Tensor Products and Exterior Algebra

⚡ Key Takeaways

The tensor product $V \otimes W$ is the universal target for bilinear maps out of $V \times W$ , linearizing multilinearity; the exterior algebra $\Lambda(V) = \bigoplus_k \Lambda^k V$ is the quotient of the tensor algebra by the ideal generated by $v \otimes v$ , encoding antisymmetric (alternating) multilinear structure. Key result: $\dim \Lambda^k V = \binom{n}{n}$ ... more precisely $\binom{n}{k}$ for $\dim V = n$ , and the top power $\Lambda^n V$ is one-dimensional — the home of the determinant. Used everywhere: differential forms, determinants, orientation, cohomology, spinors, Plücker coordinates. Foundations laid by Grassmann (1844) and Cartan (1899).

💡 The Core Insight

Multilinearity is hard; linearity is easy. The tensor product converts a $p$ -linear map $V_1 \times \cdots \times V_p \to U$ into an ordinary linear map $V_1 \otimes \cdots \otimes V_p \to U$ . The exterior power does the same for alternating multilinear maps. Thus questions about determinants, volumes, and bilinear forms reduce to linear algebra on a (larger, but still linear) space. The universal property — not any explicit formula — is the definition, and every concrete realization (matrices, index notation, basis expansions) is a corollary.

DefinitionTensor Product via Universal Property

Let $V$ , $W$ be vector spaces over a field $k$ . A tensor product of $V$ and $W$ is a pair $(T, \otimes)$ consisting of a $k$ -vector space $T$ and a bilinear map $\otimes: V \times W \to T$ , $(v,w) \mapsto v \otimes w$ , such that for every $k$ -vector space $U$ and every bilinear map $B: V \times W \to U$ there exists a unique linear map $\tilde B: T \to U$ with $B = \tilde B \circ \otimes$ :

1

\quad B(v,w) = \tilde B(v \otimes w) \quad \text{for all } v\in V,\ w\in W.

Derivation hint: This is the factorization condition; uniqueness of $\tilde B$ is part of the requirement, forcing $\{v\otimes w\}$ to span $T$ . We write $T = V \otimes_k W$ . The elements $v \otimes w$ are called simple (or decomposable) tensors; general elements are finite sums $\sum_i v_i \otimes w_i$ .

DefinitionConcrete Construction

Let $F(V \times W)$ be the free vector space on the set $V \times W$ (formal $k$ -linear combinations of pairs), and let $R$ be the subspace generated by all elements of the forms

2

\quad (v_1+v_2, w) - (v_1,w) - (v_2,w),\quad (v, w_1+w_2)-(v,w_1)-(v,w_2),\quad (\lambda v, w)-\lambda(v,w),\quad (v,\lambda w)-\lambda(v,w).

Derivation hint: these are exactly the relations that force bilinearity in the quotient. Then $V \otimes_k W := F(V\times W)/R$ , with $v\otimes w := \overline{(v,w)}$ . One checks this satisfies the universal property.

TheoremBasis and Dimension

If $\{e_i\}_{i\in I}$ is a basis of $V$ and $\{f_j\}_{j\in J}$ a basis of $W$ , then $\{e_i \otimes f_j\}_{(i,j)\in I\times J}$ is a basis of $V\otimes_k W$ . In particular, for finite dimensions,

3

\quad \dim_k(V\otimes_k W) = (\dim_k V)(\dim_k W).

Derivation hint: surjectivity follows since simple tensors span; linear independence follows by constructing, for each pair $(i_0,j_0)$ , a bilinear functional $V\times W\to k$ dual to $e_{i_0}\otimes f_{j_0}$ via coordinate projections.

DefinitionTensor Algebra and Exterior Algebra

The tensor algebra of $V$ is the graded associative algebra

4

\quad T(V) = \bigoplus_{p\ge 0} V^{\otimes p}, \qquad V^{\otimes 0}=k,\ V^{\otimes 1}=V,\ V^{\otimes p}=\underbrace{V\otimes\cdots\otimes V}_{p},

Derivation hint: multiplication is concatenation $(\alpha,\beta)\mapsto \alpha\otimes\beta$ , giving the free associative unital $k$ -algebra on $V$ . Let $I_{\wedge}\subset T(V)$ be the two-sided ideal generated by all $v\otimes v$ , $v\in V$ . The exterior algebra is the quotient

5

\quad \Lambda(V) := T(V)/I_{\wedge}, \qquad \Lambda^k V := V^{\otimes k}/(I_\wedge \cap V^{\otimes k}),

Derivation hint: $I_\wedge$ is homogeneous (generated in degree 2), so the quotient inherits the grading $\Lambda(V)=\bigoplus_k \Lambda^k V$ . with product denoted $\wedge$ . The defining relation $v\wedge v = 0$ implies, by polarization,

6

\quad v\wedge w = -\,w\wedge v \quad (\text{char }k\ne 2),\qquad \text{and generally } \alpha\wedge\beta = (-1)^{pq}\,\beta\wedge\alpha \text{ for } \alpha\in\Lambda^p,\ \beta\in\Lambda^q.

Derivation hint: expand $0=(v+w)\wedge(v+w)=v\wedge v + v\wedge w + w\wedge v + w\wedge w$ and cancel the diagonal terms.

TheoremDimension of Exterior Powers

For $\dim_k V = n$ with basis $\{e_1,\dots,e_n\}$ , the set $\{e_{i_1}\wedge\cdots\wedge e_{i_k} : 1\le i_1<\cdots<i_k\le n\}$ is a basis of $\Lambda^k V$ . Hence

7

\quad \dim_k \Lambda^k V = \binom{n}{k}, \qquad \dim_k \Lambda(V) = \sum_{k=0}^n \binom{n}{k} = 2^n,

with $\Lambda^k V = 0$ for $k>n$ and $\dim_k \Lambda^n V = 1$ . Derivation hint: antisymmetry lets one reorder any product into increasing index order (up to sign) and kills any product with a repeated index; independence follows from the universal property of $\Lambda^k$ applied to coordinate alternating forms.

TheoremUniversal Property of the Exterior Power

For each $k$ , the map $\wedge: V^k \to \Lambda^k V$ , $(v_1,\dots,v_k)\mapsto v_1\wedge\cdots\wedge v_k$ , is alternating multilinear and universal: for every alternating $k$ -linear map $\omega: V^k \to U$ there is a unique linear $\tilde\omega: \Lambda^k V \to U$ with $\omega = \tilde\omega\circ\wedge$ . Consequently

8

\quad \mathrm{Alt}^k(V;U) \;\cong\; \mathrm{Hom}_k(\Lambda^k V, U), \qquad (\Lambda^k V)^* \cong \Lambda^k(V^*) \cong \mathrm{Alt}^k(V;k).

Derivation hint: alternating maps vanish on the generators $v\otimes v$ , hence factor through $T(V)/I_\wedge$ in degree $k$ .

ProofExistence, Uniqueness, and the Determinant from

\dim\Lambda^n V = 1

Step 1 — Existence. Construct $T = F(V\times W)/R$ as in [2]. The composite $V\times W \to F(V\times W)\to T$ is bilinear because the relations in $R$ are precisely the failures of bilinearity. Given any bilinear $B: V\times W\to U$ , define $\hat B: F(V\times W)\to U$ on basis pairs by $\hat B(v,w)=B(v,w)$ and extend linearly; since $B$ is bilinear, $\hat B$ annihilates $R$ , descending to $\tilde B: T\to U$ . Simple tensors span $T$ , so $\tilde B$ is unique. ∎(existence)

Step 2 — Uniqueness up to unique isomorphism. If $(T,\otimes)$ and $(T',\otimes')$ both satisfy [1], applying universality to $\otimes'$ and $\otimes$ yields linear maps $\phi: T\to T'$ , $\psi: T'\to T$ with $\phi\circ\otimes=\otimes'$ , $\psi\circ\otimes'=\otimes$ . Then $\psi\phi$ and $\mathrm{id}_T$ both factor $\otimes$ through $T$ ; uniqueness forces $\psi\phi=\mathrm{id}_T$ , similarly $\phi\psi=\mathrm{id}_{T'}$ . ∎(uniqueness)

Step 3 — The exterior power. Repeat with $p$ -linear alternating maps and the ideal $I_\wedge$ ; the relation $v\otimes v\in I_\wedge$ is exactly what makes the induced map alternating, giving [8].

Step 4 — Determinant. Take $\dim V = n$ , $U = \Lambda^n V \cong k$ (by [7]). For a linear operator $A: V\to V$ , the map $(v_1,\dots,v_n)\mapsto Av_1\wedge\cdots\wedge Av_n$ is alternating $n$ -linear, so it factors as a scalar $\det(A)$ times $v_1\wedge\cdots\wedge v_n$ :

9

\quad Av_1\wedge\cdots\wedge Av_n = \det(A)\,\bigl(v_1\wedge\cdots\wedge v_n\bigr).

Derivation hint: $\Lambda^n A$ is an endomorphism of the 1-dimensional space $\Lambda^n V$ , hence multiplication by a unique scalar; that scalar is $\det A$ . Multiplicativity $\det(AB)=\det A\det B$ follows from functoriality $\Lambda^n(AB)=\Lambda^n A\circ\Lambda^n B$ . ∎

∎

mathematician · Hermann Grassmann | 1809–1877

A schoolteacher in Stettin with no university position, Grassmann published Die lineale Ausdehnungslehre (1844) introducing the "extensive quantities" and the alternating "outer product" $a\wedge b$ that anticipated $\Lambda(V)$ by decades. The work was so abstract and idiosyncratically philosophical that it was almost completely ignored; frustrated, Grassmann turned to Sanskrit philology, producing a celebrated translation of the Rig-Veda. Only late in his life — and posthumously, through Klein, Élie Cartan, and the rise of vector and tensor calculus — was his exterior algebra recognized as foundational. The wedge symbol and the entire apparatus of differential forms are his intellectual descendants.

🧠 Intuition

Think of $V\otimes W$ as the space of "formal products" of vectors, where the only rules are those of bilinearity — you may distribute over sums and pull scalars across the $\otimes$ , but you may not swap or simplify further. A simple tensor $v\otimes w$ is like a rank-one matrix $vw^\top$ ; a general tensor is a sum of such, i.e. an arbitrary matrix (this is literally true: $V\otimes W^* \cong \mathrm{Hom}(W,V)$ ). For the exterior algebra, picture $v\wedge w$ as the oriented parallelogram spanned by $v$ and $w$ : swapping the vectors reverses orientation (hence the minus sign), and a "degenerate parallelogram" with $v\wedge v$ has zero area. Then $v_1\wedge\cdots\wedge v_k$ is an oriented $k$ -dimensional volume element, and $\Lambda^n V$ being one-dimensional says all top-dimensional oriented volumes are scalar multiples of one another — that scalar is the determinant.

Figure

Geometric meaning of

v\wedge w

as oriented area; swapping vectors flips sign

HA

How a physicist reads this

tensors in relativity

In physics, an element of $V^{\otimes p}\otimes (V^*)^{\otimes q}$ is a type- $(p,q)$ tensor with $p$ contravariant (upper) and $q$ covariant (lower) indices, e.g. the metric $g_{\mu\nu}$ ( $\in (V^*)^{\otimes 2}$ ) or the Riemann tensor $R^\rho{}_{\sigma\mu\nu}$ . Contraction is the natural pairing $V\otimes V^*\to k$ . Antisymmetric tensors are differential forms: the electromagnetic field $F = F_{\mu\nu}\,dx^\mu\wedge dx^\nu \in \Lambda^2$ , and Maxwell's equations become $dF = 0$ , $d\star F = J$ . The exterior derivative, Hodge star, and integration of forms over manifolds are all built on this algebra — index gymnastics is just bookkeeping for tensor and exterior products.

ExampleTensor product of two operators — Kronecker product

Given: $V = W = \mathbb{R}^2$ with standard basis $e_1,e_2$ . Let $A=\begin{pmatrix}1&2\\0&3\end{pmatrix}$ , $B=\begin{pmatrix}0&1\\1&0\end{pmatrix}$ . Compute $A\otimes B$ acting on $\mathbb{R}^2\otimes\mathbb{R}^2 \cong \mathbb{R}^4$ with ordered basis $e_1\otimes e_1, e_1\otimes e_2, e_2\otimes e_1, e_2\otimes e_2$ .

Steps: The action is $(A\otimes B)(u\otimes v) = Au\otimes Bv$ . In matrix form this is the Kronecker product, replacing each entry $A_{ij}$ by the block $A_{ij}B$ :

10

\quad A\otimes B = \begin{pmatrix} 1\cdot B & 2\cdot B\\ 0\cdot B & 3\cdot B\end{pmatrix} = \begin{pmatrix} 0&1&0&2\\ 1&0&2&0\\ 0&0&0&3\\ 0&0&3&0\end{pmatrix}.

Derivation hint: expand $(A\otimes B)(e_j\otimes e_\ell)=\sum_{i,k}A_{ij}B_{k\ell}\,(e_i\otimes e_k)$ and read off columns. Check eigenvalues: $\det(A\otimes B) = (\det A)^2(\det B)^2 = 3^2\cdot(-1)^2 = 9$ . Direct $4\times4$ determinant: expanding gives $9$ . ✓

Answer:

\boxed{A\otimes B=\begin{pmatrix} 0&1&0&2\\ 1&0&2&0\\ 0&0&0&3\\ 0&0&3&0\end{pmatrix},\quad \det(A\otimes B)=9.}

ExampleWedge products and a determinant in

\Lambda^3 \mathbb{R}^3

Given: $u=(1,2,0)$ , $v=(0,1,3)$ , $w=(2,0,1)$ in $\mathbb{R}^3$ with basis $e_1,e_2,e_3$ . Compute $u\wedge v\wedge w$ in $\Lambda^3\mathbb{R}^3$ (a 1-dimensional space spanned by $e_1\wedge e_2\wedge e_3$ ).

Steps: Expand multilinearly and use antisymmetry. Only the term picking distinct basis vectors in increasing order survives, scaled by the determinant of coordinates:

11

\quad u\wedge v\wedge w = \det\!\begin{pmatrix}1&2&0\\0&1&3\\2&0&1\end{pmatrix}\,(e_1\wedge e_2\wedge e_3).

Derivation hint: by [9] with the standard volume form, the coefficient is exactly $\det$ of the matrix whose rows are $u,v,w$ . Compute the determinant: $1(1\cdot1 - 3\cdot0) - 2(0\cdot1 - 3\cdot2) + 0 = 1(1) - 2(-6) = 1 + 12 = 13$ .

Answer:

\boxed{u\wedge v\wedge w = 13\,(e_1\wedge e_2\wedge e_3)},

so the three vectors span an oriented parallelepiped of (signed) volume $13$ , positively oriented.

⚠️ Common mistake

Not every tensor is simple. A generic element of $V\otimes W$ is a sum $\sum_i v_i\otimes w_i$ and cannot be written as a single $v\otimes w$ . The minimal number of simple terms needed is the tensor rank (= matrix rank under $V\otimes W^*\cong\mathrm{Hom}(W,V)$ ). For $\ge 3$ factors, computing tensor rank is NP-hard, and rank can even jump under limits (border rank). Treating $V\otimes W$ as merely "pairs $(v,w)$ " is wrong — it is the span of such pairs modulo bilinearity.

$\otimes$ is not commutative on the nose. There is a canonical isomorphism $V\otimes W \cong W\otimes V$ , but they are not literally equal, and the swap $v\otimes w\mapsto w\otimes v$ matters (e.g. it introduces signs in the graded/super setting).

Counterexample

$v\wedge w = 0$ does NOT mean $v=0$ or $w=0$ . In $\Lambda^2\mathbb{R}^3$ , take $v=(1,0,0)$ and $w=(2,0,0)=2v$ . Then

12

\quad v\wedge w = v\wedge(2v) = 2\,(v\wedge v) = 0,

Derivation hint: parallel vectors span a degenerate (zero-area) parallelogram. even though both vectors are nonzero. In general $v_1\wedge\cdots\wedge v_k=0$ iff the $v_i$ are linearly dependent. This is exactly why exterior algebra detects independence/rank — and why it underlies Plücker embeddings of Grassmannians. (Contrast: in $V\otimes V$ , $v\otimes w=0$ does require $v=0$ or $w=0$ .)

⚠️ Common mistake

Characteristic 2 subtlety. The relation defining $\Lambda(V)$ is $v\wedge v=0$ , which implies antisymmetry $v\wedge w=-w\wedge v$ . But in characteristic 2, " $-1=1$ ", so antisymmetry alone (the alternating-sign condition) is weaker than $v\wedge v=0$ (the alternating condition). Always define $\Lambda(V)$ by the ideal generated by $v\otimes v$ , never merely by antisymmetry, to get the correct theory in all characteristics.

connection

The tensor product is the engine behind a web of canonical isomorphisms (all natural): $V^*\otimes W \cong \mathrm{Hom}(V,W)$ for finite-dimensional $V$ ; $(V\otimes W)^*\cong V^*\otimes W^*$ ; and $\mathrm{Hom}(V\otimes W, U)\cong \mathrm{Bil}(V\times W;U)$ . It makes the category of vector spaces a Symmetric Monoidal Category, with $\otimes$ as the monoidal product and $k$ as the unit. The symmetric algebra $S(V)=T(V)/(v\otimes w - w\otimes v)$ is the commutative cousin of $\Lambda(V)$ ; together they give the decomposition $V\otimes V \cong S^2V \oplus \Lambda^2 V$ (in char $\ne 2$ ). Exterior powers feed directly into Differential Forms, de Rham Cohomology, and the Hodge Star Operator; the determinant construction connects to Characteristic Polynomial and Orientation. Over rings rather than fields, $\otimes_R$ becomes the central operation of Homological Algebra (Tor functors measure its failure of exactness).

🔬 Where this appears

⏳ History

- 1844 — Hermann Grassmann publishes Die lineale Ausdehnungslehre, introducing the alternating "outer product" and

n

-dimensional linear algebra, decades ahead of its time and largely ignored.

- 1862 — Grassmann issues a rewritten, more accessible second edition; still little recognized in his lifetime.

- 1878 — William Kingdon Clifford fuses Grassmann's exterior algebra with Hamilton's quaternions into the Clifford (geometric) algebra.

- 1884–1898 — Gregorio Ricci-Curbastro develops the "absolute differential calculus" (tensor calculus), with Levi-Civita; tensors gain systematic index notation.

- 1899 — Élie Cartan introduces the exterior derivative and the calculus of differential forms, building global analysis on the exterior algebra.

- 1916 — Einstein's general relativity makes tensor calculus indispensable to physics, popularizing the framework worldwide.

- 1930s–40s — Bourbaki and Hassler Whitney formalize the tensor product via its universal property, embedding it in modern abstract algebra and category theory.

- 1945 — Eilenberg–Mac Lane found category theory; the tensor product is later recognized as the prototypical monoidal structure (1963, Bénabou/Mac Lane).

- 1957 onward — Tensor and Tor functors become central to Cartan–Eilenberg homological algebra; tensor decompositions later become workhorses in numerical analysis and data science.

open-problem

Over a field, $V\otimes W$ is well-behaved, but several deep difficulties persist in richer settings:

Tensor rank is hard. Determining the rank of a 3-tensor is NP-hard (Håstad), and the border rank can be strictly smaller than the rank — the set of rank- $\le r$ tensors need not be closed. The exact rank of matrix multiplication (the exponent $\omega$ of matrix-multiplication complexity, $2\le\omega<2.372$ ) is a famous open problem phrased entirely in terms of tensor rank.
Non-flatness over rings. For modules over a commutative ring $R$ , $-\otimes_R M$ need not preserve injections; the failure is measured by $\mathrm{Tor}^R_i$ . Determining when modules are flat is subtle and central to algebraic geometry.
Tensor products of infinite-dimensional / topological spaces require completions (projective vs. injective tensor products of Banach spaces — Grothendieck's theory), and these need not coincide.
Functoriality of $\Lambda^k$ and splitting of $V^{\otimes p}$ into irreducible $GL(V)$ - and $S_p$ -representations is governed by Schur–Weyl duality and Young tableaux — combinatorially intricate and an active research area in representation theory.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also