RDF using rerun

LAMMPS LAMMPS Mailing List Mirror
1

Hi users,

I am trying to get RDFs for distances longer than cutoff for my two component ionic simulations. As I need to calculate partial structure factor of multi-species system, a better resolution is needed at smaller wave-number (k) values and hence causes need for long distance rdfs with bigger systems.

My 3D system having 60000 particles and for them if I dump data to later work with rerun command for a longer rdf, the data size is becoming huge.
(3000 steps giving 6 GB data). I am dumping data at each step to get better statistics.

Is there any way to avoid such huge data download and even getting rdfs for longer distance than cutoff.

Or, is there any smarter way to get better rdfs with low dump cost?

I appreciate your response.

Thanks,
Sanat

2

Hi users,

I am trying to get RDFs for distances longer than cutoff for my two
component ionic simulations. As I need to calculate partial structure
factor of multi-species system, a better resolution is needed at smaller
wave-number (k) values and hence causes need for long distance rdfs with
bigger systems.

My 3D system having 60000 particles and for them if I dump data to later
work with rerun command for a longer rdf, the data size is becoming huge.
(3000 steps giving 6 GB data). I am dumping data at each step to get
better statistics.

Is there any way to avoid such huge data download and even getting rdfs for

longer distance than cutoff.

Or, is there any smarter way to get better rdfs with low dump cost?

​you are ignoring one of the basic lessons of statistics: you can improve
your result significantly with more samples *only* if the samples are
*independent*. however, between individual time steps the data is *highly
correlated*. thus there is no value at all in storing data at every time
step. how strongly the correlation decays depends very much on the
viscosity. for liquid water at room temperature for example, there is no
benefit in storing trajectory frames more frequent than 1ps apart, and even
then there is only a small improvement over using trajectory frames 10ps
apart.

​axel.

3

Thanks Axel,

Indeed it was my mistake.