The diagonal operators of the invertible endomorphisms span all of them #
A V-endomorphism f acts on the tensor power ⨂[K] (_ : ι), V diagonally, by
f^{⊗ι} = PiTensorProduct.map (fun _ ↦ f). These diagonal operators are not closed under addition,
so the span of the ones coming from invertible f is not visibly all of the span of them all.
Over an infinite field and for a finite-dimensional V the two spans nonetheless coincide
(PiTensorProduct.span_range_map_const_units_eq_span_range_map_const), because the invertible
endomorphisms are Zariski dense. That is what lets the general linear group replace the whole
endomorphism algebra in Schur-Weyl duality, where the commutant of the symmetric-group action is
the span of the diagonal operators and one wants it to be the image of K[GL(V)].
The argument #
A vector of a vector space lies in a subspace as soon as every functional vanishing on the subspace
kills it (Subspace.forall_mem_dualAnnihilator_apply_eq_zero_iff). So fix a functional φ on
End (⨂[K] (_ : ι), V) vanishing on the span of the invertible diagonal operators. Composing φ
with PiTensorProduct.mapMultilinear, which is multilinear in the family of endomorphisms, and with
the matrix-to-endomorphism isomorphism attached to a basis, produces a multilinear form Θ in ι
matrix arguments whose diagonal is Θ (Y, …, Y) = φ (Y^{⊗ι}). That diagonal vanishes on the
invertible matrices, hence identically, by
MultilinearMap.apply_const_eq_zero_of_eq_zero_on_gl.
Main results #
PiTensorProduct.map_const_mem_span_range_map_const_units: the diagonal operator of an endomorphism is a combination of the diagonal operators of the invertible ones.PiTensorProduct.span_range_map_const_units_eq_span_range_map_const: consequently the two spans agree.
The diagonal operator of an endomorphism is a combination of the diagonal operators of the
invertible ones, over an infinite field and for a finite-dimensional V.
The diagonal operators of the invertible endomorphisms span the same subspace as all the
diagonal operators, over an infinite field and for a finite-dimensional V.