Subsections

• AUTHOR

15 FAQs

Q1: My program produces a lot of messages like the following:

(g6_test_flag) scaling error 2da00
(g6calc_lasthalf) overflow for  3  2da00  24 16165 48-- change scales
(adjust_ip_scales) decrimenting scale for jerk-333

Is this something I should care?

A: This message means that the initial guess for force, jerk or potential (in the above case, it's jerk) was too small and it caused the overflow of the result. The lasthalf routine automatically recalculates the force after decrementing the scaling index by 5 (increasing the scale factor by 32).

In the above message, ``3'' and ``2da00'' are things I do not quite want to explain, and ``24'' is the index (within i-particle array) of the particle which resulted the overflow. ``16165'' is the index of that same particle specified in the argument of firsthalf, and ``48'' is the number of i-particles. The last number, ``-333'' is the new value of the scale factor (which by itself is not too useful).

It's hard to avoid the scaling error completely since jerk sometimes changes by very large factor even if time integration goes fine. However, if you see too many of them, in particular the repeated one for one force calculation, it is likely that your initial guess is too far from the true one. You might want to try a better way to make the initial guess.

Q2: It seems GRAPE-6 calculates the force from particles with index larger than nj. Why?

A: Well, ``this is a feature, not a bug''. The problem here is that $j$-particles are actually distributed to a number of chips. Thus, if we use $p$ chips to calculate the force on one particle, each chip calculates forces from $[(N+1)/p]+1$ particles. Unless $N$ is an exact integer multiple of $p$, some chips calculate the forces from particles with indices larger than $N$. GRAPE-6 library takes care of this by clearing out the unused location when a new value of nj is set in firsthalfx. So if you do not overwrite them again, the calculated result is correct. If you accidentally overwrite these locations, GRAPE-6 just calculates the forces from them. So you should not write the particle data with index larger than you want to use. (But why you want to do so?)

AUTHOR

Junichiro Makino