Kinematic formulas for sets defined by differences of convex functions (Q2400528)

The present review reproduces the articles' introduction because it is too technical to write a shorter overview. The classical principal kinematic formula (PKF) expresses, in terms of geometric quantities (intrinsic volumes) associated separately to compact subsets \(A<B\subset\mathbb{R}^{d}\), the integral of the Euler characteristic of the intersection \(A\cap\gamma B\) over all \(\gamma\) belonging to the group \(\overline{SO(d)}=SO(d)\ltimes\mathbb{R}^{d}\) of orientation-preserving Euclidean motions. However, it is necessary to restrict \(A,B\) to have ``reasonable'' smoothness: The original framework of Blaschke assumed \(A,B\) to be convex, and subsequently, \textit{S.-S. Chern} [J. Math. Mech. 16, 101--118 (1966; Zbl 0142.20704)] and \textit{L. Santaló} [Integral geometry and geometric probability. With a foreword by Mark Kac. 2nd ed. Cambridge: Cambridge University Press (2004; Zbl 1116.53050)] showed that the formula holds when \(A\) and \(B\) are smooth domains. Both these cases were subsumed by the theory of \textit{H. Federer} [Trans. Am. Math. Soc. 93, 418--491 (1959; Zbl 0089.38402)], treating the case of sets \(A\) and \(B\) of positive reach. Federer's theory has represented the state of the art for many years: the extensions by Hadwiger to sets from the ``convex ring'' (the class of finite unions of compact convex bodies), and in [\textit{M. Zähle}, Geom. Dedicata 23, 155--171 (1987; Zbl 0627.53053)] to the class (UPR) of finite unions of sets with positive reach in general position, both rely on the analysis of the convex/positive reach case; and the extension of \textit{J. H. G. Fu} [Am. J. Math. 116, No. 4, 819--880 (1994; Zbl 0818.53091)] to subanalytic sets relies on the very special finiteness properties that these sets enjoy (in fact the methods there also apply to sets definable with respect to any given o-minimal structure [\textit{L. van den Dries}, Tame topology and o-minimal structures. Cambridge: Cambridge University Press (1998; Zbl 0953.03045)]). It is natural to ask to characterize precisely the minimal amount of smoothness needed to ensure that the PKF holds. This question turns out to be subtle and elusive, and indeed it appears to evade all classical smoothness classes. The paper [Fu, loc. cit.] attempted to formulate an answer using a notion of smoothness arising from the apparatus of the proof of the (PKF) itself. The basic object of interest is the normal cycle \(N(A)\) of \(A\subset\mathbb{R}^{d}\) , viz. an integral current associated to a singular subspace \(A\) that stands for the manifold of unit normals of a smooth set \(A\). Closely related is the differential cycle \(D(f)\) of a non-smooth function \(f:\mathbb{R}^{d}\rightarrow\mathbb{R}\) , which is an integral current that stands for the graph of the differential of a smooth \(f\). A function \(f\) that admits such a differential cycle is called a Monge-Ampère (MA) function. The general theory of [Fu, loc. cit.] posits that a set \(A\) subject to the (PKF) should be given as a sublevel set of an (MA) function at a weakly regular value. Unfortunately, the theory found only limited success: In order to prove the (PKF) for pairs of such sets it was necessary to introduce additional ad hoc hypotheses on the supports of \(N(A)\) and \(D(f)\) (viz. the hypotheses on nor \((f, 0)\), nor \((g, 0)\) in Theorem 2.2.1 of [Fu, loc. cit.]). The main point of the present paper is to show that the general scheme of [Fu, loc. cit.] works completely, without these \textit{ad hoc} devices, in the case of the (WDC) sets introduced in [\textit{H. Federer}, Geometric measure theory. Berlin-Heidelberg-New York: Springer-Verlag (1969; Zbl 0176.00801)]. In this last work it was shown first of all that any (DC) function \(f\) (i.e. a function expressible locally as \(f=g-h\), where \(g\), \(h\) are convex) is (MA). A set \(A\) is (WDC) if it may be expressed as a sublevel set of a (DC) function f at a weakly regular value. In the present paper, by sharpening a theorem of \textit{G. Ewald} et al. [Mathematika 17, 1--20 (1970; Zbl 0199.57002)], and using a construction of \textit{D. Pavlica} and \textit{L. Zajíček} [J. Convex Anal. 14, No. 1, 149--167 (2007; Zbl 1127.52005)], we show that the unwanted \textit{ad hoc} hypotheses are always fulfilled in this setting. Using a characterization of sets with positive reach due to Kleinjohann and Bangert (Theorem 2.9), it is easy to see that any set with positive reach (which also include the ``differentiable polyhedra'' employed as the test objects in Alesker's general theory of smooth valuations) is a (WDC) set. Since (WDC) is closed under finite unions and intersections in general position, it follows that any UPR set (in the original sense of [Zähle, loc. cit.]) is also (WDC). Thus the theory developed here subsumes that of [loc. cit.], and indeed ventures well beyond it, covering for example in a systematic way also the case of general convex hypersurfaces. The main result (Theorem 3.1) of Section 3 states that in this setting the sets \(nor_{\epsilon}f\), defined as the set of all elements of the sphere bundle SM that arise as normalized Clarke differentials of \(f\) at \(f=0\), has locally finite \((d-1)\)-dimensional Minkowski content, where \(d=\dim(M)\). Since Minkowski content -- unlike Hausdorff measure -- behaves well under Cartesian products, this fact is the key to establishing the support condition needed to prove the kinematic formulas. The proof of Theorem 3.1 relies ultimately on a fundamental construction of Pavlica and Zajíček [loc. cit.] relating the support elements of the graph of a (DC) function \(f:\mathbb{R}^{d}\rightarrow\mathbb{R}\) to the set of support hyperplanes \(P\) common to two different convex subsets \(A,B\subset \mathbb{R}^{d+1}\). A lemma of \textit{G. Ewald} et al. [Mathematika 17, 1--20 (1970; Zbl 0199.57002)] states that the latter set has the expected Hausdorff dimension; this is the key lemma for their well-known theorem stating that the set of directions of line segments lying in the boundary of a given convex body in \(\mathbb{R}^{d}\) has Hausdorff dimension at most \((d-2)\). As in [Federer, loc. cit], this argument is enough to show that the set of tangent hyperplanes \(P\) to the graph of the (DC) function \(f\) has the expected Hausdorff dimension \(d\), which in turn is enough to show that a (WDC) set admits a normal cycle in the sense of [\textit{G. Ewald} et al., Mathematika 17, 1--20 (1970; Zbl 0199.57002)], Theorem 3.2. However, necessary for the main result (Theorem B) is the stronger assertion that the set of pairs \((x, P)\) such that \(P\) is a tangent plane for the graph of \(f\) at \((x, f(x))\) has the same dimension \(d\), and moreover that this dimension may be evaluated in the sense of Minkowski content. This result (Lemma 3.5) follows from a refinement of the Pavlica-Zajíŏek result: given convex \(A,B\subset\mathbb{R}^{d}\), the set of pairs \((x - y, P)\) such that \(P\) is a support plane both for \(A\) at \(x\) and for \(B\) at \(y\) has Minkowski dimension \(d\). As a byproduct of this analysis the authors also arrive at the following enhanced version of the theorem of Ewald-Larman-Rogers: Theorem A. Let \(K\subset\mathbb{R}^{d}\) be closed and convex. Denote by \(T_{K}\) the set of pairs \((v, w)\in S^{d-1}\times S^{d-1}\) with the property that there exists a nondegenerate segment \(\tau\subset\partial K\)with direction \(v\) and lying in a supporting hyperplane of \(K\) with outward normal direction \(w\). Then \(T_{K}\) has \(\sigma\)-finite \((d -2)\)-dimensional Minkowski content. In Section 4, the authors prove the main theorem, which states that the kinematic formulas described in [\textit{J. H. G. Fu}, Indiana Univ. Math. J. 38, No. 3, 745--771 (1989; Zbl 0668.49010)] hold for pairs of (WDC) sets in an isotropic space \((M, G)\) (classical work of Hartman [11] implies that the space of (DC) functions is stabilized by diffeomorphisms, hence this notion, and the notion of (WDC) set, make sense on a smooth manifold). To state this precisely, let \(d\) be the dimension of \(M\). Since \((M, G)\) is Riemannian isotropic, it is clear that the space of \(G\)-invariant differential forms on \(SM\) is canonically isomorphic to the subspace of \(G_{\bar{o}}\)-invariant elements of the exterior algebra \(\Lambda^{\ast}T_{\bar{o}}(SM)\), where \(G_{\bar{o}}\subset G\) is the stabilizer of an arbitrary point \(\bar{o}\in SM\). In particular, this space has finite dimension. Such an invariant form \(\beta\) of degree \((d - 1)\), together with a constant \(c\in\mathbb{R}\), gives rise to a \(G\)-invariant curvature measure on \(M\), i.e. an object that associates to each (WDC) set \(A\subset M\) the signed measure \(\Phi_{\beta,c}(A,.)\) given by \[ \Phi_{\beta,c}(A,E)=cVol_{d}(A\cap E)+\int_{N(A)\llcorner\pi^{-1}(E)}\beta, \] where \(\pi:SM\rightarrow M\) is the projection. This measure may alternatively be viewed as a linear functional on the space of bounded Borel measurable functions \(\phi\) given by \[ \Phi_{\beta,c}(A,E)=c\int_{A}\phi+\int_{N(A)}\pi^{\ast}\phi_1\beta \] \noindent Denote the space of all such \(\Phi_{\beta,c}\) by \(\mathcal{C}^{G}\). For \(\Phi,\Psi\in\mathcal{C}^{G}\) and \(A,B\in WDC(M)\) they consider \[ (\Phi\times\Psi)(A,\phi;B,\psi)=\Phi(A,\phi)\Psi(B,\psi) \] \noindent and extend it to all of \(\mathcal{C}^{G}\otimes\mathcal{C}^{G}\) by bilinearity. Theorem B. Let \((M, G)\) be a Riemannian isotropic space, and put \(d\gamma\) for the Haar measure on \(G\) that projects to the Riemannian volume of \(M\). 1. If \(A,B\in WDC(M)\), then \(A\cap\gamma B\in WDC(M)\) for a.e. \(\gamma\in G\). 2. There exists a linear map \[ K:\mathcal{C}^{G}\rightarrow\mathcal{C}^{G}\otimes\mathcal{C}^{G} \] \noindent such that, for any \(\Phi\in\mathcal{C}^{G}\), any compact \(A,B\in WDC(M)\), and any bounded Borel measurable functions \(\phi,\psi:M\rightarrow\mathbb{R}\) \[ \int_{G}\Phi(A\cap\gamma B),\phi.(\psi\circ\gamma^{-1}))d\gamma=K(\Phi)(A,\phi;B,\psi). \]