The analytic minimal rank Sard Conjecture

Let us briefly recall the main objects introduced in [4] – we equally rely on that after-mentioned paper for extended discussion on these objects.

Given an analytic totally nonholonomic distribution $\Delta$ on a real-analytic connected manifold $M$, we consider the nonzero annihilator of $\Delta$ in the cotagent bundle $T^{*}M$:

By [4, Theorem 1.1], there exists an open and dense subanalytic set $\mathcal{S}_{0}\subset\Delta^{\perp}$ called essential domain. Its complement, the set $\Sigma=\Delta^{\perp}\setminus\mathcal{S}_{0}$, is in fact a proper analytic subvariety of $\Delta^{\perp}$. Moreover, there exists a subanalytic distribution $\vec{\mathcal{K}}$ over $\Delta^{\perp}$ which is, in fact, a subanalytic isotropic foliation over $\mathcal{S}_{0}$; we denote the restriction of $\vec{\mathcal{K}}$ to $S_{0}$ by $\vec{\mathcal{K}}_{|S_{0}}$. We call $\vec{\mathcal{K}}$ the abnormal distribution because a horizontal path $\gamma:[0,1]\to M$ is singular if, and only if, it admits a lift $\xi:[0,1]\to\Delta^{\perp}$ which is horizontal with respect to $\vec{\mathcal{K}}$, see [4, Theorem 1.1(iii)].

In our main result, we will need to add an extra hypothesis on the foliation $\vec{\mathcal{K}}_{|S_{0}}$, which we call splittability. We postpone the precise definition to Section 1.3, and we anticipate in here that any line foliation is splittable. We can now state our main result:

The proof of Theorem 1.1 will consist in showing by contradiction that if the set of minimal rank singular horizontal paths from a given point reaches a set of positive Lebesgue measure in $M$, then we can, roughly speaking, lift all those horizontal paths into abnormal curves sitting in the leaves of the foliations given by $\vec{\mathcal{K}}$ on its essential domain and from here get a contradiction. This strategy requires to be able to select from a given set of positive measure contained in a transverse local section of the foliation $\vec{\mathcal{K}}_{|\mathcal{S}_{0}}$ a subset of positive measure whose all points belong to distinct leaves of $\vec{\mathcal{K}}_{|\mathcal{S}_{0}}$. A foliation subject to such a selection result will be called splittable; this is the heuristic behind the definition of splittability given in Section 1.3.

Thanks to [4, Theorem 1.1(iv)], moreover, we know that the rank of the abnormal distribution restricted to its essential domain $\vec{\mathcal{K}}_{|S_{0}}$ is less than or equal to $m-2$. Hence, the Minimal rank Sard conjecture holds true whenever $\Delta$ has rank $\leq 3$. Furthermore, the equivalence of the minimal rank Sard Conjecture with the Sard Conjecture in the case of corank-$1$ distributions yields the following immediate corollary:

In particular, the Sard Conjecture holds true when $n=4$ and $m=3$. This four dimensional result remains true in the $C^{\infty}$ category, as we show in our follow-up work [3], where we extend a few results (as much as we could) from this work to the smooth category.

Finally, our approach to the proof of Theorem 1.1 requires to lift the set of singular horizontal curves in $M$ to a subset of $\Delta^{\perp}$ of positive transverse volume with respect to $\vec{\mathcal{K}}$. As a consequence, we cannot prove the Sard conjecture for distribution of corank strictly greater than one yet. For an extended discussion, see [4, Section 1.4 and 1.5].

The proof of Theorem 1.1 follows from a combination of the description of abnormal lifts given in [4, Theorem 1.1]; differential geometry methods; and a new result on the existence of special transverse sections for singular analytic foliations, which we call witness transverse sections, see Theorem 3.2.

In fact, roughly speaking, we show that if $\mathcal{F}$ is a singular analytic foliation of generic corank $r$ in a real-analytic manifold $N$ equipped with a smooth Riemannian metric $g$, then we can construct locally, for every point $x$ in the singular set $\Sigma$ of $\mathcal{F}$, a special subanalytic set $X\subset V\setminus\Sigma$ where $V$ is an open neighborhood of $x$, called witness transverse section. This section has the property that its slices $X^{c}:=X\cap h^{-1}(c)$ ($c>0$) with respect to some nonnegative analytic function $h$ (verifying $\Sigma\cap V=\{h=0\}$) have dimension $\leq r$ with $r$-dimensional volume uniformly bounded (w.r.t $c$) and such that any point of $V$ can be connected to $X^{c}$ through a horizontal curve (w.r.t. $\mathcal{F}$) of length no greater than $\ell$ (w.r.t. $g$). We refer to Section 3 for further detail.

Let $N$ be a real-analytic manifold of dimension $n\geq 2$ equipped with a smooth Riemannian metric $h$ (not necessary assumed to be complete) and $\mathcal{F}$ a (regular) analytic foliation of constant rank $d\in[1,n-1]$. Given $\ell>0$, we say that two points $x$ and $y\in N$ are $(\mathcal{F},\ell)$-related if there exists a smooth path $\varphi:[0,1]\rightarrow N$ with length $\in[0,\ell]$ with respect to $h$ which is horizontal with respect to $\mathcal{F}$ and joins $x$ to $y$. Note that the $(\Delta,\ell)$-relation is not an equivalence relation, since it is not transitive. Moreover, given a point $\bar{x}\in N$, we call local transverse section at $\bar{x}$ any set $S\subset N$ containing $\bar{x}$ which is a smooth submanifold diffeomorphic to an open disc of dimension $n-d$ and transverse to the leaves of $\mathcal{F}$.

We provide in Section 2 a sufficient conditions for a foliation to be splittable. Indeed, we introduce the notion of foliation having locally horizontal balls with finite volume (with respect to the metric $h$ in $N$), see Definition 2.1, and we prove that this property implies the splittability, see Proposition 2.3. As a consequence, we infer that every line foliation is splittable, as well as every foliation whose leaves have Ricci curvatures uniformly bounded from below. In particular, all regular foliations in a compact manifold are splittable. An example of non-splittable analytic foliation in a non-compact manifold equipped with a smooth metric is presented in Section 2.2; we do not know if such examples do exist with an analytic metric.

The paper is organized as follows: in Section 2, we discuss the notion of spplitable foliations. In section 3 we construct Witness transverse sections for analytic foliations, see Theorem 3.2. Finally, the proof of Theorem 1.1 is given in Section 4.

The notion of splittable foliation and of $(\mathcal{F},\ell)$-related points have been provided in the Introduction, see Definition 1.3, and we follow its notation. Let us start our discussion by providing a more structured way to describe two $(\mathcal{F},\ell)$-related points. In fact, let us consider horizontal balls with respect to $\mathcal{F}$ and $h$. Given $x\in N$, we denote by $\mathcal{L}_{x}$ the leaf of $\mathcal{F}$ through $x$ in $N$. Then, for every $\ell>0$, we call horizontal ball with respect to $\mathcal{F}$ and $h$ the subset of $\mathcal{L}_{x}$ given by

We check easily that two points $x,y\in N$ are $(\mathcal{F},\ell)$-related if and only if $y\in\mathcal{L}_{x}^{\ell}$ (or $x\in\mathcal{L}_{y}^{\ell}$). Let us now introduce the following definition where $\mbox{vol}^{h,\mathcal{F}}(A)$ stands for the volume of a Borel set $A$ contained in a leaf of $\mathcal{F}$ with respect to the Riemannian metric induced by $h$ on that leaf:

The first example of foliations having locally horizontal balls with finite volume is given by foliations associated with complete Riemannian metrics. As a matter of fact, if $h$ is complete, then by the Hopf-Rinow Theorem, all balls $\mathcal{L}_{y}^{\ell}$ with $y$ close to $x$ are contained in the ball centered at $x$ with radius $\ell+1$ which happens to be compact, so all of those horizontal balls are compact sets with a volume which is finite and depends continuously upon $y$. Another example is given by foliations whose curvature satisfy a lower bound:

The proof of this result is left to the reader. We draw their attention to the fact that the comparison theorem required for the proof (of the second part) remains true for a non-complete metric (see e.g. [8, §4]). We end this section with the result that justifies the introduction of Definition 2.1 and provide many examples of splittable foliations.

As a consequence of Propositions 2.2 and 2.3, we get the following result:

We provide in Section 2.2 an example of analytic foliation ${\mathcal{F}}$ contained in an analytic manifold with boundary $M$, endowed with a (non-complete) $C^{\infty}$ metric $g$, which is non-splittable. This example illustrates the kind of qualitative behavior that we must exclude when studying the minimal rank Sard Conjecture. Nevertheless, note that the example is constructed on an abstract manifold. We do not know the answer to the following question:

If the answer to the above question is negative, then the hypothesis of Theorem 1.1 would always be satisfied provided that $M$ and $\Delta$ are analytic.

We modify a construction of Hirsch [11] in order to define a foliation which is non-splittable in a (non-compact) manifold with border $M$. As a matter of fact, Hirsch foliations are two-dimensional analytic foliations which satisfy the topological properties of a non-splittable foliation, but they lack the metric properties. In order to obtain the metric properties, we modify the original construction, and we make use of $C^{\infty}$-partitions of the unit to yield a $C^{\infty}$-metric.

We start by defining the building-blocks. Consider the double cover immersion $f:S^{1}\to S^{1}$ given by $f(t)=2t$, and choose an analytic embedding $\iota$ of the solid torus $S^{1}\times D^{2}$ onto its interior so that $\pi\circ\iota=f\circ\pi$, where $\pi:S^{1}\times D^{2}\to S^{1}$ is the projection. Let $V=S^{1}\times D^{2}\setminus\mbox{Int}(\iota(S^{1}\times D^{2}))$. Then the boundary of $V$ is two copies of $S^{1}\times S^{1}$, which we denote by $V^{-}$ and $V^{+}$ where $\iota(V^{-})=V^{+}$. Denote by $\mathcal{G}$ foliation over $V$ induced by the the fibration $\pi$. Note that the leaves of this foliation are topological pants, whose intersection with $V^{-}$ is a $S^{1}$, and whose intersection with $V^{+}$ is the disjoint union of two $S^{1}$, cf. figure 1(a).

Now, we consider a countable family of building-blocks $(V_{n},V_{n}^{-},V_{n}^{+},\iota_{n},\mathcal{G}_{n},g_{n})$, where $g_{n}$ are analytic metrics over $V_{n}$ satisfying the following property: given two points $x$ and $y$ in a leaf $L$ of $\mathcal{G}_{n}$, the distance of $x$ and $y$ in $L$ is bounded by $4^{-n}$. We denote by $M$ the manifold with boundary given by the union of all $V_{n}$, by identifying $V_{n}^{-}$ with $V_{n+1}^{+}$ via $\iota$, that is, we take the identification $x\in V_{n}^{-}$ equivalent to $\iota(x)\in V_{n+1}^{+}$. This yields an analytic manifold with border, where the border is a torus $M_{0}=V_{0}^{+}=S^{1}\times S^{1}$. This construction induces, furthermore, an analytic foliation $\mathcal{F}$ over the manifold with border $M$ which locally agrees with $\mathcal{G}_{n}$ over each $V_{n}$, because $\pi\circ f=\iota\circ\pi$, cf. figure 1(b). Furthermore, we can define a globally defined $C^{\infty}$ metric $g$ over $M$ by patching the metrics $g_{n}$ via partition of the unit. We may chose such a partition so that $g$ satisfies the following property: given two points $x$ and $y$ in a leaf $L$ of $\mathcal{G}_{n}$, the distance of $x$ and $y$ in $L$ is bounded by $2^{-n}$.

We claim that $\mathcal{F}$ is a non-splittable foliation. Indeed, consider a transverse section $\Sigma=S^{1}\subset M_{0}=S^{1}\times S^{1}$ and let us identify $\Sigma$ with the interval $[0,1]$. Given a point $x\in\Sigma$, denote by $L_{x}$ the leaf passing by $x$. First, consider the foliation $\mathcal{G}_{0}$, and note that, since $f(x)=f(x+1/2)$ and $\mathcal{G}_{0}$ is a foliation by pants, $x+1/2$ also belongs to the leaf $L$, cf. figure 1(b). Since this argument can be iterated over any $\mathcal{G}_{n}$, we get that all points $x+m/2^{n}$ with $m,n\in\mathbb{N}$ belong to $L_{x}$. Moreover, the distance on $L_{x}$ between $x$ and $x+m/2^{n}$ is bounded by:

since there exists a path between $x$ and $x+m/2^{n}$, contained in the leaf $L_{x}$, and which is contained in the union of $V_{k}$ with $k<n$, crossing each of these components at most twice. In conclusion, for every $x\in\Sigma$, the intersection of $L_{x}$, the leaf passing through $x$, with $\Sigma$ is a countable and dense set of points invariant by a countable subgroup of rotations, which are pairwise $(\mathcal{F},4)$-related. We infer that $\mathcal{F}$ is a not splittable in $(M,g)$ because there is no measurable set $E\subset\Sigma$ with positive Lebesgue measure whose intersection with each $L_{x}$ (with $x\in\Sigma$) contains only one point.