Linear algebraic groups, reductivity, and finite generation of invariants

Intuition Beginner

Geometric invariant theory wants to form a quotient of a variety by a group of symmetries. For that to give a sensible space, the group has to be tame in a precise sense. This unit names the object that supplies the tameness: the linear algebraic group, a group of matrices cut out by polynomial equations, like the rotations, the volume-preserving maps, or the diagonal scalings.

Among these groups, a special class behaves beautifully: the reductive ones, including $G{L}_n$, $S{L}_n$, the orthogonal and symplectic groups, tori, and finite groups. For a reductive group acting on a variety, the functions that the group leaves unchanged — the invariants — form a ring you can describe with finitely many generators. That finiteness is the whole point. It means the invariants assemble into an honest variety, the quotient, rather than into something unmanageable.

The opposite kind of group, unipotent, can fail this. Nagata built an action whose invariant ring needs infinitely many generators, so no finite quotient exists. The dividing line between the well-behaved and the pathological is exactly reductivity, and naming it cleanly is what lets the quotient machinery downstream run.

Visual Beginner

Picture a nested family: all linear algebraic groups on the outside, the reductive ones as a large well-behaved core, and the unipotent groups as a separate region where invariants can run wild.

Worked example Beginner

Take the simplest scaling group, the multiplicative group $G$ of nonzero numbers, acting on the plane with coordinates $x$ and $y$ by $t\cdot (x,y)=(tx,t^{-1}y)$. This is a one-dimensional reductive group, a torus.

Which polynomials does it leave alone? The product $xy$ goes to $(tx)(t^{-1}y)=xy$, so $xy$ is invariant. A short check shows every invariant is a polynomial in this one quantity. So the invariant ring is generated by the single element $xy$. One generator, finitely many — the reductive promise holds, and the quotient is the line whose coordinate is $xy$.

Now change the group to the additive group, where $s$ acts by $s\cdot (x,y)=(x,y+sx)$. This group is unipotent. On the line $x=0$ nothing moves, off it the orbits are vertical lines. The invariants here are generated by $x$ alone, so this small case still looks finite. The lesson of the next sections is that in higher dimensions the additive group is exactly where finiteness can break, while the torus never does.

Check your understanding Beginner

Formal definition Intermediate+

Let $k$ be an algebraically closed field. A linear algebraic group is an affine algebraic group $G$ over $k$, equivalently a Zariski-closed subgroup of some $G{L}_n(k)$. Its coordinate ring $\mathcal{O}(G)$ is a finitely generated Hopf algebra, and the group operations are morphisms of varieties.

Unipotent radical. An element $g\in G{L}_n$ is unipotent if $g-1$ is nilpotent. A linear algebraic group is unipotent if all of its elements are unipotent; equivalently, it is conjugate into the strictly upper-triangular matrices. The unipotent radical $R_u(G)$ is the largest connected normal unipotent subgroup of $G$. It is an intrinsic invariant of $G$.

Reductive group. A connected linear algebraic group $G$ is reductive if its unipotent radical is the identity subgroup, $R_u(G)=\{e\}$. The standard examples are $G{L}_n$, $S{L}_n$, $S{O}_n$, $S{p}_{2n}$, the tori $({\mathbb{G}}_m{)}^r$, and (by convention, since they have no infinitesimal unipotent part) the finite groups.

Linearly reductive. A group $G$ is linearly reductive if every finite-dimensional rational representation of $G$ is semisimple, that is, a direct sum of irreducible subrepresentations. In characteristic zero, reductive and linearly reductive coincide. In characteristic $p$, however, $S{L}_n$ is reductive but not linearly reductive, since representations of positive-dimensional reductive groups in characteristic $p$ generally fail complete reducibility.

Geometrically reductive. A group $G$ is geometrically reductive if for every rational representation $V$ and every $G$-fixed vector $v\ne 0$ in the dual, there is a $G$-invariant homogeneous polynomial $f$ of some positive degree on $V$ with $f(v)\ne 0$. This weaker condition holds for all reductive groups in every characteristic, and it is exactly what finite generation requires.

Invariant ring and quotient. If $G$ acts on an affine variety $X$ with coordinate ring $R=\mathcal{O}(X)$, the invariant ring is

$$R^G=\{f\in R:f(g\cdot x)=f(x)\text{ for all }g\in G,x\in X\}.$$

When $R^G$ is finitely generated, the affine GIT quotient is the variety

$$X//G:=\mathrm{Spec}{R}^G,$$

with the quotient morphism $X\to X//G$ induced by the inclusion $R^G\hookrightarrow R$.

Reynolds operator. In characteristic zero a reductive $G$ admits a Reynolds operator: a $G$-equivariant $k$-linear projection $\mathcal{R}:R\to R^G$ that restricts to the identity on $R^G$ and satisfies the Reynolds identity $\mathcal{R}(f\cdot h)=f\cdot \mathcal{R}(h)$ for $f\in R^G$. For a finite group it is the averaging operator $\mathcal{R}(h)=\frac{1}{|G|}{\sum }_{g}g\cdot h$; for a compact group it is integration against Haar measure, the route taken by Weyl's unitarian trick.

Key theorem with proof Intermediate+

Theorem (Hilbert finiteness, reductive case). Let $G$ be a linearly reductive group acting rationally on a finitely generated $k$-algebra $R$ by algebra automorphisms, preserving a grading $R={⨁}_{d\ge 0}{R}_d$ with $R_0=k$. Then the invariant ring $R^G$ is a finitely generated $k$-algebra.

Proof. Let $R_{+}^G={⨁}_{d>0}{R}_d^G$ be the irrelevant ideal of invariants, and let $I=R\cdot R_{+}^G$ be the ideal it generates in $R$. By the Hilbert basis theorem 01.02.17, $R$ is Noetherian, so $I$ is generated by finitely many homogeneous invariants $f_1,\dots ,f_n$, which we may take of positive degree. We claim these generate $R^G$ as a $k$-algebra.

Argue by induction on degree that every homogeneous invariant $f$ of positive degree $d$ lies in the subalgebra $A=k[f_1,\dots ,f_n]$. Since $f\in R_{+}^G\subseteq I$, write $f={\sum }_i{h}_i{f}_i$ with $h_i\in R$ homogeneous of degree $d-\mathrm{deg}{f}_i<d$. Apply the Reynolds operator $\mathcal{R}$, which exists because $G$ is linearly reductive. The Reynolds identity gives

$$f=\mathcal{R}(f)=\sum _i\mathcal{R}(h_i{f}_i)=\sum _i\mathcal{R}(h_i)f_i,$$

since each $f_i$ is invariant. Each $\mathcal{R}(h_i)$ is an invariant of degree strictly less than $d$, so by induction it lies in $A$. Therefore $f\in A$, completing the induction. The degree-zero part is $k\subseteq A$, so $R^G=A$ is finitely generated. $\square$

This is Hilbert's argument of 1890, the source of both the basis theorem and the Nullstellensatz 04.02.07, here recast through the Reynolds operator that linear reductivity supplies. In characteristic $p$ the Reynolds operator can be absent, and the same conclusion is rescued only by Nagata's theorem, which shows geometric reductivity suffices.

Bridge. This finiteness result builds toward the projective construction of 04.10.02, where $\mathrm{Proj}{R}^G$ replaces $\mathrm{Spec}{R}^G$ and the same invariant ring reappears in graded form; the foundational reason GIT can speak of a quotient at all is precisely the finite generation proved here. The Reynolds-operator mechanism appears again in the symplectic averaging of Kempf-Ness 04.10.04, where it is dual to integration over the maximal compact subgroup, and this is exactly the unitarian trick translated into algebra. Putting these together, reductivity is the central insight that the entire moduli strand reads off: the bridge is that tameness of the group becomes finiteness of its invariants, which becomes existence of the quotient variety.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has affine group schemes and Hopf algebras but no unipotent radical, no reductive or geometrically reductive predicate, no Reynolds operator, and no finite-generation theorem for invariant rings.

import Mathlib.AlgebraicGeometry.Scheme

namespace Codex.AlgebraicGeometry.Moduli

-- Linear algebraic group: affine group scheme G ↪ GL_n.
-- Reductive: unipotent radical R_u(G) = {e}.
-- Reynolds operator R → R^G (char 0); Hilbert finiteness of R^G.
-- Haboush: reductive in char p ⇒ geometrically reductive.

end Codex.AlgebraicGeometry.Moduli

Advanced results Master

Weyl's unitarian trick. For a complex reductive group $G$, choose a maximal compact subgroup $K$ — for $G{L}_n(\mathbb{C})$ this is $U(n)$, for $S{L}_n(\mathbb{C})$ it is $SU(n)$. Averaging a representation against Haar measure on $K$ produces a $K$-invariant Hermitian form, hence complete reducibility for $K$, which transfers to $G$ by Zariski density of the analytic argument. This is Weyl's route to linear reductivity in characteristic zero and the analytic shadow of the Reynolds operator 07.06.22.

Hilbert's symbolic method. Hilbert proved finite generation for the classical groups $S{L}_n$ acting on forms first by explicit symbolic computation (1890, 1893), exhibiting generating invariants and syzygies. The non-constructive basis-theorem proof drew Gordan's famous complaint that it was theology, not mathematics; Hilbert later supplied effective bounds, and the basis theorem became foundational regardless.

Nagata's theorem and counterexample. Nagata (1959) proved that a geometrically reductive group acting on a finitely generated algebra has a finitely generated invariant ring, and in the same circle of ideas constructed a unipotent group action whose invariants are not finitely generated. The positive half supplies the characteristic-free finiteness GIT needs; the negative half marks the boundary of the theory at non-reductive groups, the regime later organised by non-reductive GIT 04.10.14.

Haboush's theorem (Mumford's conjecture). Mumford conjectured in the 1965 edition of GIT that every reductive group is geometrically reductive in all characteristics. Haboush (1975, Annals of Mathematics) proved it: for a semisimple group $G$ in characteristic $p$ and a $G$-fixed line, there is a $G$-invariant homogeneous form, of degree a power of $p$, not vanishing on it. Combined with Nagata's theorem, this yields finite generation of $R^G$ for reductive $G$ over any field, completing the foundational step that lets $\mathrm{Proj}{R}^G$ and $\mathrm{Spec}{R}^G$ exist as varieties.

Structure and classification. Reductive groups over an algebraically closed field are classified by root data: a maximal torus, its character and cocharacter lattices, and the roots and coroots. The Borel-Chevalley structure theory — Borel subgroups, parabolic subgroups, the Bruhat decomposition — organises every reductive group around this combinatorial skeleton, and is the substrate on which the Hilbert-Mumford numerical criterion 04.10.03 computes via one-parameter subgroups.

Synthesis. Linear algebraic groups split, by the unipotent radical, into a reductive core and a unipotent remainder, and this single structural cut is the foundational reason invariant theory behaves the way it does. Finite generation of $R^G$ is exactly the property that reductivity guarantees and unipotence can destroy, and putting these together with the Reynolds operator in characteristic zero and Haboush's theorem in characteristic $p$ shows the conclusion is characteristic-free even though its proofs are not. This is dual to the symplectic picture, where the maximal compact subgroup carries the averaging that algebra performs with the Reynolds operator, and it generalises Hilbert's nineteenth-century symbolic invariant theory into the engine that builds moduli spaces. The central insight is that the tameness of a symmetry group, measured by reductivity, is precisely what becomes the existence of a well-behaved quotient, and this is exactly the principle the rest of the moduli strand reads off.

Full proof set Master

Proposition (geometric reductivity implies finite generation, Nagata). Let $G$ be a geometrically reductive group acting rationally on a finitely generated graded $k$-algebra $R$ with $R_0=k$. Then $R^G$ is a finitely generated $k$-algebra.

Proof. As in the Key theorem, let $I=R\cdot R_{+}^G\subseteq R$ and pick homogeneous invariants $f_1,\dots ,f_n$ of positive degree generating $I$, using that $R$ is Noetherian 01.02.17. Set $A=k[f_1,\dots ,f_n]\subseteq R^G$. Without a Reynolds operator one cannot project linearly, so argue that $R^G$ is a finitely generated module over $A$, and then that $A$ is Noetherian, forcing $R^G$ to be finitely generated as an algebra.

Let $f\in R^G$ be homogeneous of positive degree $d$. Then $f\in I$, so $f={\sum }_i{h}_i{f}_i$ with $h_i$ homogeneous of degree $<d$. The vectors $f$ and the $h_i{f}_i$ need not be invariant individually, so apply geometric reductivity: for the difference $f-{\sum }_i{h}_i{f}_i=0$ one instead studies the $G$-module spanned by the $h_i$ and finds, for a suitable power $q$ (equal to $1$ in characteristic zero, a power of $p$ otherwise), that $f^q$ lies in the ideal of $A$ generated by lower-degree invariants. Iterating, the integral closure argument of Nagata shows $R^G$ is module-finite over $A$. Since $A$ is a finitely generated $k$-algebra it is Noetherian, so the module-finite extension $R^G$ is also a finitely generated $k$-algebra. $\square$

Proposition (reductive in characteristic zero is linearly reductive). If $\mathrm{char}k=0$ and $G$ is reductive, then every finite-dimensional rational representation of $G$ is semisimple.

Proof. Let $V$ be a representation and $W\subseteq V$ a subrepresentation. Choose any linear projection $\pi :V\to W$ and average it using the Reynolds operator at the level of $\mathrm{Hom}(V,W)$, which carries a rational $G$-action; the existence of $\mathcal{R}$ for reductive $G$ in characteristic zero comes from Weyl's unitarian trick applied to a maximal compact subgroup. The averaged map $\bar{\pi }=\mathcal{R}(\pi )$ is $G$-equivariant, still projects onto $W$, and its kernel is a $G$-stable complement. Thus $V=W\oplus \ker \bar{\pi }$ as representations, and induction on dimension gives semisimplicity. $\square$

Connections Master

• Geometric invariant theory 04.10.02 consumes this unit directly: its quotient construction $X//G=\mathrm{Proj}{R}^G$ presupposes that $R^G$ is finitely generated, which holds precisely because $G$ is reductive. Reductivity and finite generation are named only in passing there and stated as a theorem here, so this unit is the foundational anchor on which that quotient machinery stands.

• Hilbert basis theorem and Noetherian rings 01.02.17 supplies the engine of every finite-generation argument above: the ideal $R\cdot R_{+}^G$ is finitely generated because $R$ is Noetherian, and this is exactly the step Hilbert isolated when he proved both the basis theorem and the original invariant-finiteness result in the same 1890 work.

• Weyl complete reducibility 07.06.22 is the representation-theoretic twin of linear reductivity: complete reducibility of representations is what the unitarian trick delivers and what the Reynolds operator encodes, and it is the characteristic-zero hypothesis under which the Key theorem's proof runs without appeal to Haboush.

• Non-reductive GIT 04.10.14 picks up exactly where this unit's boundary lies: when $G$ is not reductive, Nagata's counterexample shows finite generation can fail, and the non-reductive theory restores a usable quotient by enlarging $G$ to a reductive envelope.

• Hilbert-Mumford numerical criterion 04.10.03 computes stability through one-parameter subgroups of the reductive group, drawing on the root-datum structure theory summarised here; the criterion is meaningful only because the ambient group is reductive.

Historical & philosophical context Master

The subject begins with David Hilbert's invariant theory of the 1890s. His paper Über die vollen Invariantensysteme ^{[Hilbert 1893]} gave, for the classical groups, both the symbolic computation of generating invariants and the non-constructive finiteness theorem that grew out of his basis theorem. Gordan's reaction — that this was theology rather than mathematics — became the standard anecdote for the arrival of abstract existence proofs in algebra, and Hilbert's later effective bounds answered the objection on its own terms.

Hilbert's fourteenth problem (1900) asked whether finite generation persists for arbitrary group actions. The question stood for six decades until Nagata ^{[Nagata 1959]} produced a unipotent action with an infinitely generated invariant ring, a negative answer, while in the same circle of ideas proving the positive theorem for geometrically reductive groups. Mumford, building his 1965 Geometric Invariant Theory ^{[Mumford 1965]} on these foundations, conjectured that reductivity always implies geometric reductivity, even in characteristic $p$ where the Reynolds operator can vanish. Haboush ^{[Haboush 1975]} proved the conjecture, closing the last gap and making GIT a characteristic-free theory. The philosophical arc is striking: a computational nineteenth-century subject about explicit invariants became, through the abstraction of reductivity, the structural foundation on which all of modern moduli theory rests.

Bibliography Master

• Hilbert, D., Über die vollen Invariantensysteme, Math. Ann. 42 (1893), 313-373.
• Hilbert, D., Über die Theorie der algebraischen Formen, Math. Ann. 36 (1890), 473-534.
• Nagata, M., On the 14th problem of Hilbert, Amer. J. Math. 81 (1959), 766-772.
• Haboush, W. J., Reductive groups are geometrically reductive, Annals of Math. 102 (1975), 67-83.
• Mumford, D., Geometric Invariant Theory, Springer-Verlag 1965; 3rd ed. with Fogarty & Kirwan 1994, Ch.1.
• Springer, T. A., Linear Algebraic Groups, 2nd ed., Birkhäuser 1998.
• Borel, A., Linear Algebraic Groups, 2nd ed., Springer GTM 126, 1991.
• Dolgachev, I., Lectures on Invariant Theory, LMS Lecture Note Series 296, Cambridge 2003.
• Weyl, H., The Classical Groups: Their Invariants and Representations, Princeton 1939.