Galois theory turns algebraic equations into symmetry. A Galois representation takes those arithmetic symmetries and lets them act linearly, so that matrices, traces, determinants, eigenvalues and characteristic polynomials can record information that is otherwise difficult to see.
This is one of representation theory’s deepest return paths into number theory. The source group can be enormous: the absolute Galois group of a number field is infinite and topological. Yet arithmetic objects such as roots of unity, torsion points on elliptic curves and étale cohomology carry finite-dimensional linear actions. At unramified primes, Frobenius conjugacy classes become especially useful probes. Their matrix traces and determinants often recover concrete arithmetic data.
This guide works mainly over number fields and ℓ-adic coefficient spaces. It assumes basic groups, fields and linear algebra, and it builds on the BTT guides to representation theory, characters and traces, and elliptic curves over finite fields.
Definition · Cyclotomic character · Elliptic curves · Frobenius · Euler factors · Practice · BTT Mathematics Hub
Arithmetic symmetry begins with field automorphisms
Let K be a field and let K̄ be an algebraic closure. The absolute Galois group GK=Gal(K̄/K) consists of field automorphisms of K̄ that fix every element of K.
For a finite Galois extension L/K, Gal(L/K) is a finite quotient of GK. Classical Galois theory studies how this finite symmetry group permutes roots and intermediate fields. A Galois representation keeps the symmetry but changes the carrier: instead of acting on a set of roots, the group acts on vectors.
The point is not that the field automorphism itself is linear on K̄ as a vector space over every coefficient field we might want. Rather, arithmetic constructions naturally produce vector spaces or modules equipped with a compatible Galois action. Representation theory then studies that action.
The basic definition needs continuity
A finite-dimensional Galois representation is typically a continuous homomorphism
ρ:GK→GL(V),
where V is a finite-dimensional vector space over a field such as C, Qℓ, a finite extension of Qℓ, or a finite field. The topology matters. GK is profinite, and ℓ-adic coefficient groups have their own topology. Continuity forces finite-level arithmetic information to vary compatibly rather than allowing an arbitrary abstract group homomorphism.
Finite-image complex representations are often called Artin representations. ℓ-adic representations can have infinite image but remain controlled through congruence information modulo ℓ, ℓ², ℓ³ and so on.
When V has dimension n and a basis is chosen, each σ∈GK becomes an n×n invertible matrix ρ(σ). As in every representation, multiplication is preserved:
ρ(στ)=ρ(σ)ρ(τ).
Changing basis conjugates every matrix simultaneously. Therefore trace, determinant and characteristic polynomial of a fixed group element are basis-independent data.
A finite warm-up: representations that factor through a finite Galois group
Suppose L/K is finite Galois with group G. Any representation G→GL(V) can be composed with the restriction map GK→G. The resulting Galois representation sees only the finite quotient Gal(L/K).
This gives the simplest bridge from classical Galois theory to representation theory. If G permutes several embeddings, roots or cosets, we can form a permutation representation. Decomposing it may reveal a trivial invariant line, sign-like quotient data or higher-dimensional irreducible components.
The kernel tells us exactly which arithmetic symmetries have become invisible. A representation factoring through GK/N records only the quotient symmetry.
The cyclotomic character: the one-dimensional prototype
Fix a prime ℓ. For every n, the group of ℓn-th roots of unity μℓⁿ is stable under Galois action. If ζ is such a root and σ∈GK, then σ(ζ)=ζa for some unit a modulo ℓn.
As n varies, these exponents are compatible. Passing to the inverse limit gives the ℓ-adic cyclotomic character
χℓ:GK→Zℓ×.
This is a one-dimensional ℓ-adic Galois representation. It records how Galois automorphisms act on all ℓ-power roots of unity at once.
For K=Q and a prime p≠ℓ, arithmetic Frobenius at p acts on residue-field roots by x↦xp. Under the arithmetic-Frobenius convention used in this article, χℓ(Frobp)=p in Zℓ×. Some authors use geometric Frobenius, the inverse convention. Never move formulas between references without checking which Frobenius is being used.
Why the coefficient field changes the problem
An ℓ-adic representation can be reduced modulo ℓ after choosing a stable lattice. This produces a residual representation
ρ̄:GK→GLn(Fℓ)
or over a finite extension of Fℓ. The reduction can merge eigenvalues that were distinct in characteristic zero, and semisimplicity can fail. Thus an ℓ-adic representation and its mod-ℓ reduction can have sharply different decomposition behaviour.
This is a number-theoretic version of the field-dependence already seen in modular representation theory. The characteristic of the coefficient field changes what a matrix can reveal.
Elliptic curves produce natural two-dimensional representations
Let E be an elliptic curve over Q and let ℓ be prime. The ℓn-torsion group E[ℓn] over an algebraic closure is, away from characteristic ℓ, isomorphic as an abstract module to (Z/ℓnZ)².
Galois automorphisms act on the coordinates of torsion points and preserve the elliptic-curve group law. Therefore they act linearly on E[ℓn]. Passing through all n gives the Tate module
Tℓ(E)=lim← E[ℓn]≈Zℓ².
The resulting representation is
ρE,ℓ:GQ→GL₂(Zℓ).
This construction is not an arbitrary encoding of the curve. The representation records how every ℓ-power torsion point moves under arithmetic symmetry.
The Weil pairing explains the determinant
The Weil pairing is an alternating pairing E[ℓn]×E[ℓn]→μℓⁿ compatible with Galois action. Under a basis, the effect of a 2×2 matrix on this alternating volume is its determinant.
Compatibility therefore identifies the determinant of the elliptic-curve Galois representation with the cyclotomic character:
det ρE,ℓ=χℓ.
At an unramified prime p≠ℓ of good reduction, this gives detρ(Frobp)=p under our arithmetic-Frobenius convention.
Frobenius is defined up to conjugacy, and that is enough
At a prime p where the representation is unramified, the inertia subgroup acts trivially and one obtains a Frobenius conjugacy class. Different representatives are conjugate inside the Galois group.
A matrix representation sends conjugate group elements to similar matrices. Therefore trace, determinant and characteristic polynomial are well-defined for the Frobenius conjugacy class even though an individual Frobenius element may depend on a choice.
Arithmetic symmetry gives a conjugacy class; representation theory gives a similarity class; trace and determinant survive both.
This alignment is one reason Frobenius traces appear throughout arithmetic geometry.
Worked elliptic-curve example over F5
Take E:y²=x³−x and reduce modulo 5. Count points directly.
- x=0 gives y²=0: one affine point.
- x=1 gives y²=0: one affine point.
- x=2 gives y²=1: two affine points.
- x=3 gives y²=4: two affine points.
- x=4 gives y²=0: one affine point.
Add the point at infinity. The total is 8. Therefore
a₅=5+1−#E(F₅)=6−8=−2.
For every auxiliary ℓ≠5, the Frobenius matrix on the ℓ-adic Tate module has trace −2 and determinant 5. Its characteristic polynomial is
X²+2X+5.
This one calculation connects a finite-field point count to an ℓ-adic matrix invariant. MIT’s elliptic-curve notes formulate this general relation as trρE,ℓ(Frobp)=ap and detρE,ℓ(Frobp)=p at good primes p≠ℓ. [1]
Hasse gives an immediate error check
For an elliptic curve over Fp, Hasse’s bound gives |ap|≤2√p. Our value |−2|=2 is certainly below 2√5.
If a point count produced #E(F₅)=20, then a₅=−14, violating the bound. The representation-theoretic trace would be impossible before any Tate-module matrix was constructed. Good arithmetic workflows use such independent constraints early.
Ramification records exceptional local behaviour
At a prime p, the decomposition group contains an inertia subgroup Ip. An ℓ-adic representation is unramified at p when inertia acts trivially.
When inertia acts nontrivially, the local representation contains information beyond a single Frobenius conjugacy class. Bad reduction of an elliptic curve is one source of such behaviour. Local Euler factors then require the inertia-fixed subspace rather than blindly applying the good-prime quadratic polynomial.
The phrase “trace of Frobenius at p” therefore needs its hypotheses. Good reduction and unramifiedness cannot be omitted merely because the same formula worked at nearby primes.
Traces are powerful, but they do not automatically contain extension data
Two matrices can have the same trace and determinant while differing as matrices. Across a whole group, character-style trace data can determine a semisimple representation under appropriate hypotheses, but non-split extensions may have the same semisimplified trace data.
This is the same warning encountered in modular representation theory: composition-factor information is not the same as extension information. Arithmetic applications often state results for semisimplifications precisely because trace comparisons naturally live at that level.
Never infer a complete integral lattice or extension structure from Frobenius traces alone unless a theorem justifies that reconstruction.
Residual representations compress congruence information
Reducing ρE,ℓ modulo ℓ gives an action on E[ℓ]:
ρ̄E,ℓ:GQ→GL₂(Fℓ).
At a good prime p≠ℓ, its Frobenius characteristic polynomial is the reduction of X²−apX+p modulo ℓ. Thus point counts modulo different primes feed congruence information into one residual representation.
For our p=5 example and ℓ=3, X²+2X+5 becomes X²−X+2 modulo 3. The determinant becomes 2, agreeing with 5 modulo 3. This is a small but exact compatibility check between integer arithmetic and the residual matrix.
From Frobenius matrices to local Euler factors
For an unramified representation ρ and a prime p, a local Euler factor is built from the characteristic polynomial of Frobenius. With a common arithmetic-Frobenius convention, one writes a factor of the form
Lp(s,ρ)=det(I−ρ(Frobp)p−s)−1,
after including the appropriate inertia-fixed space when ramification is present. Conventions vary, so a geometric-Frobenius reference may place an inverse elsewhere.
For the elliptic curve at a good prime, the polynomial is 1−apT+pT² with T=p−s. Our p=5 example gives 1+2T+5T².
The global L-function is an Euler product of these local factors in a region where the product converges, followed in many important cases by analytic continuation obtained through deeper theory. A Galois representation therefore turns arithmetic symmetry into a sequence of local matrix polynomials.
Modular forms create another representation bridge
For suitable normalized Hecke eigenforms f=Σanqn of weight k and nebentypus ε, deep theorems attach two-dimensional ℓ-adic Galois representations whose unramified Frobenius polynomials have the form
X²−apX+ε(p)pk−1
for primes away from the level and ℓ. This places Hecke eigenvalues and Frobenius traces in the same equation. It is one of the concrete interfaces behind the wider automorphic–Galois correspondence discussed later in this batch. An expository discussion of such representations appears in the Harvard and Calegari notes listed below. [2–3]
Why Chebotarev makes Frobenius data so important
For finite Galois extensions, the Chebotarev density theorem says, roughly, that Frobenius conjugacy classes at unramified primes are distributed across the conjugacy classes of the Galois group with precise densities.
For representation comparisons, this means Frobenius elements provide a large and arithmetically accessible test set. Under suitable continuity and semisimplicity assumptions, matching Frobenius characteristic polynomials at sufficiently many primes can force strong equivalences.
The theorem does not permit us to replace a proof by checking the first ten primes. Finite computation can suggest a correspondence; density and representation-theoretic uniqueness theorems supply the global logic.
A computational verification workflow
- State the number field, coefficient field and prime ℓ.
- Record whether arithmetic or geometric Frobenius is being used.
- Identify the finite-level objects whose inverse limit produces the representation.
- At each test prime, check unramifiedness or good reduction before using a good-prime formula.
- Compute ap independently where possible, for example from #E(Fp).
- Check determinant against the cyclotomic prediction.
- Apply Hasse or another known bound before trusting a trace.
- Reduce modulo ℓ only after preserving the integral polynomial first.
- Separate semisimple trace information from integral lattice and extension information.
These checks are small enough to perform by hand in examples and strict enough to catch many convention errors.
Common mistakes
- Ignoring topology: an ℓ-adic Galois representation is normally continuous.
- Mixing Frobenius conventions: arithmetic and geometric Frobenius are inverses.
- Using the good-prime polynomial at a bad prime: ramification changes the local problem.
- Equating trace data with a full integral representation: extension and lattice information can be lost.
- Forgetting the determinant: for elliptic curves it gives an independent cyclotomic check.
- Reducing too early: mod-ℓ data can merge distinctions visible over Zℓ or Qℓ.
Practice with worked answers
1. One-dimensional representation
What is the dimension of the ℓ-adic cyclotomic character? Answer: one. Its matrices are 1×1 units in Zℓ.
2. Basis independence
Why is trρ(Frobp) independent of basis? Answer: a basis change conjugates the Frobenius matrix, and trace is invariant under similarity.
3. Point-count trace
An elliptic curve has #E(F11)=15. Find a11. Answer: 11+1−15=−3.
4. Frobenius polynomial
For the same good prime, write the elliptic-curve Frobenius polynomial. Answer: X²+3X+11.
5. Hasse check
Could a11=9 occur for an elliptic curve over F11? Answer: no, because 9>2√11≈6.63.
6. Residual determinant
At p=11 and ℓ=5, what determinant should the residual Frobenius matrix have? Answer: 11 mod 5=1.
7. Conjugacy
Two choices of Frobenius representative are conjugate. Which quantities are immediately unchanged? Answer: trace, determinant and characteristic polynomial of their representation matrices.
8. Ramification warning
Why can the formula X²−apX+p not be used blindly at every prime of an elliptic curve? Answer: bad-reduction or ramified primes require a different local analysis; the good unramified formula has hypotheses.
9. Trace versus extension
Can equal traces on every group element automatically prove two non-semisimple integral representations are the same? Answer: no. Trace data can forget non-split extension structure; additional hypotheses are required.
10. Modular-form bridge
A weight-2 eigenform with trivial nebentypus has ap=4 at an unramified prime p=7. What Frobenius polynomial is predicted by the standard formula? Answer: X²−4X+7.
What Galois representations teach about representation itself
The original arithmetic symmetry lives in field automorphisms. The representation does not replace that world; it selects a linear object on which the symmetry acts. Frobenius then supplies local conjugacy classes, and trace and determinant compress the resulting similarity classes into arithmetic scalars.
The full cycle is therefore:
arithmetic object → Galois action → linear representation → Frobenius matrix → trace/determinant → arithmetic interpretation.
This is precisely the kind of controlled translation that makes representation theory so reusable.
Mathematical references
[1] Andrew Sutherland, MIT 18.783, Lecture 26: Galois representations attached to elliptic curves, for Tate modules, Frobenius traces and determinants. [2] Samuel Marks, Galois Representations, for a concise introduction to continuous Galois representations, unramified primes and L-functions. [3] Frank Calegari, Even Galois Representations, for the modular-form and automorphic interface. The explicit F₅ point count and residual congruence checks above are worked independently in this guide.
Continue Representation Mathematics — Batch 04
Next, follow arithmetic operators into Hecke Algebras, Double Cosets and Modular Forms. Then move from classical modular forms into Automorphic Representations, L-Functions and the Langlands Program. For a categorical method of recovering a symmetry group from its representation theory, continue to Tannakian Categories and Fibre Functors. Return to the BTT Mathematics Learning Hub.