High-dimensional Menger-type curvatures. II: \(d\)-separation and a menagerie of curvatures (Q843728)

This article is concerned with measuring the goodness of approximation, by a least squares fit for local curvatures of a set \(E\) in \(\mathbb{R}^{d}\), based on hyperplanes through \(E\). It deals only with curve-like structures and the supports of \textit{regular} measures (though definitions of regularity vary for higher dimensions). The measures themselves are incidental, used only for estimation of cumulative error of least square approximations `summing' local curvatures. The use of a least squares fit originates in surveying, say, to determine the position of a beacon in geodetic triangulation; rays from three observers do not usually intersect at a point but form a small spherical triangle. The observations are weighted, primarily by the distances of the beacon to the observers and on visibility conditions. Geometrically the weights are taken into account by constructing strips of appropriate widths around the rays to minimize the area of the triangle and to estimate a final position. The best fit is estimated using squares of the perpendicular distances from the (undetermined) position. The area of the small triangle would indicate the least squares error. Denote by \(B(x,t)\) the closed ball in \(\mathbb{R}^{n}\) with centre \(x\) and radius \(t\). The radius is called the \textit{scale} of the determination of balls in \(\mathbb{R}^{n}\). The \(k\)-dimensional Hausdorff measure (a tool for studying lower-dimensional sets in \(\mathbb{R}^{n}\)) is denoted by \(H^{k}\). The authors assume their measures \(\mu\) to be (Ahlfors) \textit{regular}, i.e., locally finite Borel measures such that there exists \(C \geq 1\) and for all \(x \in \text{supp}(\mu)\), \(\theta < r\leq \text{diam}(\text{supp}(\mu))\), \(C^{-1} r^{d} \leq \mu(B(x,r)) \leq C \cdot r^d\) . Following \textit{R. Schul} [J. Anal. Math. 103, 331--375 (2007; Zbl 1152.28006)], the authors use a real separable Hilbert space \(H\) which encompasses the \(\mathbb{R}^{n}\) for all \(n\) and hence they can use multiscalar coverings of a set \(E\) in \(H\), allowing the scales to vary with \(n\) getting independence of the `ambulant' dimension. The basic geometry for sets of Lebesgue measure zero, as are the sets concerned in this article, was set out in \(\mathbb{R}^{2}\) by \textit{A. S. Besicovitch} [Proceedings L. M. S. 27, 373--388 (1928; JFM 54.0332.01)] constructing a null set in the plane which contained lines in every direction. \textit{H. Federer} [Trans. Am. Math. Soc. 62, 536--547 (1947; Zbl 0032.15001); Trans. Am. Math. Soc. 62, 114--192 (1947; Zbl 0032.14902)] was concerned with geometric content of surfaces and replaced the smoothness condition in \(\mathbb{R}^{n}\) by \(m\)-rectifiabilty, i.e., the existence of Lipschitz maps \(f_{i}:{\mathbb R}^{m} \to {\mathbb R}^{n} \) such that \(H^{n}(E \backslash \bigcup f_{i}({\mathbb R}^{k})) = 0\), or in that \(E\), modulo a negligeable set, is a countable union of Lipschitz images of subsets of \({\mathbb R}^{k}\), \(k \leq n\). The basic rectifiable sets involved are like dyadic `cubes', or grids, made up of half-open lines. Approximation of a rectifiable set is by affine `subspaces', i.e., made up of affine hyperplanes in (small) neighbourhoods; cf. [\textit{J. Léger}, Ann. Math. (2) 149, No.~3, 831--869 (1999; Zbl 0966.28003)] and others. More recently, one uses the concept of \textit{uniform rectifiability}. This uniformity has several equivalent formulations and could depend on choosing a weight \(\omega\) to reduce the error in approximation, analogous to the geodetic weighting mentioned above. A rectifiable closed Borel subset \(E \subset {\mathbb R}^{n}\) is considered to be (Ahlfors) \textit{regular} if there exists a regular measure \(\mu\) supported on \(E\). For \(d = 2\) it is sufficient for a regular measure to be uniformly rectifiable that its support is a subset of a regular curve. (The reviewer notes that the description of a set to be `within' a rectifiable set can be misleading. What is meant is that the set is a subset of the rectifiable set.) \textit{P. W. Jones} [``Rectifiable sets and the traveling salesman problem'', Invent. Math. 102, No. 1, 1--15 (1990; Zbl 0731.30018)] (a tour which covers a graph and crossing each edge at most twice) introduced \(\beta\)-numbers to describe the local flatness of a curve. These indicate how well a compact Borel set in the plane can be approximated by lines as one zooms in points on the surface. For higher dimensions one uses affine hyperplanes instead of lines. Borel sets \(E\) can also be defined as rectifiable in terms of Jones' \(\beta\)-quantities or in terms of a curvature \(c\) (squared) as by \(c^{2}(H^{d-1}|_{E})<\infty\). P. W. Jones, for one dimension, proved the existence of an extension of the measure on a line to the content of a curve in the plane, like a thin strip around the line (cf. Theorem 45 in [\textit{H. Pajot}, Analytic capacity, rectifiability, Menger curvature, and the Cauchy integral. Lecture Notes in Mathematics 1799. Berlin: Springer (2002; Zbl 1043.28002)]). \textit{J. M. Marstrand} [Proc. Lond. Math. Soc., III. Ser. 11, 91--108 (1961; Zbl 0109.03504)] obtained an extension from two dimensions to three. Speaking figuratively, such a curve has almost-tangents at \(H^{1}\)-almost every point so that one can determine Jones' flatness of a curve at suitable scale. The authors' aim is to estimate the \(\beta\)-type least squares error of approximation for a potential determination of the geometry of a set using Menger-type curvatures. \textit{K. Menger}'s [Math. Ann. 103, 466--501 (1930; JFM 56.0508.04)] curvature \(c_{n}(x,y,z)\) of a triple \((x,y,z) \in \mathbb{R}^d\) is the inverse of the radius of the circle containing the triple. The Menger curvature of \(d+2\) points in \(H\) is the inverse of the radius of the circle through the points \(v_{1} , v_{2}, \dots v_{d+2} \in H\). For a Borel set \(E \in \mathbb{R}^n\) with finite one-dimensional Hausdorff measure it is said, by Léger [loc. cit.], to have \textit{total} Menger curvature \(c\) where \(c^2(E) = \int\int\int_{E^3} c^2(x,y,z) dH^1 (x)dH^1 (y)dH^1 (z)\). Considering \(\mathbb{R}^d\), the authors are concerned with \(d+1\) simplices and corresponding parallelotopes (in three dimensions, rhomboids; \(n\)-dimensional parallelotopes having faces which are (\(n-1\))-dimensional parallelotopes). They thus need to define discrete Menger-type curvatures. The \textit{polar sine function} describes the angles between the edges of the corresponding parallelotope; this is a multi-dimensional generalization of the sine functions (dropping perpendiculars from the vertices to the opposite edges). They then define a continuous curvature (squared), denoted \(c_d^2\), involving squares of the discrete curvature, normalized dividing by the volumes (squared). The original form of Jones' \(\beta(x,t) = \beta (B(x,t))\) is as \(\inf \sup \int{(E \cap B(y,t))}\frac {\text{dist}(y,L)}{t}dy\), where the infimum is taken over all lines \(L\) in \(E\) and the supremum is over \(y \in E \cap B(x,t)\). The authors' \(\beta\)-type quantity sums the scaled errors of approximations of \(\mu\) by affine \(d-1\) hyperplanes at different scales and locations (but they do not carry out any estimates on the regular surface containing \(\text{supp} (\mu)\)). The flatness is indeed an integrated generalization of the least squares error, the discrete sum of the geodetic case. The original Jones' \(\beta\) is denoted \(\beta_{\infty}\) in this article as it corresponds to \(L^{\infty}\); the notation \(\beta_2\) corresponds to \(L^2\). The authors make essential use of local curvature being \textit{comparable} with, or \textit{controlled, above and below} (as described in Theorems 31 and 38 in [Pajot, loc. cit.]) by the \(\beta\)-quantities to get estimates of some local curvature \(c\) inside the support of a regular measure. A major result is that locally \(\beta_2^2\) is controlled by \(c_2^2\). Their definition of \textit{least squares error} \(\beta_{2}^{2}(B(x,t))\) of an approximation of geometric content by affine hyperplanes \(L\) through \(B\) with the given scale, is \(\min_{L} \int_{B} \frac{\text{dist}(y,L)} {\text{diam} (B)^2} \frac {d\mu} {\mu(B)} \). The article is hard to read as it is chock-a-block with grisly details. For Part I of this series of papers see [Rev. Mat. Iberoam. 27, No.~2, 493--555 (2011; Zbl 1232.28007)].