Past exam of the mathematics course of the University of Cambridge 2002 iii Paper 26 3 Solution Created 2026-10-03 Updated 2026-10-14
Extend by zero to integers not coprime to , and write for its Gauss sum of a Dirichlet character. We use the analytic branch of that is positive when is positive imaginary. With the convention , the Gaussian Fourier transform and its differentiated version areThe first formula follows by completing the square, first on the imaginary axis and then by analytic continuation; the second follows by differentiating the Fourier transform in . All the functions are Schwartz functions on the real axis.
For completeness, the finite Fourier transform of a primitive Dirichlet character satisfiesFor a unit this is substitution by its inverse. If , primitivity supplies a unit with : otherwise the character would factor through modulus . Multiplication by fixes the additive exponential but multiplies the sum by , so the sum vanishes. This proves the identity also on nonunits. Finite Parseval identity now gives , since exactly frequencies have nonzero character values. In particular this Gauss sum of a Dirichlet character is nonzero.
Apply Poisson summation to each residue class: . Multiply by and sum over . The formulas above give the two character theta transformationsandIn the odd case the prefactor originally reads ; using explains the factor . Opposite parity makes the corresponding series identically zero by pairing . This is why the weight-three-halves series is needed for odd Dirichlet characters.
To deduce the products from these transformations, use two specific primitive Dirichlet characters. The even character takes values on modulo twelve. It is primitive: its values do not factor through six or four, hence through no proper divisor of twelve. Its Gauss sum of a Dirichlet character is . PutThe even transformation gives ; also , since each nonzero term has . Thus transforms as a weight-twelve modular form under the generators of the modular group. It is holomorphic and starts with , so it is a normalized cusp form. The proof in Question 1 shows , giving . In particular has no interior zeros, by the valence formula for the modular group.
The Serre derivative argument in Question 2 gives . Hence . The analytic product is nonzero and has the same logarithmic derivative, since . Therefore is constant on the connected complex upper half-plane; its limit at infinity is one. This proves , without assuming either of the product identities being deduced. Pair the terms in , choosing representatives . Since , we getThis is the Euler pentagonal identity, derived by the character theta proof of the eta product.
For the second product use the odd primitive character , , with Gauss sum of a Dirichlet character . SetThe odd transformation gives and the series gives . Thus is another normalized weight-twelve cusp form, and . Because is nonzero, has eighth power one, so it is constant; its leading coefficient fixes it to one. Pairing positive and negative odd integers now givesSince , the Jacobi cubic product identity follows:All pairings and differentiations take place in locally uniformly convergent theta functions or products on .