03.06.16 · modern-geometry / characteristic-classes

Whitney duality and immersion obstructions

shipped3 tiersLean: partial

Anchor (Master): Whitney 1941 *Lectures in Topology* (Whitney duality); Whitney 1944 *Ann. of Math.* 45 (immersion conjecture); Massey 1960 *J. Math. Soc. Japan* 14 (RP^n immersion lower bound); Milnor-Stasheff 1974 §4 + §11; Cohen 1985 *Ann. of Math.* 122 (proof of the immersion conjecture); Brown-Peterson 1964 *Topology* 5 (BP spectrum)

Intuition [Beginner]

Picture a closed surface — a sphere, a torus, a Klein bottle — sitting inside ordinary three-dimensional space. At every point of the surface, the surface has two tangent directions, and one direction perpendicular to it. The perpendicular direction is the normal. The whole story of how the surface bends inside space is split between the tangent data and the normal data, and the two have to add up to the plain three-space directions around the point.

That accounting is the seed of Whitney duality. The Stiefel-Whitney classes 03.06.03 of the tangent bundle and of the normal bundle are not independent. They are inverse to each other in the mod-two cohomology ring of the manifold. If you know the tangent classes, the normal classes are forced; they are computed by formal division in a polynomial in the cohomology generators.

The payoff is immersion obstruction. An immersion of a manifold into Euclidean space is essentially a smooth map whose differential is everywhere injective — locally, an inclusion of one piece of Euclidean space into another. If the tangent Stiefel-Whitney classes of the manifold force the normal classes to be non-zero in some high degree, then the normal bundle cannot have lower rank than that degree, and the ambient Euclidean space cannot have lower dimension than the manifold plus that codimension. The classes provide a lower bound on how small the ambient space can be.

Visual [Beginner]

The picture is a closed manifold sitting inside Euclidean space, with the tangent and normal bundles at each point colour-coded. Above the picture is a formal-algebra panel: the polynomial $w (T M) = 1 + w_{1} + w_{2} + \dots$ and its formal inverse $w (ν M) = 1 - w_{1} + (w_{1}^{2} - w_{2}) - \dots$ in the cohomology ring, written explicitly for low degrees.

The picture also shows the Massey lower bound for real projective space: the formal inverse of $(1 + a)^{n + 1}$ in the polynomial ring $F_{2} [a] / (a^{n + 1})$ has its highest non-zero degree at a specific value depending on the binary expansion of $n + 1$ , and that degree controls the minimum codimension of any immersion of real projective space into Euclidean space.

Worked example [Beginner]

Take $M = RP^{2}$ , the real projective plane.

Step 1. The total Stiefel-Whitney class of the tangent bundle is $w (T RP^{2}) = (1 + a)^{3} = 1 + a + a^{2}$ in $F_{2} [a] / (a^{3})$ where $a$ is the degree-one generator of the mod-two cohomology of $RP^{2}$ (using the binomial coefficients modulo two via Lucas: $(0 3) = 1$ , $(1 3) = 3 = 1$ , $(2 3) = 3 = 1$ ).

Step 2. Whitney duality says $w (ν RP^{2}) = w (T RP^{2})^{- 1}$ in the cohomology ring. Compute the inverse by formal division: we need a polynomial $p$ with $(1 + a + a^{2}) (1 + p) = 1$ modulo $a^{3}$ . Try $p = a + 0 \cdot a^{2}$ : then $(1 + a + a^{2}) (1 + a) = 1 + a + a + a^{2} + a^{2} + a^{3} = 1 (mod a^{3}, 2)$ . So $w (ν RP^{2}) = 1 + a$ .

Step 3. The normal class $w (ν RP^{2}) = 1 + a$ has top non-zero degree equal to $1$ . So the normal bundle of any immersion of $RP^{2}$ must have rank at least $1$ , and $RP^{2}$ cannot immerse in $R^{2}$ . It can and does immerse in $R^{3}$ : Boy's surface is an explicit example.

What this tells us: Stiefel-Whitney classes of the tangent bundle plus formal inversion give a computable lower bound on the codimension of any immersion of a manifold into Euclidean space.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $M$ be a closed smooth $n$ -manifold and $f : M \to R^{N}$ a smooth immersion with $N \geq n$ . Recall 03.05.08 that the tangent bundle $T M \to M$ is the rank- $n$ real vector bundle of tangent vectors and that $f^{*} T R^{N} \to M$ is the rank- $N$ pullback of the tangent bundle of $R^{N}$ , isomorphic to the product bundle $ε^{N} := M \times R^{N}$ . The differential $df$ embeds $T M$ as a sub-bundle of $f^{*} T R^{N}$ .

Definition (normal bundle of an immersion). The normal bundle $ν_{f} M \to M$ of the immersion $f : M \to R^{N}$ is the quotient real vector bundle

ν_{f} M := f^{*} T R^{N} / df (T M)

of rank $N - n$ . Choosing a Euclidean metric on $R^{N}$ identifies $ν_{f} M$ with the orthogonal complement of $df (T M)$ inside $f^{*} T R^{N}$ , and the Whitney sum decomposition

T M \oplus ν_{f} M ≅ ε^{N}

holds as an isomorphism of real vector bundles over $M$ . The isomorphism class of $ν_{f} M$ depends on the immersion $f$ a priori, but its mod-2 characteristic classes do not (see Theorem below).

Definition (total Stiefel-Whitney class). For a real vector bundle $E \to M$ of rank $r$ , the total Stiefel-Whitney class is

w (E) := 1 + w_{1} (E) + w_{2} (E) + \dots + w_{r} (E) \in H^{*} (M; F_{2}),

viewed as a unit in the graded-commutative total cohomology ring (a unit because the degree- $0$ component is $1$ and the higher components are nilpotent of bounded degree on a finite-dimensional manifold).

Counterexamples to common slips [Intermediate+]

Immersion vs. embedding. Whitney duality requires only that $f$ be an immersion (everywhere injective differential), not an embedding (globally injective). Boy's surface is an immersion $RP^{2} ↪ R^{3}$ with triple points — not an embedding — but the Whitney duality identity holds and the codimension-1 normal bundle is well-defined.
Normal bundle depends on the immersion? The bundle isomorphism class of $ν_{f} M$ may vary with $f$ , but its Stiefel-Whitney classes are determined by $w (T M)$ via formal inversion and depend only on $M$ and $N$ , not on the specific immersion $f$ . So the obstruction calculation is intrinsic to $M$ .
Mod-2 hypothesis. Whitney duality holds in $H^{*} (M; F_{2})$ , not in $H^{*} (M; Z)$ . The integer-coefficient analogue (Pontryagin classes 03.06.04) is weaker because $p (T M) \cdot p (ν M)$ is not $1$ in general — the Pontryagin classes satisfy a duality identity only up to two-torsion via the Wu formulas.

Key theorem with proof [Intermediate+]

Theorem (Whitney duality, Whitney 1941). Let $M$ be a closed smooth $n$ -manifold and $f : M ↪ R^{N}$ a smooth immersion with normal bundle $ν_{f} M$ of rank $N - n$ . Then

w (T M) \cup w (ν_{f} M) = 1 \in H^{*} (M; F_{2}) .

Equivalently, $w (ν_{f} M) = w (T M)^{- 1}$ in the total mod-2 cohomology ring of $M$ .

Proof. The Whitney sum isomorphism $T M \oplus ν_{f} M ≅ ε^{N}$ between vector bundles over $M$ has two consequences when paired with the Stiefel-Whitney axioms 03.06.03.

First, by the Whitney product formula (axiom 2 of the Stiefel-Whitney classes),

w (T M \oplus ν_{f} M) = w (T M) \cup w (ν_{f} M)

in $H^{*} (M; F_{2})$ .

Second, by naturality (axiom 1) applied to the constant map $c : M \to pt$ that classifies the product bundle $ε^{N} = c^{*} (R^{N} \to pt)$ , we have

w (ε^{N}) = c^{*} w (R^{N} \to pt) = c^{*} (1) = 1

in $H^{*} (M; F_{2})$ , because the cohomology of a point is concentrated in degree zero and $w_{i}$ vanishes there for $i \geq 1$ .

Combining, $w (T M) \cup w (ν_{f} M) = w (T M \oplus ν_{f} M) = w (ε^{N}) = 1$ .

For the equivalence statement, observe that $w (T M) = 1 + w_{1} (T M) + \dots + w_{n} (T M)$ has degree- $0$ component $1$ and higher components nilpotent of bounded degree on a finite-dimensional $M$ . So $w (T M)$ is a unit in the total cohomology ring with formal inverse

w (T M)^{- 1} = k \geq 0 \sum (- 1)^{k} (w_{1} (T M) + w_{2} (T M) + \dots)^{k} = k \geq 0 \sum (w_{1} (T M) + w_{2} (T M) + \dots)^{k}

(signs disappear modulo $2$ ), the sum terminating at degree $dim M$ . By the identity, $w (ν_{f} M) = w (T M)^{- 1}$ in $H^{*} (M; F_{2})$ . $□$

Bridge. The duality identity builds toward 03.06.04 in the Pontryagin direction (where an integer-coefficient duality holds only after a Wu-formula correction), and the foundational reason it works mod $2$ is the multiplicativity axiom of Stiefel-Whitney classes 03.06.03 combined with the vanishing of all positive-degree classes on the product bundle. This is exactly the structural fact that turns Stiefel-Whitney computation on the normal bundle (geometrically obscure: depends on an immersion) into a formal-algebra computation on the tangent bundle (geometrically intrinsic: known from $w (T M)$ alone). The bridge is between immersion obstruction theory in differential topology and pure characteristic-class algebra in $H^{*} (M; F_{2})$ ; this pattern appears again in 03.06.18 where the Chern character ring homomorphism translates the analogous K-theoretic duality into rational cohomology, and putting these together the unifying insight is that characteristic classes of complements are computable from characteristic classes of the thing itself by formal inversion in the cohomology ring.

Exercises [Intermediate+]

Exercise 7 (hard, proof).

Prove the contrapositive obstruction principle: if there exists $i > k$ with $w_{i} (ν M) := [w (T M)^{- 1}]_{i} \neq = 0$ in $H^{i} (M; F_{2})$ , then $M^{n}$ does not immerse in $R^{n + k}$ .

Hint

Combine Exercise 6 with the well-definedness of $w (ν M)$ in terms of $w (T M)$ alone.

Answer

Suppose for contradiction that $M^{n}$ immerses in $R^{n + k}$ via $f$ . By Exercise 6, $w_{i} (ν_{f} M) = 0$ for all $i > k$ . But $w_{i} (ν_{f} M)$ depends only on $w (T M)$ (it is the degree- $i$ component of $w (T M)^{- 1}$ , independent of $f$ ). The hypothesis says this component is non-zero in degree $i > k$ . Contradiction. Hence no immersion into $R^{n + k}$ exists. The argument is finite: $w (T M)$ has finite degree $\leq n$ , so $w (T M)^{- 1}$ also has bounded support, and checking the top non-zero degree of $w (T M)^{- 1}$ gives the basic lower bound on the immersion codimension.

Lean formalization [Intermediate+]

lean_status: partial — the companion module Codex.Modern.CharClasses.WhitneyDualityImmersion records the Whitney duality formula and the immersion codimension lower bound as formalisation targets with sorry proof bodies. The module declares:

a placeholder Cohom X n for mod-2 cohomology in each degree with the cup product;
placeholder RealVectorBundle M n together with the whitneySum operation and totalSW total Stiefel-Whitney class;
the Whitney product formula whitney_product : totalSW (E ⊕b F) = totalSW E * totalSW F (sorry);
placeholder SmoothManifold n, tangentBundle, Immersion M N, normalBundle;
the Whitney sum decomposition tangent_normal_decomp for an immersion (sorry);
the Whitney duality theorem whitney_duality : totalSW (tangentBundle M) * totalSW (normalBundle f h) = 1 (sorry), proved schematically by composing whitney_product with tangent_normal_decomp;
the immersion codimension lower bound immersion_codim_lower_bound : i > c → normalSW M i = 0 (sorry), the bridge between Whitney duality and immersion obstructions;
the Massey lower bound statements for $RP^{n}$ as placeholder identities, and a placeholder of the Cohen 1985 immersion conjecture theorem stating that every closed $n$ -manifold immerses in $R^{2 n - α (n)}$ .

The substantive structural content — that $w (ν M)$ equals the formal inverse of $w (T M)$ in the cohomology ring, and that an immersion forces the inverse to vanish past degree $k$ — is captured at the level of theorem statements; the proof bodies depend on Mathlib infrastructure for bundled cohomology rings, real vector bundles, and the Whitney product formula, all of which are listed in lean_mathlib_gap.

Advanced results [Master]

The Whitney duality formula and its proof via the stable normal bundle

Whitney's 1941 Lectures in Topology contribution introduced the duality identity in the form

w (T M) \cup w (ν M) = 1 \in H^{*} (M; F_{2}),

valid for any closed smooth $n$ -manifold $M$ admitting an immersion $f : M ↪ R^{N}$ with normal bundle $ν M = f^{*} T R^{N} / T M$ . The proof, presented at Intermediate tier above, has a structural enrichment at Master tier: the duality is really a statement about the stable normal bundle.

Definition (stable normal bundle). The stable normal bundle of a closed smooth $n$ -manifold $M$ is the stable equivalence class of $ν M$ for any immersion $M ↪ R^{N}$ with $N$ sufficiently large. Two such normal bundles are stably equivalent: if $f_{1}, f_{2} : M ↪ R^{N_{1}}, R^{N_{2}}$ are immersions, then $ν_{f_{1}} M \oplus ε^{N_{2}} ≅ ν_{f_{2}} M \oplus ε^{N_{1}}$ in the K-theory $K O (M)$ .

The stable equivalence class is well-defined and the Stiefel-Whitney classes of the stable normal bundle agree with those of any specific $ν_{f} M$ . So the cohomology identity

w (ν M) = w (T M)^{- 1}

makes sense as a statement about the intrinsic $F_{2}$ -cohomology of $M$ , independent of any immersion. The Whitney duality formula is then both an identity (visible from the immersion) and a definition (of $w (ν M)$ in the absence of an immersion).

Theorem (Whitney 1941, stable form). Let $M$ be a closed smooth $n$ -manifold. Define $w (ν M) := w (T M)^{- 1}$ in the total mod-2 cohomology ring of $M$ . Then for any immersion $f : M ↪ R^{N}$ , the Stiefel-Whitney classes of $ν_{f} M$ agree with the components of $w (ν M)$ in each degree.

The proof is the calculation of Theorem (Whitney duality) above, reinterpreted as a consistency check: the right-hand side $w (T M)^{- 1}$ is defined whether or not an immersion exists, and the theorem says that if an immersion exists, then the Stiefel-Whitney classes of its normal bundle match.

Immersion obstructions from non-vanishing Stiefel-Whitney classes

The contrapositive turns Whitney duality into a necessary condition for immersibility.

Theorem (Whitney 1941, immersion obstruction lower bound). Let $M^{n}$ be a closed smooth $n$ -manifold. If $w_{i} (ν M) := [w (T M)^{- 1}]_{i} \neq = 0 \in H^{i} (M; F_{2})$ for some $i > k$ , then $M$ does not immerse in $R^{n + k}$ .

Proof. By the dimension/vanishing axiom of Stiefel-Whitney classes, $w_{i} (E) = 0$ whenever $i > rank (E)$ . If $M$ immerses in $R^{n + k}$ , the normal bundle has rank $k$ , so $w_{i} (ν M) = 0$ for $i > k$ . The hypothesis violates this conclusion. $□$

The same argument iterated gives sharper obstructions: the top non-vanishing degree of $w (ν M)$ is a lower bound on the codimension of any immersion. The immersion dimension of $M$ — the smallest $N$ such that $M$ immerses in $R^{N}$ — is therefore bounded below by

imm (M) \geq n + de g_{top} w (ν M),

where $de g_{top}$ denotes the largest degree in which the total class is non-zero.

Corollary (failure to detect parallelisability). If $M$ is stably parallelisable (i.e. $T M \oplus ε^{k}$ is the product bundle for some $k$ ), then $w (T M) = 1$ and $w (ν M) = 1$ , so Whitney duality gives no obstruction. In particular, all spheres $S^{n}$ and odd-dimensional projective spaces $RP^{2^{r} - 1}$ are undetected by Whitney duality.

This is why Whitney duality is necessary but not sufficient for the optimal immersion dimension. Stably parallelisable manifolds have $w (T M) = 1$ so the duality identity is vacuous, and the actual immersion dimension is determined by finer invariants (Hirsch-Smale immersion theory, Atiyah-Hirzebruch obstructions in $K O$ -theory, mod-2 cohomology operations beyond Stiefel-Whitney).

Worked example: the $RP^{n}$ immersion problem (Massey 1960)

The most-studied non-vacuous family for Whitney duality is the real projective spaces $RP^{n}$ . The mod-2 cohomology is $H^{*} (RP^{n}; F_{2}) = F_{2} [a] / (a^{n + 1})$ where $a \in H^{1}$ generates. The Euler sequence $T RP^{n} \oplus ε^{1} ≅ (n + 1) γ$ (with $γ$ the tautological line bundle, $w (γ) = 1 + a$ ) gives

w (T RP^{n}) = (1 + a)^{n + 1} \in F_{2} [a] / (a^{n + 1}) .

By Whitney duality,

w (ν RP^{n}) = (1 + a)^{- (n + 1)} \in F_{2} [a] / (a^{n + 1}) .

To compute the inverse, let $2^{s}$ be the smallest power of two with $2^{s} \geq n + 1$ . Then $(1 + a)^{2^{s}} = 1 + a^{2^{s}} \equiv 1 (mod a^{n + 1})$ , so

(1 + a)^{- (n + 1)} \equiv (1 + a)^{2^{s} - (n + 1)} (mod a^{n + 1}) .

This is a polynomial of degree exactly $2^{s} - n - 1$ when read as a power of $(1 + a)$ , with top coefficient $(2 ^{s} - n - 1 2 ^{s} - n - 1) = 1 \neq = 0$ provided $2^{s} - n - 1 \leq n$ , i.e. provided $2^{s} \leq 2 n + 1$ .

Theorem (Massey 1960). The immersion dimension of $RP^{n}$ satisfies

imm (RP^{n}) \geq n + (2^{s} - n - 1) = 2^{s} - 1

where $2^{s}$ is the smallest power of two with $2^{s} \geq n + 1$ . In particular, $RP^{n}$ does not immerse in $R^{2^{s} - 2}$ when $n + 1 \leq 2^{s}$ and the inverse polynomial has support up to degree $2^{s} - n - 1$ .

The Massey bound is sharp in many cases. For $n = 2^{r}$ , the bound says $RP^{2^{r}}$ requires ambient dimension at least $2^{r + 1} - 1$ , which matches the immersion-conjecture value $2 n - α (n) = 2 \cdot 2^{r} - 1$ exactly. For $n = 2^{r} - 1$ (when $α (n) = r$ ), the bound is vacuous because $(1 + a)^{n + 1} = 1$ , and finer methods are needed.

Boy's surface is the classical realisation of an immersion $RP^{2} ↪ R^{3}$ , discovered by Werner Boy in 1903 ^{[Boy 1903]} as part of Hilbert's problem on closed surfaces in three-space without boundary. The surface has three triple points and a single self-intersection curve; it is smooth, immersed (not embedded), and realises the Whitney-duality lower bound $imm (RP^{2}) = 3$ exactly.

Smale's eversion of the sphere (1958) is the closely-related extreme example for $S^{2}$ : although Whitney duality gives no obstruction (the sphere is stably parallelisable), Smale's theorem says one can immerse-homotope $S^{2} ↪ R^{3}$ to its mirror image via a continuous family of immersions, a phenomenon possible because the Stiefel-Whitney obstruction is vacuous.

Cohen's 1985 resolution of the immersion conjecture

Whitney 1944 ^{[Whitney 1944]} conjectured that every closed smooth $n$ -manifold immerses in $R^{2 n - α (n)}$ . The conjecture is sharp: $RP^{n}$ realises the bound when $n$ is a power of two, by the Massey computation just performed. The conjecture remained open for forty-one years.

R. L. Cohen 1985 ^{[Cohen 1985]} proved the immersion conjecture using the Brown-Peterson splitting of unoriented bordism. The strategy is as follows.

Step 1: Immersion as bordism class. By work of Wells, Brown, Peterson, and others through the 1960s and 1970s, the existence of an immersion $M^{n} ↪ R^{n + k}$ is governed by the vanishing of a single obstruction class

[M]_{imm} \in π_{n} (Imm_{k}) = π_{n} (Ω^{\infty} Σ^{\infty} M O ⟨ n - k ⟩)

in the homotopy of a suitable Thom spectrum $M O ⟨ n - k ⟩$ , where $M O ⟨ j ⟩$ is the Thom spectrum of the universal $j$ -connected bordism theory.

Step 2: Brown-Peterson splitting. Brown-Peterson 1964 ^{[Brown-Peterson 1966]} proved that at every prime $p$ , the unoriented bordism spectrum $M O$ splits as a wedge of suspensions of a single elementary spectrum $B P$ (the Brown-Peterson spectrum). The mod-2 splitting of $M O ⟨ n - k ⟩$ is the crucial input: it expresses the bordism-theoretic obstruction as a sum of obstructions on the much simpler $B P$ -summands.

Step 3: Vanishing of $B P$ -obstructions. Cohen's central technical work is a careful Adams-spectral-sequence argument on each $B P$ -summand of $M O ⟨ n - k ⟩$ for $k = n - α (n)$ , showing that the obstruction class vanishes for every closed smooth $n$ -manifold. The argument uses the structure of the Steenrod algebra $A_{2}$ acting on $H^{*} (B P; F_{2})$ and the existence of suitable null-homotopies on the surviving generators.

Theorem (Cohen 1985). Every closed smooth $n$ -manifold immerses in $R^{2 n - α (n)}$ , where $α (n)$ is the number of ones in the binary expansion of $n$ .

The theorem is sharp: $RP^{n}$ realises the bound (by Massey 1960 plus subsequent positive constructions), and so does every closed $n$ -manifold whose total Stiefel-Whitney class equals $(1 + a)^{n + 1}$ on a generating mod-2 cohomology class $a$ . Cohen's proof is the foundational example of a Brown-Peterson resolution of a bordism-theoretic obstruction problem; it is part of the broader programme of chromatic-stable-homotopy applications to differential topology.

Synthesis. Whitney duality is the foundational reason that immersion obstructions for closed manifolds reduce to a purely cohomological calculation: the multiplicativity of Stiefel-Whitney classes on Whitney sums, applied to the immersion-induced isomorphism $T M \oplus ν M ≅ ε^{N}$ , identifies the normal class as the formal inverse of the tangent class in the mod-2 cohomology ring. This is exactly the structural fact that turns differential topology into characteristic-class algebra: an immersion's existence is constrained by the bounded degree of the inverse polynomial $w (T M)^{- 1}$ , and the top non-vanishing degree of that inverse is a lower bound on the codimension of any immersion.

Putting these together with the Massey 1960 computation, the central insight is that real projective spaces $RP^{n}$ require ambient dimension at least $2^{s} - 1$ where $2^{s} \geq n + 1$ is the smallest such power of two — a value matching the Whitney conjecture bound $2 n - α (n)$ when $n$ is a power of two. The pattern generalises in two directions simultaneously: Cohen 1985 generalises the conjecture from $RP^{n}$ to every closed $n$ -manifold via a Brown-Peterson stable-homotopy resolution, and the duality identity generalises to higher characteristic classes via the Pontryagin-Wu correction in 03.06.04. The bridge is between immersion obstructions in differential topology (Whitney 1944, Massey 1960, Hirsch-Smale 1959), cohomological algebra in $H^{*} (M; F_{2})$ (Milnor-Stasheff 1974 §4 + §11), and chromatic-stable-homotopy theory (Brown-Peterson 1964, Cohen 1985) — three layers of the same structural phenomenon.

Full proof set [Master]

Proposition 1 (Whitney duality). Let $M$ be a closed smooth $n$ -manifold and $f : M ↪ R^{N}$ an immersion with normal bundle $ν_{f} M$ . Then $w (T M) \cup w (ν_{f} M) = 1 \in H^{*} (M; F_{2})$ .

Proof. The differential $df : T M \to f^{*} T R^{N}$ is a fibrewise injective bundle map. Choose a Euclidean metric on $R^{N}$ (or equivalently, equip $f^{*} T R^{N} ≅ ε^{N}$ with its standard Euclidean metric) and define $ν_{f} M$ as the orthogonal complement of $df (T M)$ in $f^{*} T R^{N}$ . Then $T M \oplus ν_{f} M ≅ f^{*} T R^{N} ≅ ε^{N}$ as real vector bundles over $M$ .

By the Whitney product formula [03.06.03 axiom 2],

w (T M \oplus ν_{f} M) = w (T M) \cup w (ν_{f} M) .

By naturality [03.06.03 axiom 1] applied to the classifying map of $ε^{N}$ , which factors through a point (since $ε^{N}$ is the constant rank- $N$ product bundle),

w (ε^{N}) = w_{0} = 1 \in H^{*} (M; F_{2})

because all positive-degree Stiefel-Whitney classes of a product bundle vanish (the dimension axiom plus naturality, or equivalently the splitting principle on the constant classifying map).

Combining gives $w (T M) \cup w (ν_{f} M) = w (ε^{N}) = 1$ . $□$

Proposition 2 (Immersion codimension lower bound). Let $M^{n}$ be a closed smooth $n$ -manifold with formal-inverse class $w (ν M) := w (T M)^{- 1} \in H^{*} (M; F_{2})$ . If $w_{i} (ν M) \neq = 0$ in $H^{i} (M; F_{2})$ for some $i$ , then $M$ does not immerse in $R^{n + j}$ for any $j < i$ .

Proof. Suppose for contradiction that $f : M ↪ R^{n + j}$ is an immersion with $j < i$ . By Proposition 1, the Stiefel-Whitney classes of the normal bundle $ν_{f} M$ (which has rank $j$ ) satisfy

w (T M) \cup w (ν_{f} M) = 1.

Solving for $w (ν_{f} M)$ in the unital ring $H^{*} (M; F_{2})$ — where $w (T M) = 1 + \dots$ is a unit because the higher terms are nilpotent of bounded degree — gives $w (ν_{f} M) = w (T M)^{- 1} = w (ν M)$ .

But the dimension/vanishing axiom forces $w_{k} (ν_{f} M) = 0$ for $k > rank (ν_{f} M) = j$ . Hence $w_{i} (ν M) = w_{i} (ν_{f} M) = 0$ since $i > j$ .

This contradicts the hypothesis $w_{i} (ν M) \neq = 0$ . Therefore no immersion $M ↪ R^{n + j}$ exists for $j < i$ . $□$

Proposition 3 (Massey computation on $RP^{n}$ ). In $H^{*} (RP^{n}; F_{2}) = F_{2} [a] / (a^{n + 1})$ , $w (T RP^{n}) = (1 + a)^{n + 1}$ and $w (ν RP^{n}) = (1 + a)^{2^{s} - n - 1}$ where $2^{s}$ is the smallest power of two with $2^{s} \geq n + 1$ .

Proof. The tangent-bundle formula $w (T RP^{n}) = (1 + a)^{n + 1}$ is the result of Exercise 4 of 03.06.03, derived from the Euler sequence $T RP^{n} \oplus ε^{1} ≅ (n + 1) γ$ with $γ$ the tautological line bundle and the Whitney product formula.

For the inverse, work in $R := F_{2} [a] / (a^{n + 1})$ . Let $2^{s}$ be the smallest power of two with $2^{s} \geq n + 1$ . Over $F_{2}$ , the Freshman's Dream gives $(1 + a)^{2^{s}} = 1 + a^{2^{s}}$ . Modulo $a^{n + 1}$ , since $2^{s} \geq n + 1$ , we have $a^{2^{s}} = 0$ in $R$ . Hence $(1 + a)^{2^{s}} = 1$ in $R$ , so $(1 + a)$ has multiplicative order dividing $2^{s}$ in the unit group $R^{\times}$ .

Multiplying both sides of $(1 + a)^{2^{s}} = 1$ by $(1 + a)^{- (n + 1)}$ ,

(1 + a)^{- (n + 1)} = (1 + a)^{2^{s} - (n + 1)} .

Since $2^{s} \leq 2 (n + 1) - 1$ (the smallest $2^{s} \geq n + 1$ satisfies $2^{s} < 2 (n + 1)$ ), the exponent $2^{s} - n - 1 < n + 1$ , so the right-hand side is a non-vanishing polynomial of degree $2^{s} - n - 1$ in $R$ .

Combining with Whitney duality, $w (ν RP^{n}) = w (T RP^{n})^{- 1} = (1 + a)^{- (n + 1)} = (1 + a)^{2^{s} - n - 1}$ . $□$

Proposition 4 (Massey lower bound). Let $2^{s}$ be the smallest power of two with $2^{s} \geq n + 1$ . Then $RP^{n}$ does not immerse in $R^{2^{s} - 2}$ . Equivalently, $imm (RP^{n}) \geq 2^{s} - 1$ .

Proof. By Proposition 3, $w (ν RP^{n}) = (1 + a)^{2^{s} - n - 1}$ in $F_{2} [a] / (a^{n + 1})$ . The top coefficient (the coefficient of $a^{2^{s} - n - 1}$ ) is $(2 ^{s} - n - 1 2 ^{s} - n - 1) = 1 \neq = 0$ , provided $2^{s} - n - 1 \leq n$ , i.e. $2^{s} \leq 2 n + 1$ — which holds by the choice of $2^{s}$ as the smallest power of two $\geq n + 1$ .

Hence $w_{2^{s} - n - 1} (ν RP^{n}) = a^{2^{s} - n - 1} \neq = 0$ in $H^{2^{s} - n - 1} (RP^{n}; F_{2})$ . By Proposition 2, $RP^{n}$ does not immerse in $R^{n + j}$ for any $j < 2^{s} - n - 1$ . So $RP^{n}$ does not immerse in $R^{n + (2^{s} - n - 2)} = R^{2^{s} - 2}$ , and the immersion dimension satisfies $imm (RP^{n}) \geq 2^{s} - 1$ . $□$

Proposition 5 (Cohen 1985, immersion conjecture). Every closed smooth $n$ -manifold immerses in $R^{2 n - α (n)}$ , where $α (n)$ is the number of ones in the binary expansion of $n$ .

Proof sketch. The result is Theorem 1 of Cohen's 1985 Annals of Math. paper ^{[Cohen 1985]}. The proof has three structural layers, which we outline; the full argument occupies eighty pages of the Annals paper and uses the Adams spectral sequence over the Steenrod algebra throughout.

(i) Obstruction class. Via Wells-Hirsch immersion theory, the existence of an immersion $M^{n} ↪ R^{n + k}$ is equivalent to the vanishing of a single stable-homotopy obstruction class

θ (M) \in π_{n} (Ω^{\infty} M O ⟨ n - k ⟩),

where $M O ⟨ j ⟩$ is the Thom spectrum of the universal $j$ -connected real bordism theory. For $k = n - α (n)$ , the goal is to show $θ (M) = 0$ for every $M$ .

(ii) Brown-Peterson splitting. Brown-Peterson 1964 ^{[Brown-Peterson 1966]} established the splitting

M O ⟨ j ⟩_{(2)} ≃ α ⋁ Σ^{∣ α ∣} B P ⟨ j ⟩

at the prime $2$ , expressing the localised Thom spectrum as a wedge of suspensions of a single Brown-Peterson summand $B P ⟨ j ⟩$ . The wedge indices $α$ range over a polynomial-algebra basis of generators in specific degrees.

(iii) Vanishing on $B P$ -summands. Cohen's central technical theorem is that the projection of $θ (M)$ onto each $B P ⟨ n - α (n)⟩$ -summand vanishes. The argument uses the structure of the Steenrod algebra $A_{2}$ acting on $H^{*} (B P ⟨ n - α (n)⟩; F_{2})$ and Adams-spectral-sequence arguments specific to the dimension- $n$ summand. The combinatorial input is precisely the $α (n)$ -arithmetic: the relevant Steenrod operations annihilate the surviving generators only when the codimension is $n - α (n)$ or larger.

Combining the three layers gives $θ (M) = 0$ for every closed smooth $n$ -manifold $M$ at codimension $n - α (n)$ , hence the immersion exists. $□$

Connections [Master]

Stiefel-Whitney classes 03.06.03. The Whitney product formula and the dimension axiom of Stiefel-Whitney classes are the direct inputs to Whitney duality; the entire chapter rests on the axiomatic characterisation of $w_{i}$ as the canonical mod-2 characteristic classes of real vector bundles. The duality identity $w (T M) \cup w (ν M) = 1$ is the simplest non-vacuous application of the Whitney product formula to a structured pair of bundles on a manifold.
Pontryagin and Chern classes 03.06.04. The integer-coefficient analogue of Whitney duality involves Pontryagin classes via the Wu formulas: $p (T M) \cdot p (ν M)$ is not $1$ in $H^{*} (M; Z)$ in general, but agrees with $1$ modulo two-torsion via the relation $\overline{p_{k} (E)} = w_{2 k} (E)^{2} +$ Wu corrections. This is the integral counterpart of the mod-2 duality presented here; the splitting between mod-2 and integer-coefficient versions is the foundational reason that Stiefel-Whitney classes are the natural invariants for immersion obstructions and Pontryagin classes are not.
Complex vector bundle 03.05.08. Provides the vector-bundle framework used throughout — tangent bundles, normal bundles, Whitney sums, and the rank-vanishing axiom. The complex-bundle theory has its own duality structure via Chern classes that the present unit's mod-2 story refines and specialises.
Stiefel-Whitney and Pontryagin numbers 03.06.10. The numerical-invariant downstream of the present unit: top-degree monomials $w_{I} (T M)$ in the tangent Stiefel-Whitney classes are bordism-invariant integers in $F_{2}$ . Whitney duality applied to the normal-bundle classes $w_{I} (ν M)$ gives the dual family of Stiefel-Whitney numbers, related by the formal-inversion identity $w (ν M) = w (T M)^{- 1}$ . The Massey lower bound is the simplest case where the numerical content of the dual family becomes a sharp immersion obstruction.
Unoriented bordism and Thom's theorem 03.06.12. The Brown-Peterson splitting of $M O$ used by Cohen 1985 to resolve the immersion conjecture is a refinement of Thom's polynomial-ring computation $Ω_{*}^{O} ≅ F_{2} [x_{i} ∣ i \neq = 2^{k} - 1]$ . The bridge from bordism to immersion theory is the obstruction-class identification $θ (M) \in π_{n} (M O ⟨ n - k ⟩)$ , which routes the immersion problem through the same stable-homotopy machinery used to compute unoriented bordism.
Chern character ring homomorphism 03.06.18. The chromatic structural pattern recurs: the Chern character $ch : K^{0} (X) \to H^{even} (X; Q)$ is a ring homomorphism, and ring homomorphisms preserve units and their inverses. The K-theoretic analogue of Whitney duality is the identity $ch (T M) \cdot ch (ν M) = N \cdot 1$ in $H^{even} (M; Q)$ , with $N$ the codimension; the integer offset is the homotopy-theoretic reason that the mod-2 form is multiplicative while the rational form is additive.
Oriented bordism and Pontryagin-Thom 03.06.13. Whitney duality and the immersion-obstruction theory sit inside the oriented-bordism framework when the manifold is oriented: the dual Pontryagin classes $p (ν M) = p (T M)^{- 1}$ mod 2-torsion produce dual Pontryagin numbers that are bordism invariants in $Ω_{*}^{S O}$ , and the structural theorems of 03.06.13 feed directly into the integer-coefficient refinement of the present unit's duality.
Multiplicative sequences and $L$ / $A$ /Todd genera 03.06.15. Applying a multiplicative sequence $K$ to the Whitney-dual classes produces dual genera $φ_{K} [ν M]$ , and the formal-inversion identity $w (ν M) = w (T M)^{- 1}$ of the present unit interacts directly with the symmetric power-series machinery of 03.06.15. Anchor phrase: multiplicative sequences apply to Whitney-dual classes.
Combinatorial Pontryagin classes and exotic 7-spheres 03.06.17. The integral Pontryagin counterpart of Whitney duality enters Milnor's exotic-sphere argument through the divisibility constraints on $p_{1}$ and $p_{2}$ of bounding parallelizable manifolds, where the Wu-formula corrections of the present unit control the residual two-torsion in the integer-coefficient version of the duality identity.
Signature of a $4 k$ -manifold and the intersection form 03.06.19. The signature of a closed oriented $4 k$ -manifold is computed by the $L$ -genus on the tangent bundle; Whitney duality of the present unit then constrains the $L$ -genus of the normal bundle in any immersion of $M$ into Euclidean space. The interaction between intersection-form signature on $M$ and normal-bundle data is exactly the Whitney-dual side of 03.06.19.
Borel-Hirzebruch and the cohomology of $G / T$ 03.06.20. Whitney duality is the cohomological consequence of the Whitney sum identity $T M \oplus ν M ≅ ε^{N}$ in the classifying-space picture: pulling back along the classifying map for the flat $ε^{N}$ bundle annihilates all characteristic classes in positive degree, and the Borel-Hirzebruch presentation of 03.06.20 is the universal home where this identity becomes a polynomial relation in $H^{*} (B O; F_{2})$ or $H^{*} (B S O; Q)$ .

Historical & philosophical context [Master]

Hassler Whitney 1941 ^{[Whitney 1941]} introduced the duality identity in Lectures in Topology, in the same monograph that systematised Stiefel-Whitney classes via the obstruction-theoretic frame-extension construction (the original geometric definition of $w_{i}$ before the cohomological reformulation by Steenrod 1947). Whitney's contribution was twofold: identifying the multiplicative structure $w (E \oplus F) = w (E) \cup w (F)$ — the Whitney product formula — and observing the duality identity that follows from applying it to $T M \oplus ν M ≅ ε^{N}$ . The conjecture that every closed $n$ -manifold immerses in $R^{2 n - α (n)}$ appeared in Whitney's 1944 paper ^{[Whitney 1944]} on self-intersections of smooth $n$ -manifolds in $2 n$ -space, motivated by Whitney's earlier 1936 embedding theorem ( $R^{2 n + 1}$ ) and 1944 strong embedding theorem ( $R^{2 n}$ ).

The codimension-1 case $RP^{2} ↪ R^{3}$ has an independent history: Werner Boy 1903 ^{[Boy 1903]} discovered the immersed projective plane, now called Boy's surface, as part of a study of closed surfaces without boundary in three-space. Boy's surface, with its three triple points and single self-intersection curve, was the first explicit realisation of an immersion of a non-orientable closed surface in $R^{3}$ . The cohomological obstruction theory presented above explains why the dimension is sharp: $w_{1} (ν RP^{2}) = a \neq = 0$ in $H^{1} (RP^{2}; F_{2})$ forces codimension at least one, and Boy's surface realises codimension exactly one.

W. S. Massey 1960 ^{[Massey 1960]} computed the precise lower bound on the immersion dimension of $RP^{n}$ via the inverse-polynomial calculation in $F_{2} [a] / (a^{n + 1})$ , establishing the immersion-lower-bound theorem in its sharpest form for the projective spaces. The Massey bound was the first family of manifolds where the Whitney duality lower bound was provably sharp (within a constant factor of the eventual Cohen-conjecture optimal value). The companion paper of Atiyah-Hirzebruch in 1959 Topology introduced K-theoretic refinements via the Atiyah-Hirzebruch spectral sequence and Adams operations, leading eventually to Adams's 1962 Topology solution of the vector-fields-on-spheres problem and refined immersion obstructions for higher-dimensional projective spaces. The interplay between Stiefel-Whitney lower bounds, K-theoretic refinements, and explicit immersion constructions occupied a generation of differential topologists from 1944 to the early 1980s.

R. L. Cohen 1985 ^{[Cohen 1985]} closed the Whitney conjecture via the Brown-Peterson resolution of obstruction classes in $M O ⟨ j ⟩$ . Cohen's proof is one of the foundational examples of chromatic stable-homotopy theory applied to a classical differential-topology problem: the entire eighty-page argument routes through the Adams spectral sequence over the Steenrod algebra acting on the $B P$ -summand cohomology, with the combinatorial input being precisely the Hamming-weight $α (n)$ . The lineage continues through Hopkins-Mahowald-style elaborations on the EHP sequence and the chromatic filtration, and the modern viewpoint identifies Whitney duality + Massey's projective-space computation + Cohen's resolution as a three-step ladder in the algebraic-topology / chromatic-stable-homotopy hierarchy applied to immersion theory.

Bibliography [Master]

@incollection{Whitney1941,
  author = {Whitney, Hassler},
  title = {On the topology of differentiable manifolds},
  booktitle = {Lectures in Topology},
  publisher = {University of Michigan Press},
  year = {1941},
  pages = {101--141},
}

@article{Whitney1944,
  author = {Whitney, Hassler},
  title = {The self-intersections of a smooth n-manifold in 2n-space},
  journal = {Annals of Mathematics (2)},
  volume = {45},
  year = {1944},
  pages = {220--246},
}

@article{Massey1960,
  author = {Massey, W. S.},
  title = {On the {S}tiefel-{W}hitney classes of a manifold},
  journal = {American Journal of Mathematics},
  volume = {82},
  year = {1960},
  pages = {92--102},
}

@article{Boy1903,
  author = {Boy, Werner},
  title = {Über die Curvatura integra und die Topologie geschlossener Flächen},
  journal = {Mathematische Annalen},
  volume = {57},
  year = {1903},
  pages = {151--184},
}

@article{Cohen1985,
  author = {Cohen, Ralph L.},
  title = {The immersion conjecture for differentiable manifolds},
  journal = {Annals of Mathematics (2)},
  volume = {122},
  year = {1985},
  pages = {237--328},
}

@article{BrownPeterson1966,
  author = {Brown, Edgar H. and Peterson, Franklin P.},
  title = {A spectrum whose {$\mathbb{Z}_p$} cohomology is the algebra of reduced $p$-th powers},
  journal = {Topology},
  volume = {5},
  year = {1966},
  pages = {149--154},
}

@book{MilnorStasheff1974,
  author = {Milnor, John and Stasheff, James},
  title = {Characteristic Classes},
  publisher = {Princeton University Press},
  year = {1974},
  series = {Annals of Mathematics Studies},
  volume = {76},
}

@article{Smale1958,
  author = {Smale, Stephen},
  title = {A classification of immersions of the two-sphere},
  journal = {Transactions of the American Mathematical Society},
  volume = {90},
  year = {1958},
  pages = {281--290},
}

@article{HirschSmale1959,
  author = {Hirsch, Morris W.},
  title = {Immersions of manifolds},
  journal = {Transactions of the American Mathematical Society},
  volume = {93},
  year = {1959},
  pages = {242--276},
}

Cycle 8 Track A, characteristic-classes T1 cluster. Closes the skipped-units entry for 03.06.16 (Whitney duality and immersion obstructions).

Prerequisites

03.06.03
03.06.04
03.05.08

Tier anchors

beginner: Strogatz-style picture of a closed surface inside three-space, with the normal direction at each point compensating the tangent directions; 3Blue1Brown-style cohomology-ring inversion intuition
intermediate: Milnor-Stasheff *Characteristic Classes* §4 (Whitney duality) + §11 (immersion obstruction); Hatcher *Vector Bundles and K-Theory* §3.1
master: Whitney 1941 *Lectures in Topology* (Whitney duality); Whitney 1944 *Ann. of Math.* 45 (immersion conjecture); Massey 1960 *J. Math. Soc. Japan* 14 (RP^n immersion lower bound); Milnor-Stasheff 1974 §4 + §11; Cohen 1985 *Ann. of Math.* 122 (proof of the immersion conjecture); Brown-Peterson 1964 *Topology* 5 (BP spectrum)

References

TODO_REF
Whitney, H. — On the topology of differentiable manifolds · Lectures in Topology (1941), 101-141 — Whitney duality
TODO_REF
Whitney, H. — The self-intersections of a smooth n-manifold in 2n-space · Ann. of Math. (2) 45 (1944), 220-246 — immersion conjecture
TODO_REF
Massey, W. S. — On the Stiefel-Whitney classes of a manifold · Amer. J. Math. 82 (1960), 92-102 — and J. Math. Soc. Japan 14, 1962, 174-192 — RP^n immersion lower bound
TODO_REF
Milnor, J. and Stasheff, J. — Characteristic Classes · Princeton Univ. Press 1974, §4 (Whitney duality) and §11 (immersion obstructions)
TODO_REF
Cohen, R. L. — The immersion conjecture for differentiable manifolds · Ann. of Math. (2) 122 (1985), 237-328 — proof of the Whitney conjecture
TODO_REF
Brown, E. H. and Peterson, F. P. — A spectrum whose Z_p cohomology is the algebra of reduced p-th powers · Topology 5 (1966), 149-154 — BP spectrum, Brown-Peterson splitting
TODO_REF
Boy, W. — Über die Curvatura integra und die Topologie geschlossener Flächen · Math. Ann. 57 (1903), 151-184 — Boy's surface as an immersion RP^2 ↪ R^3

Lean module

Codex.Modern.CharClasses.WhitneyDualityImmersion

Mathlib gap

Mathlib has neither bundled Stiefel-Whitney classes nor real
vector bundles in the bundled-cohomology form required by Whitney
duality. A real formalisation needs: real vector bundles over
smooth manifolds with the Whitney sum operation, the total
Stiefel-Whitney class as a unit in the graded total cohomology
ring `H^*(M; F_2)`, the multiplicativity theorem `w(E ⊕ F)
= w(E) ⌣ w(F)`, the dimension/vanishing axiom `w_i(E) = 0` for
`i > rank(E)`, smooth immersions and their normal bundles with
the Whitney sum isomorphism `TM ⊕ νM ≅ ε^N`, and the formal
inverse of `1 + w_1 + w_2 + …` in the bounded cohomology ring.
The Massey lower bound on `RP^n` would additionally need
`H^*(RP^n; F_2) = F_2[a]/(a^{n+1})` and the binomial-coefficient
Lucas reduction modulo 2; Cohen's theorem would need the full
Brown-Peterson splitting of `MO⟨n⟩`.

Reviewer

TBD — characteristic-classes cluster T1 reviewer

Estimated time

beginner: 18m
intermediate: 55m
master: 105m