A guided ascent from calculus to one of topology’s most counter-intuitive theorems.

Prologue

In 1957, a struggling graduate student named Stephen Smale proved something that most mathematicians believed was impossible: a sphere can be turned inside out through a continuous, smooth deformation — without ever cutting, tearing, or creasing the surface. The surface is allowed to pass through itself, but at every instant, it must remain a smooth immersion.

This result, known as sphere eversion, is an existence proof. Smale showed that the path exists in an abstract space of immersions; he never exhibited the deformation itself. It took decades before mathematicians constructed explicit eversions — first Bernard Morin (who was blind) in the 1960s, then William Thurston’s elegant corrugation method in the 1990s.

The animated eversion below, passing through the famous Morin surface halfway model, shows what the deformation looks like:

Sphere eversion animation

This post will build your understanding from multivariable calculus to the heart of the proof. We assume you know partial derivatives, the chain rule, and basic linear algebra. Everything else, we construct from the ground up.

Part I: The Language of Smooth Maps

1.1 What is a Sphere, Really?

In everyday life, a sphere is a rubber ball. In mathematics, we need precision. The 2-sphere $S^2$ is the set of points in $\mathbb{R}^3$ at unit distance from the origin:

\[S^2 = \{(x, y, z) \in \mathbb{R}^3 : x^2 + y^2 + z^2 = 1\}.\]

But for our purposes, we care less about where the sphere sits and more about the sphere as an abstract 2-dimensional surface — a manifold. Every point on $S^2$ has a neighborhood that looks like a patch of $\mathbb{R}^2$. We can cover $S^2$ with coordinate charts (for instance, stereographic projection from the north and south poles), and transitions between overlapping charts are smooth.

What is a manifold? A manifold is a space that locally looks like ordinary flat space $\mathbb{R}^n$, even if globally it has a more complicated shape. The idea is captured perfectly by maps of the Earth. The Earth’s surface is a sphere — it curves, it closes on itself — yet any atlas decomposes it into flat rectangular maps (a map of Europe, a map of Asia, etc.). Each individual map is a coordinate chart: a bijection between a patch of the sphere and a rectangle in $\mathbb{R}^2$. Where two maps overlap (say, the map of Europe and the map of Asia both cover Turkey), you need a rule for translating coordinates from one map to the other. If these translation rules are smooth (infinitely differentiable), the atlas defines a smooth manifold.

Formally: a smooth $n$-manifold $M$ is a topological space covered by open sets $U_\alpha$, each equipped with a homeomorphism $\varphi_\alpha: U_\alpha \to \mathbb{R}^n$ (the chart), such that on overlaps $U_\alpha \cap U_\beta$, the transition map $\varphi_\beta \circ \varphi_\alpha^{-1}: \mathbb{R}^n \to \mathbb{R}^n$ is smooth. The sphere $S^2$ is a 2-manifold: you can cover it with two charts via stereographic projection from the north and south poles, and the transition between them is smooth.

A smooth map $f: S^2 \to \mathbb{R}^3$ assigns to each point $p \in S^2$ a point $f(p) \in \mathbb{R}^3$. In local coordinates $(u, v)$ on a chart of $S^2$, we can write:

\[f(u, v) = \big(f_1(u,v),\; f_2(u,v),\; f_3(u,v)\big).\]

The standard inclusion $\iota: S^2 \hookrightarrow \mathbb{R}^3$, where each point maps to itself, is the simplest example. But there are many other smooth maps from $S^2$ into $\mathbb{R}^3$ — stretched, twisted, even self-intersecting.

1.2 Immersions: The No-Crease Condition

Not every smooth map is geometrically well-behaved. Consider crushing the entire sphere to a single point — that’s smooth, but it destroys all geometric information. We need a condition that says “the surface never collapses.”

Definition. A smooth map $f: S^2 \to \mathbb{R}^3$ is an immersion if its differential $df_p: T_pS^2 \to T_{f(p)}\mathbb{R}^3$ is injective at every point $p \in S^2$.

What is $T_pS^2$? The tangent space $T_pM$ at a point $p$ on a manifold $M$ is the set of all “velocity vectors” of smooth curves passing through $p$. Imagine an ant walking on the sphere. At any instant, the ant has a velocity — a direction and speed of motion. That velocity vector lives in a flat plane tangent to the sphere at the ant’s location. The set of all possible velocities at $p$ forms a 2-dimensional vector space, $T_pS^2$.

More formally: if $\gamma: (-\epsilon, \epsilon) \to S^2$ is any smooth curve with $\gamma(0) = p$, then $\gamma’(0)$ is a tangent vector at $p$. The tangent space $T_pS^2$ is the collection of all such velocity vectors. For $S^2 \subset \mathbb{R}^3$, it’s literally the plane tangent to the sphere at $p$. For $\mathbb{R}^3$ itself, the tangent space $T_q\mathbb{R}^3$ at any point $q$ is just $\mathbb{R}^3$ again (flat space is its own tangent space).

What is the differential $df_p$? The differential (also called the differential map or pushforward) is the generalization of “the derivative is a linear map” from multivariable calculus to manifolds. Recall that in calculus, if $f: \mathbb{R}^2 \to \mathbb{R}^3$, the derivative at a point $p$ is a linear map that sends a small displacement $\mathbf{v}$ to $J_f(p) \cdot \mathbf{v}$, where $J_f$ is the Jacobian matrix. On a manifold, $df_p: T_pS^2 \to T_{f(p)}\mathbb{R}^3$ does the same thing: it takes a tangent vector at $p$ (a velocity of a curve on $S^2$) and maps it to a tangent vector at $f(p)$ (the velocity of the image curve in $\mathbb{R}^3$). In coordinates, $df_p$ is the Jacobian matrix. The notation just makes explicit that the derivative acts between tangent spaces.

What does “injective” mean? A function is injective (one-to-one) if different inputs always produce different outputs. For instance, $f(x) = 2x$ is injective: if $f(a) = f(b)$ then $a = b$. But $f(x) = x^2$ is not injective: $f(3) = f(-3) = 9$, so two different inputs give the same output. When we say $df_p$ is injective, we mean: if two tangent vectors at $p$ are different, their images under $df_p$ are also different. Equivalently, the only tangent vector that $df_p$ sends to zero is the zero vector itself. This ensures the surface doesn’t “collapse” at $p$.

In coordinates, this means the Jacobian matrix of $f$ has full rank everywhere:

\[J_f(u,v) = \begin{pmatrix} \dfrac{\partial f_1}{\partial u} & \dfrac{\partial f_1}{\partial v} \\[8pt] \dfrac{\partial f_2}{\partial u} & \dfrac{\partial f_2}{\partial v} \\[8pt] \dfrac{\partial f_3}{\partial u} & \dfrac{\partial f_3}{\partial v} \end{pmatrix}\]

This is a $3 \times 2$ matrix. For $f$ to be an immersion, we need $\operatorname{rank}(J_f) = 2$ at every point — the two columns must be linearly independent. Geometrically, the partial derivatives $\frac{\partial f}{\partial u}$ and $\frac{\partial f}{\partial v}$ span a tangent plane to the image surface at every point.

What does rank 2 forbid? It forbids creases — points where the surface folds onto itself so sharply that the tangent plane degenerates into a line or a point. Think of folding a piece of paper: at the fold line, the two sides of the paper share the same tangent line rather than spanning a plane. An immersion is allowed to self-intersect (the surface passes through itself), but it must never crease.

Calculus checkpoint. If you’ve computed the tangent plane to a parametric surface $\mathbf{r}(u,v)$ by taking $\mathbf{r}_u \times \mathbf{r}_v$ and checking it’s nonzero, you already understand the immersion condition: $\mathbf{r}_u \times \mathbf{r}_v \neq \mathbf{0}$ everywhere is equivalent to $\operatorname{rank}(J_f) = 2$.

Worked examples. To build intuition, let’s look at curves (one dimension down). A curve $\gamma: \mathbb{R} \to \mathbb{R}^2$ is an immersion if $\gamma’(t) \neq \mathbf{0}$ for all $t$ — the curve never “stops moving.”

Figure-eight $\gamma(t) = (\sin 2t, \sin t)$: This is an immersion. We have $\gamma’(t) = (2\cos 2t, \cos t)$, and you can verify this is never the zero vector. But the curve crosses itself at the origin. Self-intersection is fine — immersion does not mean injective. What it does mean is that at the crossing point, the curve passes through cleanly, with a well-defined tangent direction on each branch.
Cusp $\gamma(t) = (t^2, t^3)$: This is not an immersion. At $t=0$, we get $\gamma’(0) = (0, 0) = \mathbf{0}$ — the velocity vanishes. The curve comes to a sharp point (a cusp). This is exactly the kind of degenerate behavior the immersion condition forbids.

The distinction for surfaces is analogous. An immersed sphere may pass through itself (creating curves of self-intersection), but at every point, the surface has a well-defined tangent plane. A non-immersion would have a point where the surface pinches down to a line or a point — a crease or a cusp.

Embedding vs. immersion. An embedding is an immersion that is also injective (one-to-one) — the surface doesn’t touch itself at all. The standard sphere $\iota: S^2 \hookrightarrow \mathbb{R}^3$ is an embedding. During an eversion, the intermediate surfaces are immersions but not embeddings — they must pass through themselves. The key constraint is that they remain immersions (no creases), even though they are not embeddings (they self-intersect).

1.3 What is Eversion?

The standard embedding $\iota$ maps $S^2$ into $\mathbb{R}^3$ with the “outside” facing outward. The antipodal map $\alpha: S^2 \to \mathbb{R}^3$ defined by $\alpha(p) = -p$ gives us the same geometric sphere, but with the orientation reversed — inside facing out.

Why does $\alpha$ reverse orientation? At any point $p$ on the sphere, the outward unit normal is $\mathbf{n}(p) = p$ (pointing away from the origin). Under the antipodal map, the point $p$ goes to $-p$. The differential of $\alpha$ maps each tangent vector $v$ to $-v$, so the cross product of two tangent vectors flips sign:

\[d\alpha(\mathbf{e}_1) \times d\alpha(\mathbf{e}_2) = (-\mathbf{e}_1) \times (-\mathbf{e}_2) = \mathbf{e}_1 \times \mathbf{e}_2\]

Wait — the cross product of two negated vectors is the same? Yes, but the normal relative to the new position $-p$ now points inward: the outward normal at $-p$ on the standard sphere is $-p$, but our surface’s normal (inherited from $\alpha$) is $+p$, which points inward at the new location. The surface has been turned inside out. More precisely, $\alpha$ is an orientation-reversing map because its Jacobian has determinant $(-1)^3 = -1$ (it negates all three coordinates).

An eversion is a smooth, continuous path of immersions connecting $\iota$ to $\alpha$. Formally:

Definition. A regular homotopy between two immersions $f_0, f_1: S^2 \to \mathbb{R}^3$ is a smooth map

\[F: S^2 \times [0,1] \to \mathbb{R}^3\]

such that:

$F(\cdot, 0) = f_0$ and $F(\cdot, 1) = f_1$,
For every $t \in [0,1]$, the map $f_t = F(\cdot, t): S^2 \to \mathbb{R}^3$ is an immersion.

A sphere eversion is a regular homotopy from $\iota$ to $\alpha$. At every instant $t$, the intermediate surface $f_t(S^2)$ must be a smooth immersion — self-intersections are allowed, but creases are not.

“Why can’t we just push it through?” A natural first attempt: push the north pole down through the south pole, like inverting a sock. The problem is that at the moment the north pole passes through the equator, the surface develops a crease — a circle of points where the tangent plane degenerates. The equator becomes a fold line where the surface doubles back on itself, violating the immersion condition. Any “obvious” physical inversion creates creases. The genius of eversion is finding a path that avoids all creases, at the cost of allowing the surface to pass through itself in more subtle ways.

To see why this is hard, note that the immersion condition $\operatorname{rank}(J_f) = 2$ must hold at every point and at every instant $t \in [0,1]$. That’s an infinite family of pointwise conditions. The eversion must thread through this infinite-dimensional needle.

An analogy from calculus. Think of the space of all immersions as an infinite-dimensional landscape. Each point in this landscape is a particular immersion $f: S^2 \to \mathbb{R}^3$. A regular homotopy is a path in this landscape. Smale’s theorem says: the standard sphere $\iota$ and the everted sphere $\alpha$ lie in the same path-connected component of this landscape.

    ■ outside  
    ■ inside
  

Deformation parameter t: 0.00

Drag to rotate. Use the slider to deform the sphere. Indigo shows the outside surface, orange shows the inside. As $t$ increases, corrugations develop, lobes push through each other, and the sphere turns inside out — the orange interior becomes the exterior. Notice how the surface always remains smooth (no creases), even as it passes through itself.

Part II: The Space of Immersions

2.1 Configuration Space

Here is where the story shifts from individual maps to the space of all possible maps. This leap — from studying objects to studying the space of objects — is one of the most powerful ideas in modern mathematics.

Let $\operatorname{Imm}(S^2, \mathbb{R}^3)$ denote the space of all $C^2$ immersions of $S^2$ into $\mathbb{R}^3$. Each point in this space is an entire immersion $f: S^2 \to \mathbb{R}^3$. The topology on this space is the $C^1$ uniform topology: two immersions are “close” if they are close in value and their first derivatives are close, at every point of $S^2$.

What does $C^2$ mean? The notation $C^k$ describes how many times a function can be differentiated. A $C^0$ function is merely continuous (no jumps). A $C^1$ function has a continuous first derivative (no sharp corners). A $C^2$ function has continuous first and second derivatives (the surface curves smoothly, with no abrupt changes in curvature). A $C^\infty$ function — also called smooth — can be differentiated infinitely many times. The immersion condition requires checking the first derivative (the Jacobian), so $C^1$ would suffice for that, but Smale works with $C^2$ to ensure the second-derivative structure needed for the analysis is available.

This is an infinite-dimensional space. Just as a curve in $\mathbb{R}^3$ can be specified by three functions of one variable, an immersion $f: S^2 \to \mathbb{R}^3$ is specified by three functions of two variables — and the “dimension” of the function space reflects the infinite degrees of freedom available.

Analogy from calculus of variations. In the calculus of variations, you study functionals like

\[\mathcal{L}[y] = \int_a^b L(x, y, y') \, dx\]

where each function $y(x)$ is a point in an infinite-dimensional function space. The Euler–Lagrange equation finds critical points of $\mathcal{L}$ in this space.

What is the Euler–Lagrange equation? Just as finding the maximum or minimum of a function $f(x)$ requires solving $f’(x) = 0$, finding the function $y(x)$ that maximizes or minimizes a functional $\mathcal{L}[y]$ requires solving a differential equation. That equation is the Euler–Lagrange equation:

\[\frac{\partial L}{\partial y} - \frac{d}{dx}\frac{\partial L}{\partial y'} = 0.\]

The analogy is: $f’(x) = 0$ says “the function $f$ is flat at $x$” (no small change in $x$ improves the value). The Euler–Lagrange equation says “$\mathcal{L}$ is flat at $y$” (no small change in the function $y$ improves the value). For example, to find the shortest path between two points in the plane, you minimize $\mathcal{L}[y] = \int_a^b \sqrt{1 + (y’)^2}\, dx$. The Euler–Lagrange equation gives $y’’ = 0$, so the critical points are straight lines — exactly what you’d expect.

Smale’s approach is similar in spirit: he studies the topology of a function space (the space of immersions) rather than optimizing a functional on it.

2.2 Paths in Configuration Space

A regular homotopy $f_t$ with $t \in [0,1]$ is simply a continuous path in $\operatorname{Imm}(S^2, \mathbb{R}^3)$. Two immersions $f$ and $g$ are regularly homotopic if they lie in the same path-connected component of $\operatorname{Imm}(S^2, \mathbb{R}^3)$.

What is a path-connected component? Take all the points (immersions) in $\operatorname{Imm}(S^2, \mathbb{R}^3)$ and declare two of them equivalent if you can draw a continuous path between them. Each equivalence class is a “path-connected component.” If the entire space has only one component, every immersion can be deformed into every other — that’s what Smale proves.

Smale’s eversion theorem can now be stated with striking simplicity:

Theorem (Smale, 1958). The space $\operatorname{Imm}(S^2, \mathbb{R}^3)$ is path-connected.

That’s it. Every immersion of $S^2$ in $\mathbb{R}^3$ can be continuously deformed into every other immersion, through immersions. In particular, the standard embedding $\iota$ and the antipodal embedding $\alpha$ are connected by a path. The sphere can be everted.

2.3 Tangential Data and the Stiefel Manifold

To classify immersions, Smale extracted an invariant from the tangential data of each immersion. At each point $p \in S^2$, an immersion $f$ determines a 2-frame in $\mathbb{R}^3$: the pair of vectors

\[\left(\frac{\partial f}{\partial u}(p),\; \frac{\partial f}{\partial v}(p)\right)\]

which are linearly independent (by the immersion condition). This pair lives in the Stiefel manifold:

\[V_{3,2} = \{(v_1, v_2) \in \mathbb{R}^3 \times \mathbb{R}^3 : v_1, v_2 \text{ linearly independent}\}.\]

Unpacking the Stiefel manifold. $V_{3,2}$ is the space of all ordered pairs of linearly independent vectors in $\mathbb{R}^3$. Think of it as the space of all possible “coordinate frames” for a tangent plane sitting inside $\mathbb{R}^3$. Here “ordered” matters: the pair $(\mathbf{e}1, \mathbf{e}_2)$ is a different point in $V{3,2}$ from $(\mathbf{e}_2, \mathbf{e}_1)$, because swapping the vectors reverses the orientation of the frame.

Concrete example. At the north pole $(0,0,1)$ of the standard sphere, the tangent plane is the $xy$-plane, and the immersion $\iota$ gives us the frame $\left(\frac{\partial \iota}{\partial u}, \frac{\partial \iota}{\partial v}\right) = (\mathbf{e}1, \mathbf{e}_2)$. At the equator point $(1,0,0)$, the tangent plane is the $yz$-plane, and the frame might be $(\mathbf{e}_2, -\mathbf{e}_3)$. As you move across the sphere, the frame varies continuously — you get a continuous map $T\iota: S^2 \to V_{3,2}$.

More precisely, we can normalize to get ordered pairs of orthonormal vectors, giving the compact Stiefel manifold $V_{3,2} \cong SO(3)$.

So each immersion $f$ induces a map $T_f: S^2 \to V_{3,2}$ — the tangential map that records the tangent frame at each point. Smale’s invariant $\Omega(f, g)$ measures the difference between the tangential maps of two immersions.

2.4 The Gauss Map Perspective

There is a closely related and perhaps more intuitive viewpoint. The Gauss map of an immersion $f$ is:

\[\mathbf{n}_f(p) = \frac{\frac{\partial f}{\partial u} \times \frac{\partial f}{\partial v}}{\left\|\frac{\partial f}{\partial u} \times \frac{\partial f}{\partial v}\right\|}\]

This assigns to each point $p \in S^2$ the unit normal vector to the image surface at $f(p)$. The Gauss map $\mathbf{n}_f: S^2 \to S^2$ has a degree — the number of times the normal vector wraps around the target sphere.

What is the degree of a map? For a continuous map $g: S^n \to S^n$ from a sphere to itself, the degree is an integer that counts “how many times the domain wraps around the target,” with orientation. A concrete example: think of a map $g: S^1 \to S^1$ from the circle to itself. If $g(\theta) = 2\theta$ (the angle doubles), then as $\theta$ traverses the circle once ($0$ to $2\pi$), the image traverses the circle twice. This map has degree $2$. The identity map $g(\theta) = \theta$ has degree $1$. A constant map (everything goes to one point) has degree $0$. The reflection $g(\theta) = -\theta$ has degree $-1$ — it wraps around once but reverses orientation. For maps $S^2 \to S^2$, the same idea applies: degree counts the signed number of times the image covers the target sphere.

A classical result states that the degree of the Gauss map equals $1$ for any immersion of $S^2$ in $\mathbb{R}^3$.

Why is the degree always 1? For the standard embedding, this is intuitive: the outward normal at a point $p$ is just $p$ itself, so the Gauss map is the identity $S^2 \to S^2$, which has degree $1$. The deeper fact is that degree is invariant under regular homotopy (the degree of a continuous family of maps can’t jump, since it’s an integer varying continuously). Since every immersion of $S^2$ in $\mathbb{R}^3$ is regularly homotopic to the standard embedding (that’s Smale’s theorem!), every immersion has Gauss map degree $1$. Alternatively, the degree of the Gauss map equals $\frac{1}{2}\chi(S^2) = 1$ by the Gauss–Bonnet theorem, where $\chi(S^2) = 2$ is the Euler characteristic, a topological invariant that doesn’t depend on the immersion.

This is reassuring: both the standard sphere (outward normals) and the everted sphere (inward normals, but now on the “outside”) have Gauss map degree $1$. The Gauss map degree does not obstruct eversion.

But this is not the whole story — the Gauss map captures only the normal direction, not the full tangent frame. Smale’s invariant $\Omega(f,g)$ is finer, living in the homotopy group $\pi_2(V_{3,2})$.

Part III: The Topology of the Proof

3.1 Why Eversion is Surprising

Before diving into Smale’s argument, let’s understand why eversion seems impossible.

The 1-dimensional analogy. Consider a circle $S^1$ immersed in $\mathbb{R}^2$. Can we evert it? An immersed circle in the plane has a well-defined turning number — the total number of times the tangent vector winds around as you traverse the curve. The standard circle has turning number $+1$; its reflection has turning number $-1$. The Whitney–Graustein theorem tells us that two immersed curves in $\mathbb{R}^2$ are regularly homotopic if and only if they have the same turning number. Since $+1 \neq -1$, a circle cannot be everted in the plane.

Concrete calculation. Parametrize the unit circle as $\gamma(t) = (\cos t, \sin t)$ for $t \in [0, 2\pi]$. The unit tangent vector is:

\[\mathbf{T}(t) = \frac{\gamma'(t)}{\|\gamma'(t)\|} = (-\sin t, \cos t).\]

The angle this tangent makes with the $x$-axis is $\theta(t) = t + \pi/2$. As $t$ goes from $0$ to $2\pi$, $\theta$ increases by $2\pi$, so the turning number is $\frac{1}{2\pi}\Delta\theta = +1$.

For the reflected circle $\bar{\gamma}(t) = (\cos t, -\sin t)$ (traversed with opposite orientation), the tangent is $\bar{\mathbf{T}}(t) = (-\sin t, -\cos t)$, with angle $\theta(t) = -(t + \pi/2)$. Now $\theta$ decreases by $2\pi$, giving turning number $-1$.

Since turning number is invariant under regular homotopy (it’s an integer and varies continuously, so it can’t jump), $\gamma$ and $\bar{\gamma}$ can never be connected by a path of immersions.

Why does this obstruction vanish in one dimension higher? The turning number for $S^1 \subset \mathbb{R}^2$ lives in $\pi_1(S^1) = \mathbb{Z}$ — the tangent direction traces a loop in the circle of unit vectors in $\mathbb{R}^2$, and $\mathbb{Z}$ has many distinct elements. For $S^2 \subset \mathbb{R}^3$, the analogous invariant lives in $\pi_2(V_{3,2})$, which turns out to be $0$. The extra dimension provides enough room for the obstruction to disappear.

    Turning Number Visualization
  

Curve type:

Left: the immersed curve with tangent vectors (cyan arrows). Right: the tangent indicatrix — the path traced by the tip of the unit tangent vector on the unit circle (yellow). The turning number is how many times the yellow curve winds around the unit circle. Select different curves to see how turning number changes.

This is the source of the intuition that eversion should be impossible: in one dimension lower, it is impossible. The surprise is that going from $S^1 \subset \mathbb{R}^2$ to $S^2 \subset \mathbb{R}^3$, the obstruction vanishes.

3.2 Smale’s Classification Theorem

Smale proved a far more general result. Let $V_{n,2}$ be the Stiefel manifold of 2-frames in $\mathbb{R}^n$.

Theorem A (Smale, 1958). Two $C^2$ immersions $f, g: S^2 \to \mathbb{R}^n$ are regularly homotopic if and only if $\Omega(f,g) = 0$ in $\pi_2(V_{n,2})$.

Moreover, for any $\Omega_0 \in \pi_2(V_{n,2})$ and any immersion $f$, there exists an immersion $g$ with $\Omega(f,g) = \Omega_0$.

This establishes a one-to-one correspondence between regular homotopy classes of immersions and elements of $\pi_2(V_{n,2})$.

3.3 Constructing the Invariant $\Omega(f,g)$

Here is how $\Omega(f,g)$ is defined. Start with two immersions $f$ and $g$ that agree on a neighborhood of some base point $z_0 \in S^2$ (we can always arrange this by a preliminary deformation).

Remove a small disk around $z_0$. On the remaining region (topologically a disk $D$), $f$ and $g$ induce two maps into $V_{n,2}$ that agree on the boundary $\partial D$. Glue these two maps together along $\partial D$ to get a single map:

\[\Phi: S^2 \to V_{n,2}.\]

Intuition for the gluing. Think of it this way: $f$ paints the disk $D$ one color pattern in $V_{n,2}$, and $g$ paints the same disk a different color pattern, but they agree on the boundary circle. If you take the disk painted by $f$ as the “northern hemisphere” and the disk painted by $g$ as the “southern hemisphere,” and glue them along their shared boundary, you get a sphere mapped into $V_{n,2}$. The homotopy class of this map — whether it can be shrunk to a point or wraps nontrivially around $V_{n,2}$ — is the invariant $\Omega(f,g)$. If it can be shrunk to a point ($\Omega = 0$), then $f$ and $g$ are “the same” up to regular homotopy.

The homotopy class $[\Phi] \in \pi_2(V_{n,2})$ is the invariant $\Omega(f,g)$.

3.4 The Key Computation: $\pi_2(V_{3,2}) = 0$

For sphere eversion, we need $n = 3$. The classification reduces to computing $\pi_2(V_{3,2})$.

Step 1. The Stiefel manifold $V_{3,2}$ is homeomorphic to $SO(3)$, the rotation group of $\mathbb{R}^3$.

What is $SO(3)$? $SO(3)$ is the group of all rotations of 3-dimensional space. Each element is a $3 \times 3$ orthogonal matrix with determinant $+1$ (the “S” stands for “special,” meaning determinant 1; the “O” stands for “orthogonal”). Concretely, every rotation is determined by an axis (a line through the origin) and an angle of rotation about that axis. $SO(3)$ is a 3-dimensional manifold: you need three numbers to specify a rotation — for instance, the two angles that determine the rotation axis (a point on $S^2$) plus the rotation angle. It’s a manifold because nearby rotations correspond to nearby parameters, and it’s a group because you can compose rotations and take inverses. As a topological space, $SO(3)$ is compact and connected but not simply connected: there is a loop in $SO(3)$ (a $360°$ rotation about a fixed axis) that cannot be shrunk to a point, but doing the same rotation twice ($720°$) gives a loop that can be shrunk to a point. This is the famous “plate trick” or “belt trick.”

Step 2. $SO(3)$ is homeomorphic to $\mathbb{RP}^3$ (real projective 3-space). This follows from the quaternion double cover: $SU(2) \cong S^3$ maps onto $SO(3)$ with kernel ${\pm 1}$, so $SO(3) \cong S^3 / {\pm 1} = \mathbb{RP}^3$.

What is $\mathbb{RP}^3$? Real projective $n$-space $\mathbb{RP}^n$ is the space of all lines through the origin in $\mathbb{R}^{n+1}$. Each “point” of $\mathbb{RP}^n$ is not a single point but a line. Equivalently, $\mathbb{RP}^n = S^n / {\pm 1}$: start with the $n$-sphere and declare each point equivalent to its antipode (diametrically opposite point), since both lie on the same line through the origin.

Start with the simplest case: $\mathbb{RP}^1$. Take the circle $S^1$ and glue each point to its antipode. The top half of the circle gets glued to the bottom half, and you get… another circle. So $\mathbb{RP}^1 \cong S^1$. For $\mathbb{RP}^2$: take the 2-sphere and glue antipodal points. You get a surface that cannot be embedded in $\mathbb{R}^3$ without self-intersection — it’s a non-orientable surface (the Boy’s surface from section 4.2 is an immersion of $\mathbb{RP}^2$ in $\mathbb{R}^3$). For $\mathbb{RP}^3$: take $S^3$ (the 3-sphere, living in $\mathbb{R}^4$) and glue each point to its antipode. The result is a compact 3-manifold. It’s harder to visualize, but you can think of it as a solid ball $B^3$ where each point on the boundary is identified with the diametrically opposite boundary point.

What does this mean concretely? A unit quaternion $q = a + bi + cj + dk$ with $a^2 + b^2 + c^2 + d^2 = 1$ represents a rotation of $\mathbb{R}^3$: the rotation by angle $2\cos^{-1}(a)$ about the axis $(b, c, d)/|(b,c,d)|$. But $q$ and $-q$ represent the same rotation (negating a quaternion flips both the axis direction and the rotation sense, which cancel out). So each rotation corresponds to a pair ${q, -q}$ of antipodal points on $S^3$. The space of such antipodal pairs is exactly $\mathbb{RP}^3 = S^3/{\pm 1}$. Hence $SO(3) \cong \mathbb{RP}^3$.

Step 3. We compute $\pi_2(\mathbb{RP}^3)$ using the long exact sequence of the double covering $p: S^3 \to \mathbb{RP}^3$ with fiber $S^0 = {+1, -1}$:

What is a covering space? A covering space of $X$ is a space $\tilde{X}$ together with a map $p: \tilde{X} \to X$ such that every point in $X$ has a neighborhood $U$ where $p^{-1}(U)$ is a disjoint union of copies of $U$, each mapped homeomorphically onto $U$ by $p$. Think of it as “stacking copies”: locally, $\tilde{X}$ looks like several identical floors of a building, and $p$ is the projection that forgets which floor you’re on.

Concrete example: The real line $\mathbb{R}$ covers the circle $S^1$ via the map $p(t) = e^{2\pi i t}$. Each point on the circle has infinitely many preimages (all differing by integers). Another example: $S^1$ covers itself via $z \mapsto z^2$ — each point has exactly 2 preimages, so this is a double cover. In our case, $S^3$ double-covers $\mathbb{RP}^3$: the map $p$ sends each point $q \in S^3$ to the line ${q, -q}$ in $\mathbb{RP}^3$, so each point of $\mathbb{RP}^3$ has exactly 2 preimages (namely $q$ and $-q$). The fiber $p^{-1}(\text{point})$ is $S^0 = {+1, -1}$, a discrete two-point set.

What is a long exact sequence? When one space covers another (like $S^3$ double-covering $\mathbb{RP}^3$), the homotopy groups of the three spaces involved — the covering space, the base, and the fiber — are linked by a chain of group homomorphisms. “Exact” means that at each position in the chain, the image of one map equals the kernel of the next. This constraint is so rigid that knowing some of the groups forces the others. In our case, it forces $\pi_2(\mathbb{RP}^3) \cong \pi_2(S^3)$.

Group homomorphism, kernel, and image. A group homomorphism is a function $\varphi: G \to H$ between groups that preserves the group operation: $\varphi(a \cdot b) = \varphi(a) \cdot \varphi(b)$. For example, $\varphi: \mathbb{Z} \to \mathbb{Z}$ defined by $\varphi(n) = 2n$ is a homomorphism: $\varphi(a + b) = 2(a+b) = 2a + 2b = \varphi(a) + \varphi(b)$.

The image of $\varphi$ is the set of all outputs: $\operatorname{im}(\varphi) = {\varphi(g) : g \in G}$. For $\varphi(n) = 2n$, the image is the even integers $2\mathbb{Z}$.

The kernel of $\varphi$ is the set of inputs that map to the identity: $\ker(\varphi) = {g \in G : \varphi(g) = e_H}$. For $\varphi(n) = 2n$, the kernel is ${0}$ (only zero maps to zero). In the exact sequence $A \xrightarrow{f} B \xrightarrow{g} C$, “exact at $B$” means $\operatorname{im}(f) = \ker(g)$: what comes in from $f$ is exactly what gets killed by $g$.

The relevant portion of the sequence is:

\[\cdots \to \pi_2(S^0) \to \pi_2(S^3) \to \pi_2(\mathbb{RP}^3) \to \pi_1(S^0) \to \pi_1(S^3) \to \cdots\]

Since $\pi_k(S^0) = 0$ for $k \geq 1$ and $\pi_1(S^3) = 0$:

\[0 \to \pi_2(S^3) \to \pi_2(\mathbb{RP}^3) \to 0\]

so $\pi_2(\mathbb{RP}^3) \cong \pi_2(S^3)$.

Step 4. $\pi_2(S^3) = 0$ because every continuous map $S^2 \to S^3$ is homotopic to a constant — intuitively, a 2-sphere in a 3-sphere has “room” to be contracted to a point (this is a special case of $\pi_k(S^n) = 0$ for $k < n$).

Intuition for $\pi_k(S^n) = 0$ when $k < n$. Think of a lower-dimensional analogy: a loop ($S^1$) drawn on the surface of a 2-sphere ($S^2$). No matter how complicated the loop is, you can always shrink it to a point — that’s $\pi_1(S^2) = 0$. The loop never gets “stuck” because the sphere has no holes for it to catch on. Similarly, a 2-sphere ($S^2$) mapped into a 3-sphere ($S^3$) can always be contracted to a point: the 3-sphere is simply connected in every dimension up to 2. The map $S^2 \to S^3$ is like drawing a surface inside a higher-dimensional space that has no “obstacles” at that dimension.

Conclusion:

\[\pi_2(V_{3,2}) \cong \pi_2(SO(3)) \cong \pi_2(\mathbb{RP}^3) \cong \pi_2(S^3) = 0.\]

Since $\pi_2(V_{3,2}) = 0$, the invariant $\Omega(f,g)$ is always zero. By Theorem A, any two immersions of $S^2$ in $\mathbb{R}^3$ are regularly homotopic. In particular, the standard embedding and its reflection are regularly homotopic.

The sphere can be everted.

A note on what this proof does and does not give us. The argument above tells us that an eversion exists. It does not show us what it looks like. The proof works by showing that the space $\operatorname{Imm}(S^2, \mathbb{R}^3)$ has only one path component (via the vanishing of $\pi_2(V_{3,2})$), so the two immersions must be connected by some path — but the path is never constructed. This is why Smale’s result was so astonishing and initially met with disbelief: mathematicians knew the eversion existed but couldn’t picture it. It took Morin (1967) and later Thurston (1990s) to find explicit constructions.

3.5 The Fiber Bundle Machinery

How did Smale prove Theorem A? The proof is a masterpiece of algebraic topology, using fiber bundles and their homotopy sequences. Here is a roadmap.

What is a fiber bundle? Before diving in, let’s build some intuition. A fiber bundle is a space $E$ (the “total space”) that locally looks like a product $B \times F$, where $B$ is the “base space” and $F$ is the “fiber.” There’s a projection map $\pi: E \to B$ such that for each point $b \in B$, the preimage $\pi^{-1}(b)$ is a copy of the fiber $F$.

Everyday example. Consider a cylinder: it’s a product $S^1 \times [0,1]$. The base is the circle $S^1$, each fiber is an interval $[0,1]$, and the projection sends each point on the cylinder to the corresponding point on the circle. A Mobius strip is also a fiber bundle over $S^1$ with fiber $[0,1]$ — but it’s twisted, so it’s not a global product.

In Smale’s context: The total space $E$ is the space of immersions of a disk. The base space $B$ records what happens on the boundary (boundary data). The fiber $\pi^{-1}(b)$ over a particular boundary condition $b$ consists of all immersions that match that boundary data. The question “are two immersions regularly homotopic?” reduces to asking whether they lie in the same path-component of a fiber.

What is the Covering Homotopy Property (CHP)? If $(E, \pi, B)$ has the CHP (also called being a “fibration”), it means: if you have a path in the base space $B$ and a starting point in $E$ above the path’s start, you can “lift” the entire path to $E$. In Smale’s setting, this means: if you continuously deform the boundary data, you can continuously deform the immersion to match. This is the key technical property that lets Smale relate questions about immersions to questions about homotopy groups.

Smale constructs a tower of function spaces:

Space	Description
$E$	Regular maps of disk $D$ into $\mathbb{R}^n$, with $C^1$ topology
$B$	Boundary data: maps from $\partial D$ into $V_{n,2} \times \mathbb{R}^n$
$\pi: E \to B$	Restriction to boundary
$E_0, B_0$	Subspaces with fixed base point $x_0$
$E’, B_0’$	Auxiliary spaces of maps $(D, p_0) \to (V, x_0)$

The key results in the proof chain are:

Theorem 2.1: $(E, \pi, B)$ has the Covering Homotopy Property (CHP) — it’s a fibration. This means homotopies in the base space $B$ can be “lifted” to homotopies in the total space $E$.
Lemma 3.2: The auxiliary space $E’$ is contractible (via the homotopy $H_t(f)(p) = f(h_t(p))$ where $h_t$ contracts $D$ to a point).

What does “contractible” mean? A space is contractible if it can be continuously shrunk to a single point. Think of a ball of clay: you can smoothly squeeze it down to a point without tearing. Formally, $X$ is contractible if the identity map $\operatorname{id}: X \to X$ is homotopic to a constant map. Equivalently, all homotopy groups vanish: $\pi_k(X) = 0$ for all $k$. Examples: $\mathbb{R}^n$ is contractible (shrink everything to the origin: $H_t(x) = (1-t)x$). The interval $[0,1]$ is contractible. The circle $S^1$ is not contractible — you can’t shrink it to a point without tearing, and indeed $\pi_1(S^1) = \mathbb{Z} \neq 0$.
Lemma 3.3: $\pi_k(E_0) = 0$ for all $k \geq 0$ — the space of immersions of a disk with fixed boundary data has trivial homotopy groups.
Theorem 3.7: The map $\phi_0: \Gamma_c \to \Omega_d$ between fibers is a weak homotopy equivalence.
Lemma 3.8: $\pi_k(\Omega_d) = \pi_{k+2}(V_{n,2})$ — this connects the homotopy of the fibers to the homotopy of the Stiefel manifold.
Theorem 3.9: Combining everything: $\pi_k(\Gamma_c) = \pi_{k+2}(V_{n,2})$. For $k=0$: the set of path components of the fiber $\Gamma_c$ is in bijection with $\pi_2(V_{n,2})$.

Unpacking the notation. Here is what these symbols mean concretely:

$\Gamma_c$ is the fiber of the immersion fibration: it consists of all immersions of the disk $D$ into $\mathbb{R}^n$ that have a fixed boundary condition $c$ (i.e., they all agree on how they behave near the boundary of the disk). Two immersions being in the same path-component of $\Gamma_c$ means they can be connected by a regular homotopy that holds the boundary fixed — which is exactly the question of regular homotopy.
$\Omega_d$ is a loop space of the Stiefel manifold $V_{n,2}$. Specifically, $\Omega_d = \Omega^2(V_{n,2})$ consists of maps from a 2-disk into $V_{n,2}$ with prescribed boundary behavior $d$. The loop space is a standard algebraic topology construction that “shifts” homotopy groups: $\pi_k(\Omega^2 X) = \pi_{k+2}(X)$, which is why Lemma 3.8 relates $\pi_k(\Omega_d)$ to $\pi_{k+2}(V_{n,2})$.
$\phi_0$ is the map that sends an immersion to its tangential data. Given an immersion $f$ in $\Gamma_c$, define $\phi_0(f)$ by recording the tangent frame $\left(\frac{\partial f}{\partial u}, \frac{\partial f}{\partial v}\right)$ at each point — this is a map from the disk into $V_{n,2}$, i.e., an element of $\Omega_d$.
A weak homotopy equivalence $g: A \to B$ is a map that induces isomorphisms $\pi_k(A) \cong \pi_k(B)$ for all $k \geq 0$. It means $A$ and $B$ have identical homotopy groups (and the isomorphisms are induced by $g$ itself), even if $A$ and $B$ are not homeomorphic. For the purpose of computing homotopy groups — which is all we need — a weak homotopy equivalence is as good as a homeomorphism.

This is the bridge: the topology of the fiber (which controls whether two immersions with the same boundary data are regularly homotopic) is captured by the homotopy group $\pi_2(V_{n,2})$.

3.6 Contrast: $S^2$ in $\mathbb{R}^4$

The magic of $\pi_2(V_{3,2}) = 0$ is specific to codimension 1. For $n = 4$:

What is codimension? The codimension of a submanifold $M^k$ inside an ambient space $\mathbb{R}^n$ is $n - k$, the number of “extra dimensions” in the ambient space beyond those of $M$. A curve ($k = 1$) in $\mathbb{R}^3$ ($n = 3$) has codimension $3 - 1 = 2$. A surface ($k = 2$) in $\mathbb{R}^3$ has codimension $3 - 2 = 1$. Our sphere $S^2$ in $\mathbb{R}^3$ has codimension 1 — there is exactly one “extra” direction at each point (the normal direction). In $\mathbb{R}^4$, the same sphere $S^2$ would have codimension 2 — two independent normal directions at each point. Higher codimension gives more room for self-intersection to be “resolved,” but it also introduces new topological invariants that can obstruct regular homotopy.

\[\pi_2(V_{4,2}) \cong \mathbb{Z}.\]

This means there are infinitely many distinct regular homotopy classes of immersions $S^2 \to \mathbb{R}^4$, indexed by the integers. Smale’s Theorem C relates these classes to the characteristic class of the normal bundle, which for $S^2$ in $\mathbb{R}^4$ equals twice the algebraic self-intersection number of the immersion. Two immersions are regularly homotopic if and only if they have the same algebraic self-intersection number.

Normal bundle and characteristic class, explained. At each point $p$ of an immersed surface in $\mathbb{R}^n$, the tangent plane occupies 2 of the $n$ available dimensions. The remaining $n - 2$ dimensions form the normal space at $p$. As $p$ varies across the surface, these normal spaces fit together into a geometric object called the normal bundle — think of it as attaching a small normal “fiber” at every point, like the pile on a carpet.

A characteristic class is a topological invariant that measures how “twisted” a bundle is. The simplest analogy: consider a strip of paper. If you glue the ends without twisting, you get a cylinder — the normal bundle is trivial (untwisted). If you glue with a half-twist, you get a Mobius strip — the bundle is nontrivial, and the characteristic class detects this twist. For surfaces in $\mathbb{R}^4$, the relevant characteristic class is an integer (the Euler class of the normal bundle), and it equals $2 \times (\text{algebraic self-intersection number})$. This integer is the complete obstruction to regular homotopy.

Part IV: Making it Concrete

4.1 Thurston’s Corrugation Method

Smale’s proof is pure existence — it gives no recipe for constructing an eversion. The first explicit eversions were found by Bernard Morin. But the most conceptually transparent method is Thurston’s corrugation technique (1990s).

The idea: introduce waves (corrugations) into the surface that create enough “slack” to push the sphere through itself. The corrugations are like the pleats in a skirt — they add local complexity but allow global rearrangement.

Intuition: why corrugations help. Imagine trying to turn a rubber glove inside out, but the glove is rigid. You can’t do it. Now imagine the glove is made of an accordion-pleated material — the pleats let different parts of the surface slide past each other without creating creases. Corrugation works the same way: by adding high-frequency waves to the sphere, you create local “slack” that gives the surface freedom to rearrange globally. The waves are always smooth (satisfying the immersion condition), and their amplitude can be tuned continuously.

The procedure, at a high level:

Start with the standard sphere.
Corrugation phase: Introduce sinusoidal waves along the surface. The surface develops flower-like “petals” — regions that bulge outward and inward in an alternating pattern. The frequency $k$ is chosen high enough that the petals have room to interleave.
Interleaving phase: The petals from the “outward” side and the “inward” side are pushed past each other. Because they are narrow corrugations (high frequency), they can slide through without ever creating a crease. This is the key step — it’s where the surface passes through itself.
De-corrugation phase: Smoothly reduce the corrugation amplitude back to zero, arriving at the everted sphere.

Each of these phases is a smooth deformation, and the composition is a regular homotopy.

The corrugation function modifies the immersion by adding oscillatory terms:

\[f_{\text{corr}}(u, v) = f(u, v) + \epsilon(t) \cdot \mathbf{n}(u,v) \cdot \sin(k \cdot u)\]

where $\mathbf{n}$ is the normal, $k$ controls the frequency of corrugations, and $\epsilon(t)$ is an amplitude that varies during the homotopy.

4.2 The Morin Surface

The Morin surface is the halfway model of a sphere eversion — the self-intersecting surface at $t = 0.5$. It has a beautiful four-fold symmetry and a single quadruple point (where four sheets of the surface meet).

    Morin Surface — drag to rotate
  

The Morin surface: a self-intersecting immersion of $S^2$ that serves as the “halfway point” of a sphere eversion. Drag to rotate.

4.3 The Synchronicity Function

To track the progress of an eversion quantitatively, we define a synchronicity function. Let $f_t$ be a regular homotopy from $\iota$ (standard sphere) to $\alpha$ (everted sphere). Define:

\[S(t) = \int_{S^2} \mathbf{n}_{\text{initial}}(p) \cdot \mathbf{n}_t(p) \, dA\]

where $\mathbf{n}_{\text{initial}}(p)$ is the outward unit normal of the standard sphere at $p$, $\mathbf{n}_t(p)$ is the unit normal of the immersion $f_t$ at $p$, and $dA$ is the area element on $S^2$.

Interpretation:

At $t = 0$: normals align perfectly, so $S(0) = \int_{S^2} 1 \, dA = 4\pi$.
At $t = 1$: normals are reversed, so $S(1) = \int_{S^2} (-1) \, dA = -4\pi$.
$S(t) = 0$ at some intermediate time: this is where the surface is, on average, “half everted.”

The synchronicity function must be continuous (since $f_t$ varies continuously) and must cross zero — by the Intermediate Value Theorem! The halfway point $S(t^*) = 0$ corresponds roughly to the Morin surface.

Notice the elegant interplay: we used the Intermediate Value Theorem — a tool from freshman calculus — to say something meaningful about a process happening in an infinite-dimensional function space.

4.4 Stationary Points and Geometric Flows

One can also study the eversion through the lens of geometric flows on the space of immersions. Consider a functional like the Willmore energy:

\[W[f] = \int_{S^2} H^2 \, dA\]

where $H$ is the mean curvature.

What is mean curvature? At each point of a surface in $\mathbb{R}^3$, the surface curves by different amounts in different directions. The two extremes — the maximum and minimum curvatures — are called the principal curvatures $\kappa_1$ and $\kappa_2$. The mean curvature $H = \frac{1}{2}(\kappa_1 + \kappa_2)$ is their average. For a sphere of radius $R$, both principal curvatures equal $1/R$, so $H = 1/R$ everywhere. For a saddle point (like the center of a Pringles chip), the curvatures have opposite signs and can cancel: $H = 0$. Surfaces with $H = 0$ everywhere are called minimal surfaces (like soap films). The Willmore energy $\int H^2 \, dA$ measures the total “bending” of a surface; a round sphere minimizes it among all closed surfaces. The Willmore flow

\[\frac{\partial f}{\partial t} = -\nabla W[f]\]

drives the surface toward a minimum of the Willmore energy (a round sphere). Stationary points of $W$ — surfaces where $\nabla W = 0$ — are called Willmore surfaces.

While the Willmore flow does not directly produce an eversion (it’s a gradient flow that seeks minima, not a path between two configurations), it illustrates the general principle: one can study dynamics on $\operatorname{Imm}(S^2, \mathbb{R}^3)$ by defining energy functionals and studying their critical points and flow lines, exactly as in the calculus of variations.

Part V: Homotopy Groups — A Primer

For readers who want a deeper understanding of the algebraic machinery, here is a self-contained introduction to the homotopy groups that power Smale’s proof.

5.1 What are Homotopy Groups?

Homotopy groups $\pi_n(X)$ are algebraic invariants that measure the “holes” in a topological space $X$ at different dimensions.

$\pi_0(X)$ counts the path-connected components of $X$. If $\pi_0(X)$ has one element, the space is connected.

$\pi_1(X)$ is the fundamental group: it classifies loops in $X$ up to continuous deformation. For $S^2$, $\pi_1(S^2) = 0$ — every loop on a sphere can be shrunk to a point (there are no holes for a loop to “catch” on).

$\pi_2(X)$ classifies maps from $S^2$ into $X$ up to homotopy. It measures “2-dimensional holes.” For example, $\pi_2(S^2) = \mathbb{Z}$ — the integer records how many times the map wraps $S^2$ around the target $S^2$ (the degree).

The pattern: $\pi_n(X)$ classifies maps $S^n \to X$ up to homotopy, detecting $n$-dimensional holes.

5.2 Isomorphism and Homeomorphism

When we write $\pi_2(SO(3)) \cong \pi_2(\mathbb{RP}^3)$, the $\cong$ denotes a group isomorphism — a structure-preserving bijection. Two groups are isomorphic when they have identical algebraic structure.

This isomorphism holds because $SO(3)$ and $\mathbb{RP}^3$ are homeomorphic — there exists a continuous bijection between them with continuous inverse. A homeomorphism preserves all topological properties, including all homotopy groups.

5.3 The Chain of Isomorphisms

Let’s trace through the computation $\pi_2(V_{3,2}) = 0$ one more time, emphasizing why each step works:

\[\pi_2(V_{3,2}) \cong \pi_2(SO(3))\]

because $V_{3,2} \cong SO(3)$ (a 2-frame determines a rotation; conversely, a rotation determines a 2-frame).

\[\pi_2(SO(3)) \cong \pi_2(\mathbb{RP}^3)\]

because $SO(3) \cong S^3/{\pm 1} = \mathbb{RP}^3$ via the quaternion covering.

\[\pi_2(\mathbb{RP}^3) \cong \pi_2(S^3)\]

from the long exact sequence of the double cover $S^3 \to \mathbb{RP}^3$.

\[\pi_2(S^3) = 0\]

because $\pi_k(S^n) = 0$ for $k < n$: there is “not enough room” for a $k$-sphere to wrap nontrivially around an $n$-sphere when $k < n$.

Therefore: $\pi_2(V_{3,2}) = 0$, and the space of immersions $\operatorname{Imm}(S^2, \mathbb{R}^3)$ has only one path component. Every pair of immersions is regularly homotopic. The eversion exists.

Epilogue

Let’s step back and appreciate the logical arc:

Calculus gave us the language: smooth maps, Jacobians, the rank condition for immersions.
Function spaces gave us the framework: the space $\operatorname{Imm}(S^2, \mathbb{R}^3)$, paths in infinite-dimensional spaces.
Algebraic topology gave us the tool: homotopy groups, fiber bundles, covering spaces.
A single computation — $\pi_2(V_{3,2}) = 0$ — gave us the answer.

The eversion of the sphere is not a trick. It is a theorem that reveals deep structure in the topology of function spaces. The fact that this structure is encoded in a single homotopy group, computable through a chain of standard isomorphisms, is one of the most beautiful confluences in modern mathematics.

Smale’s proof tells us the path exists. Morin showed us what it looks like. Thurston showed us how to construct it. And now, perhaps, you can see why it must be so.

For further exploration of tangent bundles, differential geometry, and related topics, see my companion post: Differential Topology Q&A with Phi3.

Julian Henry's Blog

✨ The Eversion of the Sphere