$. The answer, of course, is the set of all points where the first coordinate is $5$. This is the entire vertical line defined by the equation $x=5$. We pointed to a single, zero-dimensional point on the output line, and the preimage revealed an entire, one-dimensional line in the input space. The preimage isn't an "inverse point"; it's the entire collection of origins.

This "all or nothing" character can be even more stark. Consider a constant function, the most boring machine imaginable: no matter what you put in from your domain $X$, it always outputs the exact same thing, let's call it $y_0$. Now, let's use our reverse-lookup tool. If we point to a set $V$ on the output side that contains $y_0$, and ask, "What inputs landed in this set?", the answer is... everything! Every single element of $X$, because every input lands on $y_0$, which is in $V$. The preimage is the entire domain, $X$. But if we point to a set $V$ that does not contain $y_0$, the answer is... nothing! The empty set, $\emptyset$. No input ever lands there. The preimage reveals the function's behavior in a wonderfully clear way.

Here's where it gets interesting. What happens when we combine the forward machine (the function) and the reverse-lookup device (the preimage)? Let's say we take a subset of our inputs, a group of travelers $A$ from our domain $X$, and send them on their way. They arrive at a set of destinations, which we call the ​​image​​ of $A$, or $f(A)$. Now, we use our reverse-lookup device on this set of destinations. We ask: "Who are all the travelers in the entire domain $X$ that landed in one of these destinations $f(A)$?" We denote this resulting set of travelers as $f^{-1}(f(A))$.

What do you expect this set to be? It seems like we should just get our original group of travelers, $A$, back again. And indeed, we will! Every traveler in $A$ certainly landed in the destination set $f(A)$, so they must be included in the reverse lookup. This means that, always, $A \subseteq f^{-1}(f(A))$.

But will it be exactly $A$? Not necessarily! This is one of the most important subtleties of preimages.

Let's try it with our squaring machine, $f(x) = x^2$. Suppose our domain is the set of integers $X = \{1, 2, 3, 4, 5, 6\}$. Our function maps $f(1) = 1, f(2) = 4, f(3) = 9, \ldots$ wait, let's use a more interesting function from the problem: $f(1)=\text{alpha}, f(2)=\text{beta}, f(3)=\text{alpha}$. Notice that $1$ and $3$ both go to the same place. This is a "many-to-one" function. Now, let's take the set of travelers $A = \{1, 2, 4\}$.

1. ​​Forward trip:​​ Where do they go? $f(A) = \{f(1), f(2), f(4)\} = \{\text{alpha}, \text{beta}, \text{gamma}\}$.
2. ​​Reverse lookup:​​ Now we ask, who in the entire domain $X$ maps to this set of destinations $\{\text{alpha}, \text{beta}, \text{gamma}\}$?
   • Who maps to 'alpha'? Both $1$ and $3$.
   • Who maps to 'beta'? Both $2$ and $5$.
   • Who maps to 'gamma'? Just $4$. So, the preimage $f^{-1}(f(A))$ is the set $\{1, 3, 2, 5, 4\}$.

Look at that! Our original group was $A = \{1, 2, 4\}$, but the set we got back is $\{1, 2, 3, 4, 5\}$. We got our original travelers back, but we also picked up stowaways—numbers $3$ and $5$—who were not in our original group $A$ but happened to travel to the same destinations as members of $A$. So, in this case, $A$ is a proper subset of $f^{-1}(f(A))$. The only way you are guaranteed to get exactly your original set $A$ back is if the function is ​​injective​​ (one-to-one), meaning no two distinct inputs ever go to the same output. This round-trip principle is a powerful way to test and understand the nature of a function.

So far we've talked about collections of points. But the real fun begins when we apply this to continuous spaces. The preimage becomes a kind of sculptor's tool, carving out shapes in the domain based on what we select in the codomain.

Imagine a different kind of function, one that takes a point $(x,y)$ in the plane and tells you its "Manhattan distance" from the point $(0,2)$. The formula is $f(x,y) = |x| + |y-2|$. The function maps the entire 2D plane to the 1D number line of non-negative reals.

Now let's use our reverse-lookup device. What is the preimage of the single value $\{c\}$? It's the set of all points $(x,y)$ whose Manhattan distance from $(0,2)$ is exactly $c$. This isn't a circle, which you'd get with the usual Euclidean distance. It's a square, tilted by 45 degrees, centered at $(0,2)$.

What if we ask for the preimage of a whole interval of values in the codomain, say the closed interval $[3, 5]$? We're asking for all points $(x,y)$ that satisfy $3 \le |x| + |y-2| \le 5$. What shape is that? It's the region in the plane lying on or between the tilted square for $c=3$ and the tilted square for $c=5$. By simply selecting a simple interval on the output number line, we have sculpted a complex, beautiful shape—a hollow, tilted square frame—in the input plane. This is an incredibly powerful idea in geometry and physics: complicated shapes can often be understood as preimages of simple sets under some function.

Here we arrive at one of the deepest insights the preimage offers. What does it mean for a function to be ​​continuous​​? The high-school notion is "you can draw its graph without lifting your pen." This is a fine intuition, but it falls apart for more exotic spaces. The modern, powerful definition of continuity is built entirely on the concept of the preimage.

A function $f$ is continuous if and only if ​​the preimage of every open set in the codomain is an open set in the domain.​​

Why is this the right definition? Because it beautifully captures the idea of "nearness" without tearing. If you take a small open neighborhood around an output point $f(x)$, its preimage must contain a small open neighborhood around the input point $x$. Points that are close together in the output must have come from points that were, in some sense, already close together.

But this definition leads to some surprising consequences. A continuous function can't tear space apart as it goes forward, but the preimage can reveal that the space was already folded or seamed. Consider the function $f(x) = (x-2)^2 + 3$. The graph is a parabola with its vertex at $(2,3)$. Now, let's take a nice, simple, connected open interval in the codomain, say $U=(3,8)$. This is one single, connected piece. What is its preimage? We're looking for all $x$ such that $3 (x-2)^2 + 3 8$, which simplifies to $0 (x-2)^2 5$. This inequality is solved by $x$ in $(2-\sqrt{5}, 2) \cup (2, 2+\sqrt{5})$.

Look at that! The preimage of one connected interval is two separate, disconnected intervals. The continuity of $f$ is not violated; the preimage of the open set $(3,8)$ is indeed an open set, namely the union of two open intervals. But the property of being a single connected piece was lost.

This gets even more dramatic.

• ​​Is the preimage of a connected set always connected?​​ No. As we just saw, it can be disconnected. A more striking example is the function $f(x,y)=x^2$ from the connected plane $\mathbb{R}^2$ to the connected line $\mathbb{R}$. The preimage of the connected interval $[1,4]$ is the set of all points $(x,y)$ where $1 \le x^2 \le 4$, which means $x \in [-2, -1] \cup [1, 2]$. This is two infinite, disconnected vertical strips in the plane. A single connected piece in the codomain was "pulled back" to two separate pieces in the domain.
• ​​Is the preimage of a compact set always compact?​​ (In $\mathbb{R}$, "compact" is a technical term for "closed and bounded.") No! Consider a constant function $f$ that maps every integer in $\mathbb{Z}$ (with the discrete topology) to the single point $\{0\}$ in $\mathbb{R}$. The target set $K = \{0\}$ is as compact as you can get. But what is its preimage? It is the entire set of integers $\mathbb{Z}$. This set is infinite and not bounded, so it is not compact.

This reveals a fundamental asymmetry. A continuous function is like a well-behaved tour guide: it will always take a connected group ($X$) to a connected destination ($f(X)$), and it will always take a compact group (a finite tour group) to a compact destination. But the reverse lookup, the preimage, makes no such promises!

We've seen some quirky behavior. The round-trip isn't always perfect. Properties like connectedness can vanish. There's one more "quirk" that turns out to be the key to the whole story. What happens when you compose two functions? Say you have a machine $f$ that feeds into a second machine $g$. This gives a composite machine, $g \circ f$. What is the preimage of a set $U$ under this composite machine?

Let's trace it. We want to find $(g \circ f)^{-1}(U)$, which is the set of all inputs $x$ such that $g(f(x))$ is in $U$. Think about it step-by-step. For $g(f(x))$ to be in $U$, the intermediate value $f(x)$ must be in the set of things that $g$ maps into $U$. That set is just $g^{-1}(U)$. So, our condition on $x$ is now that $f(x)$ must be in $g^{-1}(U)$. But this is just the definition of the preimage of the set $g^{-1}(U)$ under the function $f$! So we have found a fundamental rule:

Or, thinking of these as operations: $(g \circ f)^{-1} = f^{-1} \circ g^{-1}$. The order is reversed! To undo a composition, you undo them one by one, starting from the last one. It's like taking off your shoes and socks: you first took off your shoes, then your socks. To reverse this, you must first put on your socks, then your shoes.

For centuries, mathematicians saw this reversal as just a handy calculational rule. But in the 20th century, a new perspective called ​​category theory​​ emerged, which focuses on objects and the maps (morphisms) between them. It looks for grand, overarching structures. From this viewpoint, the reversal property is not a quirk; it's a profound statement.

There are processes that preserve the direction of maps, which are called ​​covariant functors​​. But there is another, equally important type of process that systematically reverses the direction of maps. These are called ​​contravariant functors​​. The preimage operation is the archetypal example of contravariance. It takes an arrow $f: X \to Y$ and gives you a new arrow, $f^{-1}$, that goes from the power set of $Y$ to the power set of $X$. It runs backwards. The reversal of composition, $(g \circ f)^{-1} = f^{-1} \circ g^{-1}$, is the defining feature of this contravariant structure.

So the humble preimage, our simple "reverse-lookup" device, is actually a window into one of the most elegant and unifying concepts in modern mathematics. It shows us that moving "backwards" is not just the opposite of moving forwards; it is its own rich, structured, and beautiful way of seeing the world.

We have spent some time getting to know the preimage, perhaps as a simple matter of definition—a bit of necessary but dry bookkeeping. To stop there, however, would be like learning the alphabet but never reading a word of poetry. The true power and beauty of the preimage concept lie not in its definition, but in its application. It is a master key that unlocks doors in what appear to be completely different wings of the scientific mansion, revealing that they are all, in fact, part of one grand, interconnected structure. It is a tool for asking one of the most fruitful questions in science: "Given a certain outcome, what were the possible beginnings?"

Let us now embark on a journey to see how this one idea—the simple act of looking backward through a function—becomes a cornerstone of geometry, analysis, algebra, and beyond.

At its most intuitive level, mathematics gives us a language to describe shapes. We can talk about circles, spheres, planes, and more complex curves and surfaces. But how can we define these objects rigorously? The preimage provides a surprisingly elegant and powerful answer.

Imagine you are standing at a fixed point, let's call it $x_0$, in some space. You have a way to measure the distance to any other point $x$ in that space, a function we can call $f(x) = d(x_0, x)$. Now, ask a simple question: which points are "close" to me, say, within a distance of $\epsilon$? In the language of functions, you are asking for the set of all points $x$ such that their distance $f(x)$ falls in the interval $[0, \epsilon)$. You are asking for the preimage $f^{-1}([0, \epsilon))$. And what is this set? It is none other than the familiar open ball $B(x_0, \epsilon)$—the collection of all points within a certain radius of a center. This fundamental building block of geometry and topology is, in its essence, a preimage.

This idea scales up to breathtaking effect in the field of differential geometry. Consider a simple function from three-dimensional space to the real numbers, like $f(x, y, z) = x^2 + y^2 + z^2$. This function takes a point in space and tells you the square of its distance from the origin. Now, let's ask for the preimage of a single positive number, say $q > 0$. What is the set $f^{-1}(q)$? It is the set of all points $(x,y,z)$ such that $x^2 + y^2 + z^2 = q$. This is the equation of a sphere with radius $\sqrt{q}$!

This is the heart of the ​​Preimage Theorem​​, a giant of modern geometry. It tells us that for a "well-behaved" function (a smooth map) and a "typical" output value (a regular value), the preimage is always a beautiful, smooth geometric object (a manifold), whose dimension is precisely the dimension of the domain minus the dimension of the codomain. In our example, we start in a 3D space and map to a 1D space (the number line), so the preimage is a $(3-1)=2$-dimensional surface—a sphere. This single principle is the engine used to construct and understand a vast universe of shapes, from simple spheres to exotic surfaces in higher dimensions that are impossible to visualize directly.

This geometric insight has profound practical consequences in fields like ​​optimization​​. Many real-world problems, from designing an airline schedule to training a machine learning model, involve finding the "best" solution among a set of "allowed" solutions. This set of allowed solutions, or the feasible set, is often defined by a series of constraints, such as $a_1 x_1 + a_2 x_2 \le b$. Each such constraint defines the feasible set as the preimage of an interval $(-\infty, b]$ under a linear function. A key property in optimization is convexity—a set where the straight line between any two points in the set lies entirely within the set. The preimage of a convex set under a linear map is always convex. This fact, which is a direct consequence of how preimages work, ensures that the feasible sets for a huge class of optimization problems are "well-behaved," guaranteeing that efficient algorithms can find the single best solution.

The concept of continuity seems intuitive. A continuous function is one you can draw without lifting your pen. There are no sudden jumps, rips, or tears. For centuries, this intuitive picture was enough. But as mathematicians began to explore more abstract spaces—spaces of functions, spaces with strange topologies—they needed a definition that was more robust and universal. They found it in the preimage.

The modern definition is as simple as it is profound: ​​A function is continuous if and only if the preimage of every open set is open.​​

Why is this so powerful? It shifts the focus from checking points one by one to examining the overall structure of the mapping. Think of a function as a transformation that deforms the domain into the codomain. If the function is continuous, it might stretch or bend the space, but it won't tear it apart. An open set is like a "neighborhood" of points that are all close to each other. If the function is continuous, the source of any such neighborhood must also be a neighborhood where the points were originally close. If you find even one open set in the codomain whose preimage is not open, you've found a "tear"—a proof of discontinuity.

This definition allows us to analyze situations that would be baffling otherwise. Consider the simple identity function, $f(x) = x$. Is it continuous? The question is meaningless without specifying the "structure" of the domain and codomain—their topologies. If we map from the real numbers with the "lower limit topology" (where open sets look like $[a,b)$) to the real numbers with the standard topology (where open sets are unions of $(a,b)$), we find that the preimage of a standard open set $(a,b)$ is just $(a,b)$ itself. This set can be built from a union of lower-limit intervals, so it is open in the domain. The function is continuous! The preimage provides the definitive test.

This perspective even informs how we define parts of a space. When we consider a subset $A$ of a larger space $X$, we create the subspace topology on $A$. The open sets of $A$ are defined to be intersections of the open sets of $X$ with $A$. Why this specific rule? Because it is exactly the rule needed to make the simple inclusion map $i: A \to X$ (where $i(a)=a$) a continuous function. The open sets in the subspace are, by definition, the preimages of the open sets from the parent space. The preimage concept is baked into the very foundation of how we talk about topological subspaces.

The reach of the preimage extends far beyond geometry and topology into the abstract realms of algebra and analysis. Here, it doesn't just describe shape, but reveals deep structural relationships.

In ​​abstract algebra​​, we study groups and homomorphisms—functions that respect the group operation. We can consider the lattice (or collection) of all subgroups of a group. A homomorphism $f: G \to H$ naturally invites us to ask how the subgroups of $G$ relate to the subgroups of $H$. One way is to take the preimage of a subgroup in $H$. Does this process preserve the structure of the lattice? The answer is a beautiful "yes, and no." The preimage operation is perfectly behaved with respect to intersections: the preimage of the intersection of two subgroups is always the intersection of their preimages ($f^{-1}(T_1 \cap T_2) = f^{-1}(T_1) \cap f^{-1}(T_2)$). This is a universal property of preimages, and in the context of group theory, it establishes a precise structural correspondence. However, the preimage does not, in general, preserve the "join" (the smallest subgroup containing the union). This subtle distinction, revealed by studying preimages, teaches us about the nature of homomorphisms themselves.

This idea finds a powerful echo in ​​functional analysis​​, where the objects of study are not numbers or vectors, but entire functions living in infinite-dimensional spaces. Consider the space of all continuous functions on the interval $[0,1]$, and define a map $\phi$ that takes a function $f$ and returns a single number: its integral from 0 to 1, $\phi(f) = \int_0^1 f(t) \, dt$. This map is a "linear functional." Now, what is the preimage of $\{0\}$? It is the set of all continuous functions whose integral is zero—or, equivalently, whose average value is zero. This set, $\phi^{-1}(\{0\})$, is known as the ​​kernel​​ of the functional. The concept of a kernel is central to all of linear algebra; it represents everything that is "annihilated" or "crushed to zero" by a linear map. Here we see that this fundamental algebraic object is, once again, simply a preimage.

We have seen that preimages preserve unions, intersections, and, by extension, the property of openness that defines continuity. It can be tempting to think they preserve everything. But just as important as knowing what a tool can do is knowing what it cannot.

Consider the operation of taking the closure of a set (the set plus all its limit points). If we take the preimage of a set and then find its closure, do we get the same result as finding the closure of the set first and then taking its preimage? Not always. For a continuous function, we are only guaranteed an inclusion: the closure of the preimage is a subset of the preimage of the closure ($\overline{f^{-1}(B)} \subseteq f^{-1}(\overline{B})$). The fact that this is not always an equality is not a defect; it is an insight. It tells us about the subtle ways a continuous function can map points. A point might not be a limit point of the source set $f^{-1}(B)$, but its image could be a limit point of the target set $B$. Exploring when this inclusion is strict reveals deeper properties of the function and the topological spaces involved.

From the tangible geometry of spheres to the abstract definition of continuity and the algebraic structure of kernels, the preimage is the common thread. It is a simple concept with extraordinary depth, a unifying lens that allows us to see the same fundamental pattern repeating itself across the vast and beautiful landscape of mathematics.