logspace.add {DPQ} | R Documentation |
Compute the log(arithm) of a sum (or difference) from the log of terms without causing overflows and without throwing away large handfuls of accuracy.
logspace.add(lx, ly)
:=\log (\exp (lx) + \exp (ly))
logspace.sub(lx, ly)
:=\log (\exp (lx) - \exp (ly))
logspace.add(lx, ly) logspace.sub(lx, ly)
lx, ly |
numeric vectors, typically of the same
|
a numeric
vector of the same length as x+y
.
This is really from R's C source code for pgamma()
, i.e.,
‘<R>/src/nmath/pgamma.c’
The function definitions are very simple, logspace.sub()
using log1mexp()
.
Morten Welinder (for R's pgamma()
); Martin Maechler
set.seed(12) ly <- rnorm(100, sd= 50) lx <- ly + abs(rnorm(100, sd=100)) # lx - ly must be positive for *.sub() stopifnot(exprs = { all.equal(logspace.add(lx,ly), log(exp(lx) + exp(ly)), tol=1e-14) all.equal(logspace.sub(lx,ly), log(exp(lx) - exp(ly)), tol=1e-14) })