[R] Matrix negative fraction power

[Berend Hasselman]

On 11-03-2012, at 18:18, Berend Hasselman wrote:

> 
> On 11-03-2012, at 17:52, Spencer Graves wrote:
> 
>>     If my memory is correct, the archives of this list contains several discussions of round off error problems associated with different methods for computing things like this.  The "Matrix" package (part of the base distribution) contains a function "expm", whose help file says, "The expm package contains newer (partly faster and more accurate) algorithms for expm() and includes logm and sqrtm."  This suggests to me that the most numerically stable way to get an arbitrary power p of a matrix M in R might be as follows:
>> 
>> 
>>           expm(p*logm(M))
>> 
> 
> I have just tested and it should be this
> 
> library(expm)
> 
> expm((-1/2)*logm(A))
> expm((-3/2)*logm(A))
> 

Indeed. It's getting on and I made a muddle.

Berend