Passes the smooth compact-perturbation identity to Lipschitz maps
through local strong convergence.
Statement
Fix
and
be globally Lipschitz, with
compactly supported. Fix a set of
output indices
.
Define
and the signed selected minor
Then
Assumptions
There are finite constants
such that, for all
,
The support convention is
,
and only this set must be compact. The fixed selection
consists of
distinct output coordinates in increasing order, so its existence
entails
.
Differentiability everywhere and a boundary condition are not
assumed.
Conclusion
The same signed minor
occurs in both terms, and their difference is integrable with integral
zero. No absolute value is placed around either determinant, and
separate whole-space integrability of the two minors is not claimed.
The coordinate projection
is contractive for the sup norms. Postcomposition
is a continuous linear map on operator spaces, preserves subtraction,
and has norm at most one. Its determinant equals the maximal minor of
the standard matrix in precisely the increasing row order.
Proof route
Smooth both maps, use one compact set for all small-parameter
perturbations, pass each restricted determinant integral to the limit,
and finally identify the whole-space difference.
Proof steps
The support lemma provides a compact
containing
and also
for every sufficiently small positive
.
Lipschitz maps are locally integrable, so mollification is linear:
.
These mollifications are
.
Apply the smooth compact-perturbation theorem to
and
;
output selection preserves compact support. The chain rule identifies
their derivatives with selected square operators and gives
Outside
,
is locally zero. The two maps then agree on a neighborhood, so their
derivatives, and thus their selected minors, agree there. The previous
integral therefore equals its restriction to
.
This argument only needs local equality; it does not require
differentiability of the unsmoothed maps at every point.
Apply strong local convergence of mollified derivatives to
and to
,
on this same compact
.
For
,
contractivity and linearity give
Continuous linear postcomposition preserves eventual
membership; the limiting derivative belongs to
by its Lipschitz bound. Thus all hypotheses of determinant-integral
continuity hold with measure restricted to
.
Its output is
for each of these two maps.
The smooth selected minors are integrable on compact
.
For the limiting maps
,
the bound
and
give the same integrability. Thus both terms may be integrated
separately on
.
Using Step 3 for each term,
This proves the restricted
difference integral is zero; it does not subtract unrestricted
whole-space integrals.
Outside
,
the original perturbation
is locally zero, so
there as well. The difference is integrable on
and equals zero outside; it is therefore integrable on the whole space
and has the same integral. This proves the asserted whole-space identity
without requiring mollification to preserve any boundary
values.
theorem integral_maximalMinor_fderiv_add_sub_eq_zero_of_lipschitzWith
{m N : ℕ} (s : Matrix.MaximalMinorIndex (m + 1) (Fin N))
{g u : (Fin (m + 1) → ℝ) → (Fin N → ℝ)}
{Cg Cu : ℝ≥0} (hg : LipschitzWith Cg g) (hu : LipschitzWith Cu u) (huc : HasCompactSupport u) :
∫ x, (maximalMinorIntegrand s (fun y => g y + u y) x -
maximalMinorIntegrand s g x) = 0
In the source
Mathematical meaning
{m N : ℕ} (s : Matrix.MaximalMinorIndex (m + 1) (Fin
N))
The domain dimension
and fixed increasingly ordered selection
of
output coordinates.
{g u : (Fin (m + 1) → ℝ) → (Fin N → ℝ)}
The base map
and perturbation
from
to
,
with sup norms.
{Cg Cu : ℝ≥0} (hg : LipschitzWith Cg g) (hu : LipschitzWith Cu
u)
The global bounds
and
for all
,
with nonnegative constants.
(huc : HasCompactSupport u)
The perturbation has compact support
.
In the source
Mathematical meaning
maximalMinorIntegrand s (fun y => g y + u y) x -
maximalMinorIntegrand s g x
The signed difference
,
where
and
.
∫ x, (maximalMinorIntegrand s (fun y => g y + u y) x -
maximalMinorIntegrand s g x) = 0
The whole-space Lebesgue integral of this difference equals zero.
The same increasing selection
occurs in both terms, without absolute values.