js850 / pygmin Goto Github PK

method analogous to removeMinimum() require for transition states

• there is lots of redundant code, improve inheritance
• the creation of the basin hopping object takes too long, move initial quench to run
• update documentation

e.g. in quench if step has failed

Solution: make specialized exceptions for these problems and only catch these

We should ensure that no function or class is more than two levels deep in import statements. i.e. no more of import pygmin.potenials.lj.LJ This should instead be import potenials.LJ

Steps to get there:

every file should declare an all variable at the top. This declares which objects are public and will be imported in an import * statement.

all init.py files should have statements like "
from myfile1 import *

This has the potential to cause name conflicts. If file NEB.py has class NEB(). Thus most files should be renamed like _NEB.py.

Also, functions and classes need to have names which make their use obvious even when divorced from their file name. Example: function start() in pymolwrapper.py wouldn't make any sense if imported as from pygmin.utils import start

Functions and classes in the same directories must have unique names. Problem areas are our 12 different implementations of LJ, and two implementations of transition state search.

If we do this, almost all import statements will break, so we will need to be careful.

describe what the parameters do

we have a huge mess of mixed code. can we make this more consistent?

Also there should be only 1 lj implementation and maximum 1 fallback option (if at all)

separate rigid body transformations from potential

http://espressomd.org/

espresso soon will have a python interface. this will make it a lot easier. otherwise, espresso can be compliled as a library.

Move this to a separate sub class, interaction lists do not make sense for all potentials

e.g. for crystal structure code

Currently it's a bit inconsistent. We should probably separate initial pushoff relaxation + final quench. the latter should use a handle of the get_minimize from system class

• minima queries& insertion
• transition states
• get dist lookup

numerical one by default

vec_random generates a random unit vector of length 1. If used in a takestep routine, this should be multiplied by a random scalar! This was not done in all cases. check this

• adaptive step taking shouold be applicable in 1 line
• what about reseeding -> superclass of reseeding

As discussed in Create MolecularSystem class, it would be nice if we could create some sort of topology module for PyGMIN.

So far, we want to be able to include:

• Molecular systems with bonding, angles and dihedrals;
• Crystal systems (including information on boundary conditions);
• Angle-axis topologies (e.g. water networks).

I am wondering whether we also need to separate all this information from information about symmetry and our various forms of coarse-graining.

I think we also need to think carefully about how this module will be related to the System classes, e.g. when individual particles are bonded by harmonic potentials, such as in AMBER, they cannot dissociate and you can't meaningfully rearrange the system to some other type of connectivity.

we need routines to check for exact alignment before using minDistStochastic. This is possible without the N! scaling. We should never be missing exact alignments.

• allow for different log levels
• write to file and stdout
• timestamps

this could be done in a wrapper class NEB_driver

The database should be able to save only the N minima with the lowest energy.

use it in storage + basin hopping

If one wants to use an existing database, it often happens that a new one is created since e.g. there was a type in name. Add a flag to explicitly create a new database and check how to best integrate it at the various places in the code

do not use i % iprint == 1

this doesn't work for iprint == 1, use i % iprint == 0 instead.

are you ok with that? you can do that in the admin page for the repository.

get NEB to work with rigid bodies

• rotations with mouse are wrong
• add zoom + translate

this module calculates derivatives of rotation matrices. But it is in fortran and passes multidimensional matrices. Four 3X3 matrices are copied for each call to this function. This should be changed. Probably rewriting it in cython is the easiest.

I am planning to create a MolecularSystem class inheriting from BaseSystem. Its purpose will be to:

• Act as a superclass for AMBERSystem and any future CHARMM, GROMOS etc. style systems;
• Identify topological features of molecular symmetry;
• Allow for interconversion between AMBER, CHARMM and other potentials;
• Implement easy I/O and visual representations.

The two important classes for this will be MolecularSystem and Atom, which are outlined below:

class MolecularSystem(BaseSystem)
    atoms
    bonds
    angles
    torsions
    topology_search()
    import_pdb()
    import_xyz()

class Atom()
    mass
    element
    charge
    lj_terms[]
    bonded_atoms[]

writing new scripts has lots of redundancy, add a system class and derive from that to also define standard crystal system

it would be nice to have fast access to the number of minima, transition states and distances in the database. len(database.minima()) is problematic for large databases

Currently all the numbered examples in pygmin/examples/scripting do not use the system classes derived from BaseSystem.

We should update them to do this (and remove the old functionality?).

Don't redraw everything but update curves instead could be better?