← All writingMathematics blog

Exterior Powers 1

Exterior powers of modules, determinants, the wedge product, and duality.

This post is intended to record some elementary proofs of properties of exterior powers, with a focus on the case of finite free modules. It is heavily inspired by Keith Conrad's notes on exterior powers. Another goal is to systematically develop the theory of determinants for finite free modules. In Exterior Powers 2, we will turn to vector bundles on schemes.

Basic properties

Let AA be a commutative ring with 1≠01\neq0, and MM an AA-module, and k≥0k\geq0 an integer. We denote by Alt⁡k(M,N)\Alt^k(M,N) the AA-module of alternating maps Mk→NM^k\to N, i.e. kk-multilinear maps

ϕ ⁣:M×⋯×M→N\phi\colon M\times\cdots\times M\to N

such that ϕ(x1,…,xk)=0\phi(x_1,\dots,x_k)=0 whenever xi=xjx_i=x_j for some i≠ji\neq j. Of course, we can describe this as a submodule of Hom⁡A(M⊗k,N)\Hom_A(M^{\otimes k},N) satisfying the same condition on elementary tensor products. Then it is obvious that Alt⁡k(M,−) ⁣:Mod⁡(A)→Mod⁡(A)\Alt^k(M,-)\colon\Mod(A)\to\Mod(A) defines a functor.

Definition 1. Let MM be an AA-module and k≥0k\geq0. A kkth exterior power of MM is an AA-module ΛkM\LA^k M equipped with an alternating map M×k→ΛkMM^{\times k}\to\LA^k M making ΛkM\LA^k M a representing object of Alt⁡k(M,−)\Alt^k(M,-).

We write the universal alternating map as

(x1,…,xk)↦x1∧⋯∧xk,(x_1,\dots,x_k)\mapsto x_1\wedge\cdots\wedge x_k,

which of course factors through M⊗kM^{\otimes k} by definition. One easily checks that Λ0M=A\LA^0 M=A and Λ1M=M\LA^1 M=M.

Lemma 2. Let MM be an AA-module.

  1. ΛkM\LA^k M exists. If {xi:i∈I}\{x_i:i\in I\} generates MM, then ΛkM\LA^k M is generated by elements of the form xi1∧⋯∧xikx_{i_1}\wedge\cdots\wedge x_{i_k} for ij∈Ii_j\in I. If II is ordered, then it is furthermore generated by such expressions where i1<⋯<iki_1<\cdots<i_k in II.

  2. We have the formula

    xσ(1)∧⋯∧xσ(k)=sgn⁡(σ)x1∧⋯∧xkx_{\sigma(1)}\wedge\cdots\wedge x_{\sigma(k)}=\sgn(\sigma)x_1\wedge\cdots\wedge x_k

    for any permutation σ∈Sk\sigma\in S_k and xi∈Mx_i\in M.

Proof. We can construct it by considering the submodule J⊂M⊗kJ\subset M^{\otimes k} generated by all elementary tensors

x1⊗⋯⊗xkx_1\otimes \cdots\otimes x_k

such that xi=xjx_i=x_j for some i≠ji\neq j. Then take ΛkM=M⊗k/J\LA^k M= M^{\otimes k}/J with its natural alternating map Mk→M⊗k→ΛkMM^k\to M^{\otimes k}\to\LA^k M; it is clear that this satisfies the universal property. This construction shows that the exterior power is generated by expressions as listed, and the last statement of (1) follows from (2).

To obtain (2), write σ\sigma as a product of transpositions τr⋯τ1\tau_r\cdots\tau_1 and induct on rr. In the base case, σ=τ1=(ij)\sigma=\tau_1=(ij), and then

0=x1∧⋯∧(xi+xj)∧⋯∧(xi+xj)∧⋯∧xk=x1∧⋯∧xi∧⋯∧xj∧⋯∧xk+x1∧⋯∧xj∧⋯∧xi∧⋯∧xk\begin{aligned} 0&=x_1\wedge\cdots\wedge(x_i+x_j)\wedge\cdots\wedge(x_i+x_j)\wedge\cdots\wedge x_k \\ &=x_1\wedge\cdots\wedge x_i\wedge\cdots\wedge x_j\wedge\cdots\wedge x_k \\ &\quad+x_1\wedge\cdots\wedge x_j\wedge\cdots\wedge x_i\wedge\cdots\wedge x_k \end{aligned}

gives the result. In the general case, induction implies

xσ(1)∧⋯∧xσ(k)=(−1)r−1xτr(1)∧⋯∧xτr(k),x_{\sigma(1)}\wedge\cdots\wedge x_{\sigma(k)}=(-1)^{r-1}x_{\tau_r(1)}\wedge\cdots\wedge x_{\tau_r(k)},

and then the base case applied to the RHS completes the proof. □

A simple corollary is that if MM is generated by mm elements, then ΛkM=0\LA^k M=0 for k>mk>m by the pigeonhole principle. It is easy to see that Λk\LA^k defines a functor Mod⁡(A)→Mod⁡(A)\Mod(A)\to\Mod(A), where given ϕ ⁣:M→N\phi\colon M\to N, we define Λkϕ\LA^k\phi to be the unique linear map satisfying

Λkϕ(x1∧⋯∧xk)=ϕ(x1)∧⋯∧ϕ(xk),\LA^k\phi(x_1\wedge\cdots\wedge x_k)=\phi(x_1)\wedge\cdots\wedge\phi(x_k),

and functoriality follows from uniqueness in the universal property right away. In general, there is an important map

EmbMk ⁣:ΛkM→M⊗k\Emb^k_M\colon\LA^k M\to M^{\otimes k}

defined as the unique linear map extending

EmbMk(x1∧⋯∧xk)=∑σ∈Sksgn⁡(σ)xσ(1)⊗⋯⊗xσ(k).\Emb^k_M(x_1\wedge\cdots\wedge x_k)=\sum_{\sigma\in S_k}\sgn(\sigma)x_{\sigma(1)}\otimes\cdots\otimes x_{\sigma(k)}.

To check this is well-defined, suppose xi=xjx_i=x_j for some i≠ji\neq j; then the RHS above becomes

∑[σ]∈⟨(ij)⟩\Sk(sgn⁡(σ)xσ(1)⊗⋯⊗xσ(k)+sgn⁡((ij)σ)x(ij)σ(1)⊗⋯⊗x(ij)σ(k))\sum_{[\sigma]\in \langle(ij)\rangle\backslash S_k}\big(\sgn(\sigma)x_{\sigma(1)}\otimes\cdots\otimes x_{\sigma(k)}+\sgn((ij)\sigma)x_{(ij)\sigma(1)}\otimes\cdots\otimes x_{(ij)\sigma(k)}\big)

by summing over coset representatives. The assumption implies that x(ij)σ(m)=xσ(m)x_{(ij)\sigma(m)}=x_{\sigma(m)} for all mm, and sgn⁡((ij)σ)=−sgn⁡(σ)\sgn((ij)\sigma)=-\sgn(\sigma). By linearity, it is easy to check that Embk\Emb^k is a natural transformation from Λk→⨂k\LA^k\to\bigotimes^k. The following is the only real theorem about exterior powers.

Theorem 3. Let MM be a free AA-module with a basis {ei:i∈I}\{e_i:i\in I\} for an ordered index set II. Then ΛkM\LA^k M is free for all kk with basis

{ei1∧⋯∧eik:i1<⋯<ik∈I}.\{e_{i_1}\wedge\cdots\wedge e_{i_k}:i_1<\cdots<i_k\in I\}.

In this case, EmbMk\Emb_M^k is a monomorphism.

Proof. Let I<k={(i1,…,ik)∈Ik:i1<⋯<ik}I_<^k=\{(i_1,\dots,i_k)\in I^k:i_1<\cdots<i_k\}, then we can write an arbitrary element of ΛkM\LA^k M as

ω=∑α∈I<kcαeα∧\omega=\sum_{\alpha\in I_<^k}c_\alpha e_\alpha^{\wedge}

where eα∧=ei1∧⋯∧eike_\alpha^{\wedge}=e_{i_1}\wedge\cdots\wedge e_{i_k} if α=(i1,…,ik)\alpha=(i_1,\dots,i_k). If Emb(ω)=0\Emb(\omega)=0, then

0=∑σ∈Sk∑α∈I<ksgn⁡(σ)cαeσ(α)⊗0=\sum_{\sigma\in S_k}\sum_{\alpha\in I^k_<}\sgn(\sigma)c_\alpha e_{\sigma(\alpha)}^{\otimes}

where eα⊗=ei1⊗⋯⊗eike^\otimes_\alpha=e_{i_1}\otimes\cdots\otimes e_{i_k} and σ(α)=(iσ(1),…,iσ(k))\sigma(\alpha)=(i_{\sigma(1)},\dots,i_{\sigma(k)}). Note that the action map Sk×I<k→IkS_k\times I^k_<\to I^k is injective, and each of the elements eσ(α)⊗e^{\otimes}_{\sigma(\alpha)} are basis elements of M⊗kM^{\otimes k} with respect to the induced basis, thus sgn⁡(σ)cα=0\sgn(\sigma)c_\alpha=0, i.e. ω=0\omega=0 and Emb\Emb is injective. In particular, if ω=0\omega=0, then Emb(ω)=0\Emb(\omega)=0 and the above showed that cα=0c_\alpha=0 for each α\alpha, proving linear independence. □

Corollary 4. Let MM be an AA-module.

  1. If MM is projective, then ΛkM\LA^k M is projective for all kk, and the map EmbMk ⁣:ΛkM→M⊗k\Emb_M^k\colon\LA^k M\to M^{\otimes k} is a monomorphism.

  2. If ϕ ⁣:M→N\phi\colon M\to N is a surjection (resp. isomorphism), then Λkϕ\LA^k\phi is a surjection (isomorphism) for all kk.

  3. If ϕ ⁣:M→N\phi\colon M\to N is a monomorphism, MM is projective, and NN is flat, then Λkϕ\LA^k\phi is a monomorphism for all kk.

Proof. (1) If i ⁣:M→Fi\colon M\to F is a split monomorphism to a free module, then Λki\LA^k i is also a split monomorphism into a free module, using Theorem 3. By naturality of Embk\Emb^k, the second statement follows since ΛkM→ΛkF→F⊗k\LA^k M\to\LA^k F\to F^{\otimes k} are both monomorphisms.

(2) These are obvious from the discussions above.

(3) Naturality of Emb\Emb and (1) reduce the problem to showing that M⊗k→N⊗kM^{\otimes k}\to N^{\otimes k} is injective. Inductively, if M⊗(k−1)→N⊗(k−1)M^{\otimes (k-1)}\to N^{\otimes(k-1)} is injective then both maps in the composite

M⊗k→N⊗(k−1)⊗M→N⊗(k−1)⊗N=N⊗kM^{\otimes k}\to N^{\otimes(k-1)}\otimes M\to N^{\otimes(k-1)}\otimes N=N^{\otimes k}

are injective, first by flatness of MM and then flatness of N⊗(k−1)N^{\otimes(k-1)}. □

Corollary 5.

  1. Let ϕ ⁣:Am→An\phi\colon A^m\to A^n be a map. If ϕ\phi is injective (resp. surjective) then m≤nm\leq n (resp. m≥nm\geq n).

  2. If MM is free and {x1,…,xm}\{x_1,\dots, x_m\} is a linearly independent subset, then

    {xi1∧⋯∧xik:1≤i1<⋯<ik≤m}.\{x_{i_1}\wedge\cdots\wedge x_{i_k}:1\leq i_1<\cdots<i_k\leq m\}.

    is a linearly independent subset of ΛkM\LA^k M.

Proof. (1) follows from applying Λm\LA^m or Λn\LA^n to the map and observing that one of these is zero. (2) follows from the above by looking at the embedding Am→MA^m\to M corresponding to the linearly independent subset. □

Corollary 6. Let MM be a free module. Then a subset {x1,…,xm}\{x_1,\dots,x_m\} is linearly dependent if and only if x1∧⋯∧xmx_1\wedge\cdots\wedge x_m is a torsion element of ΛmM\LA^m M. Here, torsion means annihilated by some nonzero element of AA.

Proof. If {x1,…,xm}\{x_1,\dots,x_m\} is linearly independent, the singleton {x1∧⋯∧xm}\{x_1\wedge\cdots\wedge x_m\} is linearly independent by the above, hence this is torsion free. On the other hand, if there is a relation

a1x1+⋯+amxm=0a_1 x_1+\cdots + a_m x_m=0

in MM with ai≠0a_i\neq0, then

0=x1∧⋯∧∑jajxj∧⋯∧xm=ai(x1∧⋯∧xm),0=x_1\wedge\cdots\wedge\sum\nolimits_j a_j x_j\wedge\cdots\wedge x_m=a_i(x_1\wedge\cdots\wedge x_m),

as desired. □

Determinants

Let MM be a finite free module of rank nn. Then Theorem 3 shows that ΛnM\LA^n M is free of rank 1. Thus, if ϕ ⁣:M→M\phi\colon M\to M is an endomorphism, then Λnϕ\LA^n\phi is multiplication by a unique element of AA, which we write as det⁡ϕ∈A\det{\phi}\in A and call the determinant of ϕ\phi.

Lemma 7. Let ϕ,ψ ⁣:M→M\phi,\psi\colon M\to M be endomorphisms of a finite free module MM. Then det⁡(ϕ∘ψ)=det⁡(ϕ)det⁡(ψ)\det(\phi\circ\psi)=\det(\phi)\det(\psi). In particular, if ϕ\phi is an automorphism then det⁡(ϕ)∈A×\det(\phi)\in A^\times.

Proof. The second statement follows immediately, and the first is a simple consequence of functoriality of Λn\LA^n. □

We will revisit determinants in more detail once we have discussed the wedge product and duality of exterior powers. For now, we state a few basic computations.

Lemma 8. Let ϕ ⁣:M→M\phi\colon M\to M be an endomorphism of a finite free module MM, {e1,…,en}\{e_1,\dots,e_n\} be a basis of MM, and ϕ(ej)=∑iaijei\phi(e_j)=\sum_i a_{ij}e_i for all jj. Then

det⁡(ϕ)=∑σ∈Snsgn⁡(σ)aσ(1)1⋯aσ(n)n=∑σ∈Snsgn⁡(σ)a1σ(1)⋯anσ(n).\det(\phi)=\sum_{\sigma\in S_n}\sgn(\sigma)a_{\sigma(1)1}\cdots a_{\sigma(n)n}=\sum_{\sigma\in S_n}\sgn(\sigma)a_{1\sigma(1)}\cdots a_{n\sigma(n)}.

In particular, det⁡(ϕ)=det⁡(ϕ∨)\det(\phi)=\det(\phi^\lor), where ϕ∨ ⁣:M∨→M∨\phi^\lor\colon M^\lor\to M^\lor is the dual map.

Proof. The second equality follows from the first because we can sum over σ−1∈Sn\sigma^{-1}\in S_n, use sgn⁡(σ)=sgn⁡(σ−1)\sgn(\sigma)=\sgn(\sigma^{-1}), and the fact that

aσ−1(1)1⋯aσ−1(n)n=a1σ(1)⋯anσ(n)a_{\sigma^{-1}(1)1}\cdots a_{\sigma^{-1}(n)n}=a_{1\sigma(1)}\cdots a_{n\sigma(n)}

by simply permuting the product. For the first, we observe

det⁡(ϕ)e1∧⋯∧en=∑i1ai11ei1∧⋯∧∑inainnein=∑i1,…,inai11⋯ainnei1∧⋯∧ein=∑σ∈Snaσ(1)1⋯aσ(n)neσ(1)∧⋯∧eσ(n)=(∑σ∈Snsgn⁡(σ)aσ(1)1⋯aσ(n)n)e1∧⋯∧en,\begin{aligned} \det(\phi)e_1\wedge\cdots\wedge e_n &= \sum_{i_1}a_{i_11}e_{i_1}\wedge\cdots\wedge\sum_{i_n}a_{i_n n}e_{i_n} \\ &=\sum_{i_1,\dots,i_n} a_{i_1 1}\cdots a_{i_n n} e_{i_1}\wedge\cdots\wedge e_{i_n} \\ &=\sum_{\sigma\in S_n}a_{\sigma(1)1}\cdots a_{\sigma(n)n}e_{\sigma(1)}\wedge\cdots\wedge e_{\sigma(n)} \\ &=\big(\sum_{\sigma\in S_n}\sgn(\sigma)a_{\sigma(1)1}\cdots a_{\sigma(n)n}\big)e_1\wedge\cdots\wedge e_n, \end{aligned}

where in the third equality we are using the fact that the action map

Sn→{1,…,n}nσ↦(σ(1),…σ(n))S_n\to\{1,\dots,n\}^n\qquad\sigma\mapsto(\sigma(1),\dots\sigma(n))

identifies SnS_n with the subset of the codomain consisting of tuples with no repetition, and also that any terms in the second sum with repetition must vanish.

The statement about ϕ∨\phi^\lor follows from the fact that if {ε1,…,εn}\{\varepsilon^1,\dots,\varepsilon^n\} is the dual basis, then

ϕ∨(εi)(ej)=εi(ϕ(ej))=aij,\phi^\lor(\varepsilon^i)(e_j)=\varepsilon^i(\phi(e_j))=a_{ij},

i.e. ϕ∨(εi)=∑jaijεj\phi^\lor(\varepsilon^i)=\sum_j a_{ij}\varepsilon^j, so applying the first part to ϕ∨\phi^\lor in this basis gives the second expression in the lemma statement, completing the proof. □

We recall that an m×nm\times n matrix in AA is an element of (Am)n(A^m)^n, thought of as an array with mm rows and nn columns. If XX is an m×nm\times n matrix, then we write X=(aij)ijX=(a_{ij})_{ij} to denote the components of XX, where aija_{ij} is the iith row and jjth column. An m×nm\times n matrix is equivalently an element of Hom⁡A(An,Am)\Hom_A(A^n,A^m), where a map ϕ\phi corresponds to a matrix with components described by ϕ(ej)=∑iaijei\phi(e_j)=\sum_i a_{ij}e_i.

We also recall that if X=(aij)X=(a_{ij}) is an m×nm\times n matrix and Y=(bkl)Y=(b_{kl}) is an n×rn\times r matrix, then their product XYXY is the m×rm\times r matrix corresponding to the composite map Ar→An→AmA^r\to A^n\to A^m. In components, XY=(cij)XY=(c_{ij}) then

cik=∑jaijbjk.c_{ik}=\sum_j a_{ij}b_{jk}.

The AA-module of m×nm\times n matrices is denoted Mm×n(A)M_{m\times n}(A), and the special case of n×nn\times n is denoted Mn(A)M_n(A). The determinant of an n×nn\times n matrix is its determinant as a map An→AnA^n\to A^n.

Proposition 9. Let n≥0n\geq0 and X,Y∈Mn(A)X,Y\in M_n(A).

  1. det⁡(XY)=det⁡(X)det⁡(Y)\det(XY)=\det(X)\det(Y) and det⁡(Id)=1\det(\Id)=1. If YY is an invertible matrix, then det⁡(Y−1XY)=det⁡(X)\det(Y^{-1}XY)=\det(X).

  2. det⁡(X)=det⁡(X⊺)\det(X)=\det(X^{\intercal}), where X⊺X^\intercal denotes the transpose of XX.

  3. det⁡(X)\det(X) is annihilated by a nonzero element of AA if and only if the columns of XX are linearly dependent, if and only if the rows of XX are linearly dependent.

  4. If YY is obtained from XX by scaling a single row or column by a∈Aa\in A, then det⁡(Y)=adet⁡(X)\det(Y)=a\det(X). In particular, det⁡(aX)=andet⁡(X)\det(aX)=a^n\det(X).

  5. If YY is obtained from XX by permuting the rows or columns by σ∈Sn\sigma\in S_n, then det⁡(Y)=sgn⁡(σ)det⁡(X)\det(Y)=\sgn(\sigma)\det(X).

  6. If any row (column, resp.) of XX is a scalar multiple of another row (column), then det⁡(X)=0\det(X)=0.

  7. For v1,…,vn∈Anv_1,\dots,v_n\in A^n, we have

    det⁡(v1,…,vi+w,…,vn)=det⁡(v1,…,vn)+det⁡(v1,…,w,…,vn)\det(v_1,\dots,v_i+w,\dots,v_n)=\det(v_1,\dots,v_n)+\det(v_1,\dots,w,\dots,v_n)

    for any w∈Anw\in A^n. The corresponding statement for rows holds as well.

Proof. All of these statements are immediate for us either because of one of the previous two lemmas, or by the corresponding properties of the exterior product. □

The exterior product

Let MM be an AA-module and k,l≥0k,l\geq0. Then we have an alternating map

Mk→Alt⁡l(M,Λk+lM)M^k\to\Alt^l(M,\LA^{k+l}M)

defined by sending (x1,…,xk)(x_1,\dots,x_k) to the alternating map

(y1,…,yl)↦x1∧⋯∧xk∧y1∧⋯∧yl.(y_1,\dots,y_l)\mapsto x_1\wedge\cdots\wedge x_k\wedge y_1\wedge\cdots\wedge y_l.

Thus we obtain a bilinear map

∧ ⁣:ΛkM⊗AΛlM→Λk+lM\wedge\colon\LA^k M\otimes_A\LA^l M\to\LA^{k+l}M

which we denote (ω,η)↦ω∧η(\omega,\eta)\mapsto\omega\wedge\eta. From our description of generators, it is easy to see that this is the unique bilinear map satisfying

(x1∧⋯∧xk,y1∧⋯∧yl)↦x1∧⋯∧xk∧y1∧⋯∧yl.(x_1\wedge\dots\wedge x_k,y_1\wedge\cdots\wedge y_l)\mapsto x_1\wedge\cdots\wedge x_k\wedge y_1\wedge\cdots\wedge y_l.

We note that if kk or ll is zero, then this is just scalar multiplication.

Proposition 10. Let MM be an AA-module and j,k,l≥0j,k,l\geq0.

  1. (ω∧η)∧ξ=ω∧(η∧ξ)(\omega\wedge\eta)\wedge\xi=\omega\wedge(\eta\wedge\xi) for ω∈ΛjM,η∈ΛkM,ξ∈ΛlM\omega\in\LA^j M,\eta\in\LA^k M,\xi\in\LA^lM.

  2. ω∧η=(−1)jkη∧ω\omega\wedge\eta=(-1)^{jk}\eta\wedge\omega for ω∈ΛjM,η∈ΛkM\omega\in\LA^j M,\eta\in\LA^k M.

In particular, ΛM≔⨁i=0∞ΛiM\LA M\coloneqq\bigoplus_{i=0}^\infty\LA^iM forms a graded-commutative AA-algebra with the product ∧\wedge extended by linearity, called the exterior algebra of MM.

Proof. (1) The two ways to parenthesize give two multilinear maps

ΛjM×ΛkM×ΛlM→Λj+k+lM,\LA^j M\times\LA^kM\times\LA^lM\to\LA^{j+k+l}M,

and to check equality we can check on elementary wedge products, in which case the result is obvious from the definition.

(2) After reducing to the elementary wedge products, the question reduces to computing the sign of the permutation

(1,…,j,j+1,…,j+k)↦(j+1,…,j+k,1,…,j).(1,\dots,j,j+1,\dots,j+k)\mapsto(j+1,\dots,j+k,1,\dots,j).

Indeed, going in order it takes jj transpositions to move j+ij+i to the iith spot, and in total this is j+⋯+j=jkj+\cdots+j=jk transpositions. □

Lemma 11. If MM is generated by {x1,…,xm}\{x_1,\dots,x_m\}, then

ΛkM=∑i=1m(Λk−1M)∧xi.\LA^kM=\sum_{i=1}^m\big(\LA^{k-1}M\big)\wedge x_i.

for all k≥1k\geq1.

Proof. We know that the elements xi1∧⋯∧xikx_{i_1}\wedge\cdots\wedge x_{i_k} generate ΛkM\LA^kM for i1<⋯<iki_1<\cdots<i_k, and by incurring a sign we can always move any element in the wedge to the end, giving the representation above. □

Proposition 12. Let MM be an AA-module and ω∈ΛkM\omega\in\LA^kM for kk odd. Then ω∧ω=0\omega\wedge\omega=0.

Proof. We use standard multi-index notation. If ω=∑IcIxI\omega=\sum_I c_I x_I, then

ω∧ω=∑I,JcIcJxI∧xJ=∑I≠JcIcJxI∧xJ,\omega\wedge\omega=\sum_{I,J}c_Ic_Jx_I\wedge x_J=\sum_{I\neq J}c_Ic_J x_I\wedge x_J,

and for each term cIcJxI∧xJc_Ic_J x_I\wedge x_J with I≠JI\neq J there is a unique corresponding term cIcJxJ∧xIc_Ic_J x_J\wedge x_I, but now we are done because

xI∧xJ+xJ∧xI=(1+(−1)k2)xI∧xJ=0x_I\wedge x_J+x_J\wedge x_I=\big(1+(-1)^{k^2}\big)x_I\wedge x_J=0

as kk was assumed odd. □

Now let us briefly discuss some easy properties of the exterior algebra, in particular its rank, a duality result, and a Künneth formula. First, if MM is finite free with basis {e1,…,em}\{e_1,\dots,e_m\}, then we know that the algebra ΛM\LA M is finite free with basis

{ei1∧⋯∧eik:0≤k≤m,i1<⋯<ik}.\{e_{i_1}\wedge\cdots\wedge e_{i_k}:0\leq k\leq m, i_1<\cdots<i_k\}.

The number of such elements is the same as the number of subsets of {1,…,m}\{1,\dots,m\} (∅\varnothing corresponds to 1∈Λ0M=A1\in\LA^0M=A), so the rank of ΛM\LA M is 2m2^m. Next, we observe that the formation of the full exterior algebra Λ\LA is also functorial because of the following naturality property, for ϕ ⁣:M→N\phi\colon M\to N

ΛkM⊗AΛlM→∧Λk+lMΛkϕ⊗Λlϕ↓↓Λk+lϕΛkN⊗AΛlN→∧Λk+lN\begin{CD} \LA^kM\otimes_A\LA^lM @>{\wedge}>> \LA^{k+l}M \\ @V{\LA^k\phi\otimes\LA^l\phi}VV @VV{\LA^{k+l}\phi}V \\ \LA^kN\otimes_A\LA^lN @>{\wedge}>> \LA^{k+l}N \end{CD}

for which commutativity is easy to check on elementary wedge products.

Duality and the adjugate

Next we discuss duality. Recall that a bilinear pairing between modules MM and NN normally means a bilinear map M×N→AM\times N\to A, and it is perfect if the induced maps

M→Hom⁡A(N,A)=N∨andN→M∨M\to\Hom_A(N,A)=N^\lor\qquad\text{and}\qquad N\to M^\lor

are isomorphisms. Here we slightly generalize this definition so that AA is allowed to be any rank 1 free module LL, and perfectness means that the maps to Hom⁡(N,L)\Hom(N,L) and Hom⁡(M,L)\Hom(M,L) are isomorphisms. In particular, when NN is finite free, the statement applies to M→∼N∨⊗ALM\isoto N^\lor\otimes_A L.

Proposition 13. Let MM be a finite free module of rank nn. Then the wedge product

ΛkM×Λn−kM→ΛnM\LA^k M\times\LA^{n-k}M\to\LA^n M

forms a perfect pairing for 0≤k≤n0\leq k\leq n.

Proof. The question is well-posed since ΛnM\LA^nM is a free module of rank 1. If {e1,…,en}\{e_1,\dots,e_n\} is a basis, then for I⊂{1,…,n}I\subset\{1,\dots,n\} a subset of size kk, we can uniquely write I={i1,…,ik}I=\{i_1,\dots,i_k\} for i1<⋯<iki_1<\cdots<i_k, so write eI=ei1∧⋯∧eike_I=e_{i_1}\wedge\cdots\wedge e_{i_k}; the same thing can be done for n−kn-k. We have a bijection

{I⊂{1,…,n}:∣I∣=k}→∼{I⊂{1,…,n}:∣I∣=n−k}\{I\subset\{1,\dots,n\}:\abs{I}=k\}\isoto \{I\subset\{1,\dots,n\}:\abs{I}=n-k\}

given by I↦{1,…,n}∖II\mapsto\{1,\dots,n\}\setminus I, which induces a bijection between the bases for ΛkM\LA^kM and Λn−kM\LA^{n-k}M. By construction,

eI∧e{1,…,n}∖I=±e1∧⋯∧en≔(−1)aIe{1,…,n}e_I\wedge e_{\{1,\dots,n\}\setminus I}=\pm e_1\wedge\cdots\wedge e_n\coloneqq(-1)^{a_I}e_{\{1,\dots,n\}}

for some numbers aIa_I. On the other hand, if J⊂{1,…,n}J\subset\{1,\dots,n\} is of size n−kn-k and I∩J≠∅I\cap J\neq\varnothing, then eI∧eJ=0e_I\wedge e_J=0; it follows that the bases

{eI:I⊂{1,…,n},∣I∣=k}{(−1)aIe{1,…,n}∖I:I⊂{1,…,n},∣I∣=k}\{e_I:I\subset\{1,\dots,n\},\abs{I}=k\}\qquad \{(-1)^{a_I}e_{\{1,\dots,n\}\setminus I}:I\subset\{1,\dots,n\},\abs{I}=k\}

of ΛkM\LA^kM and Λn−kM\LA^{n-k}M are dual under the pairing. □

As a corollary of this, whenever MM is free of rank n≥1n\geq1, for each map ϕ ⁣:M→M\phi\colon M\to M there is an adjugate map Adj⁡(ϕ)\Adj(\phi) defined by commutativity of the following diagram.

Mathematical diagramThe perfect wedge pairing identifies M with Hom from its (n minus 1)st exterior power to its top exterior power. The dashed adjugate map corresponds to precomposition by the (n minus 1)st exterior power of phi.

The perfect wedge pairing identifies M with Hom from its (n minus 1)st exterior power to its top exterior power. The dashed adjugate map corresponds to precomposition by the (n minus 1)st exterior power of phi.

Diagram TeX source
\begin{tikzcd}[sep=scriptsize]
	M & {\Hom_A(\LA^{n-1}M,\LA^{n}M)} \\
	M & {\Hom_A(\LA^{n-1}M,\LA^{n}M)}
	\arrow["\cong", from=1-1, to=1-2]
	\arrow["{\Adj(\phi)}"', dashed, from=1-1, to=2-1]
	\arrow["{(\LA^{n-1}\phi)^*}", from=1-2, to=2-2]
	\arrow["\cong", from=2-1, to=2-2]
\end{tikzcd}

Explicitly, Adj⁡(ϕ)\Adj(\phi) is the unique map M→MM\to M such that

Adj⁡(ϕ)(x)∧η=x∧ϕ(η)η∈Λn−1M.\Adj(\phi)(x)\wedge\eta=x\wedge\phi(\eta)\qquad\eta\in\LA^{n-1}M.

Observe that for each η\eta, we have

Adj⁡(ϕ)ϕ(x)∧η=ϕ(x)∧ϕ(η)=ϕ(x∧η)=det⁡(ϕ)x∧η.\Adj(\phi)\phi(x)\wedge\eta=\phi(x)\wedge\phi(\eta)=\phi(x\wedge\eta)=\det(\phi)x\wedge\eta.

By non-degeneracy of the pairing, Adj⁡(ϕ)(ϕ(x))=det⁡(ϕ)x\Adj(\phi)(\phi(x))=\det(\phi)x, i.e. we have the relation Adj⁡(ϕ)∘ϕ=det⁡(ϕ)IdM\Adj(\phi)\circ\phi=\det(\phi)\Id_M. This gives an easy criterion for invertibility:

Theorem 14. Let MM be a finite free module of positive rank and ϕ ⁣:M→M\phi\colon M\to M a map. The following are equivalent:

  1. ϕ\phi is an isomorphism.

  2. ϕ\phi is surjective.

  3. det⁡(ϕ)∈A×\det(\phi)\in A^\times.

In this situation, det⁡(ϕ)−1Adj⁡(ϕ)\det(\phi)^{-1}\Adj(\phi) is the inverse of ϕ\phi.

Proof. If (2)(2) holds, then by freeness ϕ\phi is a split epimorphism, so multiplicativity of the determinant shows that det⁡(ϕ)\det(\phi) is a unit.

If (3)(3) holds, then Adj⁡(ϕ)\Adj(\phi) has a right inverse det⁡(ϕ)−1ϕ\det(\phi)^{-1}\phi and is thus surjective, while similarly ϕ\phi has a left inverse det⁡(ϕ)−1Adj⁡(ϕ)\det(\phi)^{-1}\Adj(\phi) and is therefore injective. Then det⁡(Adj⁡(ϕ))\det(\Adj(\phi)) is also a unit, so the same argument replacing ϕ\phi with Adj⁡(ϕ)\Adj(\phi) shows that Adj⁡(ϕ)\Adj(\phi) is also injective. Thus Adj⁡(ϕ)\Adj(\phi) and its right inverse are both isomorphisms, so ϕ\phi is an isomorphism. □

For the zero module, the three conditions hold as well: its unique endomorphism is an isomorphism and has determinant 1.

Remark 15. We briefly note that the adjugate as defined is of course the same thing as defined in linear algebra, and we can extract the minors and cofactor matrices from the exterior algebra as well. It is an easy exercise to compare this with the classical definition and obtain these (and e.g. the Laplace expansion) by standard properties of the wedge products.

Remark 16. In the above situation, observe that if det⁡(ϕ)\det(\phi) is not a zero divisor in AA (which we have shown happens iff ϕ\phi maps any basis to a linearly independent set), then taking determinants and cancelling det⁡(ϕ)\det(\phi) shows that

det⁡(Adj⁡(ϕ))=det⁡(ϕ)n−1.\det(\Adj(\phi))=\det(\phi)^{n-1}.

In rank 1, Adj⁡(ϕ)=IdM\Adj(\phi)=\Id_M. Thus Adj⁡(ϕ)\Adj(\phi) can be an automorphism while ϕ\phi is injective but non-surjective: multiplication by 2 on Z\ZZ is an example.

Corollary 17. Let MM be a finite free module of rank nn. Then {x1,…,xn}\{x_1,\dots,x_n\} is a basis for MM if and only if x1∧⋯∧xnx_1\wedge\cdots\wedge x_n is a basis for ΛnM\LA^nM, i.e. if it spans ΛnM\LA^nM over AA.

Proof. One direction is known; for the other, construct a map ϕ ⁣:M→M\phi\colon M\to M to send a basis of MM to x1,…,xnx_1,\dots,x_n. Taking top exterior powers, a basis of ΛnM\LA^nM is mapped to x1∧⋯∧xnx_1\wedge\cdots\wedge x_n which by assumption is an isomorphism, so det⁡(ϕ)∈A×\det(\phi)\in A^\times which completes the proof. □

The exterior algebra and direct sums

Now we conclude this section with some more basic properties of the exterior algebra on a module.

Proposition 18. Let MM be an AA-module. For any central AA-algebra BB and a map ψ ⁣:M→B\psi\colon M\to B of AA-modules for which ψ(x)2=0\psi(x)^2=0 for all x∈Mx\in M, there is a unique extension ψ~ ⁣:Λ(M)→B\wt{\psi}\colon\LA(M)\to B of ψ\psi to a map of AA-algebras.

Proof. Note that uniqueness is clear, because as an AA-algebra Λ(M)\LA(M) is generated by its submodule MM. For any x,y∈Mx,y\in M, it follows from the assumption

0=ψ(x+y)2=ψ(x)ψ(y)+ψ(y)ψ(x),0=\psi(x+y)^2=\psi(x)\psi(y)+\psi(y)\psi(x),

so any product as above can be permuted by incurring a sign. Thus for any x1,…,xk∈Mx_1,\dots,x_k\in M, ψ(x1)⋯ψ(xk)=0\psi(x_1)\cdots\psi(x_k)=0 if xi=xjx_i=x_j for i≠ji\neq j, since the terms can be brought together by a permutation.

This implies that (x1,…,xk)↦ψ(x1)⋯ψ(xk)(x_1,\dots,x_k)\mapsto\psi(x_1)\cdots\psi(x_k) is an alternating map Mk→BM^k\to B, and thus it provides a map ΛkM→B\LA^kM\to B for each k≥0k\geq0. These all glue into a map ψ~ ⁣:Λ(M)→B\wt{\psi}\colon\LA(M)\to B of AA-modules, and it is obvious that the multiplicative structure is respected, because

ψ~(x1∧⋯∧xk)=ψ(x1)⋯ψ(xk)=ψ~(x1∧⋯∧xi)ψ~(xi+1∧⋯∧xk)\wt{\psi}(x_1\wedge\cdots\wedge x_k)=\psi(x_1)\cdots\psi(x_k)=\wt{\psi}(x_1\wedge\cdots\wedge x_i)\wt{\psi}(x_{i+1}\wedge\cdots\wedge x_k)

for any i≤ki\leq k. □

We remark that the above proposition is the universal property of Λ(M)\LA(M) as a central AA-algebra, but not as a graded-commutative ring. For this, we have an upgraded picture.

Lemma 19. Let MM be an AA-module. For each graded-commutative AA-algebra BB and AA-linear map ψ ⁣:M→B1\psi\colon M\to B_1 such that ψ(x)2=0\psi(x)^2=0 for all x∈Mx\in M, there is a unique graded map ψ~ ⁣:Λ(M)→B\wt{\psi}\colon\LA(M)\to B such that ψ~∣M=ψ\wt{\psi}\rvert_M=\psi.

Proof. The situation in the proposition above still applies, so we just need to check that the resulting map ψ~\wt{\psi} is graded. Indeed, this is demonstrated by the rule

ψ~(x1∧⋯∧xk)=ψ(x1)⋯ψ(xk),\wt{\psi}(x_1\wedge\cdots\wedge x_k)=\psi(x_1)\cdots\psi(x_k),

which shows that ΛkM\LA^kM is mapped into the degree kk component of BB. □

Here is the application: recall that in the category of graded-commutative AA-algebras, the coproduct coincides with the graded tensor product where we define B⊗ACB\otimes_A C by

(B⊗AC)n=⨁i+j=nBi⊗ACj(B\otimes_A C)^n=\bigoplus_{i+j=n}B^i\otimes_A C^j

on the underlying modules, and the multiplication is given by

(a⊗b)(a′⊗b′)=(−1)∣b∣∣a′∣aa′⊗bb′.(a\otimes b)(a'\otimes b')=(-1)^{\abs{b}\abs{a'}}aa'\otimes bb'.

In this sense, we will show that the exterior algebra turns direct sums into graded tensor products. The simple tool is the functor

N1 ⁣:Alg⁡gr(A)→Mod⁡(A),\scr{N}^1\colon\operatorname{Alg}_{\mathrm{gr}}(A)\to\Mod(A),

defined by N1(B)={x∈B1:x2=0}\scr{N}^1(B)=\{x\in B^1:x^2=0\}. Note that if x,y∈N1(B)x,y\in\scr{N}^1(B), then

(x+y)2=x2+xy+yx+y2=xy+yx=0(x+y)^2=x^2+xy+yx+y^2=xy+yx=0

because xx and yy are degree 1; thus this is an AA-module. It is clear that graded maps restrict to this degree 1 submodule to make this functorial.

Theorem 20. There is a fully faithful adjunction

Λ ⁣:Mod⁡(A)⇄Alg⁡gr(A) :N1\LA\colon\Mod(A)\rightleftarrows\operatorname{Alg}_{\mathrm{gr}}(A)\,:\scr{N}^1

with unit the canonical isomorphism M→∼N1(ΛM)M\isoto\scr{N}^1(\LA M). In particular, for any AA-modules MM and NN there is a natural isomorphism

Λ(M⊕N)→∼Λ(M)⊗AΛ(N)\LA(M\oplus N)\isoto\LA(M)\otimes_A\LA(N)

to the graded tensor product, and thus an isomorphism

Λn(M⊕N)→∼⨁i+j=nΛiM⊗AΛjN\LA^n(M\oplus N)\isoto\bigoplus_{i+j=n}\LA^iM\otimes_A\LA^jN

of AA-modules for all n≥0n\geq0.

Proof. The latter statements reduce to proving the adjunction, since ⊕\oplus is the coproduct in Mod⁡(A)\Mod(A), yet the previous lemma proves the adjunction. It's fully faithful because each element in M⊂Λ(M)M\subset\LA(M) squares to zero. □

Example 21. Here is an important example of this: Let M1,…,MkM_1,\dots,M_k be a sequence of AA-modules generated by n1,n2,…,nkn_1,n_2,\dots,n_k elements respectively. Then their direct sum is generated by n≔n1+⋯+nkn\coloneqq n_1+\cdots+n_k elements, and we have

Λn(⨁i=1kMi)≅⨂i=1kΛniMi.\LA^{n}\Big(\bigoplus_{i=1}^k M_i\Big)\cong\bigotimes_{i=1}^k\LA^{n_i}M_i.

To prove this, we observe that the statement is true if we remove the nn and nin_i, and then observe that in any sum n=j1+⋯+jkn=j_1+\cdots+j_k if ji=nij_i= n_i doesn't hold for all ii, then ji>nij_i>n_i for some ii, so ΛjiMi=0\LA^{j_i}M_i=0. Of course, in the finite free case one could count ranks to verify this result, but the importance is that there is a canonical isomorphism. This will be used in Exterior Powers 2.

The transpose pairing

We prove one final duality result in this section, which relates dual modules to the exterior algebra. For an AA-module MM, we have a transpose pairing

ΛkM∨×ΛkM→A\LA^k M^\lor\times\LA^k M\to A

defined as the unique bilinear pairing such that

(φ1∧⋯∧φk,x1∧⋯∧xk)↦det⁡([φi(xj)]ij),(\varphi_1\wedge\cdots\wedge\varphi_k,x_1\wedge\cdots\wedge x_k)\mapsto\det([\varphi_i(x_j)]_{ij}),

defined using the matrix determinant from Basic properties. It is easy to verify that the corresponding map (M∨)k→Alt⁡k(M,A)(M^\lor)^k\to\Alt^k(M,A) is well-defined using properties that we have proven of the determinant, thus the above map is well-defined.

Proposition 22. Let MM be a finite free module. Then the transpose/determinant pairing

det⁡ ⁣:ΛkM∨×ΛkM→A\det\colon\LA^k M^\lor\times\LA^k M\to A

is a perfect pairing for all k≥0k\geq0.

Proof. For k=0k=0, the pairing is multiplication A×A→AA\times A\to A. Assume k>0k>0. Let {e1,…,en}\{e_1,\dots,e_n\} be a basis of MM and write {δ1,…,δn}\{\delta^1,\dots,\delta^n\} for the dual basis, and consider the corresponding bases of ΛkM∨\LA^kM^\lor and ΛkM\LA^kM corresponding to subsets I,J⊂{1,…,n}I,J\subset\{1,\dots,n\} of size kk respectively. If I=JI=J, then

det⁡(δI,eI)=det⁡([δil(eim)]lm)=det⁡([δlm]lm)=det⁡(Id)=1.\det(\delta^I,e_I)=\det([\delta^{i_{l}}(e_{i_m})]_{lm})=\det([\delta_{lm}]_{lm})=\det(\Id)=1.

On the other hand, suppose I≠JI\neq J: if iq∈I∖Ji_q\in I\setminus J, then the row

δiq(ej1),δiq(ej2),…,δiq(ejk)\delta^{i_q}(e_{j_1}),\quad\delta^{i_q}(e_{j_2}),\quad\dots\quad,\delta^{i_q}(e_{j_k})

in the corresponding matrix is identically zero, so det⁡(δI,eJ)=0\det(\delta^I,e_J)=0; we have a similar outcome if some jq∈J∖Ij_q\in J\setminus I. This shows that these two bases are dual with respect to the determinant pairing. □

Corollary 23. For a finite free module MM, the natural determinant map

Λ(M∨)→Λ(M)∨\LA(M^\lor)\to\LA(M)^\lor

is an isomorphism of graded AA-modules, where the dual on the right is taken degreewise.

We note that the determinant pairing can also be described in the following way. The matrix determinant is best described as the inverse map ΛnAn→A\LA^n A^n\to A to the isomorphism A→ΛnAnA\to\LA^n A^n sending 1↦e1∧⋯∧en1\mapsto e_1\wedge\cdots\wedge e_n. Each evaluation map M∨×M→AM^\lor\times M\to A induces

ΛkM∨×ΛkM→ΛkAk\LA^k M^\lor\times\LA^k M\to\LA^k A^k

which can be defined in two ways: either it is defined as

(λ1∧⋯∧λk,x1∧⋯∧xk)↦(∑iλ1(xi)ei)∧⋯∧(∑iλk(xi)ei),(\lambda_1\wedge\cdots\wedge\lambda_k,x_1\wedge\cdots\wedge x_k)\mapsto\big(\sum_i\lambda_1(x_i)e_i\big)\wedge\cdots\wedge\big(\sum_i\lambda_k(x_i)e_i\big),

or the same expression but where we sum over λ\lambda's instead of xx's. To check well-definedness from this definition, one needs the equivalence of these two definitions, which follows from the transpose invariance of determinant proven in Basic properties. There is also a question of naturality: given a map ϕ ⁣:M→N\phi\colon M\to N between finite free AA-modules, the diagram

ΛkN∨→≅(ΛkN)∨Λk(ϕ∨)↓↓(Λkϕ)∨ΛkM∨→≅(ΛkM)∨\begin{CD} \LA^kN^\lor @>{\cong}>> (\LA^kN)^\lor \\ @V{\LA^k(\phi^\lor)}VV @VV{(\LA^k\phi)^\lor}V \\ \LA^kM^\lor @>{\cong}>> (\LA^kM)^\lor \end{CD}

indeed commutes, because the second square in the diagram

ΛkN∨→Hom⁡(ΛkN,ΛkAk)→det⁡(ΛkN)∨Λk(ϕ∨)↓↓(Λkϕ)∗↓(Λkϕ)∨ΛkM∨→Hom⁡(ΛkM,ΛkAk)→det⁡(ΛkM)∨\begin{CD} \LA^kN^\lor @>>> \Hom(\LA^kN,\LA^k A^k) @>{\det}>> (\LA^kN)^\lor \\ @V{\LA^k(\phi^\lor)}VV @VV{(\LA^k\phi)^*}V @VV{(\LA^k\phi)^\lor}V \\ \LA^kM^\lor @>>> \Hom(\LA^kM,\LA^k A^k) @>{\det}>> (\LA^kM)^\lor \end{CD}

commutes for standard reasons, and the first is easy to check by either of the equivalent definitions above.

Back to the blog →