Rank bounds along a pencil #
A pencil is an affine line t ↦ A + t • B of matrices or linear maps. Each minor of A + t • B
is a polynomial in t, so the condition rank (A + t • B) ≤ p either holds for every t or
fails for all but finitely many t: a polynomial with infinitely many roots is zero. This is the
one-parameter case of the Zariski closedness of rank bounds in
TauCeti.LinearAlgebra.Matrix.Rank.Closed, stated over any field and without coordinates on the
source.
Dividing f + t • g by t and letting t → ∞ turns a bound on the pencil into a bound on its
direction g. Algebraically, the bound for infinitely many t passes to the reversed pencil
g + s • f, where it then holds at s = 0. The same argument tolerates a fixed summand W that
is not rescaled: a bound on range (f + t • g) ⊔ W for infinitely many t bounds
range g ⊔ W. This is the linear-algebraic heart of the Lefschetz–Whitehead rank estimate,
Lemma 2 of Łojasiewicz's triangulation of semianalytic sets, where W is spanned by the
derivative in the line parameter.
Main results #
Matrix.rank_add_smul_le_of_infinite: a rank bound on a matrix pencil holding at infinitely many parameters holds at every parameter.LinearMap.finrank_range_add_smul_le_of_infinite: the same for a pencil of linear maps into a finite-dimensional space.LinearMap.finrank_range_sup_le_of_infinite: a bound onrange (f + t • g) ⊔ Wat infinitely manytboundsrange g ⊔ W.LinearMap.finrank_range_lt_of_infinite: if moreoverv ∉ range g, then a boundponrange (f + t • g) ⊔ K ∙ vat infinitely manytforcesfinrank (range g) < p.
References #
- S. Łojasiewicz, Triangulation of semi-analytic sets, Ann. Scuola Norm. Sup. Pisa (3) 18 (1964), 449–474, §2, Lemma 2.
Rank bounds along a matrix pencil. If rank (A + t • B) ≤ p for infinitely many scalars
t, then it holds for every t: each (p + 1) × (p + 1) minor of A + t • B is a polynomial
in t with infinitely many roots, hence zero.
Rank bounds along a pencil of linear maps. If finrank (range (f + t • g)) ≤ p for
infinitely many scalars t, then it holds for every t. The source need not be
finite-dimensional: both maps factor through the quotient by ker f ⊓ ker g, which embeds into
N × N.
Rank bounds pass to the direction of a pencil. If range (f + t • g) ⊔ W has dimension at
most p for infinitely many scalars t, then so does range g ⊔ W. For t ≠ 0 the range of
f + t • g is that of g + t⁻¹ • f, so the reversed pencil g + s • f satisfies the bound at
infinitely many s and hence at s = 0.
The rank drop of a transverse direction. If v ∉ range g and range (f + t • g) ⊔ K ∙ v
has dimension at most p for infinitely many scalars t, then finrank (range g) < p.