Transactions of the American Mathematical Society, Vol. 14, No. 4 (Oct. 1913), 489-500

Introduction

Consider a system of forms f_{1}(x_{1},...,x_{m}), ...,
f_{k}(x_{1},...,x_{m}), whose coefficients a_{1},...,a_{r} (arranged in a definite order) are independent variables. Let b_{1},...,b_{r} be the coefficients (taken in the same order) of the forms derived from f_{1},...,f_{k} by applying a given linear transformation T with integral coefficients. Let F(a_{1},...,a_{r}) be a polynomial with integral coefficients and

D(a_{1},...,a_{r}) = F(b_{1},...,b_{r}) — F(a_{1},...,a_{r})

be the polynomial in a_{1},...,a_{r} with integral coefficients which is obtained from F(b) — F(a) upon replacing b_{1},...,b_{r} by their expressions in terms of a_{1},...,a_{r} and the coefficients of T. in case two or more of the terms in the initial expression for D had the same literal part a_{1}^{e1}···a_{r}^{er}, such terms are assumed to have been combined additively into a single term. Let p be a prime. According to a definition by Hurwitz (stated for a single form in two variables), F(a_{1},...,a_{r}) is an invariant of f_{1},...,f_{k} modulo p with respect to the transformation T if D(a_{1},...,a_{r}) = 0 (mod p), identically in a_{1},...,a_{r}, viz., if the coefficient of each term of D is divisible by p. He gave an interesting example of such an invariant with respect to all of the transformations T with integral coefficients whose determinant is not divisible by p, but did not construct a theory of invariants modulo p.

As a generalization we may employ transformations T with coefficients in the Galois field DF[p^{n}] composed of the p^{n} elements

c_{0} + c_{1}j + c_{2}j^{2} + c_{n-1}j^{n-1} (c_{0},...,c_{n-1} = 0,1,...,p-1),

where j is a root of a congruence of degree n irreducible modulo p. A polynomial F(a_{1},...,a_{r}) with integral coefficients is a *formal* invariant of f_{1},...,f_{k} in the field, under transformation T if D(a_{1},...,a_{r}) is identically zero in the field a_{1},...,a_{r}, i.e., if the coefficient of each term of D, when reduced to the form
c_{0} + c_{1}j + c_{2}j^{2} + c_{n-1}j^{n-1}
by means of the congruence satisfied by j, has each c_{i} a multiple of p. If F_{0}, F_{1},... are such formal invariants (with integral coefficients), then F_{0} + F_{1}j +... is a formal invariant. In this manner, or by direct extension of the previous definition, we have the concept of a formally invariant polynomial with coefficients in the GF[p^{n}].

We pass to the entirely different concept of *modular* invariants, introduced by Dickson. The coefficients a_{1},...,a_{r} of the forms are now undetermined elements of the GF[p^{n}]. A polynomial F(a_{1},...,a_{r}) with coefficients in that field is called a modular invariant of the system of forms under any given group G of linear transformations with coefficients in the field if, for each transformation of G, D(a_{1},...,a_{r}) is zero in the field. To apply this test we may first express D as a polynomial δ(a_{1},...,a_{r}) in which the exponents are all less than p^{n}, and then require that δ shall be identically zero in the field as to a_{1},...,a_{r}. We thus see clearly just how the difference in the definitions of formal and modular invariants affects the actual computations. Dickson has given a very simple and elegant theory of modular invariants. No theory has been developed for formal invariants. However, there exists between the two subjects an interesting and important relation, which I shall develop in what follows. I take this opportunity to express my gratitude to Professor Dickson for his interest and many helpful suggestions, in particular for the present formulation of this introductory section.

Full article available at JSTOR (subscription required).