The Symmetric Algebra of a Module with an Involution
Introduction
The symmetric algebra $\operatorname{sym}(M)$ of a module $M$ over a commutative ring $R$ is the free commutative $R$-algebra on $M$, the quotient of the tensor algebra by the two-sided ideal generated by the differences $x\otimes y-y\otimes x$; an involution of $M$ — a map of order two in the sense of Involutive Linear Spaces — therefore extends to an involution of $\operatorname{sym}(M)$ by the universal property, because on a commutative algebra the difference between an automorphism and an anti-automorphism disappears. The induced involution is this extension; unlike the general case of an involution of an associative algebra, no anti-multiplicativity is imposed or needed, and the induced map is an $R$-algebra automorphism of order two. The purpose of this article is to compute its fixed subalgebra and to situate it with respect to the divided-power pairing, the perfect pairing through which $\operatorname{sym}(M)$ is dual to the divided-power algebra of the dual module.
The article defines the induced involution, proves that it is the unique algebra automorphism of order two extending a given involution of $M$, and identifies its fixed subalgebra: in degree $k$ the fixed elements are the monomials with an even number of anti-fixed factors, so the fixed subalgebra is generated by the fixed part $M^+$ together with the pairwise products $xy$ of elements of the anti-fixed part $M^-$. It then derives the compatibility of the induced involution with the divided-power pairing: the transpose of the induced involution of $\operatorname{sym}(M)$ is the induced involution of $\operatorname{sym}(M^\vee)$, and the pairing is invariant. The two worked cases are the free module with a basis permuted or negated by the involution, where the fixed subalgebra is the even-degree part, and the polynomial ring with the induced involution, whose fixed symmetric part is the object of Involutions of a Polynomial Ring and the Symmetric Part in the linear-space reading.
The article assumes The Symmetric Algebra of a Module for $\operatorname{sym}(M)$ and its universal property, Tensor Algebras and Free Algebras for the tensor algebra, The Symmetric and Exterior Powers for the graded pieces, Involutive Linear Spaces for the involution of a module, Involutive Algebras for the involution of an associative algebra, The Dual Module and the Transpose for the transpose, and Divided Powers and the Symmetric Pairing for the divided-power algebra. The symmetric powers of an involutive module are the object of The Symmetric Powers of an Involutive Module, later in this group; the involution-invariant ideals are Involution-Invariant Ideals of the Symmetric Algebra. Throughout, $R$ is a commutative ring with identity, $M$ is an $R$-module, $\sigma_M$ is an $R$-linear involution of $M$ with $\sigma_M^2 = \mathrm{id}$, and $\sigma$ is the induced involution of $\operatorname{sym}(M)$; no form, norm or distance occurs.
The Induced Involution
Definition and Universal Property
Definition. Let $\sigma_M$ be an $R$-linear involution of $M$, and let $\iota : M \to \operatorname{sym}(M)$ be the canonical map. The induced involution of the symmetric algebra is the unique $R$-algebra homomorphism
$$ \sigma : \operatorname{sym}(M) \to \operatorname{sym}(M) \quad \text{with} \quad \sigma\circ\iota = \iota\circ\sigma_M . $$
Theorem. The induced map $\sigma$ exists and is unique, and it is an $R$-algebra automorphism of $\operatorname{sym}(M)$ of order two; on $\operatorname{sym}^k(M)$ it maps the product $v_1\cdots v_k$ to $\sigma_M(v_1)\cdots\sigma_M(v_k)$.
Proof. The universal property of the symmetric algebra says that an $R$-linear map $M\to B$ into a commutative $R$-algebra $B$ extends uniquely to an $R$-algebra homomorphism $\operatorname{sym}(M)\to B$. Applying it with $B = \operatorname{sym}(M)$ and the map $\iota\circ\sigma_M$ gives existence and uniqueness of $\sigma$. Since $\sigma^2\circ\iota = \iota\circ\sigma_M^2 = \iota$, uniqueness forces $\sigma^2 = \mathrm{id}$. The formula on the degree-$k$ piece is multiplicativity read on the spanning monomials.
Corollary. The correspondence $\sigma_M\mapsto\sigma$ is a homomorphism from the group of the $R$-linear involutions of $M$ to the group of the $R$-algebra involutions of $\operatorname{sym}(M)$; it is injective because $\sigma$ restricts to $\sigma_M$ on the degree-one piece, and it is an isomorphism onto the involutions of $\operatorname{sym}(M)$ that preserve the canonical copy of $M$.
Naturality
Proposition. For involutive modules $(M,\sigma_M)$, $(N,\sigma_N)$ and an $R$-linear map $f : M \to N$ with $f\sigma_M = \sigma_Nf$, the induced map $\operatorname{sym}(f) : \operatorname{sym}(M)\to\operatorname{sym}(N)$ satisfies $\operatorname{sym}(f)\circ\sigma = \sigma'\circ\operatorname{sym}(f)$, where $\sigma,\sigma'$ are the induced involutions. Hence the symmetric algebra is a functor of the involutive module.
Proof. Both sides are $R$-algebra homomorphisms agreeing on $\iota(M)$: $\operatorname{sym}(f)(\sigma(\iota v)) = \operatorname{sym}(f)(\iota\sigma_Mv) = \iota f\sigma_Mv = \iota\sigma_Nfv = \sigma'(\iota fv) = \sigma'(\operatorname{sym}(f)(\iota v))$. Uniqueness of the extension gives equality.
The Fixed Subalgebra
Degree Decomposition
Definition. The symmetric algebra is graded, $\operatorname{sym}(M) = \bigoplus_{k\ge0}\operatorname{sym}^k(M)$, and $\sigma$ preserves the degree, $\sigma(\operatorname{sym}^k(M))\subseteq\operatorname{sym}^k(M)$, because it is induced by a degree-preserving linear map.
Theorem. The fixed subalgebra of the induced involution is
$$ \boxed{\;\operatorname{sym}(M)^\sigma = \bigoplus_{k\ge 0}\operatorname{sym}^k(M)^\sigma , \qquad \operatorname{sym}^k(M)^\sigma = \operatorname{span}_R\{v_1\cdots v_k : \#\{i : v_i \in M^-\} \ \text{is even}\}\;}, $$
where $M^+ = \{v : \sigma_M(v) = v\}$ and $M^- = \{v : \sigma_M(v) = -v\}$ are the fixed and the anti-fixed parts of $M$; equivalently, $\operatorname{sym}(M)^\sigma$ is the $R$-subalgebra generated by $M^+$ and by the pairwise products $xy$ with $x, y \in M^-$.
Proof. If the involution is linear over $R$ with $2 = 1+1$ not a zero divisor, $M = M^+\oplus M^-$ and a monomial in homogeneous factors is fixed exactly when the number of factors from $M^-$ is even, which is the displayed condition; the subalgebra it generates is spanned by the even monomials, and an even monomial is a product of fixed elements, namely the factors from $M^+$ and the pairs of factors from $M^-$. In the general case the displayed span is still the fixed set, because the condition is linear and the monomials span.
Corollary (the case $M^- = 0$). If the involution of $M$ is trivial then the induced involution is trivial and $\operatorname{sym}(M)^\sigma = \operatorname{sym}(M)$, the whole algebra. If $M^+ = 0$ and $2 \neq 0$ is invertible, then $M = M^-$, the induced involution is the parity operator of the grading, and the fixed subalgebra is the even part $\bigoplus_k\operatorname{sym}^{2k}(M)$.
Corollary (generation). The fixed subalgebra is generated as an $R$-algebra by $M^+$ together with the set $\{xy : x, y \in M^-\}$; it is the image of the symmetric algebra of the fixed module under the map that adds the squares.
The Divided-Power Pairing
The Pairing
Let $M^\vee = \operatorname{Hom}_R(M,R)$ be the dual module and let $\Gamma(M^\vee)$ be the divided-power algebra of $M^\vee$, the free polynomial-like algebra in which the divided powers $\gamma_n(\varphi)$ satisfy $\gamma_m(\varphi)\gamma_n(\varphi) = \binom{m+n}{m}\gamma_{m+n}(\varphi)$. The divided-power pairing, or symmetric pairing, is the $R$-bilinear map
$$ \langle\cdot,\cdot\rangle : \operatorname{sym}(M)\times\Gamma(M^\vee) \longrightarrow R , \qquad \langle v_1\cdots v_k, \gamma_k(\varphi)\rangle = \varphi(v_1)\cdots\varphi(v_k), $$
which is a perfect pairing when $M$ is a free module of finite rank and is compatible with the gradings of the two sides.
The Transpose of the Induced Involution
Definition. The transpose of $\sigma_M$ is the unique $R$-linear map $\sigma_M^{\mathsf{T}} : M^\vee \to M^\vee$ with $\langle\sigma_M v, \varphi\rangle = \langle v, \sigma_M^{\mathsf{T}}\varphi\rangle$ for all $v \in M$, $\varphi \in M^\vee$; it is an involution of $M^\vee$ when $\sigma_M$ is an involution.
Theorem. The transpose of the induced involution $\sigma$ of $\operatorname{sym}(M)$ with respect to the divided-power pairing is the induced involution $\sigma^\vee$ of the divided-power algebra $\Gamma(M^\vee)$ generated by $\sigma_M^{\mathsf{T}}$; the pairing satisfies
$$ \boxed{\;\langle \sigma u, \sigma^{\vee}\gamma\rangle = \langle u, \gamma\rangle\;} $$
for all $u \in \operatorname{sym}(M)$ and $\gamma \in \Gamma(M^\vee)$, and $\sigma$ is the transpose of $\sigma^\vee$ in the same sense.
Proof. On the generators, $\langle\sigma(v_1\cdots v_k),\gamma_k(\varphi)\rangle = \prod_i \langle\sigma_Mv_i,\varphi\rangle = \prod_i \langle v_i,\sigma_M^{\mathsf{T}}\varphi\rangle = \langle v_1\cdots v_k,\gamma_k(\sigma_M^{\mathsf{T}}\varphi)\rangle$, and the extension to $\Gamma(M^\vee)$ is multiplicative, so the displayed invariance holds on the spanning elements and hence everywhere by bilinearity.
Corollary. The induced involution of the symmetric algebra is the transpose of the induced involution of the divided-power algebra, and the two are intertwined by the pairing; in particular the fixed subalgebra of one side pairs dually with the cofixed quotient of the other, and the invariant functionals on $\operatorname{sym}(M)$ are the self-adjoint elements of $\Gamma(M^\vee)$ under the induced involution.
Examples
Example (the free module with a permuted basis). Let $M$ be free with basis $e_1,\dots,e_n$ and let $\sigma_M$ exchange $e_1\leftrightarrow e_2$ and fix the remaining basis elements. Then $\operatorname{sym}(M)$ is the polynomial ring $R[e_1,\dots,e_n]$ and $\sigma$ is the involution exchanging the two variables $e_1,e_2$; the fixed subalgebra is $R[e_1+e_2,\ e_1e_2,\ e_3,\dots,e_n]$, generated by the fixed $M^+ = \langle e_1+e_2,e_3,\dots,e_n\rangle$ together with the product $e_1e_2$ of the two anti-fixed generators.
Example (the negated basis). Let $M$ be free with basis $e_1,\dots,e_n$ and let $\sigma_M(e_i) = -e_i$. Then $M^+ = 0$, $M = M^-$ under the assumption $2 \ne 0$, and the induced involution is the determinant of the sign of a monomial; the fixed subalgebra is the even-degree part of the polynomial ring, generated by the products $e_ie_j$.
Summary
An $R$-linear involution $\sigma_M$ of a module $M$ extends uniquely to an $R$-algebra automorphism $\sigma$ of order two of the symmetric algebra, the induced involution, which preserves the degree and is natural in the involutive module. Its fixed subalgebra is generated by the fixed part $M^+$ together with the pairwise products $xy$ with $x,y \in M^-$; in degree $k$ the fixed elements are the monomials with an even number of anti-fixed factors, so the fixed subalgebra is the even-degree part when the involution has no fixed vectors. With respect to the divided-power pairing $\operatorname{sym}(M)\times\Gamma(M^\vee)\to R$, the induced involution is the transpose of the induced involution generated by the transpose $\sigma_M^{\mathsf{T}}$ of the involution of $M$, and the pairing is invariant; the invariant functionals correspond to the self-adjoint elements of the divided-power algebra. The polynomial-ring case with the exchanged and the negated generators is worked out. No form, norm or distance occurs.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $M$ | $R$-module over a commutative ring $R$ |
| $\sigma_M$ | $R$-linear involution of $M$ |
| $M^+, M^-$ | Fixed and anti-fixed parts of $M$ |
| $\operatorname{sym}(M) = \bigoplus_k\operatorname{sym}^k(M)$ | Symmetric algebra, graded |
| $\sigma$ | Induced involution of $\operatorname{sym}(M)$ |
| $\operatorname{sym}(M)^\sigma$ | Fixed subalgebra |
| $\langle v_1\cdots v_k,\gamma_k(\varphi)\rangle = \prod\varphi(v_i)$ | Divided-power pairing |
| $\Gamma(M^\vee)$ | Divided-power algebra of the dual module |
| $\sigma_M^{\mathsf{T}}$ | Transpose involution of $M^\vee$ |
| $\sigma^\vee$ | Induced involution of $\Gamma(M^\vee)$ |
| $\langle\sigma u,\sigma^\vee\gamma\rangle = \langle u,\gamma\rangle$ | Invariance of the pairing |
Further Reading
- Nicolas Bourbaki, Algebra I (Springer, 1989), for the symmetric algebra, its universal property and the symmetric powers.
- Nicolas Bourbaki, Algebra II (Springer, 2003), for divided powers, the symmetric pairing and duality.
- Werner Greub, Multilinear Algebra (Springer, second edition, 1978), for the symmetric algebra of a module, duality and the transpose.
- Max-Albert Knus, Alexander Merkurjev, Markus Rost and Jean-Pierre Tignol, The Book of Involutions (American Mathematical Society Colloquium Publications 44, 1998), for involutions of a module and their extensions.
- Igor Shafarevich and Alexander Kostrikin, Linear Algebra and Geometry (Gordon and Breach, 1989), for the polynomial-ring involutions and their fixed subalgebras.