Thanks.
log.arg worked to make sure nonnegative value. But the other side
is overflow. Beyond the upper limit, the exp(arg) is large and bessel
overflows even if exponentially scaled one is used. This will happen at
1/500 chance. Due to I'm doing 10,000 simulations, this problem blocked
me.
Hope this bug will be fixed in the future.
Li
On Wednesday 20 August 2003 15:44, Shengqiao Li wrote:
Hi, R guys:
I'm using L-BFGS-B method of optim for minimization problem. My function
called besselI function which need non-negative parameter and the besselI
will overflow if the
I see two problems, which I handle differently:
First, if arg must be nonnegative, I program fn in terms of log.arg,
so it can never become nonpositive.
Second, is besselI always positive? If yes, then program fn to
compute log(besselI) and use that. Standard references such as