Volume 2021, Issue 1 6974292
Research Article
Open Access

Neighborhoods and Manifolds of Immersed Curves

Andrea C. G. Mennucci

Corresponding Author

Andrea C. G. Mennucci

Scuola Normale Superiore, Pisa, Italy sns.it

Search for more papers by this author
First published: 31 December 2021
Academic Editor: Luca Vitagliano

Abstract

We present some fine properties of immersions ℐ : MN between manifolds, with particular attention to the case of immersed curves c : S1n. We present new results, as well as known results but with quantitative statements (that may be useful in numerical applications) regarding tubular coordinates, neighborhoods of immersed and freely immersed curve, and local unique representations of nearby such curves, possibly “up to reparameterization.” We present examples and counterexamples to support the significance of these results. Eventually, we provide a complete and detailed proof of a result first stated in a 1991-paper by Cervera, Mascaró, and Michor: the quotient of the freely immersed curves by the action of reparameterization is a smooth (infinite dimensional) manifold.

1. Introduction

In general, let M and N be smooth finite dimensional connected Hausdorff paracompact manifolds without boundary, with dim(M) ≤ dim(N).

This paper studies properties of immersions ℐ : MN that are C1 maps such that Tx is full rank at each x.

A particular but very interesting case are closed immersed curves c : S1n that are C1 maps with c(θ) ≠ 0 at all θS1, where S1 = {x2 : |x| = 1} be the circle in the plane. They will be called planar when n = 2.

This paper is mostly devoted to this case (a forthcoming paper [15] will generalize many results in this paper to the general case of immersions ℐ : MN).

Immersed planar curves c : S12 have been used in computer vision for decades; indeed, the boundary of an object in an image can be modeled as a closed embedded curve, by the Jordan Theorem. Possibly, the first occurrence was active contours, introduced by [9] and used for the segmentation problem: the idea is to minimize energy, defined on contours or curves, that contains an image-based edge attraction term and a smoothness term, which becomes large when the curve is irregular. An evolution is derived to minimize the energy based on principles from the calculus of variations. There have been many variations to original model of [9]; for example [6] and a survey in [3].

An unjustified feature of the model of [9] was that the evolution is dependent on the way the contour is parameterized. Thereafter, the authors of [4, 11] considered minimizing a geometric energy, which is a generalization of Euclidean arclength, defined on curves for the edge-detection problem. The authors derived the gradient descent flow in order to minimize the geometric energy.

This lead to a principle: all operations related to curves should be independent of the choice of parameterizations.

Operations on the space of curves are best described and studied if the whole space of curves is endowed with a differential structure, so that it becomes a smooth manifold.

The above two remarks lead to the following question. If I is the space of curves that we are interested in and D is the action of reparameterization, then the quotient
(1)
is the space of curves up to parameterization (also called geometric curves in the following): when (and how) can we say that this quotient B is a smooth manifold?

This was discussed in [16], using a result from [5].

A purpose of this paper is to revisit the key result in [5]: indeed the proof in that paper is missing two key steps.

1.1. Plan of the Paper

In Section 2, we will define the needed topologies on the space of functions; we will present well known definitions and notations for curves, such as derivation and integration in arc parameter, length, normal vectors, and curvature; we will classify immersed and freely immersed curves and present results and examples.

In Section 3, we will present advanced results for immersed curves; we will discuss representation of nearby curves in tubular coordinates; we will show how the open neighborhood of a curve c in the space of curves can be defined using tubular coordinates, so that if c is immersed (respectively, freely immersed) then all curves in the neighborhood are immersed (respectively, freely immersed); we will show with examples what goes wrong when hypotheses are not met.

In Section 4, we will present the proof of this theorem: the quotient of the freely immersed curves by the action of reparameterization is a smooth (infinite dimensional) manifold. We will then explain, in a step by step analysis, why the original proof in [5] was incorrect.

A supplemental file contains Wolfram Mathematica code to generate some of the figures.

2. Definitions

In this section, we will present well-known definitions and results regarding immersions, with particular attention to immersed curves.

2.1. Topologies

Definition 1. We denote by Cr(M, N) the space of all maps f : MN that are of class Cr. Here, r ∈ {0,1,2, …, }.

There are classically two types of topologies for this space.
  • (i)

    The weak topology, as defined in Ch. 1 Sec. 1 in [8], that coincides with the compact-open Cr-topology as defined in 41.9 in [12]; if N = n, then the “weak topology” is the topology of the Fréchet space of local uniform convergence of functions and their derivatives up to order r

  • (ii)

    The strong topology as defined Ch. 1 Sec. 1 in [8] coincides with the Whitney Cr-topology as defined in 41.10 in [12]

If M is compact, then the two above coincide; if moreover N = n and r < , then Cr(M, n) is the usual Banach space.

Remark 1. If N = n but M is not compact, then “strong topology” does not make Cr(M, n) a topological vector space since it has uncountably many connected components; but the connected component containing f ≡ 0 contains only compactly supported functions, and it has the topology (as defined in 6.9 in Rudin [18]) that is the strict inductive limit (for the definition of strict inductive limit and its properties, we refer to 17G at page 148 in [10]) of the immersions

(2)
where for each KM compact, VK is the space of f : MN that are zero outside of K, with a standard Banach (or Fréchet, for r = ) structure.

Proposition 1. The sets of immersions, submersions, and embeddings are open in Cr(M, N) with the strong topology, for r ≥ 1.

Proofs are in Ch. 1 Sec. 1 in [8].

Definition 2. For r ∈ {1, …, }, let Diffr(M) be the family of diffeomorphisms of M: all the maps ϕ : MM that are Cr and invertible, and the inverse ϕ−1 is Cr. It is a group, the group operation being “composition of functions.”

Proposition 2. Diffr(M) is open in Cr(M, M) with the strong topology.

See Thm. 1.7 in Ch. 1 Sec. 1 in [8].

We will omit the superscript “r” from Diffr(M) in the following, for ease of notation.

2.2. Immersions

2.2.1. Free Immersion

Definition 3. An immersion ℐ : MN is called “free” if ℐ ≡ ℐ∘ϕ for ϕ ∈ Diff(M) implies that ϕ is the identity.

Proposition 3 (see [5] Lemma 1.3.)If ℐ is immersed and ℐ(ϕ(t)) = ℐ(t) for all t and ϕ(t) = t for a t, then ϕ = Id.

Proof. Indeed, it is easily seen that

(3)
is closed; and it is also open, since an immersion is also a local diffeomorphism with its image.

As a corollary, if ℐ∘ϕ ≡ ℐ and ℐ∘ψ ≡ ℐ and ϕ(t) = ψ(t) for a t, then ϕψ. Another corollary states the following.

Corollary 1 (see [5] Lemma 1.4.)If ℐ is an immersion and there is a xn s.t. ℐ(t) = x for one and only one t, then ℐ is a free immersion.

This implies that, when dim(M) < dim(N), the free immersions are a dense subset of all immersions (for all the topologies considered in this paper).

2.2.2. Reparameterizations and Isotropy Group

We first consider the general case of immersions ℐ : MN.

Definition 4. The isotropy group (a.k.a. “stabilizer subgroup” or “little group”) is the set of all ϕ ∈ Diff(M) such that ℐ ≡ ℐ∘ϕ; it is a subgroup of Diff(M).

Obviously, ℐ is freely immersed if and only if contains only the identity.

We will prove that is discrete and finite when M is compact.

Remark 2. If we reparameterize , then changes by conjugation:

(4)

Remark 3. If M is orientable, then Diff(M) has a subgroup Diff+(M) of orientation preserving diffeomorphisms; for the case of curves, then we obtain that Diff(S1) has two connected components

(5)
where
  • (i)

    Diff+(S1) is the family of diffeomorphisms with ϕ > 0, and is a normal subgroup

  • (ii)

    Diff(S1) is the family of diffeomorphisms with ϕ < 0

Consider a curve c and let be its isotropy group: we will prove in Lemma 3 that if , then ϕ ∈ Diff+(S1).

We will mostly use Diff+(S1) in the following.

Note that Diff+(S1) is a perfect group [19] (see [13] for a self-contained presentation); it is also a simple group: see Discussion in Sec. 2 in [2] for further references. (The author thanks Prof. Kathryn Mann for her help on these subjects.)

2.3. Curves

Remember that S1 = {x2 : |x| = 1} is the circle in the plane. We will often associate 2 = , for convenience. In this case, we will associate S1 = {eit, t} ⊂ .

Definition 5. A closed curve is a map c : S1n. We will always assume that the curve is of class C1 (at least). The image of the curve, or trace of the curve, is c(S1).

When convenient, we will (equivalently) view S1 as /(2π) (that is, modulus 2π translations), and consequently a closed curve will be a map c : n that is 2π-periodic.

In particular, this will be the correct interpretation when we will write the operation θ1 + θ2 for θ1, θ2S1.

Remark 4. The “distance” of points in S1 will be the intrinsic distance; this distance will be represented by the notation:

(6)
for θ1, θ2S1, and it is the length of the shortest arc in S1 connecting the two points θ1, θ2. Note that if we identify S1 to /(2π) and pick two points θ1, θ2/(2π) and represent them as real numbers, it may happen that
(7)

Definition 6. (basepoint). We will select a distinguished point θ0 in the circle S1: for S12, it will be θ0 = (1,0); for S1, it will be θ0 = 1; for S1 = /(2π), it will be θ0 = 0.

Given a curve as above, we will call c(θ0) the basepoint for the curve.

Example 1 (of a nonfreely immersed curve). The doubly traversed circle defined as

  • (i)

    c2(z) = z2 for zS1 when we consider S1, or equivalently

  • (ii)

    c2(θ) = (cos(2θ),   sin(2θ)) for θ/(2π) that we identify with S1

Setting ϕ(t) = t + π, we have that c2 = c2ϕ, so c2 is not freely immersed.

Example 2 (taken from [5]). Note that there are free immersions without a point with only one preimage: consider a “figure eight” which consists of two touching circles. Now, we may map the circle to the figure eight by going first three times around the upper circle, then twice around the lower one. This immersion c : S12 is free.

We provide a simple Example 3 that shows how such curve can be made smooth.

2.3.1. Length, Tangent, and Curvatures

In the following, let c : S1n be an immersed curve.

Definition 7. If the curve c is immersed, we can define the derivation with respect to the arc parameter

(8)

We will write (/c) instead of (/s) when we are dealing with multiple curves, and we will want to specify which curve is used.

Definition 8. We define the tangent vector

(9)

Definition 9. The length of the curve c is

(10)

Definition 10. We define the integration by arc-parameter of a function g : S1n along the curve c by

(11)

There are two different definitions of curvature of an immersed curve: mean curvature H and signed curvature κ, which is defined when c is valued in 2.

H and k are extrinsic curvatures, they are properties of the embedding of c into n.

Definition 11 (H). If c is C2 regular and immersed, we can define the (mean) curvature H of c as

(12)

It is easy to prove that HT.

Definition 12 (N). When the curve c is planar, we can define a normal vector N to the curve, by requiring that |N| = 1, NT and N is rotated π/2 degree anticlockwise with respect to T.

Definition 13. (κ) If c is in 2 and C2, then we can define a signed scalar curvature κ = 〈H, N〉, so that

(13)

There is a choice of sign in the above two definitions; this choice agrees with the choice in [21].

When we will be dealing with multiple curves, we will specify the curve as a subscript, e.g., Tc, κc, Nc will be the tangent, curvature, and normal to the curve c.

Remark 5. Note that T, κ, N, H are geometrical quantities. If ψ ∈ Diff+(S1) and , then , , and .

2.3.2. Arc Parameter

Let c be an immersed planar curve. We recall this important transformation.

Lemma 1 (constant speed reparameterization). A curve cC1 can be reparameterized to using a φ ∈ Diff+(S1) so that where = len(c)/(2π) is constant.

Proof. For simplicity, we assume that S1 = [0,2π]. Let L = len(c), let . Then ψ : [0,2π]⟶[0,2π] is a diffeomorphism, let φ = ψ−1.

Reparameterization to constant speed is a smooth operation in the space of curves, see Theorem 7 in [20].

When |c| ≡ 1, we will say that the curve is by arc parameter. A curve can be reparameterized to arc parameter without changing its domain (as done above) iff len(c) = 2π (if this is not the case, we will rescale the curve to make it so.)

2.3.3. Angle Function and Rotation Index

Proposition 4 (angle function and rotation index). If cC1 is planar and is immersed, then T = c/|c| is continuous and |T| = 1, so there exists a continuous function α : satisfying

(14)
and α(s) is unique, up to adding the constant k2π with k. α is called the angle function.

Moreover α(s + 2π) − α(s) = 2πI, where I is an integer, known as rotation index of c. This number is unaltered if c is deformed by a smooth homotopy (Figure 1).

Details are in the caption following the image
Examples of curves of different rotation index.

See 2.1.4 in [1] or Theorem 53.1 in [17], and following.

Remark 6. We can use the angle function to compute the scalar curvature κ, that was defined in Definition 13 by sT = κN, indeed deriving (14) and combining this with

(15)
we obtain
(16)

2.4. Shapes

Shapes are usually considered to be geometric objects. Representing a curve using c : S1n forces a choice of parameterization that is not really part of the concept of “shape.”

Suppose that I is a space of immersed curves c : S1n.

Definition 14 (geometric curves). The quotient space B = I/Diff(S1) is the space of curves up to reparameterization, also called geometric curves in the following. Two parametric curves c1, c2M such that c1 = c2ϕ for a ϕ ∈ Diff(S1) are the same geometric curve inside B.

B is mathematically defined as the set B = {[c]} of all equivalence classes [c] of curves that are equal but for reparameterization,
(17)

We may also consider the quotient w.r.t Diff+(S1). The quotient space I/Diff+(S1) is the space of geometric-oriented curves.

Unfortunately, the quotient of immersed curves by reparameterizations is not a manifold; but the quotient of freely immersed curve is.

Theorem 1. Suppose that I is the space of the freely immersed curves; and that I and Diff(S1) have the topology of the Fréchet space of C functions, then the quotient B = I/Diff+(S1) is a smooth manifold modeled on C.

One aim of this paper will be to give a complete proof of this result, first presented in [5]. (We remark that the theorem in [5] was presented for the case of immersions ℐ : MN). The proof is in Section 4.1. Indeed, as we will discuss in Section 4.2, the proof in [5] misses some key arguments.

3. Advanced Properties of Immersed Curves

In this section, we will present results regarding immersed curves that are either new or presented in more precise form than usually found in the literature.

Most of the results are presented, for sake of simplicity, for planar curves c : S12, but can be extended to the case of curves c : S1N taking values in a manifold N, up to some nuisance in notations.

The general case of immersions ℐ : MN requires instead some arguments that will be discussed in a future paper [15].

3.1. Examples

Definition 15. We start with some classical examples of C functions of compact support. Let

(18)
(19)
(see Figure 2 on the following page).

Details are in the caption following the image
Graph of η0 as defined in (18); graph of η1 as defined in (19).

We will use these to build some following examples.

Example 3. We present here a simple smooth formula for Example 2

(20)
this is a C function depicted at Figure 3 on the next page.

Details are in the caption following the image
x and y coordinates of curve defined in Example 2; trace of curve defined in Example 2.

3.1.1. Trace and Parameterization

If a curve is embedded then the curve is identified by its image, in these senses.
  • (i)

    If are embedded and have the same image, then there is an unique reparameterization ϕ such that .

  • (ii)

    Suppose that c0 : S1n is embedded and A = c(S1) is the trace; suppose that c0 is parameterized by constant speed parameter; let us fix a candidate basepoint vA in the trace.

  • We can state that A, v characterize the embedded curve up to a choice of direction: precisely, there are exactly two different c1, c2 : S1n, parameterized by constant speed parameter, such that c1(θ0) = c2(θ0) = v, and they satisfy

(21)
for unique choices of a, bS1 (dependant on v).

In particular, if the rotation index of c0 is r, then the latter curves have rotation indexes ±r.

Since the definition of freely immersed curve says that the curve identifies a unique parameterization, then we may be induced to think that the above two properties extend to freely immersed curves: but this is not the case.

Example 4. The following two curves have the same trace, are freely immersed, are smooth, but have rotation indexes 0 and 1.

  • (1)

    This immersed closed curve c : [0,2π]⟶2 with components

    (22)

  • (2)

    This immersed closed curve c : [0,2π]⟶2 with components

(23)

See Figure 4 on the following page.

Details are in the caption following the image
Trace of the two curves and parameterization defined in Example 4; with color scheme used for the parameter.

3.2. Reparameterizations and Isotropy Group

Lemma 2. If ϕ ∈ Diff(S1), then ϕ has two fixed points.

Proof. We represent ϕ as a map ϕ : [0,2π]⟶ that is continuous, strictly decreasing and such that

(24)
then
(25)
so the graph y = ϕ(x) must intersect both the graph y = x and the graph y = x + 2π for two different points x1, x2 ∈ [0,2π) that are the two fixed points.

Lemma 3. If c is immersed and , then ψ ∈ Diff+(S1).

Proof. Suppose that ψ ∈ Diff(S1), let uS1 be a fixed point (by Lemma 2). By deriving

(26)
setting θ = u
(27)
and this is impossible since ψ(u) < 0.

3.3. Local Embedding

3.3.1. Length of Curve Arcs

Definition 16. Suppose c : S1n is C1. Let . When there are two arcs in S1 connecting σ to . By

(28)
we will mean the minimum of the lengths of c when restricted to one of the two arcs connecting σ to .

If c is periodically extended to c : n and , then there is an unique k such that
(29)
and then, letting
(30)
we define
(31)
In particular, when c is parameterized at constant speed (i.e., |c| ≡ ), then we will (covertly) assume that are chosen (up to adding k2π) so that and then
(32)

Remark 7. When c is not parameterized by constant velocity, the above may lead to some confusion. The interval in the notation (28) implicitly refers to the choice of arc in S1 that provides the above minimum. Note that this may not be the shortest arc connecting σ to in S1. This may happen if the parameterization of c has regions of fast and slow velocity, as in this example.

Example 5. Let be the standard circle, and

(33)
(see plot in Figure 5) then smooth out the corners of ϕ so that it becomes a diffeomorphism of S1; let
(34)
let θ0 = 0, θ1 = π/3 in S1 ~ /(2π), then , and is given by the arc moving counterclockwise from θ0 to θ1, while
(35)
is given by the arc moving clockwise from θ0 to θ1.

Details are in the caption following the image
Plot of function ϕ in equation (33) in Example 5.

This never happens for small distances/lengths, though.

Theorem 2. Fix an immersed curve c : S1n; let

(36)
  • (i)

    For any θ0, θ1 in S1 such that

    (37)

  • the shortest arc connecting them in S1 is also the arc where

    (38)

  • is computed

  • (ii)

    For any θ0, θ1 in S1 such that

    (39)

  • the arc where

    (40)

  • is computed is also the shortest arc connecting them in S1, whose length is

    (41)

  • (iii)

    In any of the above cases,

(42)

3.3.2. Estimates

We begin with this estimate.

Proposition 5. Let α(s) be the angle function, for sJ = [0,2π]. The fact that the curve is closed imposes lower bounds on maxJ|α|.

  • (i)

    If the rotation index ℐ of the curve is not zero, then |α(2π) − α(0)| = 2π|I| so necessarily maxJ|α| ≥ |I|

  • (ii)

    If the rotation index of the curve is zero, then necessarily maxJ|α| ≥ 1/2

Indeed, we can prove that

(43)
otherwise, let
(44)
taking v = (cos  β,   sin  β), we would have
(45)
for all θ, hence the curve would not be closed.

Definition 17. Given c : S12, a C2 immersed closed curve, we recall that κ is the scalar curvature of c; we define

(46)
Note that since the curve is closed, κ cannot be identically zero.

Note that δc = τc(2π/3), but we define two quantities since this simplifies the notation in the following. We have 2τcδc but 3τcδc.

Remark 8. Note that if we rescale the curve c by a factor λ, then δc, τc and are multiplied by λ as well. If we rotate or translate c, then δc, τc and are unaffected. If we reparameterize, then δc, τc are unchanged, whereas if ψ ∈ Diff(S1) and , we have

(47)

In all (with the exception of relation (100) in Lemma 7) following definitions, propositions, and theorems, the formulas are built to be “geometrical”: this means that, if the curves are reparameterized, rescaled, translated or rotated, then the formulas change in predictable ways (as explained above).

This simplifies the proofs: in the proofs we can assume, with no loss of generality, that the curve is parameterized by arc parameter.

Remark 9. Note that δc ≤ len(c)/3 for curves of index zero and δc ≤ len(c)/(6|I|) for curves of index I ≠ 0.

Proof. We use Proposition 5. The formula in the thesis is invariant for reparameterizations and scaling; we rescale the curve so that len(c) = 2π and reparameterize by arc parameter so that |c| ≡ 1. For curves of index zero, the thesis π/(3maxJ|κ|) ≤ len(c)/3, that is, π ≤ maxJ|κ|len(c); since |c| ≡ 1 by (16), this last becomes 1/2 ≤ maxJ|α| that was proved above. For curves of index I, the thesis π/(3maxJ|κ|) ≤ len(c)/(6|I|), that is, 2π|I| ≤ maxJ|κ|len(c), then becomes |I| ≤ maxJ|α| that was proved above.

For I ≠ 0 the above is sharp, as in the case of c(θ) = (cos(Iθ),   sin(Iθ)).

3.3.3. Local Embedding of Curves

It is well known that a C1 immersion ℐ : MN is a local embedding. For curves of class C2, we can provide a simple quantitative statement.

Proposition 6 (local embedding). Let c : S12 a C2 immersed curve. Define δc as in Definition 17. For any a, bS1, let ; assume that L ≤ 2δc, then |c(b) − c(a)| ≥ L/2; so, c|[a, b] is embedded.

Proof. For simplicity, we assume that c is periodically extended to c : 2; then, we identify the interval in that is associated to the arc of the curve where the length len c|[a, b] is computed; for simplicity, we call this interval [a, b] again. (if the arc is short enough, then by Theorem 2, no ambiguity is possible).

Using Lemma 1 and Remark 8, assume that |c| ≡ = len(c)/(2π); then, s = (1/)θ, so T = c/ and

(48)

As noted in (16)

(49)
so δc = π/(3max|α|). Let m = (a + b)/2 be the middle point. Let α(t) be the angle function (14). Up to rotation, suppose c(m) = (, 0), T(m) = (1,0); so, we can assume α(m) = 0. Let L = len c|[a, b] = (ba) ≤ 2δc. For any θ ∈ [a, b], we have |θm| ≤ δc; hence, |θm| ≤ π/(3max|α|); hence, for all aθb, we have
(50)
and hence, cos(α(θ)) ≥ 1/2; hence, for aθ1θ2b, for the abscissa, we can write
(51)

3.4. Isotropy Group Is Discrete

Given an immersion ℐ : MN, it is possible to prove that the isotropy group is discrete (when M is paracompact) and even finite (when M is compact; this latter result appears in [5]). When considering curves, we can obtain the same results (and even more) in a more direct and geometric way.

Lemma 4. Let c : S1n be immersed.

  • (i)

    is finite

  • (ii)

    If cψcϕ and ψ(a) = ϕ(a) for an aS1, then ψϕ

  • (iii)

    If c is parameterized by constant speed (see Lemma 1), then there is a k, k ≥ 1 s.t. is the set of all ϕ(t) = t + (2πj/k) for j = 0, …, k − 1 (the proof of this is a special case of the 2nd step of the proof of Lemma 9)

Proof.

  • (i)

    We prove the third point. Indeed deriving c = cϕ and noting that |c| ≡ , we obtain ϕ ≡ 1 so ϕ(t) = t + β; hence, , for all j; if β/π is irrational, then jβ would be dense in S1 = /(2π), and this is denied by Proposition 6. Moreover, if ϕ(t) = t + (2π/k) and c = cϕ, then c(0) = c(2π/k); but by Proposition 6, (2π)/(3max|κ|) ≤ (2π/k), that is, k ≤ 3max|κ|. So, there is an unique k such that any can be written as ϕ(t) = t + (2πj/k).

  • (ii)

    The above characterization shows that if ϕ ≡ 1 ≡ ψ and ϕ(a) = ψ(a), then ϕψ. By Remark 2, this is valid for any curve (even when it is not parameterized by constant speed). This proves the second point.

  • (iii)

    The first point follows again from Remark 2.

3.5. Tubular Neighborhoods

Existence of tubular neighborhood is well known; we provide a quantitative result for C2 planar immersed curves.

Proposition 7 (tubular neighborhood). Define δc, τc as in Definition 17. Fix a, bS1 with len c|[a, b] ≤ 2δc. Let

(52)
then Φ is a diffeomorphism with its image. Moreover, if the arc [s1, s2] is contained in the arc [a, b] identified above, then
(53)
whereas (obviously)
(54)

Proof. Assume that the curve has length 2π and is parameterized in arc parameter; with no loss of generality (recalling Remark 8); let α be the angle function (14).

Extend c to a periodic function c : 2 and identify the interval in that is associated with the arc of the curve where the length len c|[a, b] is computed. For simplicity, we call this interval [a, b] again (if the arc is short enough, then by Theorem 2, no ambiguity is possible).

The Jacobian of Φ is

(55)
so its determinant is (1 − κt) ≥ 1/2 by the hypothesis |t| ≤ τc.

We will then prove that Φ is injective, so it will be an homeomorphism with its image, and since it is a local diffeomorphism, it will be a diffeomorphism.

Choose (s1, t1) and (s2, t2) with as1 < s2b and |t1| ≤ τc, |t2| ≤ τc.

We set m = (s2 + s1)/2. Up to rotation, we assume that T(m) = e1 = (1,0) and α(m) = 0, so that N(m) is perpendicular to the x axis. As in Proposition 6, we can prove that cos(α(s)) ≥ 1/2 for all s1ss2.

We write

(56)
and then for the abscissa
(57)
note that Φ(m, t)1 = c(m)1. Deriving, we obtain
(58)

We then obtain that

(59)
while
(60)
and recalling that and summing, we obtain
(61)

We will call tubular coordinates around c the formula (52).

3.5.1. Counterexample

The hypothesis “cC2” in the previous proposition may be broadened to “cC1,1”; but the results fails if we only assume that “cC1,α” with α ∈ (0,1), as seen in this example (adapted from [1]).

Example 6. Let 0 < α < 1 and c(θ) = (θ, |θ|1+α); then, for θ > 0,

(62)
and this meets the y axes for
(63)

So, by symmetry,

(64)
and at the same time,
(65)

3.5.2. Nearby Points

Suppose c : S12 is a C2 immersed curve. Let 0 < ττc and Vτ = S1 × [−τ, τ] and
(66)
and Uτ = Φ(Vτ).

Proposition 8. Uτ is also the set of points at distance at most τ from the trace c(S1).

Proof. Let K = c(S1) be the trace of the curve (it is a compact subset of 2). We use the distance function dK : 2 defined as

(67)
(for an introduction to this object, see [14] and references therein).

Let xUτ, there is a θ, tVτ such that

(68)

So,

(69)
then let be a minimum for
(70)

So, clearly,

(71)

Vice versa if dK(x) ≤ τ let be a minimum as above, then geometrical considerations tell that the segment from x to is orthogonal to the tangent at .

As a corollary of Proposition 7, for any such neighborhood of the image of c, the “projection to c” is a C1 multivalued map (with finitely many projections in S1).

3.6. Not a Covering Map

By looking at the previous Proposition 7, we may think that Φ is the universal covering map of U = Φ(V) (see [17] for the definition). This would be very convenient, and indeed, we could use the lifting lemma to ease some of the following proofs.

Suppose are C2 immersed curves. Consider this statement, that is usually called lifting lemma:

If the trace of is contained in Uτ, then there is a choice of continuous φ : , a : ⟶[−τ, τ] such that
(72)

Unfortunately, this is not the case, as seen in this example in Figure 6, where the curve c is blue and the curve is red. The trace of the curve is all contained in the open set , but representation (72) cannot hold. We can though prove a version of the lifting lemma useful in the following.

Details are in the caption following the image
Figure for example in Section 3.6; the curve c is blue and the curve is red.

3.7. Neighborhoods

3.7.1. Nearby Projection

Lemma 5 (nearby projection). Fix a CR immersed curve c, with R ≥ 2.

  • (1)

    If x2 and and

    (73)

  • then there is an a with |a| ≤ d and a σS1 with

    (74)

  • such that

    (75)

  • Note also that a is uniquely identified by σ.

  • (2)

    They are unique in the family of σ, a such that |a| ≤ τc and

    (76)

  • so we can see σ, a as functions of x, as follows.

  • (3)

    Consider and for which

(77)
let ε > 0 small such that d + ε < δc/4 and let for convenience. There is a choice of function a, φ : B of class CR−1 such that
(78)
for all xB, and they are unique as specified above.

Note that τc/2 < δc/4 < τc.

Proof. Suppose c is by arc parameter (with no loss of generality as explained in Remark 8); so we write (recalling (32)) |ab| instead of len c|[a, b].

  • (i)

    Choose as in the statement and let , then consider any minimum point of

    (79)

  • note that the minimum value has to be less than d; so,

    (80)

  • but at the same time (since and are at arc distance at most δc), by the previous Proposition 6.

    (81)

  • So, combining the two

    (82)

  • but 4d < δc so is not at extremes. Then, any providing the minimum must be internal in the interval : by geometrical reasoning the segment from x to is orthogonal to the curve so there is a a such that

    (83)

  • (ii)

    Recall that δc/4 ≤ τc; the map Φ is injective for and |a| ≤ τc, so are unique.

  • (iii)

    For any xB, we have

    (84)
    and since d + ε < δc/4, then there is an unique and a with |a| ≤ τc such that
    (85)
    and we denote them by σ = φ(x), a = a(x). Moreover, we can invert the function
    (86)
    and write
    (87)
    for xB. This proves that φ, aC1(B).

3.7.2. Global Lifting

Proposition 9 (global lifting). Suppose c : S12 is a CR immersed curve and is CR−1, with R ≥ 2. Fix 0 < τ < δc/4. Suppose that we have for all θ. There exists choice of a : S1 and φ : S1S1 such that

(88)
with |a(σ)| ≤ τ and
(89)
holding for all σ. And, they are unique in the class of CR−1 functions such that |a| ≤ τc and
(90)

Note also that a is uniquely identified by φ.

Proof. We just substitute in the previous Lemma. By the second point, we can define functions φ(θ) and a(θ) uniquely as prescribed. By the third point, they are C1.

Remark 10. Suppose we are given two curves c1, c2 and we know that there exists a choice (a, φ) such that

(91)
as in (88). If we rotate or translate the two curves, then the above relation will hold, with the same (a, φ). If we rescale the two curves by λ > 0, then the relation will hold with (λa, φ).

If we choose ψ ∈ Diff+(S1) and we reparameterize all curves at the same time by , then

(92)
holds for
(93)

This follows from direct computation and Remark 5.

Example 7. So far so good, but φ may fail to be a diffeomorphism, as in this simple example in Figure 7. where the curve c is blue and the curve is red.

But some simple lemmas can help.

Details are in the caption following the image
Curves in Example 7; curve c is blue and curve is red.
Details are in the caption following the image
Scheme for Lemma 6.

Lemma 6. Let 0 < α < 1. If we have w, vn such that

(94)
then the angle β between v and w satisfies |β| ≤   arcsin(α), and moreover,
(95)
See Figure 8 on the next page.

Remark 11. Let α > 0. Suppose c1, c2, c3 : S12 are C1 maps and

(96)
if ψ ∈ Diff+(S1) and we reparameterize all curves at the same time by , then
(97)

Similarly, if we rescale, rotate, or translate all curves at the same time.

So, (96) is a geometric estimate, indeed we may rewrite it as
(98)
where .

Lemma 7. Assume all hypotheses in Proposition 9. Assume moreover that ττc/4 and assume that

(99)
then is immersed and φ is a diffeomorphism. Moreover, when c is parameterized at constant speed, we can state that
(100)

Proof. We rescale and reparameterize c to arc parameter using a reparameterization ψ, and at the same time, we rescale and reparameterize using the same rescaling and ψ (note that is not necessarily by arc parameter); with no loss in generality, as explained in Remarks 10 and 11.

Let β be the angle between and c(σ): by Lemma 6, β = arcsin(α), so it is at most π/6. Let θ = φ(σ); we know that |σθ| ≤ 4ττc; the angle γ between c(σ) and c(θ) is at most |σθ|max|k| so γ ≤ 1/2: so the angle β + γ between and is at most 1/2 + π/6, and this is less than π/2.

Deriving in σ and assuming that c is by arc parameter

(101)
where T, N, κ are evaluated at (φ(σ)); then,
(102)
now if |a| ≤ τc then (1 − κa) ≥ 1/2; moreover, by the above reasoning , so φ(σ) > 0. Moreover, we note that , 1/2 ≤ (1 − κa) ≤ 3/2 and cos(1/2 + π/6) ≤ 6/10 to prove (100). Relation (93) tells then that φ will always be a diffeomorphism, for any curve satisfying the hypotheses.

We summarize all the above: we show sufficient hypothesis such that may be represented in tubular coordinates around c.

Theorem 3 (representation theorem). Suppose c : S12 is a CR immersed curve and is CR−1, with R ≥ 2. Define δc, τc as in Definition 17. Fix 0 < ττc/4. Suppose that we have and

(103)
for all θ.

Then, is immersed, there are φ ∈ Diff+(S1) and a : S1⟶[−τ, τ] of class CR−1 such that

(104)
with |a(σ)| ≤ τ and
(105)
holding for all σ.

They are unique in the class of C1 functions such that |a| ≤ τc and

(106)

Note also that a is uniquely identified by φ.

Proof. We can rescale and reparameterize c to arc parameter, and we rescale and reparameterize at the same time ( will not be by arc parameter in general); as discussed in Remarks 8, 10, and 11, the hypotheses and theses are unaffected by this action. Then we apply all previous results. Just note that

(107)
for any diffeomorphism.

Remark 12. Actually, rerunning on the above proofs with some patience, we can improve the above thesis a bit. We add to the previous theorem these hypotheses: fix 0 < α ≤ 1/2 and then 0 < τατc/2 and suppose that we have and

(108)
for all θ.

Then, all above thesis hold, moreover, there are two continuous functions f, g : [0, ]⟶ (independent on α, τ) with f(0) = g(0) = 1, such that

(109)

3.7.3. Asymmetry

Warning: the previous theorem seems symmetric, but it is not. The caveat is in the constants τc, δc: it may be the case that they are quite different from . In Figure 9, we see a piece of the two curves: curve c is blue and is flat; curve is red, and it has two inflections points A, B where the tangents are at an angle β which is as small as we would like, but then the inflection points can also be so close that the normals will cross before reaching curve c. So, while there is an easy way of representing using tubular coordinates around c, there is no way to find so as to write
(110)
Details are in the caption following the image
Curves for the example in Section 3.7.3; curve c is blue and curve is red.

3.7.4. Vice Versa

We have also a sort of vice versa of the previous Theorem 3.

Proposition 10 (derepresentation). Suppose c : S12 is a C2 immersed curve and c1, c2 : S12 are given by tubular coordinates

(111)
where i = 1,2 and a1, a2 : S1 are of class C1; define
(112)
(where κ is the curvature of c). Then,
(113)
and (obviously) , for all θ.

Proof. We rescale and reparameterize c by arc parameter, and we rescale and reparameterize c1, c2 along with c, as explained in Remark 11; this operation is justified by Remarks 8 and 10; in particular, note that α, β are scale invariant; then,

(114)
so
(115)

Hence,

(116)

3.7.5. Loss of Regularity

Unfortunately, the representation discussed above suffers from a loss of regularity. Indeed, if c : S12 is C2 and a : S1 is C, it may be the case that
(117)
is of class C1 but not of class C2.

This can be seen in very simple examples.

Example 8. Suppose that, for t near t = 0, we have

(118)
such a curve is C2 but not C3; then, for t ≥ 0,
(119)
(that can be easily computed using a standard formula for curvature of planar curves, see Sec. 1.7.1 in [21]).

Choose then a ≡ 1, so

(120)
but then
(121)
and for t > 0,
(122)

So,

(123)
but for t < 0.

3.7.6. Neighborhoods of a Curve

The above results encode two different but equivalent ways to define a topology on the “manifold of immersed curves.” We specify them by describing the local bases of neighborhoods of a curve c.
  • (i)

    The “Banach way” in which a local base of open neighborhoods of a curve c is given by the sets

    (124)

  • where ε1 > 0 is small, and

    (125)

  • (ii)

    The “geometric way” in which a neighborhood in the local base is defined, for ε2 > 0 small, as the set of all that can be expressed as in (88), namely,

    (126)

  • for all choices of a : S1 and ϕ ∈ Diff(S1) with

    (127)

  • where

    (128)
    moreover derivatives of a may be computed in arc-parameter.

The above are “equivalent” in this sense. Assume that the curve c is CR+1.
  • (i)

    For any ε1 that defines neighborhood U of the first type, there is small enough ε2 that defines a neighborhood V of the second type, so that VU; this is easily proved (by using Leibnitz and Faa di Bruno formulas).

  • (ii)

    Consider now a neighborhood V of the second type, for an ε2 > 0 small; for ε1 small enough, the previous results Theorem 3 and Remark 12 tells us that any curve can be expressed in tubular coordinates; since tubular coordinates are a local diffeomorphism, similar arguments as above (plus Theorem 2) show that (for ε1 even smaller) UV.

We skip details for sake of brevity.

In all the above there is, however, an annoying condition: to prove equivalence of CR neighborhoods, we have to assume that the curve c is CR+1. For this reason, this works well for defining topologies in the “manifold of smooth immersed curves”; in this case, we will use neighborhoods of the first kind (or, respectively, of the second kind) for all ε > 0 and all R. This is the common approach, see [12].

3.7.7. Local Injectivity

So far, we have considered parametric curves. We have seen in Theorem 3 that we can represent nearby curves in a unique way using tubular coordinates, i.e., the map Φ.

What happens when we consider geometric curves, that is, curves up to parameterization?

Lemma 8 (local injectivity). Let c be a C2 freely immersed planar curve. There exists a r = rc > 0 such that, if

(129)
with
(130)
where (/c) is the arc derivation: then ab and φ is the identity.

Proof. If rescale the curves and the functions a, b by a constant λ > 0, and we rescale the constant rc by the same constant λ, then all hypotheses and theses are unaffected. So, we can assume with no loss of generality that c has length 2π.

If we reparameterize then (cf the relation (93) in Remark 10) the functions a, b are reparameterized as well; having , then ; and similarly for b, again hypotheses and theses are unaffected.

So, we can assume that c is parameterized by arc parameter with no loss of generality.

Suppose that

(131)
with |a| ≤ τc and |a| ≤ 1/2, then
(132)

So,

(133)

Suppose moreover,

(134)
with |b| ≤ 1/2, then
(135)
and then
(136)

By contradiction, we may write

(137)
with
(138)
where φn is not the identity: then, when 1/n < τc/4, the uniqueness condition (90) is contradicted, so there is a σn such that
(139)

So, using Theorem 2,

(140)

Up to a subsequence, we can assume that and

(141)

We know that

(142)
up to a subsequence φnφ uniformly and uniformly, where φ is a bi-Lip homeomorphism. Then, uniformly, and passing to the limit
(143)
so φ is a diffeomorphism. Moreover,
(144)
so φ cannot be the identity; hence, c is not freely immersed.

3.7.8. Auto Representation (This Section May Be Skipped on a First Read, Since It Is Not Needed in the following)

The above result is very important, but the proof gives no hint on what is going on. To this end, we drop the requirement that the curve be freely immersed and look at an easy question. Is it possible for a curve to represent itself locally?

Lemma 9. Let c be a C2 immersed planar curve. There are only finitely many ways in which the curve can represent itself geometrically, that is, finitely many choices a, ϕ

(145)
with ϕ ∈ Diff(S1) and a : S1 continuous with ‖aτc/2. In particular, there is a ρc > 0 such that, if ‖aρc then a ≡ 0.

In particular, if c is freely immersed, then ‖aρc implies a ≡ 0 and ϕ ≡ Id.

Proof. Define δc, τc as in Definition 17 (see also Proposition 7). Let Φ(s, t) = c(s) + tN(s). Consider the family P of all the pairs a, φ with ‖aτc/2 and a≢0 and

(146)
we will prove that there are only finitely many such pairs.

Hence, we will let Pc be smaller than the minimum of ‖a for all such pairs:

(147)

In the example, in Figure 10, there are 3 pairs in P.

We have some very strong properties.

  • (i)

    If (a1, φ1), (a2, φ2) ∈ P and there is a s.t. , then φ1φ2 and a1a2. Indeed, there is a small interval J containing , where we can invert the map Φ and

    (148)

  • that is, the first component of Φ−1c is , so φ1φ2 in J; so, the set {sS1 : φ1(s) = φ2(s)} is both open and closed. The previous argument also proves that a1a2.

  • (ii)

    Figure 11 on the following page can be used as a visual guide in the following proof.

  • Let I0S1 be an open interval such that the length of is less than δc and more than δc/2. As noted in Remark 9, δc ≤ len(c)/2. Let U0 be the image of Φ for sI0 and |t| < τc. Choose (ai, φi) ∈ P with i ∈ 1,2, let

    (149)

  • we will prove that either I1I2 = ∅, or I1 = I2, φ1φ2, and a1a2. Assume that , let , then s1, s2I0; so,

    (150)

  • using the relation (146),

    (151)

  • and the fact that Φ is a diffeomorphism for sI0, |t| < τc and , we obtain that so ; hence, by the previous point φ1φ2, a1a2.

  • (iii)

    The differential of Φ (computed using arc-length derivative) is

    (152)
    so the smallest principal value is at least 1/2. We can estimate the length of c for sIi to be at least δc/4. Hence, there can be only finitely many such intervals.

Details are in the caption following the image
Example of curve with 3 pairs in P.
Details are in the caption following the image
Helper scheme for proof of Lemma 9.

Example 9. The constants ρc in Lemma 9 and rc in Lemma 8 though cannot be estimated apriori by using differential quantities such as max|κ|. These constants may arbitrarily smaller than the quantity τc (defined in Definition 17) that provides the width of the tubular neighborhood (Proposition 7). They really depend on how the curve is drawn.

This is seen in simple examples as follows:

  • (i)

    c(z) = εz + z2 for zS1, or equivalently c(θ) = (ε  cos(θ) + cos(2θ), ε  sin(θ) + sin(2θ)) for θ/(2π) ~ S1

that are small C perturbations of the doubly traversed circle c2 seen in Example 1; see Figure 12. This curve is freely immersed, but it is quite near to the doubly traversed circle that is not freely immersed. For ε small, the curvature κc of c is approximatively 1; so, τc ~ 1/2, and
(153)
while
(154)

Hence,

(155)
so we can use Theorem 3 to express
(156)
using tubular coordinates when ε < 1/8, with a ~ ε: so ; similarly, we can represent c(θ + π) using c, so we have .

Details are in the caption following the image
Curve in the Example 9.

3.8. Free Immersions Are Open

We have thus come to a fundamental result.

Theorem 4. Free immersions are an open subset of immersions.

We can detail and prove this fact in two ways.
  • (i)

    We can see it as a corollary of Lemma 8. Consider a neighborhood V defined using the tubular coordinates; precisely, define V as in the second definition in Section 3.7.6, choosing ε2 < min{rc, 1/2, τc/4} and R = 2; knowing that , we could choose ab, but then by Lemma 8, φ would be the identity. So, each and any curve in V is freely immersed.

  • As discussed in Section 3.7.6, this proves the result in the manifold of smooth immersed curves, where the above neighborhoods define a topology.

  • (ii)

    If we instead want to prove this for the standard Banach C2 topology (first definition in Section 3.7.6), we can proceed as follows. Suppose that cn is a sequence of immersed curves that are not free; and suppose that cnc in C2. We may rescale and reparametrize all the curves so that all have length 2π and , and still cnc in C2; we skip the details (see Theorem 7 in [20]); these assumptions simplify the following arguments. Let ϕn be a sequence such that cn = cnϕn and ϕn is not the identity; as above, we know that and we may choose ϕn to be a generator of the isotropy group, so that ϕn(t) = t + (2π)/jn; we know that jn is bounded by 4 times the curvature, so up to a subsequence jn is constant, let us call it j, and then ϕn = ϕ with ϕ(t) = t + (2π)/j, and passing to the limits cϕ = c, so c is not freely immersed.

Note that both proofs need a compactness argument; this seems unavoidable since the size of the neighborhood cannot be estimated by using differentiable quantities, as explained in Example 9.

4. The Manifold of Free Geometric Curves

Definition 18 (classes of curves).

  • (i)

    Imm(S1, n) is the class of immersed curves c: curves such that c ≠ 0 at all points

  • (ii)

    Immf(S1, n) is the class of freely immersed curve, the immersed c such that, moreover, if ϕ : S1S1 is a diffeomorphism and c(ϕ(t)) = c(t) for all t, then ϕ = Id

  • (iii)

    Emb(S1, n) are the embedded curves, maps c that are diffeomorphic onto their image c(S1); and the image is an embedded submanifold of n of dimension 1

Each class contains the one following it (this follows from the propositions seen in Section 2.3).

4.1. Proof of Theorem 1

Definition 19.

(157)
is the quotient of Immf(S1, 2) (free immersions) by the positive diffeomorphisms Diff(S1) (reparameterization).

We now provide the complete proof of Theorem 1, namely, that this Bi,f is a manifold, for the case of smooth freely immersed planar curves; afterward, we will show in Section 4.2 how and where the proof in [5] misses some key arguments.

The following proof is for immersed curves c : S12, in a forthcoming paper [15], we will explain how it can be generalized to the case of immersions ℐ : MN.

4.1.1. Quotient Topology

We discuss the topological aspect of Theorem 1.

Let π : ImmfBi,f be the canonical projection of the positive quotient that defines Bi,f in (157).

The definition of the quotient topology is as follows. A set is open in Bi,f when the union of its orbits
(158)
is open in Immf, that is, it is open in C.

We endow Immf with the C topology described earlier and Bi,f with the induced quotient topology.

Now, we want to describe a specific family of open neighborhoods that will be quite useful. Fix c1 smooth freely immersed. Let , where rc was defined in Lemma 8.

Proposition 11. Consider the set

(159)
and this set is open in C(S12).

Proof. This is a simple case of the arguments of Section 3.7.6. Let , let and

(160)
we know that α < τ, β < 1/3; if c3 is a smooth curve and satisfies
(161)
then by the results in Section 3.

By Theorem 4, all curves in are freely immersed.

Now, let us reparameterize all the curves in and define
(162)
since the above conditions are reparameterization invariant, then
(163)
that is an union of open sets; hence, it is open in C. Moreover, it contains all the orbits of all of its curves: in the language of [5], we may say that “ is saturated for the action of Diff +(S1).”
So, we define
(164)
and we have
(165)

Hence, is open in Bi,f.

4.1.2. Geometric Representation

We discuss the representation aspect of Theorem 1. Consider again the set defined in (162). For any curve in this set, by Theorem 3, we have a representation
(166)
with |a| ≤ τ.
Let for convenience; by the derepresentation result Theorem 10 (setting α = 1/8, β = 1/3 in that proposition and noting ), we have
(167)
so there is an unique reparameterization of that can be expressed in tubular coordinates around c; by Lemma 8, this means that φ is uniquely identified by ; so, we will concentrate on a.

Proposition 12. Let

(168)
be the set of all such a. This set is open.

Proof. The map (φ, a) ↦ (c + aN)∘φ−1 is smooth, and is the projection on the second component of the counterimage of that is open.

All of the above can be stated in the language of [5] as follows: the set is an open neighborhood of c1 in C, it is composed only of freely immersed curves, it is saturated for the Diff +(S1)-action and the map
(169)
splits it smoothly as
(170)

4.1.3. Charts

Choose a curve c1; consider the map (it is not the same map Φ defined in Proposition 7)
(171)
and we already proved in Lemma 8 that it is injective; is also smooth as a map from C to C.
If we compose
(172)
then the composition
(173)
is bijective: indeed, if , then we proved in Section 3 that, picking a in the equivalence class C1,
(174)
for an unique φ, and then
(175)
for a unique a.

4.1.4. Atlas

To conclude, we discuss the Atlas of charts needed for Theorem 1. For i = 1,2, consider now two equivalence classes CiBi,f and choose a curve ciCi in each; we consider the maps
(176)

We want to check that these are charts of an Atlas for the manifold.

Suppose that
(177)
then
(178)
and we need to check that
(179)
is smooth in a neighborhood of . We can change variable in the previous one, that is, reparameterize c1, c2, so that
(180)
but for a1 near we know that
(181)
by Lemma 10, so by the representation Theorem 3, there are φ and a2 dependent on a1 such that
(182)

The representation theorem’s proof shows that the dependency of a2 on a1 is smooth (it is given by the inverse of the tubular coordinates, as discussed in the nearby projection Lemma 5).

This concludes the desired proof of Theorem 1.

4.2. Comparison with [5]

We endow N with a Riemannian metric such that scalar curvatures are bounded and convexity radius is bounded uniformly away from zero; see [7].

The proof in [5] is presented for generic immersions ℐ : MN; we fix one such immersion.

We follow notations and definitions in [5], here copied for convenience of the user (parts copied from [5] will be in italic, and enclosed in ≪…≫).

≪We choose connected open sets and such that , is an open cover of M, each is compact, is a locally finite open cover of M, and such that is an embedding.≫

≪Let be the normal bundle of ℐ, defined in the following way: for xM, let
(183)
that is, the orthogonal complement of tangent at ℐ(x) of the immersed manifold in Tℐ(x)N.≫

≪The following diagramimage

is a vector bundle homomorphism over ℐ, which is fiberwise injective.≫

≪Let expN : TNN be the exponential map on N. Now, there is a neighbourhood U of the zero section in the previous bundle that is small enough so that
(184)
when restricted to U|Uα is a diffeomorphism with its image. The restriction of to U is called τ. It will serve us as a substitute for a tubular neighborhood of ℐ(M).≫

The notation U|Uα is not described in [5], but by its usage it should be equivalent to Up−1(Uα). Note also that, later on, the paper adds the superscript ℐ to Uα and will write it as .

Consider planar immersed curves c : S12: normal vectors at c(θ) are a one dimensional space tNc (where Nc is the normal vector to the curve c as defined in Definition 12); hence, the fibre of is one dimensional, so a point in the bundle can be represented by a pair θS1, t, and the map τ becomes the map Φ defined in (52) in Proposition 7. Fix a τ > 0, small, as we will discuss later on. We cover S1 by arcs UαS1 each shorter than ε, and WαUα subarcs that can be chosen so that they are an open cover; for small ε we can ensure that tubular coordinates can be perused. The open set U will include normal vectors tNc with |t| < ε.

The statement of main Theorem 1.5 in [5] starts as follows.

Let ℐ be a free immersion MN. Then, there is an open neighborhood in Imm (M, N) which is saturated for the Diff (M)-action and which splits smoothly as
(185)

Here, is a smooth splitting submanifold of Imm (M, N), diffeomorphic to an open neighborhood of 0 in C(N(ℐ)). In particular, the space Immf(M, N) is open in C(M, N).≫

The proof covers also the case when M is not compact; we will assume that M is compact so that some arguments can be simplified.

The proof goes as follows.

Define
(186)
(the proof then goes on showing that this is an open set—we skip details).
For each jU(i), we define
(187)

Indeed, we know that is a diffeomorphism onto its image, and that when , by definition of .

Then is a mapping which is bijective onto the open set
(188)
in . Its inverse is given by the smooth mapping
(189)

The proof then goes on showing that this φ is smooth (we skip details).

We now translate the above objects into the language of Section 4.1. If c = ℐ : S12 is a freely immersed curve, and is a curve such that
(190)
then ; so, for ε small, for each σS1, there are θS1, t such that
(191)
and this pair θS1, t is exactly associated to correct point in the above bundle, that is, we can write
(192)
Possibly reducing the width ε of the tubular neighborhoods, we can also use Proposition 9 to ensure that the above representation is “unique.” So, Proposition 9 can be applied, and this means that, for each σ, we can write
(193)
so, in conclusion, we can explicitly write the map h above defined as
(194)
where the first component in h encodes a position in the base space M = S1 of the bundle , the second encodes a normal vector aNc to the curve c at φ(θ). Indeed, the inverse τ is exactly the map (193).

The proof continues as follows.

We have for those f ∈ Diff(M) which are near enough to the identity so that (that was defined in (188)). We consider now the open set (we added the notation for ease of reference)
(195)
Obviously, we have a smooth mapping from it into
(196)
given by
(197)
where C(U) is the space of sections of UiM.≫
Here comes the first mistake in the original proof. Consider Example 8, where c = ℐ is the blue curve and is the red curve. There is a choice of Uα, Wα such that . We can express
(198)
and then define h by (194); but when we apply the rule (197) to this h, we obtain that
(199)
and φ is a map φ : S1S1 that is not a diffeomorphism.
The correct statement is that map (197) achieves a splitting of the open set described in (195) into
(200)

Some conditions must be added to the definition of to make sure that φ is a diffeomorphism; as was done in (159) to define , by adding the condition : this condition is necessary to apply the representation Theorem 4. No similar condition is present in the proof in [5].

The proof afterwards proceeds as follows.

So, if we let
(201)
We have
(202)
since the action of Diff (M) on (ℐ) is free. Consequently, Diff (M) acts freely on each immersion in , so Immf(M, N) is open in C(M, N).

(this is not the same as the defined in (168), but it has the same scope; the latter one is an open neighborhood of 0 in )

This is the second mistake in the proof.

Even if we restrict the open set described in (195) by adding a first order condition, so that the map (197) properly splits , we have not guarantee that all curves in the associated neighborhood are freely immersed. This is shown in Example 9.

This is why, in our proof, we added the condition τrc to the definition of .

We have shown that the proof in [5] does not prove the desired result.

Conflicts of Interest

The author declares no conflicts of interest.

Acknowledgments

The author thanks Prof. Kathryn Mann for her help on the properties of Diff+(S1).

    Data Availability

    No data were used to support this study.

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.