Immersion and Submersion

In the study of smooth manifolds—the curved spaces that form the language of modern geometry and physics—a central question arises: how do we relate one manifold to another? While we can visualize a sphere or a torus, understanding the maps between them requires a more powerful tool than intuition alone. Just as the derivative unlocks the local behavior of functions in calculus, the ​​differential​​ serves as a "local magnifying glass" for maps between manifolds, transforming complex, curved interactions into simple linear algebra at every point. This powerful linearization is the key to a deeper understanding.

This article addresses the foundational geometric behaviors that emerge when we impose consistent conditions on this differential. Specifically, we explore two of the most important classes of maps in all of geometry: ​​immersions​​ and ​​submersions​​. These concepts arise from asking simple questions about the differential map: is it injective (one-to-one) or surjective (onto)? The answers reveal a profound duality in how manifolds can interact.

In the "Principles and Mechanisms" chapter, we will precisely define immersions and submersions, exploring their local structure through the Constant Rank Theorem and their global consequences, such as the distinction between an immersion and a well-behaved embedding. We will uncover the "manifold-making machine" of the Regular Value Theorem, a direct consequence of the submersion condition. Following this, the "Applications and Interdisciplinary Connections" chapter will demonstrate how these abstract principles are used as master tools by geometers, physicists, and engineers to define surfaces, induce geometric structures, organize spaces into fiber bundles, and even control complex robotic systems.

After our brief introduction to the world of smooth manifolds, you might be wondering how we actually do anything with them. How do we compare them? How do we map one to another? In the familiar world of functions on the real line, we have a wonderful tool: the derivative. It tells us, at any given point, the local behavior of a function—is it stretching, shrinking, or holding steady? We want a similar tool for the far richer landscape of curved spaces.

Imagine you're standing on a curved surface, say a sphere, and you're about to take a tiny step. That step can be represented by a "tangent vector"—an infinitesimal arrow pointing in the direction you're moving. Now, suppose a smooth map $f$ transforms your sphere into another shape, perhaps a torus. What happens to your tiny step?

The map $f$ takes the point you were standing on, $p$, to a new point $f(p)$ on the torus. It also transforms your tiny step-vector at $p$ into a new tiny step-vector at $f(p)$. The rule that governs this transformation of vectors is called the ​​differential​​ (or ​​derivative​​) of the map $f$ at the point $p$. We denote it as $df_p$.

This differential, $df_p$, is the best possible linear approximation of the map $f$ right around the point $p$. It's like putting a powerful magnifying glass over the point $p$ on the sphere. When you look through it, the curved surface looks flat (this is the tangent space, $T_pM$), and the complicated map $f$ looks like a simple linear transformation—the kind you studied in linear algebra—sending vectors from the flat patch at $p$ to another flat patch at $f(p)$ (the tangent space $T_{f(p)}N$). This magnificent tool, the differential, is the key to understanding the local geometry of any smooth map.

When we look at this linear map, $df_p$, a natural question arises: what are its fundamental properties? From linear algebra, we know two of the most important properties of a linear map are whether it is injective (one-to-one) or surjective (onto). When we demand that one of these properties holds at every single point of a manifold, the map $f$ acquires a profound and consistent geometric character. This gives rise to two of the most important classes of maps in all of geometry: ​​immersions​​ and ​​submersions​​.

Immersions: Drawing Without Crushing

Let's start with injectivity. What does it mean for the differential $df_p$ to be injective? It means that if you take two different tangent vectors at $p$ (two different directions to step in), they will always be mapped to two different tangent vectors at $f(p)$. No non-zero vector at $p$ gets "crushed" into the zero vector at $f(p)$. A map $f: M \to N$ is called an ​​immersion​​ if its differential $df_p$ is injective at every point $p \in M$.

Since the differential preserves the dimension of the tangent space, this immediately tells us that the dimension of the domain manifold $M$ can be no larger than the dimension of the target manifold $N$ (i.e., $\dim(M) \le \dim(N)$). You can't fit a 3D space into a 2D plane without crushing something!

What does an immersion look like? The beautiful ​​Constant Rank Theorem​​ gives us a stunningly simple local picture. It says that for any immersion, you can always find a set of "magic coordinates" around any point $p$ and its image $f(p)$ such that the map looks like a standard coordinate inclusion. For example, a 2D manifold immersing into a 3D one looks locally just like the map from $\mathbb{R}^2$ to $\mathbb{R}^3$ given by $(x_1, x_2) \mapsto (x_1, x_2, 0)$. Locally, an immersion is just laying down the domain manifold flatly within the higher-dimensional target manifold.

A wonderful example is the parametrization of a torus in 3D space. The map $F(u, v) = ( (R + r \cos u)\cos v, (R + r \cos u)\sin v, r \sin u )$ takes a flat 2D rectangle (the $(u,v)$ plane) and wraps it into a 3D donut shape. At every point, the differential is injective—it maps the 2D tangent plane of the flat sheet to a 2D tangent plane on the surface of the torus. It never crushes the sheet. However, since the target space is 3D, there is always a direction "off the torus" that the differential cannot point to. Thus, the differential is not surjective, and this map is an immersion but not a submersion.

Submersions: Projecting Without Missing

Now, let's consider surjectivity. What if we require the differential $df_p$ to be surjective at every point? This means that for any direction you want to move in the target space at $f(p)$, there is at least one direction you can move at $p$ that will take you there. The map "covers" all possible directions locally. A map $f: M \to N$ with this property is called a ​​submersion​​. This requires that the dimension of the domain is at least as large as that of the target, $\dim(M) \ge \dim(N)$.

Think of a movie projector. A complex object (the film, which we can think of as a manifold of dimensions space + space + time) is projected onto a 2D screen. Every point on the screen is illuminated; the projection is surjective. A map from $\mathbb{R}^3$ to $\mathbb{R}^2$ given by $f(x,y,z) = (x,y)$ is a simple submersion. The differential is a $2 \times 3$ matrix that clearly has rank 2, so it is surjective.

Once again, the Constant Rank Theorem provides a clear local picture. Near any point, a submersion looks just like a standard coordinate projection. A map from a 3D manifold to a 2D one, for instance, looks locally like $(x_1, x_2, x_3) \mapsto (x_1, x_2)$. It simply forgets the extra coordinates. This simple local structure has profound consequences. Because projections are ​​open maps​​ (they send open sets to open sets), all submersions are open maps. This is a powerful topological constraint.

So, locally, immersions are just nice inclusions. But what happens when we zoom out and look at the global picture? Here, things can get wonderfully complicated. An immersion guarantees that the map is locally one-to-one, but it does not guarantee that it is globally one-to-one. The map can loop back and cross over itself.

Consider the map of a circle into the plane given by $\gamma(t) = (\sin(t), \sin(2t))$. The velocity vector is never zero, so this is a perfectly valid immersion. But if you trace it out, you'll see it draws a figure-eight. The map is not injective, as it visits the origin $(0,0)$ twice (at $t=0$ and $t=\pi$). This is an immersed circle, but it's not "nicely" sitting in the plane; it intersects itself.

Even if an immersion is injective, it can still fail to be "nice". Consider a line wrapping around a torus with an irrational slope, like $\gamma(t) = (e^{it}, e^{i\alpha t})$ for irrational $\alpha$. This is an injective immersion of the real line $\mathbb{R}$ into the torus. However, the image of this line is a dense "scribble" that covers the entire torus. If you take a small open neighborhood around a point on the image, it doesn't look like a simple line segment. The topology of the image (as a subspace of the torus) is vastly more complicated than the topology of the original line.

This leads us to a crucial distinction. An ​​embedding​​ is an immersion that is well-behaved globally. Formally, an embedding is an injective immersion that is also a homeomorphism onto its image. This just means that the global topology is also preserved; the way the object sits inside the larger space is simple. The image of an embedding is called an ​​embedded submanifold​​. It's what we intuitively think of as a "nice" subspace. A standard circle in the plane is an embedding. A paraboloid in 3D space is an embedding.

A beautiful and useful fact is that if the domain manifold $M$ is ​​compact​​ (finite in a certain topological sense, like a sphere or a torus), then any injective immersion from $M$ into a standard Euclidean space is automatically an embedding. The compactness prevents the map from doing wild things like wrapping infinitely densely.

We've seen that immersions are about faithfully representing one space inside another. Submersions, it turns out, have a completely different and arguably even more powerful role: they are tools for creating new manifolds.

This magical ability is captured by the ​​Regular Value Theorem​​. Let's say we have a map $F: M \to N$. A point $c \in N$ is called a ​​regular value​​ if for every point $p$ in its preimage $F^{-1}(c)$, the differential $dF_p$ is surjective. (Note that this is a weaker condition than being a submersion, which requires surjectivity everywhere, not just on one level set). The theorem then makes a spectacular claim:

If $c$ is a regular value of $F$, then the level set $F^{-1}(c)$ is a smooth, embedded submanifold of $M$. Its dimension is precisely $\dim(M) - \dim(N)$.

This is a manifold-making machine! Think of the function $F: \mathbb{R}^3 \to \mathbb{R}$ given by $F(x,y,z) = x^2+y^2+z^2$. The differential is the gradient vector $(2x, 2y, 2z)$, which is surjective (non-zero) for any point other than the origin. So, any value $c>0$ is a regular value. The theorem tells us that the level set $F^{-1}(c) = \{ (x,y,z) \mid x^2+y^2+z^2=c \}$ is a smooth submanifold of $\mathbb{R}^3$ of dimension $3-1=2$. And of course, it is: it's the sphere of radius $\sqrt{c}$!

How does this miracle work? The proof is a beautiful application of the ​​Implicit Function Theorem​​. The condition that the differential is surjective at a point $p$ is exactly what the Implicit Function Theorem needs to guarantee that, locally, you can "solve" for some of the coordinates in terms of the others. This means that near $p$, the level set $F^{-1}(c)$ can be described as the graph of a smooth function. And the graph of a smooth function is the very definition of a well-behaved, embedded submanifold. The surjectivity condition ensures you have enough "directions" to project away, leaving behind a lower-dimensional, well-defined surface.

So we see the beautiful duality. Immersions let us place manifolds inside other manifolds, while submersions let us carve new manifolds out of existing ones. These two simple conditions on the differential—injectivity and surjectivity—open the door to the entire constructive and descriptive framework of modern geometry.

After our exploration of the principles and mechanisms of immersions and submersions, you might be left with a feeling akin to learning the rules of chess. You understand the moves, the definitions are crisp, but you’re left wondering, "What kind of game can I play with these pieces?" It is a fair question. The true power and beauty of a mathematical idea are revealed not in its definition, but in the richness of the world it unlocks. So, let us embark on a journey to see what games we can play. We will see that these twin concepts are not merely abstract classifications; they are the master tools of the geometer, the physicist, and even the engineer, used to build, dissect, and control the very fabric of the spaces they study.

Let's start with a seemingly simple question: what is a surface? You might think of a sphere or a donut. But how do you describe one mathematically, with absolute precision? There are two great schools of thought, and wonderfully, they correspond exactly to our two main concepts.

The first approach is to build a surface from the inside out, like a tailor stitching together flat pieces of cloth to make a shirt. We can imagine taking small, open patches of the flat Euclidean plane, $\mathbb{R}^2$, and mapping them into three-dimensional space, $\mathbb{R}^3$. To form a smooth, "regular" surface without any ugly pinches or sharp creases, we must insist that this mapping is an ​​immersion​​. This condition ensures that at every point, the map's differential is injective—it takes the two-dimensional tangent plane of our flat patch and faithfully embeds it as a two-dimensional plane in the higher space. It doesn't crush the patch down to a line or a point. By covering our entire target shape with a collection of these well-behaved immersed patches (an "atlas"), we have parametrically defined a regular surface. An immersion, then, is the guarantee that we are truly "sculpting" a 2D object inside a 3D world.

The second approach is to define a surface from the outside in, like a sculptor carving a statue from a block of marble. Instead of building it up, we can specify it as the set of all points in $\mathbb{R}^3$ that satisfy an equation, such as $F(x,y,z) = 0$. Think of the sphere, defined by $x^2+y^2+z^2-1=0$. For the resulting level set to be a smooth surface, we need a condition. The condition is that the function $F$, which maps from $\mathbb{R}^3$ to $\mathbb{R}$, must be a ​​submersion​​ on that level set. This means its differential must be surjective, which for a function to the real numbers simply means its gradient vector must not be zero. A non-zero gradient ensures that the function is genuinely "moving" through that level, and the level set is consequently a nice, regular surface. This is the essence of the Regular Value Theorem.

Here we see a beautiful duality. An immersion builds a world by expanding a smaller one into a larger one. A submersion carves a world out of a larger one by projecting it down. One builds up, the other cuts down, and together they give us the complete language for describing the geometric objects that surround us.

Now that we have a surface, how do we do geometry on it? How do we measure the distance between two points on a sphere if we are constrained to walk on its surface? We can't just use a ruler in the ambient 3D space, as that would mean tunneling through the sphere! We need a way for the surface to inherit a notion of distance and angle from the space in which it lives.

Once again, the concept of an immersion provides the key. A Riemannian metric is the mathematical tool that defines an inner product, or a "dot product," on the tangent vectors at each point. It's our infinitesimal ruler. If our ambient space, like $\mathbb{R}^3$, has a metric (the familiar Euclidean dot product), we can use the immersion $f: S \to \mathbb{R}^3$ that defines our surface $S$ to pull back this metric.

The procedure is delightfully simple: to find the inner product of two tangent vectors $u$ and $v$ on the surface, the pullback metric $(f^*g)_p(u,v)$ simply says, "Push them forward into the ambient space using the differential $df_p$, and then take their dot product there!".

But why is the immersion condition so crucial? Suppose our map was not an immersion at some point. This would mean that some non-zero tangent vector $u$ on our surface gets crushed to the zero vector in the ambient space. If we then tried to measure its length using our pullback procedure, we would find its length to be zero! Our ruler would be broken. An immersion guarantees that $df_p$ is injective, so no non-zero vector is ever crushed. It guarantees that we can use the geometry of the larger space to endow our submanifold with its own consistent, non-degenerate geometry. Every time a physicist or geometer talks about the "induced metric" on a surface, they are implicitly relying on the fact that its inclusion into spacetime is an immersion.

Submersions, on the other hand, are the great organizers of the mathematical universe. They allow us to take a complex space and simplify it by collapsing parts of it in a controlled way. A simple, visual example is the central projection that creates perspective in a drawing. Imagine a light source at the origin projecting the upper half-space onto a flat screen at height one. This map, which takes a point and maps it to where the line through it and the origin hits the screen, is a perfect example of a submersion. It is a "regular squashing" of a 3D world onto a 2D plane.

This idea of "regular squashing" finds its most profound expression in the ​​Quotient Manifold Theorem​​. Often, we have a space $M$ with some symmetry, described by the action of a Lie group $G$. For example, a sphere is symmetric under rotations. We might want to consider all points that are related by a symmetry to be "the same." The set of these equivalence classes is the quotient space, $M/G$. A vital question is: when is this new space $M/G$ also a nice, smooth manifold? The theorem gives the answer: if the group action is "free" and "proper" (technical conditions ensuring the symmetries don't overlap or accumulate badly), then the quotient is indeed a manifold. And the punchline? The natural projection map $\pi: M \to M/G$ that sends each point to its equivalence class is a ​​submersion​​.

The fibers of this submersion—the preimages of a single point in the quotient—are precisely the orbits of the group action. The submersion provides a perfect "foliation" of the larger space into these symmetric fibers. This structure is known as a ​​fiber bundle​​, and it is one of the most important concepts in modern geometry and physics.

The most famous example is the ​​Hopf fibration​​, a submersion from the 3-sphere $S^3$ to the 2-sphere $S^2$. It presents the higher-dimensional sphere as a beautiful collection of disjoint circles (the fibers) perfectly arranged over the familiar 2-sphere (the base). In gauge theories, which form the basis of the Standard Model of particle physics, spacetime is the base of a fiber bundle, and the fibers represent the "internal symmetries" of the fundamental particles. A submersion is the map that lets us look at the underlying spacetime while systematically ignoring the internal state.

The magic of this structure, particularly in a Riemannian submersion where the metric is also respected in a certain way, leads to astonishing connections. Consider a curve $\gamma$ on the base sphere $S^2$. If we look at its preimage in $S^3$, we get a surface called a "Hopf tube." An incredible result states that this surface is a ​​minimal surface​​—an object that locally minimizes its area, just like a soap film—if and only if the original curve $\gamma$ was a ​​geodesic​​ (the straightest possible path) on the base sphere. This direct link between the geometry of the base (straightness) and the geometry of the total space (minimal area) is a testament to the deep organizing power of the submersion.

Similarly, these maps have a clean interaction with other geometric objects like differential forms. A pullback by a map of rank $r$ will always annihilate any form of degree higher than $r$, providing a simple algebraic tool to understand the geometric constraints imposed by the map.

Lest you think these ideas are confined to the ethereal realms of pure mathematics and theoretical physics, let us bring them down to Earth—and use them to control a robot.

Many real-world systems, from aircraft to chemical reactors to robotic arms, are described by complicated nonlinear differential equations. Controlling them is notoriously difficult. Our toolkit for control engineering is most powerful for linear systems, whose behavior is much simpler and more predictable. This raises a tantalizing question: could we find a clever change of coordinates, a new way of looking at the system's state, that makes the dynamics appear linear?

This is the goal of ​​feedback linearization​​, and its foundation is purely geometric. To transform the state $x$ of a nonlinear system into a new state $z$ where the dynamics are linear, the required transformation $z = T(x)$ must be a ​​local diffeomorphism​​. A diffeomorphism is a map that is smooth and has a smooth inverse; it is locally both an immersion and a submersion for maps between spaces of the same dimension.

Why is this necessary? The Inverse Function Theorem tells us that a map $T$ is a local diffeomorphism around a point if and only if its Jacobian matrix is invertible there. This invertibility is the mathematical guarantee that the change of coordinates is legitimate: it's a one-to-one mapping in that neighborhood, so no information is lost, and the transformation is smooth in both directions, so the transformed dynamics remain well-posed. Without a local diffeomorphism, we couldn't uniquely translate our control commands from the simple linear world of $z$ back to the complex nonlinear world of $x$ where the actual motors and actuators live. The abstract condition of an invertible differential becomes the concrete requirement for building a working control system.

From sculpting surfaces and measuring distances on them, to building new spaces from symmetry, uncovering deep links between spheres and soap films, and designing controllers for complex machinery, the concepts of immersion and submersion run as a unifying thread. They are fundamentally about how a map between spaces treats local, infinitesimal information. Does it preserve it faithfully, like an immersion? Or does it project it down in a regular fashion, like a submersion? This simple question, when asked with mathematical precision, provides the bedrock for an astonishing amount of modern science and engineering. It is a beautiful illustration of how the pursuit of simple, elegant ideas can give us a powerful lens through which to understand, and even control, our world.