search for: _asymp

...a-x)/x) = -x * log((lambda-x)/x) - (lambda-x) and naively assumes that this is small enough to use a power series expansion in 1/x with coefficients as powers of pt_. To make matters worse, consider > pgamma(0.9*1e25, 1e25, log=T) pgamma_raw(x=9e+024, alph=1e+025, low=1, log=1) using ppois_asymp() pt_ = 5.3605156578263e+022 pp*_asymp(): f=-2.0803930924076e-013 np=-5.3605156578263e+022 nd=-5.3605156578263e+022 f*nd=11151979746.284 [1] -Inf Hmm, how did that manage to lose *all* the accuracy here? Hubris again appears in the comments. Here np and nd are on log scale and if they are lar...