Motivation
A field extension lets us multiply and add elements that did not exist in the base field. Trace and norm bring such an element back to the base: they collect its conjugates by addition and multiplication. These two operations turn questions about an extension into concrete equations over a smaller field, where computations may be easier to organize. In finite fields they also reveal an asymmetry: the norm is multiplicative and the trace is additive, but neither operation generally preserves both operations.
Throughout, let be a prime power, and with . We use the earlier results that is a degree- Galois extension, its Galois group is generated by Frobenius , and is cyclic of order . The trace and norm are
For the norm is zero. When , both operations are the identity. The formulas use all automorphisms, even when some conjugate values coincide: a base-field element is repeated times in the trace and norm.
The values really lie in the base field
Theorem 4.2.1
(Additivity and multiplicativity)
The trace and norm take values in . Trace is -linear, while norm satisfies . For , and .
Proof
(Proof)
Frobenius permutes the summands in the trace cyclically, because . It also cyclically permutes the factors of the norm. Both results are therefore fixed by , and its fixed field is . Each power is additive in characteristic and fixes every , so the trace is additive and -homogeneous. Multiplicativity follows by regrouping the factors in the finite product. Finally for each , giving identical summands or factors. The integer in is interpreted in the characteristic of .
Worked example 4.2.2
(A quadratic extension of )
Let with . Its four elements are , and Frobenius sends to . Thus and . On the other hand . This last calculation does not mean that the trace map is zero: already has trace one.
Common mistake
(Trace of one does not decide the trace map)
If the characteristic divides , then in . It is still possible, and in fact always true for a finite-field extension, that another element has nonzero trace. A linear map can vanish on a particular nonzero vector without vanishing everywhere.
Surjectivity and the size of each fiber
Theorem 4.2.3
(The finite-field trace map is onto)
The trace is a nonzero -linear map and hence is surjective. Every element of has exactly preimages.
Proof
(Proof)
Consider the polynomial
It is nonzero and has degree : for it is , and otherwise its highest coefficient is one. If the trace vanished at every element of , this polynomial would have roots, more than its degree. This contradicts the root bound for a nonzero polynomial over a field. Thus some element has nonzero trace. Its image spans the one-dimensional -space , so linearity gives surjectivity. Rank–nullity gives a kernel of dimension , containing elements. Each fiber is a coset of the kernel, hence has the same size.
Worked example 4.2.4
(Trace fibers in four elements)
In the preceding quadratic example, and . Each fiber contains two elements, as predicted by . The kernel is an additive subspace, although it is not an ideal of the field .
The norm has a different but equally explicit structure. For nonzero ,
The exponent is an integer even when . This formula makes the map on the cyclic multiplicative groups visible.
Theorem 4.2.5
(The norm map is onto)
The norm maps onto . Each nonzero value has exactly preimages, and zero has the single preimage zero.
Proof
(Proof)
Choose a generator of , whose order is , and put . The order of is . It is the norm of and lies in , so it generates the group of order . Thus the norm homomorphism is surjective. Among the possible exponents of , exactly lie in each congruence class modulo ; these are precisely the fibers. A product of field elements is zero exactly when at least one factor is zero, so only zero has norm zero.
Counterexample mode
Norm surjectivity needs a finite field
Claim: in every finite Galois extension the norm maps onto . For the Galois group is , so . Thus is not a norm; the image is exactly the positive reals, since is the norm of . The proof above needed a generator of , and is not cyclic. Repair: the claim holds for extensions of finite fields.
Worked example 4.2.6
(Norms from to )
Let where . The Frobenius conjugate of is . For ,
For instance , while . The nonzero norm fibers each have elements. In particular has norm even though its two coordinate squares both equal one.
The trace pairing
Define . It is a symmetric -bilinear form: bilinearity follows from the linearity of trace and distributivity in . Nondegeneracy means that for each nonzero there is a with ; it does not mean for every nonzero .
Theorem 4.2.7
(Nondegenerate trace pairing)
For every nonzero , the map is a nonzero -linear functional. Consequently the map
is an isomorphism of -vector spaces.
Proof
(Proof)
Choose with , which exists by the trace-surjectivity theorem. Since , the element satisfies . Thus the displayed map has zero kernel. Its source and target both have dimension over , since the dual of an -dimensional vector space has dimension . An injective linear map between finite-dimensional spaces of equal dimension is an isomorphism.
Worked example 4.2.8
(A nonzero vector with zero self-pairing)
For , . Nevertheless . Thus the form is nondegenerate even though a nonzero vector pairs to zero with itself. This example is especially useful in characteristic two, where familiar positive-definite intuition from real inner products does not apply.
If is any -basis of , the isomorphism yields a unique dual basis with . Indeed, each coordinate functional is represented by a unique field element under the isomorphism. The duality gives the reconstruction formula
This is a basis-dependent coordinate formula, not a claim that every chosen basis is self-dual.
Worked example 4.2.9
(A dual basis in )
Take the basis over above. The dual basis is : the four traces are respectively . For , the formula gives coefficients and , reconstructing .
A trace test for a Frobenius equation
The map is -linear. Its image describes which equations are solvable in . The trace provides a complete obstruction, not merely a necessary condition. This is a finite-field version of an additive cohomological identity, and here it follows from dimension counting alone.
Theorem 4.2.10
(Solving a Frobenius difference equation)
For , the equation has a solution if and only if . When it has a solution, it has exactly solutions, forming a coset of .
Proof
(Proof)
The kernel of is the fixed field , a one-dimensional -space. Rank–nullity therefore gives . If , then Frobenius invariance of trace gives
The trace kernel also has dimension because trace is a nonzero linear map onto . Thus is contained in the trace kernel and the two subspaces have the same finite dimension; they are equal. If is one solution, is another exactly when , or . The solution set is and contains elements.
Proof X-ray
Containment plus equal dimension
The proof never constructs a solution of . It proves one containment, , from the Frobenius invariance , and then compares dimensions. Both subspaces have dimension : the first by rank–nullity with , the second because the trace is onto . Equal finite dimensions turn the containment into equality. The surjectivity theorem is therefore an essential input: without it, the trace kernel could a priori be all of , and the containment would say nothing about solvability.
Worked example 4.2.11
(Solvability over )
In with , the equation has the two solutions and : their squares differ from themselves by one. This agrees with . In contrast, has no solution, because . Checking a single unsuccessful candidate would not establish the latter conclusion; the trace obstruction excludes every candidate.
The analogous multiplicative equation also has a clean criterion, although its proof uses cyclicity rather than vector-space dimension. Division requires nonzero elements, so zero must be kept outside the statement.
Proposition 4.2.12
(A multiplicative Frobenius equation)
For , there is an with if and only if . If such an exists, exactly nonzero elements solve the equation.
Proof
(Proof)
Write , , and choose a generator of . If then . Conversely write . The norm-one condition is , equivalent to , or . Thus for some integer . The homomorphism has kernel , because is equivalent to for nonzero . Every nonempty fiber is a kernel coset and therefore has elements. This includes , where the kernel has one element.
Worked example 4.2.13
(Solving a norm-one equation in )
In , has norm one, so the theorem predicts exactly two nonzero solutions of . They are and : indeed in characteristic three, and the second is its negative. By contrast has norm two, so it is not a square of the form for .
Intermediate fields and nested operations
Suppose divides and let , the unique intermediate field of that order. Write . The trace and norm from to use powers of the -Frobenius . They compose with the corresponding operations from to :
To see the first equality, expand the left side as . The indices run once through . The product proof is identical with sums replaced by products: applying a Frobenius automorphism to a product distributes over the factors. No division by an extension degree is involved, so these identities remain valid when the characteristic divides or .
Worked example 4.2.14
(Two steps down a tower)
Let , and . Here and . For , the first trace is . Adding its square gives , precisely the six-term trace from to . This identity describes every element of the extension; choosing a generator or factoring its minimal polynomial is unnecessary.
These calculations also separate conjugacy from the number of distinct conjugates. If lies in a proper intermediate field, its orbit under -Frobenius has fewer than elements, and the -term trace repeats each orbit value. For instance a base-field element has only one distinct conjugate but contributes copies of . Omitting repetitions would make the tower formula false in general.
Relating the formulas to a minimal polynomial
Let have minimal polynomial of degree . By the finite-field theory proved earlier, and its distinct roots are . Thus
where and . The conjugates in the extension formulas repeat this orbit exactly times. Consequently
Here is an integer subsequently interpreted as a scalar in ; this is multiplication, not division by an element of . In positive characteristic the scalar can vanish. The product formula also covers , whose minimal polynomial is .
Concept lensAlgebraic
Trace and norm as trace and determinant of multiplication
Multiplication by , , is a -linear map . As a -space, is a direct sum of copies of , and on the characteristic polynomial of is . Hence the characteristic polynomial of on is , and the formulas above say and . For and the basis , the matrix of has columns and : trace , determinant .
Worked example 4.2.15
(Repeated conjugates in a larger extension)
Take in . Its minimal polynomial over is , so , and . The trace is , while the norm is . Reading only the coefficient of without accounting for multiplicity would incorrectly give trace one. The minimal-polynomial coefficients do directly give the trace and norm from to ; the further power and scalar account for the larger ambient extension.
Practice and solutions
Quick check
(Compute and explain why it does not make the trace map zero.)
Solution · Answer
The trace is , but has trace one in the stated model of .
Quick check
(How many elements of have trace over ?)
Solution · Answer
The extension has degree three, so the fiber has elements.
Quick check
(How many elements of have norm over ?)
Solution · Answer
The nonzero norm fiber has elements.
Quick check
(In with , compute the trace and norm of .)
Solution · Answer
The trace is in ; the norm is .
Quick check
(Why does a zero self-pairing not imply degeneracy?)
Solution · Answer
Nondegeneracy asks for some partner with nonzero pairing, not necessarily . In , and provide an example.
Quick check
(If , how can a preimage of under trace produce a with ?)
Solution · Answer
Set . Then and .
Quick check
(How many solutions does have in ? Give them explicitly.)
Solution · Answer
Exactly three: the kernel of is . These are all the solutions, not just three examples.
Quick check
(A nonzero element of has norm over . Can it equal for any nonzero in ? Explain using the norm criterion.)
Solution · Answer
No. Every nonzero square has the form for , hence has norm one. Since in , the norm obstruction rules it out.