The radicals of gl n K and sl n K #
TauCeti/Algebra/Lie/GeneralLinear/Basic.lean identifies the centre of gl n R (the scalar
matrices) and its derived ideal (sl n R), and shows that the two are complementary submodules as
soon as the size of the matrices is invertible. This file proves the reductivity criterion itself,
Mathlib's LieAlgebra.HasCentralRadical: over a field in which 2 ≠ 0, the solvable radical of
gl n K is its centre. This makes the reductive vocabulary applicable to gl n, whose Killing
form is degenerate.
The same structure shows that sl n K has trivial radical when 2 and the nonempty matrix size
are nonzero. In characteristic zero this is registered as an instance, so Cartan's criterion makes
the Killing-form API available by typeclass synthesis.
The proof is the structure of the Lie ideals of gl n K. The centre is always an abelian, hence
solvable, ideal, so it is contained in the radical. For the other inclusion, every solvable ideal
must consist of scalar matrices, and that is a consequence of the sharper statement
TauCeti.slIdeal_le_of_notMem_center: a Lie ideal containing a single non-central matrix already
contains all of sl n K. Since sl n K is perfect (TauCeti.lie_slIdeal_slIdeal) and nonzero as
soon as there are two indices, it is not solvable, so no solvable ideal reaches outside the centre.
The ideal-generation argument runs through matrix units, and only two brackets are needed.
Bracketing against a diagonal matrix rescales a matrix unit,
⁅diagonal d, Eₚq c⁆ = Eₚq ((dₚ - d_q) c) (TauCeti.lie_single_of_mem_diagonalCartan, from the
diagonal Cartan subalgebra file), which extracts Eₚq from any diagonal element separating the
indices p and q. Bracketing twice against Eⱼᵢ annihilates everything except one entry,
(ad Eⱼᵢ)² x = Eⱼᵢ (-2 xᵢⱼ) (TauCeti.lie_single_lie_single_of_ne), which extracts Eⱼᵢ from any
element with a nonzero (i, j) entry. A non-central matrix is either non-diagonal, and then the
second bracket applies, or diagonal with two distinct entries, and then the first does. Once one
off-diagonal matrix unit lies in the ideal, so does the difference of diagonal units
E_bb - Eₐₐ = ⁅E_bₐ, Eₐb⁆, which separates b from every other index; that produces every unit in
the row and the column of b, then every difference of diagonal units, and finally, by the first
bracket again, all the remaining units.
Main results #
TauCeti.slIdeal_le_of_notMem_center: a Lie ideal ofgl n Kcontaining a non-central matrix containssl n K; equivalentlyTauCeti.slIdeal_le_or_le_center, every Lie ideal ofgl n Keither containssl n Kor consists of scalar matrices.TauCeti.lie_slIdeal_slIdeal:sl n Ris perfect,⁅sl n R, sl n R⁆ = sl n R, whenever2is invertible inR, whenceTauCeti.not_isSolvable_slIdeal: it is not solvable whenRis nontrivial and there are at least two indices.TauCeti.radical_matrix_eq_centerandTauCeti.hasCentralRadical_matrix:gl n Kis reductive, its radical being its centre.TauCeti.hasTrivialRadical_sl:sl n Khas trivial radical when2and the nonempty matrix size are nonzero; the characteristic-zero case is a named instance.
Implementation notes #
The hypothesis (2 : K) ≠ 0 is genuinely needed, and not an artefact of the bracket computations:
in gl 2 (ZMod 2) the span of E₁₂ and the identity matrix is a Lie ideal containing a non-central
element but not E₂₁, and there sl 2 (ZMod 2) is nilpotent, so it is a solvable ideal strictly
larger than the centre. No invertibility of the size of the matrices is needed, however: the
statement is proved uniformly in the index type, the case of at most one index being the abelian
one, where the radical and the centre are both everything.
Only the results about ideals are stated over a field, because the generation argument divides by an
arbitrary nonzero matrix entry. The two bracket identities in Basic.lean need only a ring, and
perfectness of sl n R — which divides by 2 alone — is stated over a commutative ring in which
2 is a unit, non-solvability adding only Nontrivial R; the radical computation specialises these
to a field.
The bracket against a diagonal matrix comes from TauCeti.lie_single_of_mem_diagonalCartan.
The matrix-unit bracket identities live alongside the other ring-level computations in
TauCeti/Algebra/Lie/GeneralLinear/Basic.lean.
References #
- J. Dixmier, Enveloping Algebras, AMS GSM 11 (1996), Section 1.6 (reductive Lie algebras).
- J. E. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer GTM 9 (1972), Section 1.2 (the ideals of the classical linear Lie algebras).
Matrix units inside a Lie ideal #
A diagonal element of a Lie ideal separating two indices contributes their matrix unit.
An element of a Lie ideal with a nonzero off-diagonal entry contributes the transposed matrix unit there.
One off-diagonal matrix unit generates sl n K: a Lie ideal of gl n K containing Eₐb
for some a ≠ b contains every trace-zero matrix.
A Lie ideal of gl n K containing a non-central matrix contains sl n K.
sl n K is perfect, hence not solvable #
sl n R is perfect whenever 2 is invertible in R: ⁅sl n R, sl n R⁆ = sl n R.
Each off-diagonal matrix unit is Eₚq c = ⁅Eₚₚ - E_qq, Eₚq (t c)⁆ for an inverse t of 2, and
each difference of diagonal units is Eₚₚ c - E_qq c = ⁅Eₚq c, E_qₚ⁆; all four factors have trace
zero.
sl n R is nonzero over a nontrivial ring as soon as there are two indices.
sl n R is not solvable when 2 is invertible in a nontrivial R and there are at least
two indices: it is perfect and nonzero, so its derived series is constant.
The radical of gl n K #
The radical of gl n K is its centre, for any field in which 2 ≠ 0: gl n K is
reductive.
Every solvable Lie ideal consists of scalar matrices, since an ideal containing a non-central matrix
contains the non-solvable sl n K (TauCeti.slIdeal_le_of_notMem_center), and conversely the
centre is itself a solvable ideal.
gl n K is reductive, for any field in which 2 ≠ 0.
Over a field of characteristic zero — the setting of the reductive structure theory — gl n K
is reductive with no side hypothesis.
Semisimplicity of the special linear Lie algebra #
The special linear Lie algebra has trivial radical. If 2 and, for nonempty n, the
cardinality of n are nonzero, a solvable ideal of sl n maps to a solvable ideal of gl n,
hence lies in the scalar matrices. Its image also lies in the complementary derived ideal sl n,
so the original ideal vanishes.
In characteristic zero, sl n has trivial radical.