Rough differentiability of planar Gaussian random fields (Q1587164)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Rough differentiability of planar Gaussian random fields
scientific article

    Statements

    Rough differentiability of planar Gaussian random fields (English)
    0 references
    19 April 2001
    0 references
    The paper poses the problem of ``rough differentiability'' at a given point \(O\) (the origin) of a Gaussian random field \(\xi(Q)\) defined in \(\mathbb R^2\). Essentially, rough differentiability is the condition of existence of (random) tangent plane to a random surface \(\xi(Q)\) at the point \(O\), expressed in a geometrical form of coplanarity (with probability 1) of lines that are tangent (at \(O\)) to restrictions of \(\xi(Q)\) to lines containing \(O\). The points \(Q\in\mathbb R^2\) are described by pseudopolar coordinates \(Q=(\varphi, u)\): the azimuth \(\varphi\) is determined by the condition that \(Q\in g(\varphi)\) while \(u\) is one-dimensional coordinate on \(g(\varphi)\), the sensed line containing the origin \(O\in \mathbb R^2\). So, \(\varphi\in (0,\pi)\) and \(u\in (-\infty, +\infty)\). A random Gaussian field \(\xi(Q)\) defined in \(\mathbb R^2\) is said to satisfy Rice condition, if there exists some function (Rice function) \(h(\varphi)>0\) depending on the azimuth \(\varphi\in (0,\pi)\) such that asymptotically \(E\xi(O) \xi(Q)= 1- h^2(\varphi) u^2+o(u^2)\), whenever \(Q=(\varphi,u)\) tends to \(O\) along the line \(g(\varphi)\) (\(E\) stands for expectation). The Rice class is not empty, for it contains translation invariant Gaussian fields with ellipto-exponential covariance function. In that case \(h^2(\varphi)= a^2\sin^2(\varphi-\alpha)+ b^2\cos^2(\varphi-\alpha)\), where \(\alpha\in[0,\pi)\) is a rotation parameter, i.e. \(h(\varphi)\) coincides with the support function of an ellipse. The paper shows that within the Rice class, translation invariant Gaussian fields are in fact never roughly differentiable, except for the case where \(h(\varphi)\) is the support function of an ellipse. In other words, rough differentiability of a Gaussian field satisfying Rice condition implies the above form of \(h^2(\varphi)\). Also, in this case, the random spatial direction \(\Omega\) turns out to be independent of \(\xi(O)\), and the probability density for \(\Omega\) is written down explicitly. This result opens the way for calculation of the intensity of the point process of intersections of a given horizontal line with the random surface \(\xi(Q)\) on any level \(x\). For this purpose the author uses rather skillfully the standard Jacobian formulae of integral geometry. The unexpected result concerning ellipticity of \(h(\varphi)\) naturally gives rise to a problem formulated at the end of the paper: do there exist (necessarily roughly non-differentiable) Gaussian fields in \(\mathbb R^2\) that satisfy planar Rice condition with nonelliptical \(h(\varphi)\)?
    0 references
    0 references
    0 references
    0 references
    0 references
    Gaussian random fields
    0 references
    Rice condition
    0 references
    rough differentiability
    0 references
    support function of an ellipse
    0 references