The soul conjecture in Alexandrov geometry in dimension 4 (Q2671891)

In [``Alexandrov's spaces with curvatures bounded from below II'', Preprint] \textit{G. Perelman} extended to Alexandrov geometry the soul construction by \textit{J. Cheeger} and \textit{D. Gromoll} [Ann. Math. (2) 96, 413--443 (1972; Zbl 0246.53049)]. He proved that any complete non-compact Alexandrov space \(X\) of non-negative curvature and finite dimension contains a compact totally convex subset \(S\), called a soul, which is an Alexandrov space of non-negative curvature with empty boundary and \(0 \leq \dim S < \dim X\). He also proved that \(S\) is a deformation retract of \(X\) via a distance non-increasing map \(\pi : X \to S\). The set \(S\) is constructed as the least element \(C_n\) of a maximal chain of compact, totally convex subsets of \(X\) \[ C_{n} \subset C_{n-1} \subset \dots \subset C_1 \subset C_0 \] in which \(C_{i+1}\) is the set of points of \(C_i\) at maximal distance from \(\partial C_i\) for \(i=0,\dots,n-1\) (here \(\partial C_i\) denotes the intrinsic boundary of \(C_i\) viewed as an Alexandrov space on its own) and \(C_0\) is the set of maximum points of some proper concave Busemann-type function \(f : X \to \mathbb{R}\). The chain must be finite since \(\dim C_i < \dim C_{i+1}\) for \(i=0,\dots,n-1\). The map \(\pi\) is defined as the limit for \(t\to+\infty\) of the flow \(\{\phi_t\}\) obtained by patching together the gradient flows of \(f\) and of the (concave) functions \(f_i = \mathrm{dist}(\,\cdot\,,\partial C_i)_{|C_i}\), according to a construction described in the Riemannian case by \textit{V. A. Sharafutdinov} [Mat. Zametki 26, 129--136 (1979; Zbl 0423.53031)]. Differently from the Riemannian case (where \(X\) also happens to be diffeomorphic to the total space of the normal bundle of \(S\)), in the Alexandrov setting the space \(X\) may not admit a structure of vector bundle over \(S\). In his resolution of the Soul Conjecture of Cheeger-Gromoll, \textit{G. Perelman} [J. Differ. Geom. 40, No. 1, 209--212 (1994; Zbl 0818.53056)] proved a structure theorem for \(\pi\) in case \(X\) is a Riemannian manifold. In particular he showed that \(\pi\) is a \(C^1\) Riemannian submersion (later, \textit{B. Wilking} [Geom. Funct. Anal. 17, No. 4, 1297--1320 (2007; Zbl 1139.53014)] showed that \(\pi\) is actually \(C^\infty\)) and a section of the normal exponential map \(\exp^\nu\) of \(S\), and proved the so-called flat strip property, from which the Soul Conjecture follows: if \(\gamma : [a,b] \to S\) is a minimizing geodesic and \(N\) is a parallel vector field along \(\gamma\) normal to \(S\), then the geodesics \([0,+\infty) \ni t \mapsto \exp^\nu(tN_{\gamma(s)})\), \(s\in[a,b]\), are all rays and if \(\gamma\) is not constant they span a flat strip in \(X\). He also raised the question whether a similar structure theorem for \(\pi\) also holds in the Alexandrov setting. It is not known whether \(\pi\) is more generally a submetry when \(X\) is an Alexandrov space, that is, whether \(\pi(B_r(x)) = B_r(\pi(x))\) for all \(x\in X\) and for all sufficiently small \(r>0\). As also explained in the paper under review, an affirmative answer to this question would imply that the Soul Conjecture holds as well in Alexandrov geometry in the following formulation: If \(X\) is a complete non-compact non-negatively curved Alexandrov space with strictly positive curvature on an open subset, then the soul of \(X\) is a single point. \textit{T. Shioya} and \textit{T. Yamaguchi} [J. Differ. Geom. 56, No. 1, 1--66 (2000; Zbl 1036.53028)] established a partial analogue of Perelman's flat strip property in Alexandrov geometry and showed that \(\pi\) is a submetry in the cases \(\dim S = 1\) and \(\dim S = \dim X - 1\), so that in particular \(\pi\) is always a submetry when \(\dim X = 3\). \textit{X. Li} [Trans. Am. Math. Soc. 367, No. 6, 3901--3928 (2015; Zbl 1326.53046)] proved that \(\pi\) is a submetry in case \(\dim S = \dim X - 2\) under the assumption that all iterated spaces of directions at any point of \(X\) are homeomorphic to spheres. For a general complete non-negatively curved Alexandrov space \(X\), Li also proved the following fact: letting \(\mathcal{F}\) be the union of rays \(\gamma_{v,p}\) issuing from points \(p\in S\) with the same initial direction \(v\) of a geodesic realizing the distance from \(p\) to a level set of the Busemann function \(f\) (call them special normal rays), the restriction \(\pi_{|\mathcal{F}} : \mathcal{F} \to S\) is a submetry. This happens because, as in the Riemannian case, special normal rays can be grouped into flat strips over minimizing geodesic arcs contained in \(S\). On the other hand, differently from the Riemannian case, these flat strips do not span the whole space in general (that is, it may be \(\mathcal{F}\subsetneq X\)) and so a key difficulty in showing that \(\pi\) is a submetry is to prove that \(\pi(B_r(x)) = B_r(\pi(x))\) also for points \(x\in X\setminus\mathcal{F}\). In the article under review the authors prove that \(\pi\) is a submetry when \(\dim X = 4\) and \(\dim S = 2\), with no further assumptions on the geometry of \(X\). From this result it follows that \(\pi\) is always a submetry when \(\dim X = 4\) and that the Soul Conjecture holds in Alexandrov geometry in dimension \(4\) (Theorem A). In the proof, the authors restrict themselves to consider points \(x\in X\setminus\mathcal{F}\) such that \(\mathrm{dist}(x,\mathcal F) = \mathrm{dist}(x,y)\) for some \(y\in\mathcal{F}\setminus S\), since such points are dense in \(X\setminus\mathcal{F}\). They prove that for a soul of codimension \(2\) the function \(\mathrm{dist}(\,\cdot\,,\mathcal{F})\) is concave on \(X\setminus\mathcal{F}\) (the proof is rather delicate and spans Sections 2--6 of the paper), and in case \(\dim X = 4\) and assuming \(X\) to be simply connected and locally orientable (a case to which the general one is reduced by passing to a suitable cover space) they use the gradient flow of \(\mathrm{dist}(\,\cdot\,,\mathcal{F})\) to map a subset of \(B_r(y)\cap\mathcal{F}\) onto a subset of \(B_r(x)\) which in turn gets mapped onto \(B_r(\pi(x))\) via \(\pi\).