Extending smooth data from an embedded submanifold #
A smooth embedding of a boundaryless manifold admits local smooth retractions.
In particular a smooth map which near a point takes values in the image of the embedding factors
there through the embedding by a map smooth at that point
(Manifold.IsImmersionAt.exists_contMDiffAt_comp_eventuallyEq), which does not need the source
to be boundaryless. Consequently,
vector-valued smooth data on its source extend to the ambient
manifold along any closed portion of its image. The ambient extension can be
supported inside a prescribed open neighbourhood, and has compact support when
the portion is compact. Neither the full embedded image nor the source needs to
be compact or closed.
The same holds for sections of a smooth vector bundle on the ambient manifold
given along the embedding, such as tangent vectors at the points of the image. Near
an image point, the section is moved to nearby fibres by reading it at the retracted
point in a local trivialization. The local extensions are glued by Mathlib's
exists_contMDiffSection_forall_mem_convex_of_local, with the convex constraint of
agreeing with the given section on the image portion and vanishing off the
prescribed neighbourhood. Here the embedding need only be a topological embedding
which is an immersion at the points of the portion.
These are the local-coordinate and partition-of-unity steps in extending an
isotopy's velocity to an ambient vector field. The local retraction uses
Mathlib's immersion normal form, projecting (u, v) to u; gluing uses
IsClosed.exists_contMDiffMap_eqOn. The source must be boundaryless because
that projection must land in an open source-chart target.
Reference: M. Hirsch, Differential Topology, Chapter 8, ยง1 (isotopy extension).
A topological embedding with a C^n immersion normal form at x admits a
local C^n retraction to its boundaryless source. Its left-inverse law holds for
every source point whose image is in the chosen neighbourhood, not just for one
source chart. No immersion or smoothness hypothesis is needed away from x.
A map g which is C^n at p, with g p = f x, and which near p takes values in the image
of a topological embedding f with a C^n immersion normal form at x, factors near p through
f by a map which is C^n at p and sends p to x: compose g with a local retraction.
Near the image of a point where a topological embedding has a C^n immersion normal form,
a C^n section of a vector bundle along the embedding extends to a C^n local section of the
bundle on an open neighbourhood, agreeing with it at every source point whose image lies there.
Extend smooth vector-valued data along a closed portion of an embedded boundaryless manifold, with topological support inside a prescribed open set. Closedness is required only of the portion of the image where agreement is requested.
Data on a compact portion of an embedded boundaryless manifold admit a compactly supported smooth ambient extension inside any prescribed open neighbourhood.
A C^n section of a vector bundle along a topological embedding of a boundaryless manifold
extends to a global C^n section of the bundle. The extension agrees with the given section on a
portion K of the source with closed image, at whose points the embedding is a C^n immersion,
and vanishes outside a prescribed open neighbourhood of that image.