[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [FreeON-devel] The FreeON wiki is open for business!

From: Matt Challacombe
Subject: Re: [FreeON-devel] The FreeON wiki is open for business!
Date: Mon, 30 Apr 2012 12:06:20 -0600

Hi Gao,

FreeON is gamma point only, for all periodic calculations.   For
large, non-metalic and heterogeneous
systems, gamma point should converge pretty quickly with respect to N.
 If you have a small system
with lots of symmetry, well, FreeON is probably not the right tool.

While FreeON (MondoSCF) used to be fairly well parallelized with MPI,
only the NEB part over clones (beads)
is currently MPI parallel.   We dropped support for that part of the
code because (1) it was really clumsy and
difficult to support and (2) we are redoing the code from the ground
up in a different HPC paradigm.  See
for example, Nicks recent paper on SpAMM.  This will replace the BCSR
structures with quad and oct-trees
and a task parallel approach.  This will be much, much easier to
maintain and develop around (+faster).

The dynamic response is in RQI.F90 as you correctly understand. You
would probably need to implement in
both RQI.F90 and some other "backend" files.  Its important to
understand the front/back end way FreeON
is built.  It has a lot of advantages, especially for very large
systems, but can be a pain at first.  The advantage
is, if something goes wrong with ie a 2000 atom calculation, you can
for instance easily re-run/debug/profile
etc the most important/costly/buggy part of the code, instead of
starting from the beginning.
RQI is somewhat unique in that the linear algebra is in the front end,
while it makes backend calls to the J+K builds.
This is pretty different than the rest of the code.  We could probably
back end RQI, and nest the J+K builds.
We should consider that, along with ways to let some of the linear
algebra live in the front end, as it is now, but
maybe in a more friendly way.   Our task parallel approach we are
working towards will help and inform
these choices a lot.

So maybe let me know what properties you are interested in, and if HF
is going to be enough, and we can
discuss more.

Best, Matt

On Mon, Apr 30, 2012 at 11:08 AM, Gao Bin <address@hidden> wrote:
> Hi, Matt
> Thank you so much for your detailed reply. I have read your paper about 
> static response before :-) I will read the dynamic paper later on, thank you!
> By the way, may I ask you some more questions?
> (1) You said the calculations is gamma-point only for static response. Is it 
> also true for dynamic response? May I ask if it is easy or not to extend 
> FreeON for any reciprocal space points?
> (2) If I want to calculate dynamic response, is this part of code 
> parallelized using MPI over several nodes, or only on one node using shared 
> memory (for instance, the matrix operations)?
> (3) I have read the code, if I understand correctly, the dynamic response is 
> implemented in FreeON/RQI.f90. If I am interested in how it works, or want to 
> implement calculations of some new properties, should I work on 
> FreeON/RQI.f90 or other files? Thank you so much.
> Cheers
> Gao
> ________________________________________
> From: address@hidden address@hidden on behalf of Matt Challacombe 
> address@hidden
> Sent: Friday, April 27, 2012 8:30 PM
> To: address@hidden
> Cc: address@hidden
> Subject: Re: [FreeON-devel] The FreeON wiki is open for business!
> Hi Gao,
> There are examples in the freeon/Validate directory.  For the static
> response there is the
> file: Validate/Response/Static/Peroxide.inp    Note that only the HF
> model has been implemented
> so far, and for static response, only for polarization.  Also, the
> polarization will not work
> properly if the chosen axis is periodic, because it is gamma-point
> only  (you can do z axis
> polarization in a 2D slab though, with slab along xy plane).
> There is also a TDSCF directory next door to static with the file:
> Validate/Response/TDSCF/C2H4.inp  Again, this is only HF (RPA), but
> you can do 1000's of atoms if you like.  The ref paper for dynamic
> response is here:  I can get you the papers
> for static response too.  This method also works very well, ie if you have
> a local excitation, like a chromophore in a protein matrix.  Then, the cost
> of RPA is O(1) while cost of SCF is O(N).
> I think the question for you is, will TD-HF (RPA) be enough or do you need
> a DFT component?  I'm pretty sure this is the only O(N) method for TD-SCF,
> so it might be worth thinking about trying to implement the DFT part.
> Programming
> it is a bit time consuming, and I didn't have the time to do it yet.
> Maybe you can live
> without it though, or maybe you or someone you know is interested in
> programming it a bit (I'd be willing to help but couldnt do it all).
> Also, if you need
> more than the first eigenvalue, you would need to do some extra programming.
> I have that working somewhere in a prototype, but not strongly implemented.
> I hope that helps.
> Best, Matt
> On Thu, Apr 26, 2012 at 3:58 PM, Gao Bin <address@hidden> wrote:
>> Hi, Nick
>> Thank you. Since I just started to use FreeON (indeed I am still struggling
>> to get it be compiled), I am afraid I could not contribute immediately, but
>> I will do my best :-)
>> Now, I am trying to use FreeON. To me, the first step is to understand the
>> keywords. I notice all the keywords are parsed in subroutines of FreeON
>> directory, but there are ParseXX.f90 and XXKeys.f90. I wonder which files I
>> should look into? Thank you so much.
>> Another question is related to the feature mentioned in wiki page "Advanced
>> features include O(N) static and dynamic response, as well as time
>> reversible Born Oppenheimer Molecular Dynamics (MD)". I am interested in
>> dynamic response, but I did not find any routine in FreeON/Response.f90
>> related to dynamic response. Did I misunderstand or did I look into a wrong
>> file? Moreover, may I ask what the reference papers for static and dynamic
>> response? I notice there are some reference papers in
>>, but I am not sure which one and if they
>> are enough. Therefore, I think it would be better to put the reference list
>> in the current wiki page (well, just my opinion).
>> Cheers
>> Gao
>> ________________________________
>> From: Nicolas Bock address@hidden
>> Sent: Wednesday, April 25, 2012 8:40 PM
>> To: address@hidden; freeon-devel; Gao Bin; Jose R. Valverde
>> Subject: The FreeON wiki is open for business!
>> Hi,
>> please visit
>> and check out our new wiki. There is nothing really on there yet, but we are
>> hoping that with your help there will be soon. The wiki is completely open,
>> which means that you can add, edit, and delete content as you wish, with or
>> without a wiki account.
>> I suggest we think about what information a user or a developer would like
>> to find on a wiki. My guess is (in no particular order of importance):
>> * API
>> * Input file syntax
>> * Example inputs, illustrating all features of FreeON
>> * A feature list
>> I would like to encourage you to add to this list and of course to start
>> adding content to the wiki.
>> Thanks already for your help,
>> nick
>> _______________________________________________
>> FreeON-devel mailing list
>> address@hidden
> _______________________________________________
> FreeON-devel mailing list
> address@hidden
> _______________________________________________
> FreeON-devel mailing list
> address@hidden

reply via email to

[Prev in Thread] Current Thread [Next in Thread]