[MODEST] toy model

[Christopher Tout]

Dear All,
         Once again my apologies for not making it to the workshop in
New York and hence for anything I say that may have been said before.
I have thought about the interface problem and favour your option 2)
myself.  I think some of the cons for all cases are relatively easily
overcome.

Step too long: SE must return a maximum step based on estimates of mass
loss, time to RLOF etc.  We have much of this in Nbody4 already.

Step too short: SE does not have to evolve everytime it is told to.  If
the step is less than SE would normally take nothing of imortance would
occur and SE need do no evolution.  It must of course keep track of the
elapsed time and evolve when the cumulative step is large enough.

Your mode 3) does not strike me as difficult to implement because it
allows SE to run completely independently.  The thing most against it
is that we would end up evolving some stars (among which I include
binaries) beyond the time at which they may have interacted through
SD.  Of course this happens sufficiently rarely that it probably does
not slow things down significantly.  As such 3) may be the most
efficient way to go because unlike 2) stars' internal structures are
loaded into memory and stellar evolution is dealt with sequentially.
Mode 2) requires continual swapping of stars in and out of memory.  Of
course 3) requires a very long initial period of SE before SD can
begin - certainly for primordial binaries (single stars could be held
in a large database that need only be evolved once for all models).

With best wishes,
Chris Tout