Immersion normal forms from split differentials #
A map from a real or complex normed space to a Banach space with a continuous left inverse to
its derivative has the local normal form of an immersion. The map need only be C^n on
an open neighbourhood, with 1 ≤ n. This turns a differential splitting into
actual slice charts, as needed to extend data from the image of an embedded map.
The construction adds a closed complement to the source and applies the inverse
function theorem to the resulting map. It uses Mathlib's
ContinuousLinearMap.HasLeftInverse.complement and
ContinuousLinearMap.coprodSubtypeLEquivOfIsCompl, and Tau Ceti's smooth inverse
function theorem on an open neighbourhood.
The same criterion holds for maps between manifolds without boundary. A normal form of a map read
in the preferred extended charts is carried back to the manifolds by composing its charts with the
preferred charts (Manifold.IsImmersionAtOfComplement.of_writtenInExtChartAt); on a boundaryless
model the manifold derivative is the derivative of the map read in these charts. So, for a
finite-dimensional target model, a smooth map with injective differential everywhere is an
immersion (ContMDiff.isImmersion_of_injective_mfderiv). A parameter-preserving map
(p, x) ↦ (p, G (p, x)) is then an immersion at every point where its slice x ↦ G (p, x) is
one in the sense of differentials, as for the space-time track of a moving embedded submanifold.
Likewise, postcomposing an immersion with a diffeomorphism onto a manifold
with a different finite-dimensional model gives an immersion (Diffeomorph.isImmersion_comp), for
example after exchanging the factors of a product.
In particular, a C^n map with a left inverse near a point, differentiable at the image of that
point, is an immersion there (ContMDiffOn.isImmersionAt_of_comp_eventuallyEq_id), and a C^n
map with a global left inverse differentiable along its image is an immersion
(ContMDiff.isImmersion_of_leftInverse). Pairing an immersion with any C^n map into a
finite-dimensional manifold gives an immersion (Manifold.IsImmersion.prodMk).
Reference: M. Hirsch, Differential Topology, Chapter 1, §3 (the local immersion theorem), and Chapter 8, §1 (its use in isotopy extension).
A C^n map on an open set has an immersion normal form at any point where its
derivative admits a continuous linear left inverse. Both spaces may be infinite dimensional.
For a C^n map on an open subset of a normed space with values in a Banach space, the
immersion normal form is equivalent to the derivative having a continuous linear left inverse.
For a finite-dimensional target, a C^n map on an open set has the immersion
normal form exactly when its derivative is injective.
An immersion normal form of a map between boundaryless manifolds, read in the preferred extended charts, is an immersion normal form of the map itself, with the same complement.
A C^n map on an open subset of a boundaryless manifold, with values in a boundaryless
manifold, is an immersion at any point where its manifold derivative admits a continuous linear
left inverse.
For a C^n map on an open subset of a boundaryless manifold, with values in a boundaryless
manifold, the immersion normal form is equivalent to the manifold derivative having a continuous
linear left inverse.
For a finite-dimensional target model, a C^n map on an open subset of a boundaryless
manifold, with values in a boundaryless manifold, has the immersion normal form exactly when its
manifold derivative is injective.
A C^n map between boundaryless manifolds, with a finite-dimensional target model, is an
immersion when its manifold derivative is injective at every point.
A C^n map on an open subset of a boundaryless manifold, with values in a boundaryless
manifold, is an immersion at any point x near which it has a left inverse g differentiable at
f x: the manifold derivative of g at f x is then a continuous linear left inverse of that of
f at x.
A C^n map between boundaryless manifolds, with 1 ≤ n, which has a left inverse
differentiable along its image is an immersion.
A C^n parameter-preserving map (p, x) ↦ (p, G (p, x)) between boundaryless manifolds is an
immersion at every point where its slice at the fixed parameter is an immersion in the sense of
differentials.
Postcomposing an immersion from a boundaryless manifold with a diffeomorphism of the target onto a boundaryless manifold gives an immersion, also when the new target has a different model, provided that model is finite dimensional.
Pairing an immersion from a boundaryless manifold with a C^n map, both with values in
boundaryless manifolds with finite-dimensional models, gives an immersion
x ↦ (f x, g x).