Two generalizations of Alford rotation have been proposed for processing 2
x 2-component data containing nonorthogonal split shear waves: singular val
ue decomposition (SVD) and eigenvector-eigenvalue decomposition (EED). Usin
g a simple crossed-dipole synthetic model, we demonstrate that the physical
model behind the EED method is invalid. It incorrectly assumes that a vect
or source aligned with the particle motion of an anisotropic pure mode will
excite only that one mode. Ray theory shows that a vector point-force sour
ce embedded in a homogeneous anisotropic medium instead excites all those m
odes with particle motions that are not perpendicular to the direction of t
he applied force, just as a vector point receiver detects all modes with po
larizations that are not perpendicular to the receiver. Correctly generaliz
ed Alford rotation synthesizes vector sources and receivers such that each
component is perpendicular to all but one of the pure modes of the medium.
Although this ray-theory result does not allow for the possibility of a sou
rce or receiver on a free surface and so is not yet completely general, it
does apply to the idealized homogeneous crossed-dipole geometry of our exam
ple. The new method, symmetric Alford diagonalization, differs from previou
s methods by becoming unstable when applied over excessively short time win
dows. This behavior is consistent with the physics of the problem: If nonor
thogonal modes are allowed, then there is not enough information at a singl
e time sample to determine a unique solution. Any method that can find a un
ique solution at a single time sample, including both the EED and SVD metho
ds, does not respect the physics of the nonorthogonal problem. Although the
re appears to be no problem that recommends the EED method over standard Al
ford rotation for its solution, the SVD method is still applicable to the p
roblem it was originally designed to solve: orthogonal modes with an unknow
n source or receiver orientation.