Maximum B1 for P+1 = 4294967295?

[akruppa]

That might solve it - or maybe not. On a 64 bit machine, simply using a long int for those variables might do the trick, but I don't really understand the PRAC algorithm yet - it uses floating point approximations to some real constants to control the add/double (double here in the sense of: multiply by two) sequence and those constants in double (data type here) precision might not be enough for multipliers >2^32. If that is the case, we'll pretty much have to rewrite all the PRAC code.

Alex

[Joe O]

Originally Posted by akruppa

That might solve it - or maybe not. On a 64 bit machine, simply using a long int for those variables might do the trick, but I don't really understand the PRAC algorithm yet - it uses floating point approximations to some real constants to control the add/double (double here in the sense of: multiply by two) sequence and those constants in double (data type here) precision might not be enough for multipliers >2^32. If that is the case, we'll pretty much have to rewrite all the PRAC code.

Alex

So you are using an "eliptic curve" method instead of Williams' Lucas Sequence. Interesting!

As far as the constants, (so called "magic" numbers") my experience has been that they have to change when going from 2^32 to 2^64, but it all depends on what they really are.

[akruppa]

We're merely using PRAC for computing U_k(x) from U₁(x) because it uses fewer modular multiplications than the usual binary ladder that keeps both U_i(x) and U_i-1(x) around.

Alex