A finite Fourier transform is built from additive phases. A multiplicative character instead records products in a finite field. Their interaction is unexpectedly rigid: a nontrivial multiplicative character transforms into its conjugate, multiplied by one Gauss sum. That fact turns a finite-field counting problem into a short Fourier calculation.
Let be an odd prime and . Write , where is any integer representative of a residue. For , use the unnormalized additive transform
Definition 1.12.1
(Multiplicative character and Gauss sum)
A multiplicative character is a homomorphism , extended by . It is nontrivial if it is not identically one on . Its Gauss sum is
Because the multiplicative group is finite, every nonzero character value is a root of unity and has modulus one; in particular for .
Theorem 1.12.2
(A character transforms into its conjugate)
For a nontrivial multiplicative character and every ,
At both sides use . The formula gives the magnitude of the Gauss sum, not generally its complex phase.
Proof
(Rescale the additive frequency)
Nontriviality supplies with . Multiplication by permutes the nonzero residues, so
and the sum is zero. Thus (2) holds at . For , substitute . Since ,
Finite Fourier orthogonality gives Parseval in this normalization:
The zero-frequency term is zero, and each of the other terms has magnitude . Hence .
Proof X-ray
Addition and multiplication play different roles
The exponential is an additive character in , while is multiplicative on nonzero residues. The key move, rescaling by , converts one into the other: is a bijection of fixing only because is invertible, and pulling out of the sum uses multiplicativity together with .
Nontriviality is used exactly once, at , where the substitution is unavailable and the transform reduces to the plain character sum. The magnitude then closes by counting: Parseval fixes the total energy , the formula at removes one term, and the remaining terms all have the same modulus because . If the term did not vanish, the energy would no longer split evenly and the value would not follow.
Worked example 1.12.3
(A quartic character on the five-element field)
In the element generates the group: are its successive powers. Define ; multiplicativity forces , and . With ,
using , and . Since and ,
as (2) predicts. The same theorem gives , since .
Common mistake
(Omitting the conjugate in the transform)
It is tempting to write . For the quartic character above this would give , while the substitution gives . The two agree only when is real, as for the quadratic character below.
Quadratic characters and a phase constraint
Definition 1.12.4
(Quadratic character)
Let , and for set if is a square in and otherwise.
This is a nontrivial multiplicative character: using a generator of , squares are precisely the even powers of , so . Moreover
because . The prime-field generator was constructed in the finite-field Fourier-transform chapter.
Theorem 1.12.5
(A quadratic Gauss sum has a determined square)
With ,
Thus is one of when , and one of when . This argument does not select the sign.
Proof
(Conjugate and change a to minus a)
Since is real,
The first theorem gives , and . Hence , proving (4). Formula (3) gives the two cases.
Worked example 1.12.6
(The three-element field)
For , and , so direct evaluation gives
Here , and indeed . The direct calculation selects a sign in this one example; (4) alone does not select it for all primes.
Counterexample mode
The trivial character has a different transform
Let for and . Then , while . Thus (2) fails at zero and the Gauss-sum magnitude is , not . Nontriviality is necessary.
Counting a quadratic equation
Theorem 1.12.7
(Count points on a finite-field circle)
For , let be the number of ordered pairs satisfying . Then
Proof
(Enforce the equation with additive orthogonality)
For any , is one if and zero otherwise. Therefore, writing ,
Clearly . For every residue the equation has solutions, including . If , summing first over gives
The constant part is for by additive orthogonality, and (2) applies to the real quadratic character. Thus for every nonzero . Substituting in (6) leaves . The last sum is for and for , proving (5).
For instance, when , (5) gives and for either nonzero . When it gives and for each nonzero . The number being counted is an ordinary integer; the equation itself is in .
Worked example 1.12.8
(Checking the count directly for )
Here , so (5) predicts and for . The squares in are (from ), (from ), (from ) and (from ).
For , the pairs of squares with sum are , and , since ; the remaining first entry would need the second entry , which is not a square. Counting roots gives ordered pairs. For the edge case , a solution with would give , which is not a square, so only remains and . Both values agree with (5).
Common mistake
(Treating as one)
The convention is what makes count the roots of for every , including , which has the single root . Setting would count two roots there, and it would also make instead of , so the case of (2) would fail.
Practice
Quick check
(Why does a nontrivial multiplicative character sum to zero over ?)
Solution · Answer
Choose with . Multiplication by permutes the field, so the sum equals times itself and must be zero.
Quick check
(For , determine , and . Does (4) determine the sign of ?)
Solution · Answer
, and . Equation (4) leaves the two possibilities .
Quick check
(For , use (5) to find and when .)
Solution · Answer
, so and for .
Quick check
(Why does have solutions for every ?)
Solution · Answer
For a nonzero square there are exactly two roots because is odd; a nonsquare has none, and zero has the single root zero. These numbers are respectively , and .
Quick check
(What fails in the first proof if is the trivial character on ?)
Solution · Answer
There is no with , so the cancellation argument cannot start. In fact the sum over the field is , not zero.