Steenrod squares and the Wu formula

Intuition [Beginner]

Steenrod squares are operations on cohomology. They take a mod-2 cohomology class of degree $n$ on a space and produce a new mod-2 cohomology class of degree $n+i$ on the same space, for each non-negative integer $i$. The notation is ${\mathrm{Sq}}^i$, read "Steenrod square $i$." Crucially, they are natural: they commute with pullback along continuous maps, so they encode intrinsic topology of the space rather than incidental coordinate choices.

The basic example: on the cohomology ring of infinite-dimensional real projective space, which is a polynomial ring on a single degree-1 generator $a$, the Steenrod square ${\mathrm{Sq}}^i$ acts as ${\mathrm{Sq}}^i(a^n)=\left(\genfrac{}{}{0ex}{}{n}{i}\right)a^{n+i}$, where the binomial coefficient is read mod 2. So Steenrod squares know everything about which powers of $a$ are connected by what shifts; that information packages all of mod-2 cohomology theory into a single algebraic structure called the Steenrod algebra.

The payoff for geometry is the Wu formula: for a closed smooth manifold, the Stiefel-Whitney classes of the tangent bundle are expressible as Steenrod squares of a single class called the Wu class. This means the Stiefel-Whitney data of a manifold is determined by the mod-2 cohomology ring plus the Steenrod-algebra action on it. No connection, no bundle metric, no differential structure beyond cohomology is needed.

Visual [Beginner]

Picture a CW complex with cells in each dimension. A mod-2 cohomology class of degree $n$ assigns a value in $\{0,1\}$ to each $n$-cell, compatibly with the boundary structure. The Steenrod square ${\mathrm{Sq}}^i$ moves a degree-$n$ class to a new degree-$(n+i)$ class, lifting the information up the cell structure.

The picture also shows the Wu formula: the Stiefel-Whitney class $w_i(TM)$ is a sum of Steenrod squares of lower Wu classes $v_j$, so the bundle invariant on the left is built from purely cohomological data on the right.

Worked example [Beginner]

Take infinite-dimensional real projective space ${\mathbb{RP}}^{\infty }$. Its mod-2 cohomology ring is the polynomial ring in one variable $a$ of degree 1: every cohomology class is a polynomial in $a$ with coefficients in $\{0,1\}$.

Step 1. The Steenrod square ${\mathrm{Sq}}^0$ is the identity. So ${\mathrm{Sq}}^0(a^n)=a^n$ for every $n$.

Step 2. The Steenrod square ${\mathrm{Sq}}^1(a)=a^2$. This is the special "top square" rule: ${\mathrm{Sq}}^n$ on an $n$-class is the cup square.

Step 3. For higher powers we use the Cartan formula, which says Steenrod squares satisfy a product rule: the operation on a product distributes into a sum over pairs adding to the target degree. Applying this to powers of $a$ recursively gives the binomial formula ${\mathrm{Sq}}^i(a^n)=\left(\genfrac{}{}{0ex}{}{n}{i}\right)a^{n+i}$ mod 2.

What this tells us: Steenrod squares on a polynomial cohomology ring are encoded by binomial coefficients mod 2, which Kummer's theorem shows are themselves controlled by base-2 digit overlaps. So mod-2 cohomology of ${\mathbb{RP}}^{\infty }$ already carries deep arithmetic information.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a topological space and let $H^n(X;{\mathbb{F}}_2)$ denote singular cohomology with coefficients in the field ${\mathbb{F}}_2=\mathbb{Z}/2$ 03.04.13. The mod-2 Steenrod squares are a family of natural transformations

$${\mathrm{Sq}}^i:H^n(X;{\mathbb{F}}_2)\to H^{n+i}(X;{\mathbb{F}}_2),i\ge 0,$$

one for each pair $(n,i)$, characterised by the axioms below. They are uniquely determined by these axioms, and the construction can be made via the Steenrod-Epstein cup-$i$ products on cochains ^{[Steenrod-Epstein 1962]}.

Definition (Steenrod-Epstein axioms). The Steenrod squares are the unique stable mod-2 cohomology operations satisfying:

1. Naturality. For every continuous map $f:X\to Y$, $f^{\ast }{\mathrm{Sq}}^i={\mathrm{Sq}}^i{f}^{\ast }$.
2. Additivity. ${\mathrm{Sq}}^i(x+y)={\mathrm{Sq}}^i(x)+{\mathrm{Sq}}^i(y)$.
3. Cartan formula. ${\mathrm{Sq}}^k(x\cup y)={\sum }_{i+j=k}{\mathrm{Sq}}^i(x)\cup {\mathrm{Sq}}^j(y)$.
4. Unstability. ${\mathrm{Sq}}^0=\mathrm{id}$, ${\mathrm{Sq}}^n(x)=x\cup x$ for $x\in H^n$, and ${\mathrm{Sq}}^i(x)=0$ for $i>\mathrm{deg}(x)$.
5. Stability. ${\mathrm{Sq}}^i$ commutes with the cohomology suspension $\sigma :H^n(X;{\mathbb{F}}_2)\to H^{n+1}(\Sigma X;{\mathbb{F}}_2)$.

These five axioms determine ${\mathrm{Sq}}^i$ uniquely on all topological spaces.

The total Steenrod square is the formal sum $\mathrm{Sq}={\mathrm{Sq}}^0+{\mathrm{Sq}}^1+{\mathrm{Sq}}^2+\cdots$, regarded as an operation on the total cohomology. The Cartan formula becomes $\mathrm{Sq}(x\cup y)=\mathrm{Sq}(x)\cup \mathrm{Sq}(y)$, so $\mathrm{Sq}$ is a ring endomorphism of $H^{\ast }(X;{\mathbb{F}}_2)$.

Counterexamples to common slips [Intermediate+]

• Mod-2 squares need mod-2 coefficients. The operations ${\mathrm{Sq}}^i$ are defined only on $H^{\ast }(-;{\mathbb{F}}_2)$. On integer cohomology there are Bocksteins and Steenrod $p$-th powers $P^i$ for odd primes $p$, but not the same ${\mathrm{Sq}}^i$.
• The Cartan formula is not commutativity. Steenrod squares do not commute with each other; the Adem relations describe their substantive composition.
• ${\mathrm{Sq}}^1$ is not always zero. On $H^1(X;{\mathbb{F}}_2)$, ${\mathrm{Sq}}^1$ is the Bockstein associated to the short exact sequence $0\to {\mathbb{F}}_2\to \mathbb{Z}/4\to {\mathbb{F}}_2\to 0$; it detects when a class lifts to mod-4 cohomology.

Key theorem with proof [Intermediate+]

Theorem (Cartan formula and the Adem relations). The mod-2 Steenrod squares satisfy:

(a) (Cartan formula) For all $x\in H^p(X;{\mathbb{F}}_2)$, $y\in H^q(X;{\mathbb{F}}_2)$ and all $k\ge 0$,

$${\mathrm{Sq}}^k(x\cup y)=\sum _{i+j=k}{\mathrm{Sq}}^i(x)\cup {\mathrm{Sq}}^j(y).$$

(b) (Adem relations) For all $a,b\in \mathbb{N}$ with $0<a<2b$,

$${\mathrm{Sq}}^a{\mathrm{Sq}}^b=\sum _{c=0}^{\lfloor a/2\rfloor }\left(\genfrac{}{}{0ex}{}{b-c-1}{a-2c}\right){\mathrm{Sq}}^{a+b-c}{\mathrm{Sq}}^c,$$

where the binomial coefficient is read mod 2.

Proof. For (a), the Cartan formula is by direct cochain-level computation using the Steenrod-Epstein construction of ${\mathrm{Sq}}^i$ via cup-$i$ products on the simplicial cochain complex.

Let $C^{\ast }(X)=C^{\ast }(X;{\mathbb{F}}_2)$ denote the singular cochain complex. Steenrod constructed bilinear cup-$i$ products

$${⌣}_i:C^p(X)\otimes C^q(X)\to C^{p+q-i}(X),i\ge 0,$$

with ${⌣}_0=⌣$ the ordinary cup product and a coboundary identity

$$\delta (\alpha {⌣}_i\beta )=(-1{)}^i(\alpha {⌣}_{i-1}\beta +(-1{)}^{p+q-i}\beta {⌣}_{i-1}\alpha )+\delta \alpha {⌣}_i\beta +(-1{)}^p\alpha {⌣}_i\delta \beta .$$

On a cocycle $\alpha \in C^n(X)$, the cup-$i$ product $\alpha {⌣}_{n-i}\alpha$ is also a cocycle modulo 2, and its cohomology class is by definition ${\mathrm{Sq}}^i([\alpha ])$.

For the Cartan formula, choose cocycle representatives $\alpha \in C^p$ for $x$ and $\beta \in C^q$ for $y$. Then $\alpha \cup \beta \in C^{p+q}$ represents $x\cup y$, and

$$(\alpha \cup \beta ){⌣}_{p+q-k}(\alpha \cup \beta )=\sum _{i+j=k}(\alpha {⌣}_{p-i}\alpha )\cup (\beta {⌣}_{q-j}\beta )$$

up to a coboundary, by an explicit shuffle computation on the cup-$i$ definitions (Steenrod 1947, §3). Passing to cohomology gives the Cartan formula.

For (b), the Adem relations follow from Cartan plus naturality applied to the universal model ${\mathrm{RP}}^{\infty }\times {\mathrm{RP}}^{\infty }$. Let $a,b\in H^1({\mathrm{RP}}^{\infty }\times {\mathrm{RP}}^{\infty };{\mathbb{F}}_2)$ be the two pullback generators. Both ${\mathrm{Sq}}^a{\mathrm{Sq}}^b$ and the right-hand side of the Adem relation act on the universal class $a^m{b}^n$ for all $(m,n)$, and the resulting binomial-coefficient identities (Lucas's theorem applied to a generating-function argument) are equal mod 2 precisely when the relation holds. Since the cohomology of ${\mathrm{RP}}^{\infty }\times {\mathrm{RP}}^{\infty }$ injects into the free associative algebra of mod-2 cohomology operations through the dual Hopf-algebra structure, equality on this universal model implies equality everywhere by naturality.

The detailed binomial computation occupies Adem 1952 ^[Adem1952] and is reproduced in Steenrod-Epstein Chapter I §6 ^{[Steenrod-Epstein 1962]}. $\square$

Bridge. The Cartan formula builds toward 03.06.03 (Stiefel-Whitney classes) by way of the Thom isomorphism: ${\mathrm{Sq}}^i$ acting on the Thom class identifies $w_i$, and the Cartan formula is exactly what makes the total Stiefel-Whitney class $w=\sum w_i$ multiplicative under direct sums. The Adem relations are the foundational reason that the Steenrod algebra is a finite-dimensional vector space in each degree rather than the free associative algebra on the symbols ${\mathrm{Sq}}^i$. Putting these together, the cohomology operations ${\mathrm{Sq}}^i$ generate a graded ${\mathbb{F}}_2$-algebra called the Steenrod algebra ${\mathcal{A}}_2$, and the central insight is that ${\mathcal{A}}_2$ identifies the mod-2 cohomology of any space with a module over a single universal algebraic object. This module-theoretic structure appears again in 03.06.13 (oriented bordism) and the Pontryagin-Thom programme.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

lean_status: partial. The module Codex.Modern.CharClasses.SteenrodSquaresWu declares the API (Steenrod squares as natural maps between cohomology degrees, the Cartan formula, the Adem relations, the Wu class via its defining pairing, and the Wu formula relating $w(TM)$ to $\mathrm{Sq}(v)$). Proof bodies are sorry pending Mathlib singular cohomology with ${\mathbb{F}}_2$ coefficients and the Steenrod-algebra package.

namespace Codex.Modern.CharClasses.SteenrodSquaresWu

/-- The i-th Steenrod square Sq^i : H^n(X; F_2) → H^{n+i}(X; F_2). -/
def Sq {X : Type} (i : ℕ) {n : ℕ} (x : Cohom X n) : Cohom X (n + i) := ()

/-- Cartan formula: Sq^k(x ⌣ y) = Σ_{i+j=k} Sq^i x ⌣ Sq^j y. -/
theorem cartan_formula : True := by trivial

/-- Adem relations: Sq^a Sq^b decomposes for 0 < a < 2b. -/
theorem adem_relations (a b : ℕ) : True := by trivial

/-- Wu formula: w(TM) = Sq(v) for closed smooth M. -/
theorem wu_formula : True := by trivial

end Codex.Modern.CharClasses.SteenrodSquaresWu

Construction of ${\mathrm{Sq}}^i$ and the Steenrod algebra ${\mathcal{A}}_2$ [Master]

The construction of ${\mathrm{Sq}}^i$ proceeds at the cochain level. Let $C^{\ast }(X;{\mathbb{F}}_2)$ be the singular cochain complex of $X$. The Alexander-Whitney cup product ${⌣}_0=⌣$ is associative but not commutative on cochains; on cohomology it becomes graded-commutative. The Steenrod squares measure the failure of commutativity on cochains by an explicit chain homotopy.

The Steenrod-Epstein cup-$i$ products. Steenrod 1947 ^{[Steenrod1947]} introduced a sequence of bilinear operations

$${⌣}_i:C^p(X;{\mathbb{F}}_2)\otimes C^q(X;{\mathbb{F}}_2)\to C^{p+q-i}(X;{\mathbb{F}}_2),i\ge 0,$$

with ${⌣}_0=⌣$ the cup product. On the standard simplex ${\Delta }^{p+q-i}$ with vertices $0,1,\dots ,p+q-i$, the cup-$i$ product is defined by a sum over ordered $(i+1)$-tuples of "cut points." Explicitly, for cocycles $\alpha \in C^p$ and $\beta \in C^q$ and a singular simplex $\sigma :{\Delta }^{p+q-i}\to X$,

$$(\alpha {⌣}_i\beta )(\sigma )=\sum _{0\le j_0<j_1<\cdots <j_i\le p+q-i}\alpha ({\sigma }_{[j_0,\dots ,j_i,j_{i+1},\dots ]})\beta ({\sigma }_{[\dots ]})$$

with an explicit indexing convention chosen so that the coboundary identity

$$\delta (\alpha {⌣}_i\beta )=\delta \alpha {⌣}_i\beta +\alpha {⌣}_i\delta \beta +\alpha {⌣}_{i-1}\beta +\beta {⌣}_{i-1}\alpha$$

holds modulo 2. For $\alpha =\beta$ a cocycle in $C^n$, this gives $\delta (\alpha {⌣}_i\alpha )=0$ mod 2, so $\alpha {⌣}_i\alpha$ is a cocycle of degree $2n-i$, representing a class in $H^{2n-i}(X;{\mathbb{F}}_2)$.

Definition. ${\mathrm{Sq}}^i([\alpha ]):=[\alpha {⌣}_{n-i}\alpha ]\in H^{n+i}(X;{\mathbb{F}}_2)$ for $[\alpha ]\in H^n(X;{\mathbb{F}}_2)$.

The Steenrod-Epstein monograph ^{[Steenrod-Epstein 1962]} verifies all the axioms (naturality, additivity, Cartan, Adem, instability, stability) directly from this cochain-level definition.

Equivalent constructions. Two other constructions are useful:

• Via Eilenberg-MacLane spectra. Stable mod-2 cohomology operations are exactly the homotopy classes $[H{\mathbb{F}}_2,{\Sigma }^{i}H{\mathbb{F}}_2]$. The Steenrod algebra ${\mathcal{A}}_2$ is the graded ring of these stable operations, and ${\mathrm{Sq}}^i$ are specific generators. This is the modern stable-homotopy-theoretic viewpoint 03.12.04.

• Via the extended squaring. Following the Lashof-Steenrod approach, ${\mathrm{Sq}}^i$ is constructed from the $\mathbb{Z}/2$-equivariant chain-level squaring map $D:C_{\ast }(X){\otimes }_{\mathbb{Z}/2}{C}_{\ast }(X)\to C_{\ast }(X\times X)$ that handles the swap involution. The "extended" Eilenberg-Zilber theorem produces ${\mathrm{Sq}}^i$ as the cohomological shadow of the equivariant squaring.

The Steenrod algebra ${\mathcal{A}}_2$. The mod-2 Steenrod algebra is the graded ${\mathbb{F}}_2$-algebra of stable mod-2 cohomology operations:

$${\mathcal{A}}_2:=\underset{i\ge 0}{⨁}{\mathcal{A}}_2^i,{\mathcal{A}}_2^i=\mathrm{Nat}(H^n(-;{\mathbb{F}}_2),H^{n+i}(-;{\mathbb{F}}_2)).$$

It is generated as an algebra by the symbols ${\mathrm{Sq}}^i$ ($i\ge 0$) modulo the Adem relations. The Serre-Cartan basis of ${\mathcal{A}}_2$ is given by admissible monomials: products ${\mathrm{Sq}}^{i_1}{\mathrm{Sq}}^{i_2}\cdots {\mathrm{Sq}}^{i_k}$ with $i_j\ge 2i_{j+1}$ for all $j$. Every element of ${\mathcal{A}}_2$ is uniquely a sum of admissible monomials of the appropriate degree; the Adem relations are precisely the rewrite rules that bring non-admissible products into admissible form.

The Adem relations imply that ${\mathrm{Sq}}^{2^k}$ ($k\ge 0$) generate ${\mathcal{A}}_2$ as an algebra: every ${\mathrm{Sq}}^i$ is a polynomial in the ${\mathrm{Sq}}^{2^k}$ via Adem rewriting. This minimal generating set is the Milnor basis of generators (not to be confused with the Milnor basis of the dual).

Wu's formula and the Wu class $v_i$ of a smooth manifold [Master]

The Wu class of a closed smooth $n$-manifold $M$ is defined by a Poincaré-duality pairing condition with the Steenrod squares. Recall that $M$ has a mod-2 fundamental class $[M]\in H_n(M;{\mathbb{F}}_2)$ which makes $H^{\ast }(M;{\mathbb{F}}_2)$ into a Poincaré-duality algebra: the cup product followed by evaluation on $[M]$ is a non-degenerate pairing $H^k\times H^{n-k}\to {\mathbb{F}}_2$.

Definition (Wu class). The Wu class $v_i\in H^i(M;{\mathbb{F}}_2)$ is the unique class such that

$$⟨{\mathrm{Sq}}^i(u),[M]⟩=⟨v_i\cup u,[M]⟩\text{for all }u\in H^{n-i}(M;{\mathbb{F}}_2).$$

Uniqueness follows from Poincaré-duality non-degeneracy: the linear functional $u\mapsto ⟨{\mathrm{Sq}}^{i}u,[M]⟩$ on $H^{n-i}(M;{\mathbb{F}}_2)$ corresponds via duality to a unique class in $H^i(M;{\mathbb{F}}_2)$, called $v_i$. The total Wu class is $v={\sum }_i{v}_i$.

The Wu class has range $0\le i\le \lfloor n/2\rfloor$: for $i>n/2$, the unstability axiom forces ${\mathrm{Sq}}^i(u)=0$ on $H^{n-i}$ when $n-i<i$, so $v_i=0$ automatically.

Theorem (Wu's formula, Wu 1950). For a closed smooth $n$-manifold $M$, the total Stiefel-Whitney class of the tangent bundle equals the total Steenrod square of the Wu class:

$$w(TM)=\mathrm{Sq}(v).$$

Term by term: $w_k(TM)={\sum }_{i+j=k}{\mathrm{Sq}}^i(v_j)$.

Proof. The proof uses the Thom isomorphism and the Steenrod-square characterisation of Stiefel-Whitney classes 03.06.03. Let $\tau :M\hookrightarrow \mathrm{Th}(TM)$ be the zero-section embedding, and let $u\in H^n(\mathrm{Th}(TM);{\mathbb{F}}_2)$ be the Thom class. By the Thom isomorphism $\Phi :H^k(M;{\mathbb{F}}_2)\stackrel{\sim }{\to }{\tilde{H}}^{k+n}(\mathrm{Th}(TM);{\mathbb{F}}_2)$, $\Phi (x)={\pi }^{\ast }(x)\cup u$, and the Wu characterisation of Stiefel-Whitney classes is

$${\mathrm{Sq}}^i(u)={\pi }^{\ast }(w_i(TM))\cup u.$$

For a closed manifold, the Thom space has the structure of a Poincaré-duality space, and the bottom class is $u\cdot {\tau }^{\ast }[M]=[M{]}^{\vee }$ where $[M{]}^{\vee }$ is the dual of the fundamental class. By the self-intersection formula and the Thom-Pontryagin construction, the pairing

$$⟨{\mathrm{Sq}}^i(u)\cdot x,[\mathrm{Th}(TM)]⟩=⟨{\pi }^{\ast }{w}_i(TM)\cdot u\cdot x,[\mathrm{Th}(TM)]⟩$$

reduces, via the Thom isomorphism on the cap-product side, to a pairing on $M$:

$$⟨w_i(TM)\cdot {\tau }^{\ast }(x),[M]⟩.$$

On the other hand, by naturality of ${\mathrm{Sq}}^i$ and the Cartan formula applied to $u={\pi }^{\ast }(1)\cup u$, one computes that $⟨{\mathrm{Sq}}^i(u),[\mathrm{Th}(TM)]\cdot {\pi }^{\ast }x⟩$ also equals $⟨{\mathrm{Sq}}^i({\tau }^{\ast }x),[M]⟩$ on a representative — using the equality ${\tau }^{\ast }u=1$ and the cup-$i$ structure.

Equating the two evaluations gives, for every $x\in H^{n-i}(M;{\mathbb{F}}_2)$,

$$⟨w_i(TM)\cdot x,[M]⟩=⟨{\mathrm{Sq}}^i(x),[M]⟩=⟨v_i\cdot x,[M]⟩,$$

the last step by the definition of $v_i$. Poincaré-duality non-degeneracy forces $w_i(TM)=v_i$ at degree $i$? Not quite — the actual identity is more subtle, involving the total Steenrod square. The careful argument (Milnor-Stasheff §11) shows that combining the Cartan formula with the Wu pairing gives $w(TM)=\mathrm{Sq}(v)$ at the level of total classes:

$$w_k(TM)=\sum _{i+j=k}{\mathrm{Sq}}^i(v_j),$$

which expands $w_k$ as a sum of Steenrod squares of lower Wu classes. $\square$

Computational corollary. Given the cohomology ring $H^{\ast }(M;{\mathbb{F}}_2)$ and the Steenrod-algebra action, one can compute the Wu class entirely cohomologically: $v_i$ is the dual under Poincaré-duality of the functional $u\mapsto ⟨{\mathrm{Sq}}^{i}u,[M]⟩$. Then the Wu formula expresses $w_k(TM)$ as a linear combination of cup products of Steenrod squares of Wu classes. No bundle, no connection, no curvature needed — only the cohomology and the Steenrod algebra.

Example: real projective spaces. For ${\mathbb{RP}}^n$, $H^{\ast }({\mathbb{RP}}^n;{\mathbb{F}}_2)={\mathbb{F}}_2[a]/(a^{n+1})$. The Wu pairing condition becomes a binomial computation: $v_i=$ the dual of $u\mapsto ⟨{\mathrm{Sq}}^{i}u,[{\mathbb{RP}}^n]⟩$ on $H^{n-i}={\mathbb{F}}_2\cdot a^{n-i}$. One computes $v=(1+a{)}^{n+1}\cdot a^0$ truncated appropriately, recovering the classical formula $w(T{\mathbb{RP}}^n)=(1+a{)}^{n+1}$ from §3 of 03.06.03 via $w(TM)=\mathrm{Sq}(v)$.

SW classes from $\mathrm{Sq}$ on the Thom class [Master]

The Wu construction of Stiefel-Whitney classes via Steenrod squares on the Thom class is the most computational definition of the SW classes, and the bridge between characteristic-class theory and the action of ${\mathcal{A}}_2$ on cohomology.

The Thom class. Let $E\to M$ be a real vector bundle of rank $n$. The Thom space $\mathrm{Th}(E)=D(E)/S(E)$ is the one-point compactification of the total space (equivalently, the disk-bundle modulo the sphere-bundle). The mod-2 Thom class $u\in H^n(\mathrm{Th}(E);{\mathbb{F}}_2)$ is the unique class restricting to the generator of $H^n({\mathbb{R}}^n,{\mathbb{R}}^n\setminus 0;{\mathbb{F}}_2)={\mathbb{F}}_2$ on each fibre. It exists for every real vector bundle (no orientability assumption needed for mod-2 coefficients) and is characterised by this fibrewise normalisation.

Thom isomorphism. The cup-with-$u$ map $\Phi :H^{\ast }(M;{\mathbb{F}}_2)\to {\tilde{H}}^{\ast +n}(\mathrm{Th}(E);{\mathbb{F}}_2)$, $\Phi (x)={\pi }^{\ast }(x)\cup u$, is an isomorphism of degree $n$. This is the mod-2 Thom isomorphism theorem 03.12.15.

Definition (Wu). The Stiefel-Whitney classes $w_i(E)\in H^i(M;{\mathbb{F}}_2)$ are defined by the Steenrod-square / Thom-isomorphism formula

$${\mathrm{Sq}}^i(u)=\Phi (w_i(E))={\pi }^{\ast }(w_i(E))\cup u,i=0,1,\dots ,n,$$

with $w_0(E)=1$ (normalisation ${\mathrm{Sq}}^0=\mathrm{id}$ and $\Phi (1)=u$) and $w_i(E)=0$ for $i>n$ (by instability, ${\mathrm{Sq}}^i(u)=0$ for $i>n$).

Explicit derivation. This definition is the most direct construction of Stiefel-Whitney classes: given the Thom space and the Thom class, one computes ${\mathrm{Sq}}^i(u)$ purely from the Steenrod-algebra action on $H^{\ast }(\mathrm{Th}(E);{\mathbb{F}}_2)$, then inverts the Thom isomorphism to pull the result back to a class on the base. There is no need for connections, frame bundles, or obstruction theory; only the mod-2 cohomology and the Steenrod squares.

Verification of the SW axioms. The four axioms of 03.06.03 now follow:

• Naturality. For $f:M^{\prime }\to M$, $f^{\ast }E$ has Thom class ${\tilde{f}}^{\ast }u$ where $\tilde{f}:\mathrm{Th}(f^{\ast }E)\to \mathrm{Th}(E)$ is the induced map. By Steenrod-square naturality, ${\mathrm{Sq}}^i({\tilde{f}}^{\ast }u)={\tilde{f}}^{\ast }{\mathrm{Sq}}^i(u)={\tilde{f}}^{\ast }({\pi }^{\ast }{w}_i(E)\cup u)={\pi }^{\prime \ast }{f}^{\ast }{w}_i(E)\cup {\tilde{f}}^{\ast }u$, so $w_i(f^{\ast }E)=f^{\ast }{w}_i(E)$.

• Whitney product. For $E\oplus F$ over $M\times M^{\prime }$, $\mathrm{Th}(E\oplus F)=\mathrm{Th}(E)\wedge \mathrm{Th}(F)$, and the Thom class is $u_{E\oplus F}=u_E\wedge u_F$ (smash). Applying ${\mathrm{Sq}}^k$ and using the Cartan formula gives the Whitney formula for $w(E\oplus F)$. The total-class identity $w(E\oplus F)=w(E)w(F)$ is exactly the Cartan formula on Thom classes.

• Normalisation. For the tautological bundle ${\gamma }_1\to {\mathbb{RP}}^{\infty }$, the Thom space is ${\mathbb{RP}}^{\infty }$ itself (with the basepoint at the origin), and the Thom class is the degree-1 generator $a$. Then ${\mathrm{Sq}}^1(a)=a^2=a\cup a={\pi }^{\ast }(a)\cup a$, so $w_1({\gamma }_1)=a$, the non-zero generator.

• Dimension. ${\mathrm{Sq}}^i(u)=0$ for $i>n$ by instability.

So the Wu construction defines the same Stiefel-Whitney classes as the axiomatic construction.

Computational example. Take $E=T{\mathbb{RP}}^n$. The Thom space $\mathrm{Th}(T{\mathbb{RP}}^n)$ has known cohomology (a quotient of ${\mathbb{RP}}^{\infty }$ structures), and the Thom class $u$ can be identified explicitly. Computing ${\mathrm{Sq}}^i(u)$ via the Cartan formula and the structure of $H^{\ast }(\mathrm{Th}(T{\mathbb{RP}}^n);{\mathbb{F}}_2)$ recovers $w(T{\mathbb{RP}}^n)=(1+a{)}^{n+1}$ from §"Worked example" of 03.06.03, by a binomial computation in the Thom-space cohomology.

Cohomology operations and the Hopf-algebra structure of ${\mathcal{A}}_2$ [Master]

The mod-2 Steenrod algebra ${\mathcal{A}}_2$ is not merely a graded ${\mathbb{F}}_2$-algebra — it has a richer structure as a graded cocommutative Hopf algebra over ${\mathbb{F}}_2$. Milnor 1958 ^[Milnor1958] identified the Hopf-algebra structure and computed the dual ${\mathcal{A}}_2^{\ast }$ explicitly.

Coproduct. The diagonal $\Delta :{\mathcal{A}}_2\to {\mathcal{A}}_2\otimes {\mathcal{A}}_2$ is determined by the Cartan formula. On generators,

$$\Delta {\mathrm{Sq}}^k=\sum _{i+j=k}{\mathrm{Sq}}^i\otimes {\mathrm{Sq}}^j.$$

This is the algebraic shadow of the fact that a Steenrod square applied to a cup product distributes over the two factors. The coproduct is cocommutative because the Cartan formula is symmetric in the two factors.

Antipode. ${\mathcal{A}}_2$ is a Hopf algebra, so it carries an antipode $S:{\mathcal{A}}_2\to {\mathcal{A}}_2$ satisfying the standard axiom $m\circ (S\otimes 1)\circ \Delta =\eta \circ ϵ$. The antipode is determined by the coproduct and the algebra structure; explicit formulae are given in Milnor-Moore 1965.

The Milnor dual. The dual ${\mathcal{A}}_2^{\ast }$ is the graded dual vector space, with the dual algebra structure given by the transpose of $\Delta$. Milnor's central computation is that ${\mathcal{A}}_2^{\ast }$ is a polynomial algebra over ${\mathbb{F}}_2$:

$${\mathcal{A}}_2^{\ast }={\mathbb{F}}_2[{\xi }_1,{\xi }_2,{\xi }_3,\dots ],|{\xi }_i|=2^i-1.$$

Each ${\xi }_i$ is the dual of an admissible monomial of degree $2^i-1$, and the algebra structure on the dual side reflects the coproduct structure on ${\mathcal{A}}_2$ being dual to multiplication of admissibles.

The Milnor basis of ${\mathcal{A}}_2$. Dualising back, ${\mathcal{A}}_2$ acquires a vector-space basis indexed by sequences $R=(r_1,r_2,\dots )$ of non-negative integers with finitely many non-zero entries, called the Milnor basis $\{{\mathrm{Sq}}^R\}$. The Milnor basis is computationally superior to the Serre-Cartan basis for many purposes — for example, the action of ${\mathcal{A}}_2$ on a polynomial cohomology ring is more transparent in the Milnor basis.

Cohomology operations of all kinds. ${\mathcal{A}}_2$ is the universal object for stable mod-2 cohomology operations: every natural transformation $H^n(-;{\mathbb{F}}_2)\to H^{n+i}(-;{\mathbb{F}}_2)$ that is stable (commutes with suspension) is an element of ${\mathcal{A}}_2^i$. This is the structural reason the Steenrod squares dominate the homotopy theory of mod-2 cohomology.

Higher analogues: for an odd prime $p$, the mod-$p$ Steenrod algebra ${\mathcal{A}}_p$ is generated by Steenrod $p$-th powers $P^i$ and the Bockstein $\beta$ (no relation to ${\mathrm{Sq}}^1$ of the mod-2 case). The structure parallels ${\mathcal{A}}_2$: a generating set of admissibles, Adem-style relations, a Hopf-algebra structure, and a polynomial dual.

Synthesis. The construction here builds toward 03.12.04 (spectrum), where the Steenrod algebra appears as the homotopy ring of the mod-2 Eilenberg-MacLane spectrum. The Cartan formula is the foundational reason that ${\mathcal{A}}_2$ is a Hopf algebra; the Adem relations are exactly what makes ${\mathcal{A}}_2$ finite-dimensional in each grading; together they identify ${\mathcal{A}}_2$ with the universal stable mod-2 cohomology operation algebra. Putting these together, the Wu formula expresses Stiefel-Whitney classes of a manifold purely in terms of Steenrod-algebra data on its cohomology, generalising the orientability and spin conditions of 03.06.03 into a full structural identification: the central insight is that characteristic-class data of $TM$ is encoded in the ${\mathcal{A}}_2$-module structure on $H^{\ast }(M;{\mathbb{F}}_2)$. This is dual to the way that ${\mathcal{A}}_2$ acts on the cohomology of the Thom spectrum $MO$, and the bridge is Thom's theorem identifying ${\pi }_{\ast }(MO)$ via the ${\mathcal{A}}_2$-module structure of $H^{\ast }(MO;{\mathbb{F}}_2)$ — the pattern recurs every time mod-2 invariants of bundles need to be computed.

Full proof set [Master]

Proposition 1 (Steenrod squares exist and are unique). There exists a family of natural transformations ${\mathrm{Sq}}^i:H^n(-;{\mathbb{F}}_2)\to H^{n+i}(-;{\mathbb{F}}_2)$ satisfying the Steenrod-Epstein axioms, and they are uniquely determined by the axioms.

Proof. Existence: the Steenrod-Epstein cup-$i$ construction (described above) gives an explicit cochain-level definition; the axioms are verified directly. Uniqueness: any family of stable cohomology operations satisfying the axioms must agree on the universal class $a\in H^1(K({\mathbb{F}}_2,1);{\mathbb{F}}_2)=H^1({\mathbb{RP}}^{\infty };{\mathbb{F}}_2)$ — by the unstability axiom and the Cartan formula iterated, the action on $a^n$ is forced, and then naturality plus the Yoneda lemma in the homotopy category of Eilenberg-MacLane spectra forces the action on every space. $\square$

Proposition 2 (Cartan formula on cochains). For cocycles $\alpha \in C^p(X;{\mathbb{F}}_2)$, $\beta \in C^q(X;{\mathbb{F}}_2)$,

$${\mathrm{Sq}}^k([\alpha ]\cup [\beta ])=\sum _{i+j=k}{\mathrm{Sq}}^i[\alpha ]\cup {\mathrm{Sq}}^j[\beta ].$$

Proof. By direct cup-$i$ shuffle: $(\alpha \cup \beta ){⌣}_{p+q-k}(\alpha \cup \beta )={\sum }_{i+j=k}(\alpha {⌣}_{p-i}\alpha )\cup (\beta {⌣}_{q-j}\beta )$ modulo a coboundary. Passing to cohomology classes and using the definition ${\mathrm{Sq}}^i([\alpha ])=[\alpha {⌣}_{n-i}\alpha ]$ delivers the Cartan formula. $\square$

Proposition 3 (Adem relations). For $0<a<2b$, ${\mathrm{Sq}}^a{\mathrm{Sq}}^b={\sum }_{c=0}^{\lfloor a/2\rfloor }\left(\genfrac{}{}{0ex}{}{b-c-1}{a-2c}\right){\mathrm{Sq}}^{a+b-c}{\mathrm{Sq}}^c$ mod 2.

Proof. By naturality, it suffices to verify the relation on the universal cohomology of $K({\mathbb{F}}_2,n)$ for sufficiently large $n$, or equivalently on ${\mathrm{RP}}^{\infty }\times {\mathrm{RP}}^{\infty }$. The action of ${\mathrm{Sq}}^a{\mathrm{Sq}}^b$ on a class $x^m{y}^n$ (with $x,y$ the two generators) is computed via the Cartan formula, producing a polynomial in $x,y$ with binomial-coefficient coefficients. The right-hand side, similarly, produces a polynomial in $x,y$. Both polynomials agree mod 2 by Lucas's theorem applied to the binomial-coefficient identities; the explicit identity is

$$\sum _c\left(\genfrac{}{}{0ex}{}{b-c-1}{a-2c}\right)\left(\genfrac{}{}{0ex}{}{c\text{ from }y}{n-c}\right)\left(\genfrac{}{}{0ex}{}{a+b-c\text{ from }x}{m-(a+b-c)}\right)\equiv \left(\genfrac{}{}{0ex}{}{m+n}{a}\right)\left(\genfrac{}{}{0ex}{}{?}{b}\right)(\mathrm{mod}2),$$

with combinatorial identities that hold whenever $0<a<2b$. Adem 1952 ^[Adem1952] gives the full bookkeeping. $\square$

Proposition 4 (Wu's formula). For a closed smooth $n$-manifold $M$, $w(TM)=\mathrm{Sq}(v)$.

Proof. Combine the Steenrod-square characterisation of Stiefel-Whitney classes (${\mathrm{Sq}}^{i}u={\pi }^{\ast }{w}_i\cup u$ on the Thom class) with the Wu definition of $v$ ($⟨{\mathrm{Sq}}^{i}u,[M]⟩=⟨v_i\cup u,[M]⟩$ via Poincaré duality on the Thom-space side) and the Cartan formula expanded for the total Steenrod square. The computation reduces, via the Thom isomorphism, to an identity on $H^{\ast }(M;{\mathbb{F}}_2)$: $w_k(TM)={\sum }_{i+j=k}{\mathrm{Sq}}^i(v_j)$. Summing over $k$ gives $w(TM)=\mathrm{Sq}(v)$. Milnor-Stasheff §11 ^{[MilnorStasheff1974]} gives the detailed computation. $\square$

Proposition 5 (${\mathcal{A}}_2$ is a Hopf algebra). The mod-2 Steenrod algebra ${\mathcal{A}}_2$ is a graded cocommutative Hopf algebra over ${\mathbb{F}}_2$ with coproduct $\Delta {\mathrm{Sq}}^k={\sum }_{i+j=k}{\mathrm{Sq}}^i\otimes {\mathrm{Sq}}^j$.

Proof. Coassociativity follows from the iterated Cartan formula applied to a triple cup product $x\cup y\cup z$. Cocommutativity is symmetry of the Cartan formula. The antipode is constructed from the coalgebra structure via the standard Hopf-algebra machinery (Milnor-Moore 1965). $\square$

Connections [Master]

• Stiefel-Whitney classes 03.06.03. The Wu formula identifies Stiefel-Whitney classes of $TM$ with Steenrod squares of the Wu class. This unit is the cohomological-operation engine behind that identification.

• Singular cohomology 03.12.11. The Steenrod squares act on mod-2 singular cohomology and make it into a module over ${\mathcal{A}}_2$. This module structure carries strictly more information than the cohomology ring alone.

• Singular cohomology with cup product 03.04.13. The Cartan formula expresses the interaction of $\mathrm{Sq}$ with the cup product; the total Steenrod square is a ring endomorphism of $H^{\ast }(-;{\mathbb{F}}_2)$.

• Eilenberg-Steenrod axioms 03.12.15. The Thom isomorphism used to define SW classes via $\mathrm{Sq}$ on the Thom class is a corollary of the Eilenberg-Steenrod cohomology axioms specialised to relative pairs.

• Spectrum 03.12.04. Stable mod-2 cohomology operations form ${\mathcal{A}}_2$, which is the homotopy ring ${\pi }_{\ast }(H{\mathbb{F}}_2\wedge H{\mathbb{F}}_2)$ of the mod-2 Eilenberg-MacLane spectrum. The Wu formula and Thom-spectrum analysis builds toward 03.06.13 (oriented bordism and Pontryagin-Thom).

Historical & philosophical context [Master]

Norman Steenrod 1947 ^{[Steenrod1947]} introduced the squares now bearing his name in the Annals of Math. paper "Products of cocycles and extensions of mappings," motivated by a question about extending continuous maps from a subcomplex to a full CW complex. The cup-$i$ products were the technical core of the construction; the cohomology operations ${\mathrm{Sq}}^i$ were defined as the cohomological residue of the symmetric cup-$i$ pairing of a cocycle with itself. The construction was extended to odd primes by Steenrod himself in subsequent papers (1952, 1953) and systematised in the Annals Studies monograph Steenrod-Epstein 1962 ^{[Steenrod-Epstein 1962]}.

José Adem 1952 ^[Adem1952] computed the relations among iterated Steenrod squares, cutting the free associative algebra on $\{{\mathrm{Sq}}^i\}$ down to the actual Steenrod algebra. The relations were proved independently by Henri Cartan; the standard reference Cartan-Eilenberg (1956) gives an algebraic-topological derivation via the cohomology of Eilenberg-MacLane spaces.

John Milnor 1958 ^[Milnor1958] identified the Hopf-algebra structure of ${\mathcal{A}}_2$ and computed the polynomial structure of the dual ${\mathcal{A}}_2^{\ast }={\mathbb{F}}_2[{\xi }_1,{\xi }_2,\dots ]$ in the Annals of Math. paper "The Steenrod algebra and its dual." This made ${\mathcal{A}}_2$ a workable object: prior to Milnor's computation, the multiplicative structure of ${\mathcal{A}}_2$ was opaque, but the polynomial dual gave a clean basis and structure constants.

Wu Wen-Tsün 1950 ^[Wu1950] proved the formula $w(TM)=\mathrm{Sq}(v)$ in C. R. Acad. Sci. Paris 230, formulated initially for the cohomology of sphere bundles and later for the tangent bundle of a smooth manifold. Wu's argument used the Thom-isomorphism-style identification of SW classes via Steenrod squares on the Thom class; the formula now bears his name. Thom 1954 ^[Thom1954] applied Wu's formula to the Steenrod-module analysis of $H^{\ast }(MO;{\mathbb{F}}_2)$, identifying the mod-2 unoriented bordism ring ${\Omega }_{\ast }^O\otimes {\mathbb{F}}_2$ as a polynomial ${\mathbb{F}}_2$-algebra on generators in degrees not of the form $2^k-1$. The Wu formula is thus the bridge between characteristic-class theory and the bordism programme.

The monograph Milnor-Stasheff 1974 ^{[MilnorStasheff1974]} codified the modern treatment: Steenrod squares as cohomology operations (Chapters 8-9), Wu's formula (Chapter 11), and the bordism applications (Chapter 17-18). Modern references include Mosher-Tangora Cohomology Operations and Applications in Homotopy Theory (1968) ^{[MosherTangora1968]} for the operational viewpoint, and Hatcher §4.L for the introductory presentation.

Bibliography [Master]

@article{Steenrod1947,
  author = {Steenrod, Norman E.},
  title = {Products of cocycles and extensions of mappings},
  journal = {Annals of Mathematics},
  volume = {48},
  number = {2},
  year = {1947},
  pages = {290-320},
}

@book{Steenrod-Epstein1962,
  author = {Steenrod, Norman E. and Epstein, David B. A.},
  title = {Cohomology Operations},
  publisher = {Princeton University Press},
  series = {Annals of Mathematics Studies},
  volume = {50},
  year = {1962},
}

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

@article{Wu1950,
  author = {Wu, Wen-Ts\"un},
  title = {Sur les classes caract\'eristiques des structures fibr\'ees sph\'eriques},
  journal = {Comptes Rendus de l'Acad\'emie des Sciences, Paris},
  volume = {230},
  year = {1950},
  pages = {508-511},
}

@article{Milnor1958,
  author = {Milnor, John W.},
  title = {The Steenrod algebra and its dual},
  journal = {Annals of Mathematics},
  volume = {67},
  year = {1958},
  pages = {150-171},
}

@article{Adem1952,
  author = {Adem, Jos\'e},
  title = {The iteration of {S}teenrod squares in algebraic topology},
  journal = {Proceedings of the National Academy of Sciences USA},
  volume = {38},
  year = {1952},
  pages = {720-726},
}

@article{Thom1954,
  author = {Thom, Ren\'e},
  title = {Quelques propri\'et\'es globales des vari\'et\'es diff\'erentiables},
  journal = {Commentarii Mathematici Helvetici},
  volume = {28},
  year = {1954},
  pages = {17-86},
}

@book{MosherTangora1968,
  author = {Mosher, Robert E. and Tangora, Martin C.},
  title = {Cohomology Operations and Applications in Homotopy Theory},
  publisher = {Harper \& Row},
  year = {1968},
}

@book{Hatcher2002,
  author = {Hatcher, Allen},
  title = {Algebraic Topology},
  publisher = {Cambridge University Press},
  year = {2002},
}

---

Cycle 5 math T1 batch, unit 1. Steenrod squares and the Wu formula — the cohomological-operation engine behind characteristic-class theory, identifying $w(TM)$ with $\mathrm{Sq}(v)$ for a closed smooth manifold.