Scaling universalities of \(k\)th-nearest neighbor distances on closed manifolds (Q1271871): Difference between revisions
From MaRDI portal
Added link to MaRDI item. |
Set profile property. |
||
Property / MaRDI profile type | |||
Property / MaRDI profile type: MaRDI publication profile / rank | |||
Normal rank |
Revision as of 02:46, 5 March 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Scaling universalities of \(k\)th-nearest neighbor distances on closed manifolds |
scientific article |
Statements
Scaling universalities of \(k\)th-nearest neighbor distances on closed manifolds (English)
0 references
8 March 1999
0 references
This paper contains results in the field of stochastic differential geometry. Consider \(N\) sites placed randomly, with a uniform distribution, on a 2-D surface of a fixed area. Let the random variable \(D_k(N)\) represent the distance between a given point \(x\) and its \(k\)th nearest site. The authors present the expected distance \(\langle D_k(N)\rangle\) taken over the ensemble of randomly placed sites in terms of the function \(A(l)\) giving the area of a dice of radius \(l\) about that point. The two universalities are discovered. First, for a flat surface, where \(A(l)= \pi l^2\), \(\langle D_k(N)\rangle\) is separable in \(k\) and \(N\): \[ \langle D_k(N)\rangle\approx {1\over\sqrt\pi} {(k- 1/2)!\over (k- 1)!} {N!\over(N+ 1/2)!}, \] the \(N\)-dependence in \(\langle D_k(N)\rangle\) is the same, regardless of \(k\). All \(k\)th nearest neighbor distances thus scale the same way in \(N\). Furthermore, the property is not restricted to two dimensions, and turns out to be equally valid for flat manifolds of any dimension \(d\): \[ \langle D_k(N)\rangle\approx {[(d/2)!]^{1/d}\over \sqrt\pi} {(k- 1+ 1/d)!\over(k- 1)!} {N!\over(N+ 1/d)!}. \] This universality applies equally well to higher moments of the distances, and to Euclidean manifolds in dimension greater that \(2\). In the second part, the more general case of an arbitrary closed and smooth 2-D surface with Gaussian curvature \(K\) is considered. The relation between \(A(l)\) and the Gaussian curvature \(K\) is proposed: \[ \begin{multlined} A(l)= \pi l^2\Biggl\{1- {l^2\over 12} K+{l^4\over 720} (2K^2- 3\nabla^2 K)-\\ {l^6\over 161280} (8k^3- 3[10(\nabla K)^2+ 14K\nabla^2 K- 5\nabla^4 K])+ O(l^8)\Biggr\},\end{multlined} \] where \(\nabla\) is the gradient operator. The second universality, for a curved surface, \(\langle\widetilde D_k(N)\rangle\) averaged over the surface is a topological invariant at leading and subleading order in a large \(N\) expansion \[ \int\langle\widetilde D_k(N)\rangle d\mu= 1+{\chi(2k+ 1)- 9\over 24N}+ O\Biggl({1\over N^2}\Biggr), \] where \(\langle\widetilde D_k(N)\rangle\) is the reduced variable, \(d\mu\) is the differential surface element, \(\chi\) is the Euler characteristic of the surface. The \(1/N\) scaling series then depends, up through \(O(1/N)\), only on the surface's topology and not on its precise shape. The case of higher dimensions \((d>2)\), and interpretations of the results using Regge calculus are discussed.
0 references
stochastic differential geometry
0 references
expected distance
0 references
Gaussian curvature
0 references
universality
0 references
Regge calculus
0 references