Invariant Theory Notes by Bernd Sturmfels for the lecture on June 19, 2018, in the IMPRS Ringvorlesung Introduction to Nonlinear Algebra We fix the polynomial ring K[x] = K[x1; : : : ; xn] over a field K of characteristic zero. The group GL(n; K) of invertible n × n matrices acts on Kn. This induces an action by G on n the ring of polynomial functions on K . Namely, if σ = (σij) is a matrix in GL(n; K) and f is a polynomial in K[x] then σf is the polynomial that is obtained from f by replacing the Pn variable xi by the linear form j=1 σijxj for i = 1; : : : ; n. Let G be a subgroup of GL(n; K). A polynomial f 2 K[x] is an invariant of the group G if σf = f for all σ 2 G. We write K[x]G for the set of all such invariants. This set is a subring because the sum of two invariants is again an invariant, and same for the product. In this lecture we discuss two scenarios. First we consider finite groups G, and later we consider representations of nice infinite groups like SL(d; K) and SO(d; K). Such groups are called reductive. A celebrated theorem of Hilbert shows that the invariant ring is finitely generated in this case. After two initial examples, we begin by proving this for finite groups G. Example 1. Let G be the group of n × n permutation matrices. The invariant ring K[x]G consists of all polynomials f that are invariant under permuting the coordinates, i.e. f(xπ1 ; xπ2 ; : : : ; xπn ) = f(x1; x2; : : : ; xn) for all permutations π of f1; 2; : : : ; ng. The invariant ring K[x]G is generated by the n elementary symmetric polynomials E1;:::;En. These are the coefficients of the following auxiliary polynomial in one variable z: n n X n−i (z + x1)(z + x2) ··· (z + xn) = z + Ei(x)z : (1) i=1 G We also set E0 = 1. Alternatively, K[x] can also be generated by the power sums j j j Pj(x) = x1 + x2 + ··· + xn for j = 1; 2; : : : ; n: The formulas which translate between the Ei and the Pj are known as Newton's Identities: kE = Pk (−1)i−1E P k i=1 k−i i (2) k−1 Pk−1 k−1−i and Pk = (−1) kEk + i=1 (−1) Ek−iPi for 1 ≤ k ≤ n: 1 Invariants are polynomial functions that are constant along G-orbits on Kn. They offer an algebraic view on the space of orbits. Namely, we think of the spectrum of K[x]G as a quotient space Kn==G, whose points are these orbits. This interpretation is only informal, as the details are very subtle. Making it all precise is the aim of Geometric Invariant Theory. Example 2. For n = 2, consider the following representation of the cyclic group of order 4: 1 0 −1 0 0 1 0 −1 G = ; ; ; : (3) 0 1 0 −1 −1 0 1 0 These are the rotational symmetries of the square. Its invariant ring is generated by 2 2 2 2 3 3 I1 = x1 + x2 ;I2 = x1x2 ;I3 = x1x2 − x1x2: These three invariants are algebraically dependent. Using their relation we can write G 2 2 2 K[x1; x2] = K[I1;I2;I3] ' K[y1; y2; y3]= y1y2 − 4y2 − y3 : (4) 3 2 2 2 The spectrum of the ring (4) is the cubic surface in K defined by the equation y1y2 = 4y2+y3. The points on this surface are in one-to-one correspondence with the G-orbits on K2. In what follows, we assume that G is a finite subgroup of GL(n; K). One can create invariants by averaging polynomials. The Reynolds operator, denoted by a star, is defined as G 1 X ∗ : K[x] ! K[x] ; f 7! f ∗ := σf: (5) jGj σ2G Each of the following properties of the Reynolds operator is easily verified: Lemma 3. The Reynolds operators ∗ has the following three properties: (a) The map ∗ is K-linear, i.e. (λf + νg)∗ = λf ∗ + νg∗ for all f; g 2 K[x] and λ, ν 2 K. (b) The map ∗ restricts to the identity on K[x]G, i.e. I∗ = I for all invariant polynomials I. (c) The map ∗ is a K[x]G-module homomorphism, i.e. (fI)∗ = f ∗I for all f 2 K[x] and I 2 K[x]G. The following result from 1890 is seen as the beginning of modern Commutative Algebra. Theorem 4 (Hilbert's Finiteness Theorem). The invariant ring K[x]G of any finite matrix group G ⊂ GL(n; K) is finitely generated as a K-algebra. We present the proof under the hypothesis that K has characteristic zero. However, the result holds for every field K. For a proof see [3]. This is known as modular invariant theory. 2 G Proof. Let IG = hK[x]+i be the ideal in K[x] that is generated by all homogeneous in- variants of positive degree. By Lemma 3 (a), every invariant is a K-linear combination of symmetrized monomials (xa)∗. These homogeneous invariants are the images of monomials a ∗ n under the Reynolds operator. Thus IG is generated by the set f (x ) : a 2 N nf0gg. By n Hilbert's Basis Theorem, the ideal IG is finitely generated, so that a finite subset of a in N suffices. In conclusion, there exist invariants I1;I2;:::;Im such that IG = hI1;I2;:::;Imi. We claim that these m invariants generate the invariant ring K[x]G as a K-algebra. Suppose the contrary, and let I be a homogeneous element of minimal degree in G Pm K[x] nK[I1;I2;:::;Im]. Since I 2 IG, we have I = j=1 fjIj for some homogeneous polynomials fj 2 K[x] whose degrees are all strictly less than deg(I). Pm Applying the Reynolds operator on both sides of the equation I = j=1 fjIj, we obtain m m ∗ X ∗ X ∗ I = I = fjIj = fj Ij: j=1 j=1 ∗ Here we are using the properties (b) and (c) in Lemma 3. The new coefficients fj are homogeneous invariants whose degrees are less than deg(I). From the minimality assumption ∗ on the degree of I, we get fj 2 K[I1;:::;Im] for j = 1; : : : ; m. This implies I 2 K[I1;:::;Im], which is a contradiction to our assumption. This completes the proof of Theorem 4. Hilbert's Finiteness Theorem also holds for an infinite group G ⊂ GL(n; K) that has a Reynolds operator ∗ satisfying the properties (a), (b) and (c) in Lemma 3. Such G are known as reductive groups. We shall return to this important point later in the lecture. Corollary 5. Consider any reductive group G of n×n-matrices. If fg1; g2; : : : ; gmg is any col- ∗ ∗ ∗ lection of homogeneous polynomials that generates the ideal IG then its image fg1; g2; : : : ; gmg under the Reynolds operator generates the invariant ring K[x]G as a K-algebra. Proof. Let M = hx1; : : : ; xni be the homogeneous maximal ideal in K[x], and consider the finite-dimensional vector space IG=MIG. It has a basis of invariants since IG is generated by invariants. This means that the Reynolds operator acts as the identity on IG=MIG. The images of g1; g2; : : : ; gm also span IG=MIG as a vector space, and hence so do the invariants ∗ ∗ ∗ ∗ ∗ ∗ g1; g2; : : : ; gm. By Nakayama's Lemma, we find that g1; g2; : : : ; gm generate the ideal IG. As ∗ ∗ ∗ G in the proof of Theorem 4, we conclude that g1; g2; : : : ; gm generate the K-algebra K[x] . Theorem 6 (Noether's Degree Bound). If G is finite and char(K) = 0 then the invariant ring K[x]G is generated by homogeneous invariants of degree at most jGj. Proof. Let u = (u1; : : : ; un) be new variables. For any d 2 N, we consider the expression d ∗ Sd(u; x) = (u1x1 + ··· + unxn) 1 P d = jGj σ2G u1(σx1) + ··· + un(σxn) : This is a polynomial in u whose coefficients are polynomials in x. Up to a multiplicative constant, they are the invariants (xa)∗ where jaj = d. All polynomials in u are fixed under ∗. 3 Consider the jGj expressions u1(σx1) + ··· + un(σxn) , one for each group element σ 2 G. The polynomial Sd(u; x) is the dth power sum of these expressions. The power sums for d > jGj are polynomials in the first jGj power sums. Such a representation is derived from Newton's Identities (2). It implies that all u-coefficients of Sd(u; x) for d > jGj are a ∗ polynomial functions in the u-coefficients of Sd(u; x) for d ≤ jGj. Hence all invariants (x ) with jaj > jGj are polynomial functions (over K) in the invariants (xb)∗ with jbj ≤ jGj. This proves the claim. We note that Example 2 attains Noether's degree bound. The cyclic group in that example has order 4, and the invariant ring requires a generator of degree 4. Our next theorem is a useful tool for constructing the invariant ring. It says that we can count invariants by averaging the reciprocal characteristic polynomials of the group elements. Theorem 7 (Molien series). The Hilbert series of the invariant ring K[x]G equals 1 X 1 X 1 dim K[x]Gzd = : (6) K d jGj det(id − zσ) d=0 σ2G The coefficient of zd in this series is the number of linearly independent invariants of degree d. Proof. See [4, Theorem 2.2.1].