FERN: Fast Explicit Reaction Networks SVNCode

Intermediate steps in converting timestepping routine to exactly that in the Java code.

Intermediate steps in converting timestepping routine to exactly that in the Java code.

Intermediate steps in converting timestepping routine to exactly that in the Java code.

Intermediate steps in converting timestepping routine to exactly that in the Java code.

Intermediate steps in converting timestepping routine to exactly that in the Java code.

Intermediate steps in converting timestepping routine to exactly that in the Java code.

Small changes

Fixed max flux algorithm so now it is taking difference of FplusSum and FminusSum, which is more what the Java program does. Seems to be giving the same max flux as Java program now, if in the Java program we define the flux to be maxdpopp/nT. However, the actual current timestepper in Java program is effectively defining the max flux to be maxdpopp/totalSeeds, which is not the same as maxdpopp/nT. I think the difference is that nT is the correct normalization to convert raw test particle count in the Java program to Y, but totalSeeds accounts only for the nucleons in the heavy seeds and not in the light ions (alpha particles in this case). In particular nT is the total number of nucleons in the network (conserved), but totalSeeds is the total number of heavy (A>4) seed nuclei at the beginning of the calculation.

Renamed SIZE -> REACTIONS, ISOTOPES -> SPECIES, and moved the constants REACTIONS and SPECIES to top header material so they are accessible to CPU and GPU code. Changed arrays in shared GPU memory to dimension SPECIES instead of hard-wired 16 (for the test case).

Tweaks and tests