New feature: single configuration file across all nodes

[Lucas Wagner]

Hi all,

I've implemented a first stab at a single configuration file. It
works for the commonly used methods of VMC, OPTIMIZE, OPTIMIZE2, DMC,
and RMC_CORR, but not for others yet. It's not perfect, but I think
it's better than what we had. For those that are interested, check
out the write_configurations() and read_configurations() functions in
qmc_io.h. There's still a lot of debugging information being printed
as well; that will change soon enough.

Lucas

[Kevin Rasch]

Lucas,

I pulled this newest version and ran my usual testing calculations.
There was/is a problem with multi-method calculations. If I try
something such as

method { VMC
nconfig 1000
storeconfig 1k.config
. . .
}

method { OPTIMIZE
nconfig 1000
readconfig 1k.config
. . .
}

method { VMC
nconfig 1000
readconfig 1k.config
storeconfig 1k.config
. . .
}

method { DMC
nconfig 200
readconfig 1k.config
storeconfig 200.config
. . .
}

then the calculation bombs out when it reaches the DMC because 200 !=
1000. That's not a very useful constraint since I frequently need to
optimize with many walkers and then calculate DMC with fewer. It
seems like the expected behavior should be to choose 200 of the
thousand saved walkers (especially if the 200 walkers are being save
to a different config file that the 1k).

Cheers

Kevin

[Lucas Wagner]

Thanks for the comments, Kevin. I've fixed the behavior for DMC and
VMC, I think. Let me know if there are any further problems.

I've also added the ability for the RMC_CORR method to use
average_generators in the correlated sampling. In this way, we can
likely discern very small changes in dipole moment, magnetic moment,
etc, using the pure estimator! If anyone would like to use this
capability, then I'd be happy to help out. Error bars aren't
calculated for differences yet, although it shouldn't be too hard to
do (just some modifications in Properties_average; you wouldn't have
to rerun anything).

So far, RMC_CORR seems to work in terms of giving the right answers,
particularly if you use pc_gf. Not sure on efficiency yet, though.

Cheers,

Lucas

[Kevin Rasch]

Lucas,

I ran the same test and everything works fine if I make with
PLATFORM=Darwin. However, if I compile with PLATFORM=icc and run, I
get an error even before VMC begins:

. . .
Pseudopotential
Error Cannot resize sub-array
terminate called after throwing an instance of 'Qmc_error'

I just installed the intel dev suite for OS X this week so this could
be a mistake on my part in the icc.mk file.

Cheers,

Kevin

[Lucas Wagner]

Kevin,

Try doing a 'make clean PLATFORM=icc' and then recompiling. I usually
only get that error when something wasn't clean with the object files.

Lucas

[Kevin Rasch]

Lucas,

Thanks! That fixed it. All appears to work as designed :)

Cheers,

Kevin