SO3FunMLS.SO3FunMLS edit page

A class representing a function on the rotation group SO(3).

Syntax

SO3F = SO3FunMLS(nodes, values);
SO3F = SO3FunMLS(nodes, values, 'degree', 3, 'oF', 4);
SO3F = SO3FunMLS(nodes, values, 'delta', 5*degree, 'weight', @(t)(...));
SO3F = SO3FunMLS(nodes, values, 'centered', true, 'monomials', true, 'subsample', 'tangent', true);
SO3F = SO3FunMLS(nodes, values, 'detectOutliers', 'use_vor_weights', 'use_smooth_delta');

Input

nodes orientation, rotation (data points)
values array of function values assigned to the nodes

Output

SO3F SO3FunMLS

Options

degree the polynomial degree used for approximation
oF oversampling Factor. the number of neighbors nn (dependent) is the dimension of the ansatz space, times this factor
oF_max maximum oversampling factor in case of range search. At most the closest SO3F.dim * SO3F.oF_max neighbors will be used.
delta support radius of the weight function when searching for outliers
monomials use monomial basis if true, otherwise use spherical harmonics
centered evaluate the basis functions only around the identity, if true
tangent use monomials on the tangent space (only if centered == true) (in this case the a-coordinate of the neighbors is ignored)
(NOTE: 'centered' and 'tangent' trigger the monomial option to be true)
w @function_handle (weight function)
predefined weight function can be chosen via the following strings: 'C1hat', 'const', 'cos', 'hat', 'indicator', 'squared hat', 'wendland' (default)
use_smooth_delta make the support radius delta(x) a smooth function with close to SO3F.nn neighbors at each center
use_vor_weights additionally multiply w(x,x_i) by the Voronoi Volumne of x_i, as in 'Stable Moving Least Squares Approximation'
distance specify which metric to use (default: 'euclidean')
run 'help rangesearch' for available options
s symmetry of the nodes
regularize use regularization for solving the lsq-systems
maxcond max regularization threshold of condition of the gram matrix
mincond start regularizing threshold of condition of the gram matrix
basis_weights regularization weights of basis coefficients, should punish higher degrees (Sobolev-like)
basis_weights_scale degree-selectivity of the regularization weights; values around 1 give a moderate preference for lower degrees
lambda_geom_rel relative strength of geometric regularization; for a value of 1, geometryScore is used without additional scaling
outlierDetectionRange specify how many neighbors are taken into account

Flags

detectOutliers find outliers in the data and reduce their weight in the local least squares problems depending on how bad they are
subsample use subset of neighbors that minimizes the lebesgue constant