On the sums of three and four squares (Q1214457)

Fermat stated that every natural integer is the sum of (at most) three triangular numbers, four squares, five pentagonal numbers, etc. The proof of the statement concerning the four squares was first given by Lagrange, then by Euler and again by Hurwitz; next, Jacobi and Dirichlet gave proofs that included also the determination of the number of such representations. The method of proof is based on infinite descent (due to Fermat) so that the author sees no reason to doubt Fermat's assertion [Fermat, Œuvres II, p. 403] that he was in possession of a proof. The situation is somewhat different in the case of sums of three squares. This problem was first treated successfully by Gauss, who used the theory of quadratic forms. Now, Fermat's statement concerning the sum of three triangular numbers is essentially equivalent to the fact that every integer \(m\equiv 3\pmod 8\) is the sum of three squares, and it seems clear that Fermat's methods did not permit him to treat the general case of decomposition into three squares. In fact, none of the presently known proofs of Fermat's statements concerning the general polygonal numbers are likely to have been accessible to Fermat. There exists, however, a little known paper by \textit{L. Kronecker} [Abh. d. Kgl. Preussischen Akad. d. Wiss. 1883, pp. 1--60; Werke II, pp. 425--495], written in the spirit of the previously mentioned elementary proofs of Jacobi and of Dirichlet, and from which follows not only the proof of Fermat's statement concerning the decomposition into triangular numbers, but also the number of representations by sums of three squares. The balance of the paper is a streamlined presentation of part of Kronecker's paper. Let \(N_i(m)\) denote the number of solutions \((x_1,\ldots,x_i)\) of \(m= \sum_{h=1}^i x_h^2\), \(x_h>0\), \(x_h\equiv 1\pmod 2\). Knowing \(N_i(m)\) for \(m\equiv i\pmod 8\) \((N_i(m) =0\) if \(m\not\equiv i\pmod 8)\), it is rather easy to determine the number of representations of an arbitrary integer \(m\) by \(i\) squares, when \(i=2\) or \(i=4\), but not so easy when \(i=3\). The determination of \(N_i(m)\) is``elementary'' (in spite of its flavor of reciprocity laws) in the sense that no appeal is made to either quadratic forms, or to theta functions and the appeal made to Dedekind's zeta function of the Gaussian field is only for convenience and can be avoided.