Re: Another proposal for a "split" function to complement "mid"
Arnold
On 19 Mar 2012, at 13:37, Arnold Neumaier wrote:
> What is the use? You only apply it with various fixed L. Variable L is needed only if it is useful to change L during a computation - which doesn't make sense, I think.
I didn't change L during a computation. I did a whole computation with fixed L, for each of L = 1, 1e2, 1e4 and 1e6 in turn.
For [0,inf), I must be missing something. One can't *do* anything with numbers > REALMAX so what's wrong with chopping it to [0, REALMAX] and carrying on from there? I showed briefly what the asinh method does to that interval, with various L values.
Dan has done some manipulation to show that the asinh (with its implicit log) and sinh can be eliminated. Then it only needs basic arithmetic operations, and some care to avoid cancellation.
But I like Michel's take on your idea, once he tells me how to implement the cutoff.
John