Dear all,
I encounter some covariance matrix with quite small eigenvalues
(around 1e-18), which are smaller than the machine precision. The
dimension of my matrix is 17. Here I just fake some small matrix for
illustration.
a-diag(c(rep(3,4),1e-18)) # a matrix with small eigenvalues
On 30-May-05 huang min wrote:
Dear all,
I encounter some covariance matrix with quite small eigenvalues
(around 1e-18), which are smaller than the machine precision. The
dimension of my matrix is 17. Here I just fake some small matrix for
illustration.
a-diag(c(rep(3,4),1e-18)) # a
Maybe I should state more clear that I define b to get the orthogonal
matrix bb$vectors.
We also can define diag(b)-diag(b)+100, which will make the
eigenvalues of b much bigger to make sure the orthogonal matrix is
reliable.
My intention is to invert the covariance matrix to perform some
On 30-May-05 huang min wrote:
Maybe I should state more clear that I define b to get the
orthogonal matrix bb$vectors.
OK. Certainly bbv-bb$vectors is close to orthogonal: bbv%*%bbv
differs from the unit matrix only in that the off-diagonal
terms are O(10^(-16)).
We also can define
On Mon, 30 May 2005, huang min wrote:
My intention is to invert the covariance matrix to perform some
algorithm which is common in the estimating equations like GEE.
In that case there is no benefit in being able to invert very extreme
covariance matrices. The asymptotic approximations to the
