Submersions

In the landscape of differential geometry, we often seek to understand the relationships between different mathematical worlds, or manifolds. A central question is how we can map from a higher-dimensional space to a lower-dimensional one in a way that is smooth, structured, and predictable. This is where the concept of a submersion arises—a powerful type of smooth map that acts as a "perfect projection," ensuring that no local dimension of the target space is missed. Submersions resolve the problem of how to rigorously define and construct new, lower-dimensional manifolds from existing ones. This article provides a comprehensive overview of this fundamental tool.

The first chapter, "Principles and Mechanisms," will unpack the formal definition of a submersion, contrasting it with immersions and exploring the pivotal Submersion Theorem, which simplifies the local picture of these maps. We will see how this leads to the Preimage Theorem, a veritable "manifold-making machine." The second chapter, "Applications and Interdisciplinary Connections," will demonstrate the power of submersions in action. We will explore their role in constructing quotient manifolds to study symmetries, defining the elegant structure of fiber bundles, and, in the context of Riemannian geometry, even connecting the curvatures of different spaces.

Imagine you are a movie director. You are filming a vast, three-dimensional world (let's call it $M$), but your final product will be a two-dimensional movie on a screen ($N$). Your camera, at any given moment, is pointed from a specific location $p$ in your world $M$ to a specific point on the screen $f(p)$. The "differential" of your filming process, $df_p$, is a concept that describes what happens to the immediate vicinity of point $p$. It's a linear map that tells you how a tiny step in any direction in $M$ translates to a movement on your screen $N$. It's the mathematical description of your camera's lens and orientation.

Now, you have two fundamental ways your camera can behave. It could be that you're filming a tiny, intricate detail, like an ant walking on a leaf. You are following it precisely. Every distinct movement the ant makes corresponds to a distinct movement on your screen. This is an ​​immersion​​. Mathematically, the differential $df_p$ is ​​injective​​ (one-to-one). You're capturing a lower-dimensional action without losing any of its detail, but you are by no means capturing the entirety of the world around it. The image of an immersion doesn't have to fill up the whole screen; it can be just a curve traced on it.

But what if your goal is different? What if you want to ensure that from your vantage point $p$, you can pan, tilt, and zoom in such a way that you can create movement in any direction on your 2D screen? If for every possible tiny movement on the screen, there is some movement in your 3D world that produces it, then you have achieved a ​​submersion​​. Mathematically, a map $f$ is a submersion at a point $p$ if its differential $df_p$ is ​​surjective​​ (onto). It means you're not "missing" any dimension of your target screen; your map locally covers everything. This immediately tells us something about dimensions: to map onto an $n$-dimensional space, you must be coming from a space of at least that dimension. For a submersion from a manifold $M$ of dimension $m$ to one of dimension $n$, we must have $m \ge n$.

This local "covering" property of submersions leads to a remarkably beautiful and powerful result, a cornerstone of differential geometry often called the ​​Submersion Theorem​​ or the Rank Theorem. It tells us that if a map is a submersion at a point, then no matter how twisted or complicated the map might seem globally, you can always find a special local coordinate system—a "perfect camera angle"—in which the map becomes beautifully simple: a standard projection.

Imagine our map $F$ from a 3D space ($\mathbb{R}^3$) to a 2D plane ($\mathbb{R}^2$) is a submersion at a point $p$. The theorem guarantees that we can invent new coordinates $(u, v, w)$ for the space around $p$ such that the map $F$, described in these new coordinates, is just $F(u, v, w) = (v, w)$. It's as if we've aligned our world so perfectly that the complicated map simply becomes "ignore the first coordinate". All the complexity is bundled up in the change of coordinates; the mapping itself is trivial. This is a recurring theme in physics and mathematics: find the right perspective, and the problem becomes simple.

What is this "straightening out" trick good for? It turns out to be the key to one of the most elegant ways to construct new mathematical worlds. It allows us to build what are called ​​submanifolds​​.

Let's stick with our map $f: M \to N$. Instead of looking at where all points go, let's ask the reverse question: what is the set of all points in our world $M$ that get mapped to a single, specific point $q$ in our target $N$? This set is called the ​​fiber​​ or ​​level set​​ over $q$, denoted $f^{-1}(q)$.

Now, suppose we pick a point $q$ in $N$ that is a ​​regular value​​. This is a technical term with a simple meaning: it just means that for every point $p$ in its fiber $f^{-1}(q)$, the map $f$ is a submersion at $p$. If this condition holds, the ​​Preimage Theorem​​ (or Regular Value Theorem) springs into action. It tells us that the entire level set $f^{-1}(q)$ is a perfect, smooth submanifold of our original world $M$. It's a new, smaller world living inside the old one.

And what is its dimension? The Submersion Theorem gives us the answer directly. If in local coordinates the map is $(u_1, \dots, u_m) \mapsto (u_1, \dots, u_n)$, the fiber over the origin is the set of points where the first $n$ coordinates are zero. This is a space defined by the remaining $m-n$ coordinates. So, the dimension of the fiber is always $\dim(M) - \dim(N)$.

Let's make this beautifully concrete with the simplest kind of submersion: a smooth function from a manifold $M$ to the real number line, $f: M \to \mathbb{R}$. Here, the target dimension is $n=1$. The map is a submersion at $p$ as long as its differential, $df_p$, is not the zero map. In this case, the Preimage Theorem tells us that level sets $f^{-1}(c)$ are submanifolds of dimension $m-1$, also known as hypersurfaces.

This should feel familiar. If you are in ordinary 3D space ($M = \mathbb{R}^3$) and $f(x,y,z)$ is a smooth function, the condition $df_p \neq 0$ is the same as saying the ​​gradient vector​​ $\nabla f(p)$ is not the zero vector. The level set $f^{-1}(c)$ is then a surface (a 2D submanifold). The tangent space to this surface at $p$, $T_p(f^{-1}(c))$, is the set of all directions you can move while staying on the surface. Moving in such a direction means the value of $f$ does not change. In other words, the directional derivative is zero. This is precisely the definition of the ​​kernel​​ of the differential, $\ker(df_p)$. So, $T_p(f^{-1}(c)) = \ker(df_p)$. And from multivariable calculus, we know exactly what this is: it's the plane of vectors orthogonal to the gradient vector $\nabla f(p)$. The abstract theorem of differential geometry lands us squarely back on a familiar and intuitive concept: a surface's tangent plane is perpendicular to its normal vector.

The submersion property provides a magnificent structure to the tangent space $T_pM$ at every point. Since the tangent space to the fiber is the kernel of the differential, we give it a special name: the ​​vertical space​​, $\mathcal{V}_p = \ker(df_p)$. These are the directions of motion "along the fiber," movements that are invisible to the map $f$.

What about the other directions? The First Isomorphism Theorem of linear algebra tells us that the differential $df_p$ provides a natural isomorphism between the quotient space $T_pM / \mathcal{V}_p$ and the target tangent space $T_{f(p)}N$. This quotient space represents all the directions in $T_pM$ once we agree to ignore movements along the fiber. This gives a canonical, choice-free decomposition of movement into "vertical" (staying in the fiber) and "what's left over" (moving between fibers).

If our world $M$ is equipped with a Riemannian metric—a way to measure lengths and angles—we can go a step further. We can define a unique ​​horizontal space​​, $\mathcal{H}_p$, as the orthogonal complement to the vertical space $\mathcal{V}_p$. Now, the tangent space splits neatly into a direct sum: $T_pM = \mathcal{V}_p \oplus \mathcal{H}_p$. Every tangent vector has a unique vertical part and a unique horizontal part. When a submersion has the special property that it preserves the lengths of these horizontal vectors, we call it a ​​Riemannian submersion​​. But this is a special, rigid condition; a general submersion doesn't need to preserve lengths in this way.

This local mechanism of submersions—this simple rule about surjective differentials—gives rise to some of the most profound global structures in geometry.

One powerful application is the construction of ​​quotient manifolds​​. Imagine a Lie group $G$ acting on a manifold $M$. For example, the group of rotations $SO(3)$ acting on the plane $\mathbb{R}^3$. If this action is "nice" (smooth, free, and proper), then the map $\pi$ that sends each point $x \in M$ to its entire orbit $G \cdot x$ is a submersion. The set of all orbits, $M/G$, becomes a new, smooth manifold in its own right, and its dimension is $\dim M - \dim G$. This is a factory for creating new spaces. The sphere $S^2$ can be seen as the quotient of the rotation group $SO(3)$ by its subgroup $SO(2)$ of rotations around a fixed axis. The submersion framework provides the formal language to build these worlds.

The final apotheosis of this idea is the ​​fiber bundle​​. A theorem by Charles Ehresmann tells us that any proper, surjective submersion $f: M \to N$ is a ​​locally trivial fiber bundle​​. This means that while $M$ might be globally very complex, if you look at the part of $M$ that lies over a small open set $U \subset N$, it looks just like a simple product space: $f^{-1}(U) \cong F \times U$, where $F$ is the fiber.

The classic example is a Möbius strip. The base space $N$ is the central circle, and the fiber $F$ is a line segment. Locally, any small piece of the Möbius strip looks like a flat rectangle (a product of an interval of the circle and a line segment). But globally, it has a twist. It is not globally a cylinder ($S^1 \times F$). The submersion is the local rule, the engine that drives the geometry. The global topology is the rich, often surprising structure that emerges. The simple, local, and seemingly technical condition of a surjective differential is the seed from which entire, complex worlds and the beautiful, unifying theory of fiber bundles grow.

Now that we have grappled with the precise definition of a submersion, we can begin to appreciate its true power. Like a master artist's perspective drawing, a submersion is a special kind of projection from a higher-dimensional world to a lower-dimensional one. But unlike a simple shadow, which can distort and hide information, a submersion is a "perfect" projection. It is a smooth, continuous mapping that never tears, folds, or creates sharp corners. At every point, it is "fully expansive" in some directions, ensuring that no local information about the lower-dimensional target space is lost. This simple, elegant idea turns out to be one of the most powerful tools in the geometer's arsenal, allowing us to chisel new mathematical worlds from old ones, understand the deep structure of symmetry, and even relate the curvature of different universes.

Perhaps the most fundamental application of a submersion is its ability to create new manifolds. Imagine you have a smooth function $F$ that maps a manifold $M$ of dimension $n$ to a manifold $N$ of dimension $k$. You might ask: what does the set of all points in $M$ that map to a single, specific point $c$ in $N$ look like? This set is called the preimage or level set of $c$, denoted $F^{-1}(c)$.

Without any further information, this set could be a disaster—a tangled mess of points, a curve that crosses itself, or something with sharp corners. But if the map $F$ is a submersion, something miraculous happens. The ​​Preimage Theorem​​ (also known as the Regular Level Set Theorem) guarantees that this level set, $F^{-1}(c)$, is itself a beautifully smooth, perfectly-formed submanifold of $M$ with dimension $n-k$. The submersion condition acts like a guarantee of quality control; it ensures that at every point on the level set, the map $F$ is "pointing away" from the set in enough independent directions that the set itself must be smooth.

Think of a smooth temperature function on a 3D block of metal. The set of all points with a temperature of exactly 100 degrees forms a 2D surface. As long as the temperature gradient is non-zero (a condition related to the submersion property), this "isothermal surface" will be smooth. If the gradient were zero, we might have a hot spot, and the level set could shrink to a single point.

This idea can be extended to view the entire manifold $M$ as being neatly organized by the submersion. The whole space becomes a stack of these submanifold "slices," with each slice corresponding to a different point in the target space $N$. This structure, where a manifold is partitioned into a collection of smaller, disjoint submanifolds, is called a ​​foliation​​. A submersion provides one of the most natural and important ways to generate such a foliation, where the fibers (the level sets) form the leaves of the foliation. In local coordinates, this structure is beautifully simple: the submersion just looks like a projection $(x, y) \mapsto x$, and the fibers are the sets where $x$ is constant.

Nature and mathematics are filled with symmetries. We often want to consider objects that are "the same" under a certain group of transformations. For instance, in mechanics, the state of a rotating rigid body might be described by its orientation in space, but we understand that all orientations that differ by a rotation around the axis of symmetry are physically equivalent. How can we construct a space that represents these equivalence classes?

This is where the ​​Quotient Manifold Theorem​​ comes into play, and submersions are at its very heart. If a Lie group $G$ (a group that is also a smooth manifold, like the group of rotations) acts on a manifold $M$ in a "nice" way—specifically, the action is free (no element other than the identity fixes any point) and proper (a technical condition preventing orbits from accumulating in strange ways)—then the set of all orbits, denoted $M/G$, can be given the structure of a new, smooth manifold.

The magic that stitches this quotient space together into a manifold is the fact that the natural projection map $\pi: M \to M/G$, which sends each point to its orbit, is a smooth submersion. The fibers of this submersion are precisely the orbits of the group action. This means our powerful Preimage Theorem applies: each orbit is a beautiful, embedded submanifold of the original space $M$. A classic example is the action of a compact group, like the circle group $S^1$, which always acts properly; if the action is also free, the quotient is guaranteed to be a manifold.

This construction is ubiquitous. The complex projective space $\mathbb{C}P^n$, a cornerstone of algebraic geometry and quantum mechanics, can be constructed as the quotient of a sphere by a circle action. Moduli spaces, which parameterize families of geometric objects, are often constructed as quotients. Submersions provide the theoretical foundation for creating these new and often exotic mathematical worlds.

When we equip our manifolds with metrics, allowing us to measure distances and angles, the concept of a submersion deepens into that of a ​​Riemannian submersion​​. This is a submersion between Riemannian manifolds that respects the metric structure in a special way: it acts as an isometry for distances measured in directions "horizontal" to the fibers.

At each point of the total space $M$, the tangent space splits beautifully into two orthogonal subspaces: the ​​vertical space​​, which is tangent to the fiber passing through that point (and is simply the kernel of the differential, $\ker d\pi$), and the ​​horizontal space​​, its orthogonal complement ([@problem_s_id:3044245, 3053769]). The differential $d\pi$ annihilates the vertical vectors but maps the horizontal space isomorphically onto the tangent space of the base manifold $B$.

This decomposition is not just a neat local picture; it has profound global consequences. For example, if the total space $M$ is ​​geodesically complete​​ (meaning you can extend geodesics, or "straight lines," indefinitely), then the base space $B$ must also be complete. Any path in the base can be "lifted" to a unique horizontal path in the total space. This ensures that the global properties of the larger universe are inherited by the quotient world.

The most spectacular consequence arises when we study curvature. One might naively guess that if a space is "flat" (zero curvature), then any space it projects onto should also be flat. This is not true! The relationship is more subtle and far more beautiful. ​​O'Neill's curvature formula​​ reveals that the curvature of the base space $B$ is determined by two things: the curvature of the total space $M$ and a term that measures the "twisting" of the horizontal distribution. This twisting term, related to a tensor $A$, measures the failure of the horizontal planes to join up and form their own foliation.

A stunning application of this is the ​​Hopf fibration​​, a Riemannian submersion from the 3-sphere $S^3$ to the 2-sphere $S^2$. The space $S^3$ can be given a metric of constant sectional curvature $1$. The fibers of the fibration are great circles. Using O'Neill's formula, one can calculate that the induced metric on the base space $S^2$ must have a constant sectional curvature of exactly $4$. The extra curvature appears out of thin air, a direct consequence of the way the fibers are twisted as we move across the 3-sphere. The geometry of the projection is encoded in the curvature of the projected space.

The idea of a submersion is so fundamental that it continues to appear at the forefront of modern mathematics, often in generalized or approximate forms.

In the study of ​​collapsing manifolds​​, geometers analyze sequences of manifolds whose metric structure shrivels in some directions. A key result in this field, the Fibration Theorem, states that under certain curvature bounds, a manifold that "Gromov-Hausdorff converges" to a lower-dimensional space must locally look like a fiber bundle. The maps that realize this convergence are shown to be ​​almost Riemannian submersions​​, a notion that quantifies how close a map is to being a true Riemannian submersion. The submersion concept provides the very language needed to describe these dramatic geometric limits.

The concept has also been abstracted into higher categorical structures. A ​​Lie groupoid​​ is a generalization of a Lie group that can be thought of as a group where multiplication is not always defined. It consists of a manifold of "arrows" and a manifold of "objects." A crucial part of its definition is that the source and target maps, which assign an object to the beginning and end of each arrow, are required to be submersions. This single requirement ensures that the structure is well-behaved and allows the powerful machinery of differential geometry to be applied. This framework finds applications in areas like Poisson geometry and the study of foliations, where the holonomy groupoid captures the transverse geometry of the leaves.

From the simple, intuitive picture of a smooth projection wrapping the real line around a circle, the concept of a submersion unfolds into a deep and unifying principle. It is the tool we use to build new manifolds from old ones, to understand the structure of symmetry, to relate the intricate geometries of different spaces, and to frame the questions that drive modern geometric analysis. It is a testament to the power of a simple, elegant idea to illuminate the hidden structures of our mathematical universe.