Re: [Gretl-users] mysterious error discrepancy between two systems

Am 15.10.2015 um 12:54 schrieb Sven Schreiber: [Sorry, that was quicker than expected, could have sent it together with the previous email.] On the Linux server I see an interesting difference between the scalar and the matrix case: <gretlcli-session> ? cc = cdf(N,-37.64) Generated scalar cc = 2.38421e-310 ? eval cc^2 0 Warning: pow: Numerical result out of range ? printf "%e", cc^2 0.000000e+00Warning: pow: Numerical result out of range ? eval {cc}.^2 gretl_matrix_dot_op: Numerical result out of range </gretlcli-session> It seems to me that if it works for the scalar case (even with a warning) it should also work for the matrix case, or vice versa (not-work -> not-work). The latter expression "{cc}.^2" evaluates to "0" without error here on Windows. (And no warning is printed out for cc^2, either.) Apart from that I wonder how to deal with the situation, as there are some functions that converge to zero by design (like cdf, pvalue, logistic etc.) Somehow it feels inefficient --even though it's easy in the concrete case-- to introduce ad-hoc checks-and-replacements for small values every time I want to process the results of these functions further. Thanks, sven