def test_merge_minima(self):
graph = TSGraph(self.db)
# select two minima with at least one edge
minima = []
for n in graph.graph.nodes():
if graph.graph.degree(n) > 0:
minima.append(n)
if len(minima) == 2:
break
nedges = [graph.graph.degree(n) for n in minima]
min1, min2 = minima
neighbors = graph.graph.neighbors(min2)
graph.mergeMinima(min1, min2)
self.db.mergeMinima(min1, min2)
self.assertNotIn(min2, graph.graph.nodes())
# make sure the edges were copied
for n in neighbors:
if n == min1: continue
edge_exists = (graph.graph.has_edge(min1, n)
or graph.graph.has_edge(n, min1))
self.assert_(edge_exists)
# make sure the new edges have transition state data attached
for v in graph.graph.neighbors(min1):
data = graph.graph.get_edge_data(v, min1)
self.assertIn("ts", data.keys())
# make sure min2 is not in database
self.assertNotIn(min2, self.db.minima())
def test_merge_minima(self):
graph = TSGraph(self.db)
# select two minima with at least one edge
minima = []
for n in graph.graph.nodes():
if graph.graph.degree(n) > 0:
minima.append(n)
if len(minima) == 2:
break
nedges = [graph.graph.degree(n) for n in minima]
min1, min2 = minima
neighbors = graph.graph.neighbors(min2)
graph.mergeMinima(min1, min2)
self.db.mergeMinima(min1, min2)
self.assertNotIn(min2, graph.graph.nodes())
# make sure the edges were copied
for n in neighbors:
if n == min1: continue
edge_exists = (graph.graph.has_edge(min1, n)
or graph.graph.has_edge(n, min1))
self.assert_(edge_exists)
# make sure the new edges have transition state data attached
for v in graph.graph.neighbors(min1):
data = graph.graph.get_edge_data(v, min1)
self.assertIn("ts", list(data.keys()))
# make sure min2 is not in database
self.assertNotIn(min2, self.db.minima())
class DoubleEndedConnect(object):
"""
Find a connected network of minima and transition states between min1 and min2
Parameters
----------
min1, min2 : Mimumum() objects
the two minima to try to connect
pot : potential object
the potential
mindist : callable
the function which returns the optimized minimum distance between
two structures
database : pele Database object
Used to store the new minima and transition states found.
niter : int, optional
maximum number of iterations
verbosity : int
this controls how many status messages are printed. (not really
implemented yet)
merge_minima : bool
if True, minima for which NEB finds no transition state candidates
between them will be merged
max_dist_merge : float
merging minima will be aborted if the distance between them is greater
than max_dist_merge
local_connect_params : dict
parameters passed to the local connect algorithm. This includes all
NEB and all transition state search parameters, along with, e.g.
now many times to retry a local connect run. See documentation for
LocalConnect for details.
fresh_connect : bool
if true, ignore all existing minima and transition states in the
database and try to find a new path
longest_first : bool
if true, always try to connect the longest segment in the path guess
first
conf_checks : list of callables
a list of callable function that determine if a configuration is valid.
They must return a bool, and accept the keyword parameters
conf_check(energy=energy, coords=coords)
If any configuration in a minimum-transition_state-minimum triplet fails
a test then the whole triplet is rejected.
Notes
-----
The algorithm is iterative, with each iteration composed of
While min1 and min2 are not connected:
1) choose a pair of known minima to try to connect
2) use NEB to get a guess for the transition states between them
3) refine the transition states to desired accuracy
4) fall off either side of the transition states to find the two
minima associated with that candidate
5) add the transition states and associated minima to the known
network
Of the above, steps 1 and 2 and 3 are the most involved. 2, 3, 4 are
wrapped into a separate class called LocalConnect. See this class and
the NEB and FindTransitionState classes for detailed descriptions of
these procedures.
An important note is that the NEB is used only to get a *guess* for the
transition state. Thus we only want to put enough time and energy into
the NEB routine to get the guess close enough that FindTransitionState
can refine it to the correct transition state. FindTransitionState is
very fast if the initial guess is good, but can be very slow otherwise.
Choose a pair:
Here I will describe step 1), the algorithm to find a pair of known
minima to try to connect. This choice will keep in mind that the
ultimate goal is to connect min1 and min2.
In addition to the input parameter "graph", we keep a second graph
"Gdist" (now wrapped in a separate class _DistanceGraph) which also has
minima as the vertices. Gdist has an edge between every pair of nodes.
The edge weight between vertices u and v
is
if u and v are connected by transition states:
weight(u, v) = 0.
elif we have already tried local_connect on (u,v):
weight(u, v) = Infinity
else:
weight(u, v) = dist(u, v)**2
This edge weight is set to Infinity to ensure we don't repeat
LocalConnect runs over and over
again. The minimum weight path between min1 and min2 in Gdist gives a
good guess for the best way to try connect min1 and min2. So the
algorithm to find a pair of know minima (trial1, trial2) to try to
connect is
path = Gdist.minimum_weight_path(min1, min2)
trial1, trial2 = minima pair in path with lowest nonzero edge weight. (note:
if parameter longest_first is True) then the edgepair with the largest
edge weight will be selected)
todo:
allow user to pass graph
See Also
--------
DoubleEndedConnectPar : parallel version of this class
LocalConnect : the core algorithm of this routine
"""
def __init__(self, min1, min2, pot, mindist, database,
use_all_min=False, verbosity=1,
merge_minima=False,
max_dist_merge=0.1, local_connect_params=dict(),
fresh_connect=False, longest_first=True,
niter=200, conf_checks=None, load_no_distances=False
):
self.minstart = min1
assert min1._id == min1, "minima must compare equal with their id %d %s %s" % (min1._id, str(min1), str(min1.__hash__()))
self.minend = min2
self.pot = pot
self.mindist = mindist
self.pairsNEB = dict()
self.longest_first = longest_first
self.niter = niter
if conf_checks is None:
self.conf_checks = []
else:
self.conf_checks = conf_checks
self.verbosity = int(verbosity)
self.local_connect_params = dict([("verbosity",verbosity)] + local_connect_params.items())
self.database = database
self.fresh_connect = fresh_connect
if self.fresh_connect:
self.graph = TSGraph(self.database, minima=[self.minstart, self.minend], no_edges=True)
else:
self.graph = TSGraph(self.database)
self.merge_minima = merge_minima
self.max_dist_merge = float(max_dist_merge)
self.load_no_distances = load_no_distances
self.dist_graph = _DistanceGraph(self.database, self.graph, self.mindist, self.verbosity)
#check if a connection exists before initializing distance graph
if self.graph.areConnected(self.minstart, self.minend):
logger.info("minima are already connected. not initializing distance graph")
return
if use_all_min or load_no_distances:
print "warning: distances were removed from the database but you requested distances to be loaded"
self.dist_graph.initialize(self.minstart, self.minend)
if self.verbosity > 0:
logger.info("************************************************************")
logger.info("starting a double ended connect run between")
logger.info(" minimum 1: id %d energy %f" % (self.minstart._id, self.minstart.energy))
logger.info(" minimum 2: id %d energy %f" % (self.minend._id, self.minend.energy))
logger.info(" dist %f" % self.getDist(self.minstart, self.minend))
logger.info("************************************************************")
def mergeMinima(self, min1, min2):
"""merge two minimum objects
This will delete min2 and make everything that
pointed to min2 point to min1.
"""
#prefer to delete the minima with the large id. this potentially will be easier
if min2._id < min1._id:
min1, min2 = min2, min1
debug = False
dist = self.getDist(min1, min2)
logger.info( "merging minima %s %s %s %s %s", min1._id, min2._id, dist, "E1-E2", min1.energy - min2.energy)
#deal with the case where min1 and/or min2 are the same as minstart and/or minend
#make sure the one that is deleted (min2) is not minstart or minend
if ((min1 == self.minstart and min2 == self.minend) or
(min2 == self.minstart and min1 == self.minend)):
logger.error( "ERROR: trying to merge the start and end minima. aborting")
return
if min2 == self.minstart or min2 == self.minend:
min1, min2 = min2, min1
#print "min1 min2", min1, min2
if dist > self.max_dist_merge:
logger.info( " minima merge aborted. distance is too large %s", dist)
return
# if debug:
# #testing
# if min2 in self.database.minima():
# print "error, min2 is still in database"
# for ts in self.database.transition_states():
# if min2 == ts.minimum1 or min2 == ts.minimum2:
# print "error, a transition state attached to min2 is still in database", ts.minimum1._id, ts.minimum2._id
# merge minima in transition state graph
self.graph.mergeMinima(min1, min2)
# merge minima in the database also
self.database.mergeMinima(min1, min2)
if debug:
#testing
if min2 in self.graph.graph.nodes():
logger.error( "error, min2 is still in self.graph.graph")
logger.debug( "self.graph.graph. nnodes %s", self.graph.graph.number_of_nodes() )
#merge minima in distance graph
self.dist_graph.mergeMinima(min1, min2)
def getDist(self, min1, min2):
"""
get the distance between min1 and min2.
"""
return self.dist_graph.getDist(min1, min2)
def _addTransitionState(self, ts_ret, min_ret1, min_ret2):
"""
add a transition state to the database, the transition state graph and
the distance graph
"""
# if isinstance(min_ret1, tuple): # for compatability with old and new quenchers
# min_ret1 = min_ret1[4]
# if isinstance(min_ret2, tuple): # for compatability with old and new quenchers
# min_ret2 = min_ret2[4]
# if isinstance(min1_ret)
# sanity check for the energies
me1, me2 = min_ret1.energy, min_ret2.energy
if ts_ret.energy < me1 or ts_ret.energy < me2:
logger.warning("trying to add a transition state that has energy lower than its minima.")
logger.warning(" TS energy %s %s %s %s", ts_ret.energy, "minima energy", me1, me2 )
logger.warning(" aborting" )
return False
# check the minima and transition states are valid configurations.
# if any fail, then don't add anything.
configs_ok = True
for ret in [ min_ret1, min_ret2, ts_ret ]:
for check in self.conf_checks:
if not check(energy=ret.energy, coords=ret.coords):
configs_ok = False
break
if not configs_ok:
break
if not configs_ok:
return False
# Add the minima to the database
min1 = self.database.addMinimum(min_ret1.energy, min_ret1.coords)
min2 = self.database.addMinimum(min_ret2.energy, min_ret2.coords)
# Add the minima to the transition state graph.
self.graph.addMinimum(min1)
self.graph.addMinimum(min2)
if min1 == min2:
logger.warning( "stepping off the transition state resulted in twice the same minima %s", min1._id)
return False
logger.info("adding transition state %s %s", min1._id, min2._id)
# add the transition state to the database
ts = self.database.addTransitionState(ts_ret.energy, ts_ret.coords, min1, min2, eigenvec=ts_ret.eigenvec, eigenval=ts_ret.eigenval)
# update the transition state graph
self.graph.addTransitionState(ts)
# self.graph.refresh()
#update the distance graph
self.dist_graph.addMinimum(min1)
self.dist_graph.addMinimum(min2)
self.dist_graph.setTransitionStateConnection(min1, min2)
if self.verbosity > 1:
#print some information
dse = self.getDist(self.minend, self.minstart)
msid = self.minstart._id
meid = self.minend._id
m1id = min1._id
m2id = min2._id
if min1 != self.minstart and min1 != self.minend:
ds = self.getDist(min1, self.minstart)
de = self.getDist(min1, self.minend)
if ds < dse > de:
triangle = ""
else:
triangle = ": new minima not in between start and end"
logger.info(" distances: %4d -> %4d = %f %4d -> %4d = %f %4d -> %4d = %f %s" % (msid, m1id, ds, m1id, meid, de, m1id, m2id, dse, triangle))
#print " dist new min 1 to minstart, minend ", ds, de, dse
if min2 != self.minstart and min2 != self.minend:
ds = self.getDist(min2, self.minstart)
de = self.getDist(min2, self.minend)
if ds < dse > de:
triangle = ""
else:
triangle = ": new minima not in between start and end"
logger.info( " distances: %4d -> %4d = %f %4d -> %4d = %f %4d -> %4d = %f" % (msid, m2id, ds, m2id, meid, de, m2id, m2id, dse))
return True
def _getLocalConnectObject(self):
return LocalConnect(self.pot, self.mindist, **self.local_connect_params)
def _localConnect(self, min1, min2):
"""
do a local connect run between min1 and min2
Notes
-----
1) NEB to find transition state candidates.
for each transition state candidate:
2) refine the transition state candidates
3) if successful, fall off either side of the transition state
to find the minima the transition state connects. Add the new
transition state and minima to the graph
"""
#Make sure we haven't already tried this pair and
#record some data so we don't try it again in the future
if self.pairsNEB.has_key((min1, min2)):
logger.warning("WARNING: redoing NEB for minima %s %s", min1._id, min2._id)
logger.warning(" aborting NEB")
#self._remove_edgeGdist(min1, min2)
self.dist_graph.removeEdge(min1, min2)
return True
self.pairsNEB[(min1, min2)] = True
self.pairsNEB[(min2, min1)] = True
#Make sure they're not already connected. sanity test
if self.graph.areConnected(min1, min2):
logger.warning("in _local_connect, but minima are already connected. aborting %s %s %s", min1._id, min2._id, self.getDist(min1, min2))
self.dist_graph.setTransitionStateConnection(min1, min2)
self.dist_graph.checkGraph()
return True
#do local connect run
local_connect = self._getLocalConnectObject()
res = local_connect.connect(min1, min2)
#now add each new transition state to the graph and database.
nsuccess = 0
for tsret, m1ret, m2ret in res.new_transition_states:
goodts = self._addTransitionState(tsret, m1ret, m2ret)
if goodts:
nsuccess += 1
#check results
#nclimbing = len(climbing_images)
#print "from NEB search found", nclimbing, "transition state candidates"
if nsuccess == 0:
dist = self.getDist(min1, min2)
if dist < self.max_dist_merge:
logger.warning("local connect failed and the minima are close. Are the minima really the same?")
logger.warning(" energies: %s %s %s %s", min1.energy, min2.energy, "distance", dist)
if self.merge_minima:
self.mergeMinima(min1, min2)
else:
logger.warning(" set merge_minima=True to merge the minima")
return False
#remove this edge from Gdist so we don't try this pair again again
self.dist_graph.removeEdge(min1, min2)
return nsuccess > 0
def _getNextPair(self):
"""
return a pair of minima to attempt to connect
Notes
-----
this is the function which attempts to find a clever pair of minima to try to
connect with the ultimate goal of connecting minstart and minend
this method can be described as follows:
make a new graph Gnew which is complete (all vetices connected). The edges
have a weight given by
if self.graph.areConnected(u,v):
weight(u,v) = 0.
else:
weight(u,v) = mindist(u,v)
if an NEB has been attempted between u and v then the edge is removed.
we then find the shortest path between minstart and minend. We return
the pair in this path which has edge weight with the smallest non zero value
update: find the shortest path weighted by distance squared. This penalizes finding
the NEB between minima that are very far away. (Does this too much favor long paths?)
"""
logger.info("finding a good pair to try to connect")
#get the shortest path on dist_graph between minstart and minend
if True:
logger.debug("Gdist has %s %s %s %s", self.dist_graph.Gdist.number_of_nodes(),
"nodes and", self.dist_graph.Gdist.number_of_edges(), "edges")
path, weights = self.dist_graph.shortestPath(self.minstart, self.minend)
weightsum = sum(weights)
if path is None or weightsum >= 10e9:
logger.warning("Can't find any way to try to connect the minima")
return None, None
#get the weights of the path segements
weightlist = []
for i in range(1,len(path)):
min1 = path[i-1]
min2 = path[i]
# w = weights.get((min1,min2))
# if w is None:
# w = weights.get((min2,min1))
w = weights[i-1]
weightlist.append( (w, min1, min2) )
if True:
#print the path
logger.info("best guess for path. (dist=0.0 means the path is known)")
for w, min1, min2 in weightlist:
if w > 1e-6:
dist = self.getDist(min1, min2)
else:
dist = w
logger.info(" path guess %s %s %s", min1._id, min2._id, dist)
#select which minima pair to return
if self.longest_first:
weightlist.sort()
w, min1, min2 = weightlist[-1]
else:
weightlist.sort()
for w, min1, min2 in weightlist:
if w > 1e-6:
break
return min1, min2
def connect(self):
"""
the main loop of the algorithm
"""
self.NEBattempts = 2;
for i in range(self.niter):
#stop if we're done
if self.graph.areConnected(self.minstart, self.minend):
logger.info("found connection!")
return
logger.info("")
logger.info("======== starting connect cycle %s %s", i, "========")
#get pair of minima to try to connect
min1, min2 = self._getNextPair()
#fail if we can't find a good pair to try
if min1 is None or min2 is None:
break
#try to connect those minima
from pele.optimize.optimization_exceptions import LineSearchError
try:
local_success = self._localConnect(min1, min2)
except LineSearchError as err:
print err
print "caught line search error, aborting connection attempt"
break
if False and i % 10 == 0:
#do some sanity checks
self.dist_graph.checkGraph()
logger.info("failed to find connection between %s %s", self.minstart._id, self.minend._id)
def success(self):
return self.graph.areConnected(self.minstart, self.minend)
def returnPath(self):
"""return information about the path
Returns
-------
mints : list of Minimum and TransitionStates
a list of Minimum, TransitionState, Minimum objects that make up
the path
S : list of float
numpy array of the distance along the path. len(S) == len(mints)
energies : list of float
numpy array of the energies along the path
If the minima are not connected, return (None, None, None)
"""
if not self.graph.areConnected(self.minstart, self.minend):
return None, None, None
minima = nx.shortest_path(self.graph.graph, self.minstart, self.minend)
transition_states = []
mints = [minima[0]]
for i in range(1,len(minima)):
m1 = minima[i-1]
m2 = minima[i]
ts = self.database.getTransitionState(m1, m2)
transition_states.append(ts)
mints.append(ts)
mints.append(m2)
S = np.zeros(len(mints))
for i in range(1,len(mints)):
coords1 = mints[i-1].coords
coords2 = mints[i].coords
dist, c1, c2 = self.mindist(coords1, coords2)
S[i] = S[i-1] + dist
energies = np.array([m.energy for m in mints])
return mints, S, energies
class DoubleEndedConnect(object):
"""
Find a connected network of minima and transition states between min1 and min2
Parameters
----------
min1, min2 : Mimumum() objects
the two minima to try to connect
pot : potential object
the potential
mindist : callable
the function which returns the optimized minimum distance between
two structures
database : pele Database object
Used to store the new minima and transition states found.
niter : int, optional
maximum number of iterations
verbosity : int
this controls how many status messages are printed. (not really
implemented yet)
merge_minima : bool
if True, minima for which NEB finds no transition state candidates
between them will be merged
max_dist_merge : float
merging minima will be aborted if the distance between them is greater
than max_dist_merge
local_connect_params : dict
parameters passed to the local connect algorithm. This includes all
NEB and all transition state search parameters, along with, e.g.
now many times to retry a local connect run. See documentation for
LocalConnect for details.
fresh_connect : bool
if true, ignore all existing minima and transition states in the
database and try to find a new path
longest_first : bool
if true, always try to connect the longest segment in the path guess
first
conf_checks : list of callables
a list of callable function that determine if a configuration is valid.
They must return a bool, and accept the keyword parameters
conf_check(energy=energy, coords=coords)
If any configuration in a minimum-transition_state-minimum triplet fails
a test then the whole triplet is rejected.
Notes
-----
The algorithm is iterative, with each iteration composed of
While min1 and min2 are not connected:
1) choose a pair of known minima to try to connect
2) use NEB to get a guess for the transition states between them
3) refine the transition states to desired accuracy
4) fall off either side of the transition states to find the two
minima associated with that candidate
5) add the transition states and associated minima to the known
network
Of the above, steps 1 and 2 and 3 are the most involved. 2, 3, 4 are
wrapped into a separate class called LocalConnect. See this class and
the NEB and FindTransitionState classes for detailed descriptions of
these procedures.
An important note is that the NEB is used only to get a *guess* for the
transition state. Thus we only want to put enough time and energy into
the NEB routine to get the guess close enough that FindTransitionState
can refine it to the correct transition state. FindTransitionState is
very fast if the initial guess is good, but can be very slow otherwise.
Choose a pair:
Here I will describe step 1), the algorithm to find a pair of known
minima to try to connect. This choice will keep in mind that the
ultimate goal is to connect min1 and min2.
In addition to the input parameter "graph", we keep a second graph
"Gdist" (now wrapped in a separate class _DistanceGraph) which also has
minima as the vertices. Gdist has an edge between every pair of nodes.
The edge weight between vertices u and v
is
if u and v are connected by transition states:
weight(u, v) = 0.
elif we have already tried local_connect on (u,v):
weight(u, v) = Infinity
else:
weight(u, v) = dist(u, v)**2
This edge weight is set to Infinity to ensure we don't repeat
LocalConnect runs over and over
again. The minimum weight path between min1 and min2 in Gdist gives a
good guess for the best way to try connect min1 and min2. So the
algorithm to find a pair of know minima (trial1, trial2) to try to
connect is
path = Gdist.minimum_weight_path(min1, min2)
trial1, trial2 = minima pair in path with lowest nonzero edge weight. (note:
if parameter longest_first is True) then the edgepair with the largest
edge weight will be selected)
todo:
allow user to pass graph
See Also
--------
LocalConnect : the core algorithm of this routine
"""
def __init__(self,
min1,
min2,
pot,
mindist,
database,
verbosity=1,
merge_minima=False,
max_dist_merge=0.1,
local_connect_params=None,
fresh_connect=False,
longest_first=True,
niter=200,
conf_checks=None):
self.minstart = min1
assert min1.id(
) == min1, "minima must compare equal with their id %d %s %s" % (
min1.id(), str(min1), str(min1.__hash__()))
self.minend = min2
self.pot = pot
self.mindist = mindist
self.pairsNEB = dict()
self.longest_first = longest_first
self.niter = niter
if conf_checks is None:
self.conf_checks = []
else:
self.conf_checks = conf_checks
self.verbosity = int(verbosity)
if local_connect_params is None:
local_connect_params = dict()
self.local_connect_params = dict([("verbosity", verbosity)] +
local_connect_params.items())
self.database = database
self.fresh_connect = fresh_connect
if self.fresh_connect:
self.graph = TSGraph(self.database,
minima=[self.minstart, self.minend],
no_edges=True)
else:
self.graph = TSGraph(self.database)
self.merge_minima = merge_minima
self.max_dist_merge = float(max_dist_merge)
self.dist_graph = _DistanceGraph(self.database, self.graph,
self.mindist, self.verbosity)
# check if a connection exists before initializing distance graph
if self.graph.areConnected(self.minstart, self.minend):
logger.info(
"minima are already connected. not initializing distance graph"
)
return
self.dist_graph.initialize(self.minstart, self.minend)
if self.verbosity > 0:
logger.info(
"************************************************************")
logger.info("starting a double ended connect run between")
logger.info(" minimum 1: id %d energy %f" %
(self.minstart.id(), self.minstart.energy))
logger.info(" minimum 2: id %d energy %f" %
(self.minend.id(), self.minend.energy))
logger.info(" dist %f" %
self.getDist(self.minstart, self.minend))
logger.info(
"************************************************************")
def mergeMinima(self, min1, min2):
"""merge two minimum objects
This will delete min2 and make everything that
pointed to min2 point to min1.
"""
# prefer to delete the minima with the large id. this potentially will be easier
if min2.id() < min1.id():
min1, min2 = min2, min1
debug = False
dist = self.getDist(min1, min2)
logger.info("merging minima %s %s %s %s %s", min1.id(), min2.id(),
dist, "E1-E2", min1.energy - min2.energy)
# deal with the case where min1 and/or min2 are the same as minstart and/or minend
# make sure the one that is deleted (min2) is not minstart or minend
if ((min1 == self.minstart and min2 == self.minend)
or (min2 == self.minstart and min1 == self.minend)):
logger.error(
"ERROR: trying to merge the start and end minima. aborting")
return
if min2 == self.minstart or min2 == self.minend:
min1, min2 = min2, min1
if dist > self.max_dist_merge:
logger.info(" minima merge aborted. distance is too large %s",
dist)
return
# merge minima in transition state graph
self.graph.mergeMinima(min1, min2)
# merge minima in the database also
self.database.mergeMinima(min1, min2)
if debug:
# testing
if min2 in self.graph.graph.nodes():
logger.error("error, min2 is still in self.graph.graph")
logger.debug("self.graph.graph. nnodes %s",
self.graph.graph.number_of_nodes())
# merge minima in distance graph
self.dist_graph.mergeMinima(min1, min2)
def getDist(self, min1, min2):
"""
get the distance between min1 and min2.
"""
return self.dist_graph.getDist(min1, min2)
def _addTransitionState(self, ts_ret, min_ret1, min_ret2):
"""
add a transition state to the database, the transition state graph and
the distance graph
"""
# if isinstance(min_ret1, tuple): # for compatability with old and new quenchers
# min_ret1 = min_ret1[4]
# if isinstance(min_ret2, tuple): # for compatability with old and new quenchers
# min_ret2 = min_ret2[4]
# if isinstance(min1_ret)
# sanity check for the energies
me1, me2 = min_ret1.energy, min_ret2.energy
if ts_ret.energy < me1 or ts_ret.energy < me2:
logger.warning(
"trying to add a transition state that has energy lower than its minima."
)
logger.warning(" TS energy %s %s %s %s", ts_ret.energy,
"minima energy", me1, me2)
logger.warning(" aborting")
return False
# check the minima and transition states are valid configurations.
# if any fail, then don't add anything.
configs_ok = True
for ret in [min_ret1, min_ret2, ts_ret]:
for check in self.conf_checks:
if not check(energy=ret.energy, coords=ret.coords):
configs_ok = False
break
if not configs_ok:
break
if not configs_ok:
return False
# Add the minima to the database
min1 = self.database.addMinimum(min_ret1.energy, min_ret1.coords)
min2 = self.database.addMinimum(min_ret2.energy, min_ret2.coords)
# Add the minima to the transition state graph.
self.graph.addMinimum(min1)
self.graph.addMinimum(min2)
if min1 == min2:
logger.warning(
"stepping off the transition state resulted in twice the same minima %s",
min1.id())
return False
logger.info("adding transition state %s %s", min1.id(), min2.id())
# add the transition state to the database
ts = self.database.addTransitionState(ts_ret.energy,
ts_ret.coords,
min1,
min2,
eigenval=ts_ret.eigenval)
# update the transition state graph
self.graph.addTransitionState(ts)
# self.graph.refresh()
# update the distance graph
self.dist_graph.addMinimum(min1)
self.dist_graph.addMinimum(min2)
self.dist_graph.setTransitionStateConnection(min1, min2)
if self.verbosity > 1:
# print some information
dse = self.getDist(self.minend, self.minstart)
msid = self.minstart.id()
meid = self.minend.id()
m1id = min1.id()
m2id = min2.id()
if min1 != self.minstart and min1 != self.minend:
ds = self.getDist(min1, self.minstart)
de = self.getDist(min1, self.minend)
if ds < dse > de:
triangle = ""
else:
triangle = ": new minima not in between start and end"
logger.info(
" distances: %4d -> %4d = %f %4d -> %4d = %f %4d -> %4d = %f %s"
% (msid, m1id, ds, m1id, meid, de, m1id, m2id, dse,
triangle))
if min2 != self.minstart and min2 != self.minend:
ds = self.getDist(min2, self.minstart)
de = self.getDist(min2, self.minend)
# if ds < dse > de:
# triangle = ""
# else:
# triangle = ": new minima not in between start and end"
logger.info(
" distances: %4d -> %4d = %f %4d -> %4d = %f %4d -> %4d = %f"
% (msid, m2id, ds, m2id, meid, de, m2id, m2id, dse))
return True
def _getLocalConnectObject(self):
return LocalConnect(self.pot, self.mindist,
**self.local_connect_params)
def _localConnect(self, min1, min2):
"""
do a local connect run between min1 and min2
Notes
-----
1) NEB to find transition state candidates.
for each transition state candidate:
2) refine the transition state candidates
3) if successful, fall off either side of the transition state
to find the minima the transition state connects. Add the new
transition state and minima to the graph
"""
# Make sure we haven't already tried this pair and
# record some data so we don't try it again in the future
if (min1, min2) in self.pairsNEB:
logger.warning("WARNING: redoing NEB for minima %s %s", min1.id(),
min2.id())
logger.warning(" aborting NEB")
# self._remove_edgeGdist(min1, min2)
self.dist_graph.removeEdge(min1, min2)
return True
self.pairsNEB[(min1, min2)] = True
self.pairsNEB[(min2, min1)] = True
# Make sure they're not already connected. sanity test
if self.graph.areConnected(min1, min2):
logger.warning(
"in _local_connect, but minima are already connected. aborting %s %s %s",
min1.id(), min2.id(), self.getDist(min1, min2))
self.dist_graph.setTransitionStateConnection(min1, min2)
self.dist_graph.checkGraph()
return True
# do local connect run
local_connect = self._getLocalConnectObject()
res = local_connect.connect(min1, min2)
# now add each new transition state to the graph and database.
nsuccess = 0
for tsret, m1ret, m2ret in res.new_transition_states:
goodts = self._addTransitionState(tsret, m1ret, m2ret)
if goodts:
nsuccess += 1
# check results
if nsuccess == 0:
dist = self.getDist(min1, min2)
if dist < self.max_dist_merge:
logger.warning(
"local connect failed and the minima are close. Are the minima really the same?"
)
logger.warning(" energies: %s %s %s %s", min1.energy,
min2.energy, "distance", dist)
if self.merge_minima:
self.mergeMinima(min1, min2)
else:
logger.warning(
" set merge_minima=True to merge the minima")
return False
# remove this edge from Gdist so we don't try this pair again again
self.dist_graph.removeEdge(min1, min2)
return nsuccess > 0
def _getNextPair(self):
"""
return a pair of minima to attempt to connect
Notes
-----
this is the function which attempts to find a clever pair of minima to try to
connect with the ultimate goal of connecting minstart and minend
this method can be described as follows:
make a new graph Gnew which is complete (all vetices connected). The edges
have a weight given by
if self.graph.areConnected(u,v):
weight(u,v) = 0.
else:
weight(u,v) = mindist(u,v)
if an NEB has been attempted between u and v then the edge is removed.
we then find the shortest path between minstart and minend. We return
the pair in this path which has edge weight with the smallest non zero value
update: find the shortest path weighted by distance squared. This penalizes finding
the NEB between minima that are very far away. (Does this too much favor long paths?)
"""
logger.info("finding a good pair to try to connect")
# get the shortest path on dist_graph between minstart and minend
if True:
logger.debug("Gdist has %s %s %s %s",
self.dist_graph.Gdist.number_of_nodes(), "nodes and",
self.dist_graph.Gdist.number_of_edges(), "edges")
path, weights = self.dist_graph.shortestPath(self.minstart,
self.minend)
weightsum = sum(weights)
if path is None or weightsum >= 10e9:
logger.warning("Can't find any way to try to connect the minima")
return None, None
# get the weights of the path segements
weightlist = []
for i in range(1, len(path)):
min1 = path[i - 1]
min2 = path[i]
# w = weights.get((min1,min2))
# if w is None:
# w = weights.get((min2,min1))
w = weights[i - 1]
weightlist.append((w, min1, min2))
if True:
# print the path
logger.info(
"best guess for path. (dist=0.0 means the path is known)")
for w, min1, min2 in weightlist:
if w > 1e-6:
dist = self.getDist(min1, min2)
else:
dist = w
logger.info(" path guess %s %s %s", min1.id(), min2.id(),
dist)
# select which minima pair to return
if self.longest_first:
weightlist.sort()
w, min1, min2 = weightlist[-1]
else:
weightlist.sort()
for w, min1, min2 in weightlist:
if w > 1e-6:
break
return min1, min2
def connect(self):
"""
the main loop of the algorithm
"""
self.NEBattempts = 2
for i in range(self.niter):
# stop if we're done
if self.graph.areConnected(self.minstart, self.minend):
logger.info("found connection!")
return
logger.info("")
logger.info("======== starting connect cycle %s %s", i, "========")
# get pair of minima to try to connect
min1, min2 = self._getNextPair()
# fail if we can't find a good pair to try
if min1 is None or min2 is None:
break
# try to connect those minima
from pele.optimize.optimization_exceptions import LineSearchError
try:
self._localConnect(min1, min2)
except LineSearchError as err:
print err
print "caught line search error, aborting connection attempt"
break
if False and i % 10 == 0:
# do some sanity checks
self.dist_graph.checkGraph()
logger.info("failed to find connection between %s %s",
self.minstart.id(), self.minend.id())
def success(self):
return self.graph.areConnected(self.minstart, self.minend)
def returnPath(self):
"""return information about the path
Returns
-------
mints : list of Minimum and TransitionStates
a list of Minimum, TransitionState, Minimum objects that make up
the path
S : list of float
numpy array of the distance along the path. len(S) == len(mints)
energies : list of float
numpy array of the energies along the path
If the minima are not connected, return (None, None, None)
"""
if not self.graph.areConnected(self.minstart, self.minend):
return None, None, None
minima = nx.shortest_path(self.graph.graph, self.minstart, self.minend)
transition_states = []
mints = [minima[0]]
for i in range(1, len(minima)):
m1 = minima[i - 1]
m2 = minima[i]
ts = self.database.getTransitionState(m1, m2)
transition_states.append(ts)
mints.append(ts)
mints.append(m2)
S = np.zeros(len(mints))
for i in range(1, len(mints)):
coords1 = mints[i - 1].coords
coords2 = mints[i].coords
dist, c1, c2 = self.mindist(coords1, coords2)
S[i] = S[i - 1] + dist
energies = np.array([m.energy for m in mints])
return mints, S, energies