Glossary

idx

short for index

idxs

short for indices

weighted ensemble

A class of enhanced sampling algorithms that iteratively perform parallel sampling coupled with resampling of the replicas, called walkers.

A converged weighted ensemble simulation gives the exact probability distribution being sampled.

walker

The individual simulatio replicas of a weighted ensemble simulation. Each walker has a state and a weight corresponding to the probability.

trajectory

A contiguous block of trajectory frames that is not necessarily (or even usually) logically contiguous.

In other words an implementation detail.

trajectory fields

The types of data associated with a trajectory is left largely unspecified making it a general purpose container for any type of runner or state that is produced through simulation.

Typically, this includes positions, box vectors etc.

Fields are named by strings, e.g. “positions” and special compound fields are separated by the HDF5 path separator e.g. “alt_reps/alt_rep1”, “observables/rmsd”.

records

Data associated with walkers and a run.

Takes the form of a list of tuples (usually namedtuple records).

continual records

Records for a wepy simulation for which one is produced every cycle.

sporadic records

Records for a wepy simulation which may have none or many produced for a cycle.

resampling records

Records that report on the resampling of walkers i.e. cloning and merging. Contains the fields of a decision record. A destructured resampling panel.

resampler records

Records that report on the state of the resampler itself.

resampling step

A single cycle of resampling may be broken into several steps that are applied in order. A single step is a collection of decision records for each walker in the cohort. This, for example, allows for recursive clones, which would be very difficult to encode in a simple decision set.

resampling panel

This is the datastructure that describes the full process of resampling for an entire run or contig.

A single step of resampling results in a resampling record for each walker and the list of these records (in order of walker ordering) is one “step” of a “cycle” resampling.

A resampling panel is a list of lists where each element of the list contains the resampling steps of that cycle.

The cycle element is a list of records.

The structure for the resampling panel looks like this roughly:

[
  cycle_0 [
            step_0 [rec0, rec1, ...],
            step_1 [rec0, rec1, ...],
            ...
          ]
  ...
]

parent table

A listing for each cycle snapshot of a weighted ensemble contig that specifies the parents of the next generation walkers. There can be different number of walkers in each cycle and thus is a list of lists of ints.

Specifically:

For a parent table P with M cycles, N_i is the number of walkers in cycle i has a listing P_i which has N_{i+1} elements. For walker j, P_i^j, has a domain 0 to N_i and specifies the index of the walker in cycle i that is the parent of walker j in cycle i+1.

parent panel

A parent panel is similar in structure to the resampling panel except that instead of resampling records as the atomic elements there are simply integers with the same meaning as in the parent table.

warping records

Records that report on events of walkers satisfying boundary conditions in the last cycle.

Warping records are typically treated as either discontinuous (e.g. the walker’s state is set to a specified other state) or continuous (e.g. the “color” of a walker changes as it reaches a boundary).

The interpretation of this however is entirely domain specific.

bc (boundary conditions) records

Records that report on the state of the boundary conditions themselves.

progress records

Continual records that a boundary condition reports that provide some metrics on the walkers, such as their position along a progress coordinate.

run

A single weighted ensemble simulation. Is a collection of walker trajectories and records specifying parental relations, boundary condition events etc.

Runs may be logically connected to other runs (through continuations) and can be thought of as a unit of computation.

continuation

A 2-tuple of the form (continuation_run_idx, continued_run_idx) that establishes a logical relationship between the end of one run (continued_run_idx) and the beginning of another run (continuation_run_idx).

A collection of continuations is the underlying specification for a contig tree for run contigs.

contig tree

For a collection of weighted ensemble runs with continuations specified between them, the contig tree imposes the tree structure over them.

Technically a forest since there can be multiple roots.

contig

A logically contiguous list of weighted ensemble cycle snapshots.

spanning contig

A contig from a contig tree that extends from a root to a leaf

run contig

A logically contiguous list of weighted ensemble runs. The only continuation relationship allowed is end-to-end.

trace

list of tuples which indexes things over either a contig tree, contig, or a run.

run trace

list of tuples of the form (run_idx, traj_idx, cycle_idx) over the contig tree

contig trace

list of tuples of the form (run_idx, cycle_idx) over the contig tree

contig walker trace

list of tuples of the form (traj_idx, cycle_idx) over a contig

run cycle idx

An index of an in-run cycle

contig cycle idx

An index of an in-contig cycle index (can include multiple runs).

resampler

An object that resamples walkers each cycle of a weighted ensemble simulation.

runner

The object that propogates the sampling of an individual walker.

reporter

A reporter is an object that receives all of the walker states and records at the end of the cycle and is allowed to write it out in some way.

resampling decision

A specification of a type of action to take on a sample within a sample set.

decision record

A specific key-value data structure that represents a resampling decision. Must contain a representation of the decision value (enum or its integer value) and any additional fields (instructions) that are needed to perform the resampling, such as the number of clones or which walker to merge with. Contained within a resampling record which adds metadata which places it within the context of an entire simulation.

decision instruction record

See decision record.