Gauss sums refresher¶

Gauss sums are finite Fourier transforms of Dirichlet characters. In Phase 2 they serve two purposes:

A correctness/structure check for characters: for primitive characters, the Gauss sum magnitude obeys a clean law.
A bridge to analytic number theory: Gauss sums appear in functional equations for Dirichlet \(L\)-functions and in explicit evaluations of some special values.

This page gives the definitions and the specific facts that the experiments use, with enough context to interpret the plots.

Additive characters and the Fourier viewpoint¶

Fix a modulus \(q\ge 1\). The basic additive character modulo \(q\) is

\[ e_q(n) := \exp\!\left(\frac{2\pi i n}{q}\right). \]

It is periodic mod \(q\) and satisfies \(e_q(m+n)=e_q(m)e_q(n)\). You can view \(e_q(an)\) as a Fourier mode on the finite group \(\mathbb{Z}/q\mathbb{Z}\).

A Dirichlet character \(\chi\) modulo \(q\) is (after restricting to units) a multiplicative character on \((\mathbb{Z}/q\mathbb{Z})^\times\). Gauss sums mix the multiplicative character \(\chi\) with the additive phase \(e_q(\cdot)\), i.e. they are “multiplicative objects seen through additive Fourier glasses”.

Core definitions¶

Let \(\chi\) be a Dirichlet character modulo \(q\). The (basic) Gauss sum is

\[ \tau(\chi) := \sum_{a=0}^{q-1} \chi(a)\, e_q(a). \]

A more general “twisted” Gauss sum is

\[ \tau(\chi, b) := \sum_{a=0}^{q-1} \chi(a)\, e_q(ba)\qquad (b\in\mathbb{Z}). \]

Immediate properties¶

If \(b\equiv 0\pmod q\), then \(\tau(\chi,b)=\sum_a\chi(a)\).
If \(\gcd(b,q)=1\), then a change of variables implies the relation

\[ \tau(\chi,b) = \overline{\chi}(b)\,\tau(\chi). \]

This identity is the main reason experiments can focus on \(\tau(\chi)\) without losing information: all invertible twists are just rotations/scalings by a root of unity.

The magnitude law \(|\tau(\chi)|=\sqrt{q}\) for primitive characters¶

The central numerical pattern is:

If \(\chi\) is primitive modulo \(q\), then \(|\tau(\chi)| = \sqrt{q}\).

A good way to remember why this is plausible is to compute the squared magnitude and watch orthogonality collapse the double sum. Start with

\[ |\tau(\chi)|^2 = \tau(\chi)\,\overline{\tau(\chi)} = \sum_{a=0}^{q-1}\sum_{b=0}^{q-1} \chi(a)\overline{\chi}(b)\, e_q(a-b). \]

For primitive characters, the sum reorganizes into complete additive character sums with perfect cancellation, giving exactly \(q\). For imprimitive characters (induced from a smaller conductor), the same cancellation can fail; magnitudes may be smaller (and in some cases can vanish). This is why Phase 2 experiments should either restrict to primitive characters for “\(\sqrt{q}\) laws” or explicitly label imprimitive cases.

Quadratic Gauss sums (cleanest special case)¶

For an odd prime \(p\) and the quadratic character (Legendre symbol) \(\chi(n)=\left(\frac{n}{p}\right)\), the Gauss sum has a famous closed form:

\[\begin{split} \tau(\chi) = \varepsilon_p\,\sqrt{p},\qquad \varepsilon_p = \begin{cases} 1, & p\equiv 1\pmod 4,\\ i, & p\equiv 3\pmod 4. \end{cases} \end{split}\]

So not only is the magnitude \(\sqrt{p}\), but the argument depends on \(p\bmod 4\). This is a strong “sanity target” for Phase 2: if your implementation of \(\chi\) is correct and you use a consistent residue convention, the plotted points should land on the expected circle with the expected symmetry.

Why Gauss sums matter for Dirichlet \(L\)-functions (high level)¶

For a primitive character \(\chi\), the completed \(L\)-function satisfies a functional equation of the form

\[ \Lambda(s,\chi) = W(\chi)\,\Lambda(1-s,\overline{\chi}), \]

where the root number \(W(\chi)\) has absolute value \(1\) and is built from \(\tau(\chi)\) (up to normalization by \(\sqrt{q}\) and a parity factor).

You do not need the full functional equation machinery for Phase 2 plots, but it explains why Gauss sums appear naturally whenever you connect characters, \(L\)-functions, and oscillations in arithmetic progression counts.

Practical numerical notes for experiments¶

Residue conventions: be consistent about whether you sum \(a=0,\dots,q-1\) or \(a=1,\dots,q\); both are valid but should not be mixed.
Non-units: if you extend \(\chi\) by \(0\) on \(\gcd(a,q)>1\), those terms contribute \(0\), but you must ensure your implementation really does this.
Floating-point cancellation: Gauss sums are dominated by cancellation. Use complex128 and avoid rounding before taking magnitudes.

References¶

See References.

[Berndt et al., 1998, Davenport, 2000, Iwaniec and Kowalski, 2004]

Experiments in this repository¶

E067 — Gauss sums: magnitude vs. \(\sqrt{q}\) (Phase 2 core).
E109 — Gauss sums \(\tau(\chi)\): magnitude law and geometry for characters modulo \(q\).