More accurate upper tail quantiles for small survival probabilities of the normal distribution

[pbrod]

For small lower tail probabilities the ppf function of the normal distribution gives

In [83]: import scipy.stats.distributions as dst
In [84]: dst.norm.ppf(1e-40)
Out[84]: -13.31092137142517

But for small upper tail probabilites the isf function gives infinite quantile:

In [85]: dst.norm.isf(1e-40)
Out[85]: 1.#INF

where the true value is 13.31092137142517. A solution that fix this so that small upper tail probabilities have the same accuracy for the quantile as the lower tail probabilites, is to add the following _isf method to norm_gen class:

    def _isf(self,q):
        return -_norm_ppf(q)

[josefpktd]

changed in r5602, together with kstwobign

new results:

stats.norm.isf(1e-40)

13.31092137142517

stats.kstwobign.sf(10)

2.7677930534734751e-087