The four conjugacy normal forms of GL₂ over a finite field #
TauCeti/LinearAlgebra/Matrix/GeneralLinearGroup/ConjugacyClasses.lean classifies the conjugacy
classes of GL₂(F) by trace and determinant, with the companion matrix of the characteristic
polynomial as the representative of each non-scalar class. That representative is uniform but
anonymous. This file replaces it, over a finite field, by the four named normal forms the
character theory of GL₂(𝔽_q) is written against:
- a central scalar matrix
a • 1; - a split semisimple
TauCeti.diagGL ![a, b]witha ≠ b; - a non-semisimple Jordan block
TauCeti.jordanGL a 1; - an elliptic
TauCeti.GL2NonSplitTorusHom F E x, multiplication by an element of a degree-2extensionE/Fthat does not lie inF.
TauCeti.exists_isConj_normalForm is the resulting statement that every element of GL₂(F) is
conjugate to one of the four. It is what turns the four character values computed at these normal
forms in TauCeti/RepresentationTheory/CharacterTable/GL2/CharacterValues.lean into a full row of
the character table of GL₂(𝔽_q): a character is a class function, so a row is determined once the
normal forms exhaust the classes.
The two non-central split forms need no finiteness and no extension. Each is the same check
against TauCeti.isConj_iff_of_notMem_range_scalar: the normal form is not scalar, and its trace
and determinant are the prescribed ones. What the roots of X² - t X + d are is the only thing
that distinguishes them — two distinct roots give diagGL, a repeated root gives jordanGL —
and only in the repeated-root case need non-scalarness be assumed: a scalar matrix has a repeated
root, so prescribing two distinct roots, or a root outside F, already rules it out.
The elliptic case is the one that needs a quadratic extension, and it is where the finiteness of
F enters. Multiplication by x : E has trace Tr_{E/F} x and determinant N_{E/F} x
(TauCeti.GL2NonSplitTorus.trace_gl2NonSplitTorusHom and
TauCeti.GL2NonSplitTorus.val_det_gl2NonSplitTorusHom), so the normal form is pinned by the
quadratic-extension lemmas of TauCeti/FieldTheory/Quadratic.lean:
TauCeti.Algebra.trace_eq_of_mul_self_eq and TauCeti.Algebra.norm_eq_of_mul_self_eq identify
those with t and d for an x outside F satisfying x² = t x - d, and
TauCeti.exists_mul_self_eq_of_finite supplies such an x over a finite field.
Uniqueness — that the four families are pairwise disjoint and that the parameters are determined up
to the evident symmetries (a, b) ↦ (b, a) and x ↦ x^q — is the second half of the file, and it
runs on the same classification. Being scalar is a conjugacy invariant on its own
(TauCeti.not_isConj_of_mem_range_scalar), which separates the central family from the other
three; that is the one separation trace and determinant do not make, a scalar and a Jordan block
with the same eigenvalue having the same characteristic polynomial. Among the three non-central
families the characteristic polynomial does all the work, and what separates them is how it
factors: two distinct roots in F for diagGL, a repeated root for jordanGL, and no root in F
for the elliptic form. Reading that off is one equation per pair, and in the elliptic case it needs
the torus parameter x to satisfy the quadratic built from its own trace and norm,
Algebra.IsQuadraticExtension.sq_eq_trace_smul_sub_norm: an eigenvalue of a split or a Jordan form
lies in F, and x does not.
None of those statements needs F finite, and the split and non-semisimple ones need no extension
at all. What does need finiteness is the symmetry x ↦ x^q of the elliptic parameter, which is
TauCeti.GL2NonSplitTorus.isConj_gl2NonSplitTorusHom_iff in
TauCeti/LinearAlgebra/Matrix/GeneralLinearGroup/ConjugacyClasses.lean. Together with the results
here it says that the four families, with their parameters read up to the two symmetries, index the
conjugacy classes of GL₂(𝔽_q) without repetition — the description by named data of the indexing
that TauCeti.conjClassesGLFinTwoEquiv supplies anonymously.
Main results #
TauCeti.isConj_diagGL_of_trace_of_det,TauCeti.isConj_jordanGL_one_of_trace_of_detandTauCeti.isConj_gl2NonSplitTorusHom_of_trace_of_det: the three non-central normal forms, each characterized by the roots of its characteristic polynomial, read off the trace and the determinant.TauCeti.exists_isConj_normalForm: every element ofGL₂(F), forFfinite with a degree-2extensionE, is conjugate to one of the four normal forms.TauCeti.not_isConj_scalar_diagGL,TauCeti.not_isConj_scalar_jordanGL_one,TauCeti.not_isConj_scalar_gl2NonSplitTorusHom,TauCeti.not_isConj_diagGL_jordanGL_one,TauCeti.not_isConj_diagGL_gl2NonSplitTorusHomandTauCeti.not_isConj_jordanGL_one_gl2NonSplitTorusHom: the four families are pairwise disjoint.TauCeti.isConj_diagGL_iffandTauCeti.isConj_jordanGL_one_iff: the parameters inside a family are determined, the split family up to the transposition of its two eigenvalues and the non-semisimple family outright. For the central family this isTauCeti.isConj_scalar_iff, which needs nothing aboutGL₂and lives with the other scalar invariants inTauCeti/LinearAlgebra/Matrix/GeneralLinearGroup/ConjugacyClasses.lean.
References #
- C. Bonnafé, Representations of
SL₂(𝔽_q)(2011), Chapter 1. - J.-P. Serre, Linear Representations of Finite Groups, GTM 42 (1977), §5.2.
Shape adapters for the two split normal forms #
The normal forms below are written against the pair ![a, b] and the Jordan parameter 1, while
TauCeti/LinearAlgebra/Matrix/GeneralLinearGroup/Diagonal/Basic.lean and
TauCeti/LinearAlgebra/Matrix/GeneralLinearGroup/ScalarUnipotent.lean state non-scalarity, the
trace and the determinant for a general family and a general off-diagonal entry. These six private
lemmas do that adaptation once, so that no proof below recomputes the trace, the determinant or the
non-scalarity of a normal form from its matrix entries.
The two split normal forms #
The split semisimple normal form. An element of GL₂(F) whose characteristic polynomial
has the two distinct roots a and b is conjugate to diagGL ![a, b]. Distinct roots already
force the element to be non-scalar.
The non-semisimple normal form. A non-scalar element of GL₂(F) whose characteristic
polynomial is (X - a)² is conjugate to the Jordan block jordanGL a 1.
The elliptic normal form #
The elliptic normal form. An element of GL₂(F) whose characteristic polynomial
X² - t X + d is satisfied by an element x of a degree-2 extension E/F lying outside F is
conjugate to the element x of the non-split torus. A root outside F already forces the element
to be non-scalar.
The classification #
Every element of GL₂(F) is conjugate to one of the four normal forms, for F a finite
field with a supplied degree-2 extension E: a central scalar, a split semisimple
diagGL ![a, b] with a ≠ b, a non-semisimple Jordan block jordanGL a 1, or an elliptic
element GL2NonSplitTorusHom F E x with x outside F.
This is what turns a computation of a character at these four normal forms into a full row of the
character table of GL₂(𝔽_q).
Uniqueness of the normal form #
The four families are pairwise non-conjugate, and inside each family the parameters are determined
up to the evident symmetry. Only the order shown is proved; the reverse order follows from
IsConj.symm.
A scalar is not conjugate to a Jordan block. The two have the same characteristic
polynomial (X - a)² when their eigenvalues agree, so this is exactly where trace and determinant
fail to separate the conjugacy classes of GL₂(F).
A split semisimple normal form is not conjugate to a Jordan block. For distinct a and b
equal traces and determinants would force (a - b)² = 0; for a = b the left-hand side is a
scalar, which is not conjugate to a Jordan block either.
Two split semisimple normal forms are conjugate exactly when their eigenvalue pairs agree up
to order. The transposition (a, b) ↦ (b, a) is the Weyl group of the split torus acting on the
torus by permuting the diagonal entries, and it is the only identification among these normal forms.
No distinctness is asked of either pair: when a pair repeats, its normal form is the corresponding
scalar, and the statement specializes to TauCeti.isConj_scalar_iff.
A scalar is not conjugate to an elliptic normal form.
A split semisimple normal form is not conjugate to an elliptic one. Equal traces and
determinants make x a root of (X - a) (X - b), so x would be a or b, and both lie in
F.
A Jordan block is not conjugate to an elliptic normal form. Equal traces and determinants
make x a double root of (X - a)², so x would be a, which lies in F. No hypothesis is
needed on x: a torus parameter inside F gives the corresponding scalar, which is not conjugate
to a Jordan block either.