The matrix being rotated contains the values of the component functional data objects computed in either a principal components analysis or a canonical correlation analysis. The values are computed over a fine mesh of argument values.
varmx(amat, normalize=FALSE)
the matrix to be rotated. The number of rows is
equal to the number of argument values nx
used
in a fine mesh. The number of columns is the number of
components to be rotated.
either TRUE
or FALSE
. If TRUE
, the columns of
amat
are normalized prior to computing the rotation
matrix. However, this is seldom needed for functional data.
a square rotation matrix of order equal to the number of components that are rotated. A rotation matrix $T$ has that property that $T'T = TT' = I$.
The VARIMAX criterion is the variance of the squared component values. As this criterion is maximized with respect to a rotation of the space spanned by the columns of the matrix, the squared loadings tend more and more to be either near 0 or near 1, and this tends to help with the process of labelling or interpreting the rotated matrix.