Table of Contents

Ensemblefile


The ensemblefile.uq output by the uncertainty quantification ensemble method compilation contains the list of accepted MCMC potentials, see uncertainty quantification.

Format

The file begins by outputting the calculated Hessian followed by a row listing its eigenvalues. This is followed by the corresponding matrix of eigenvectors, with the eigenvectors of each principle axes in each column, directly underneath it's related eigenvalue. Then the accepted MCMC steps are listed, with the final three output columns for performance analysis purposes.

id
an identifier for the potential (the best fit potential parameter set is the first output in the following part of the output file, with an id of 0.)

param_1 param_2 … param_N
the potential parameter set for each ensemble member

cost
the fitting cost for the potential

weight
the corresponding potential weight (this is the number of MCMC trial steps proposed before the next potential is accepted)

accepted
a flag of value 1 denotes that the step was accepted (users can have potfit output all unaccepted steps when compiling with the -debug option, see compiling, with unsuccessful attempts referred to by a flag of 0 in this column)

attempts
the total number of attempted MCMC moves, used to calculate the acc_prob (obviously when generating a potential ensemble users should let the Markov chain burn is to attain the correct acceptance probability before attempting to tune this using the R value, acc_rescaling input parameter, see input parameters)

acc_prob
proportion of accepted MCMC potential parameter sets

Example

An example output for a fictitious LJ potential fit is illustrated below:

# hessian:
# 804480.9 713749.4
# 713749.4 1411542.1
#
# eigenvalues: (l1, l2, ..., lN)
# 332402.6 1883620.4
#
# eigenvectors; (v1, v2, ..., vN)
# 0.83406   0.55165
# -0.551658 0.834069
#
# id param_1 param_2 cost weight accepted attempts acc_prob
0    0.6830  2.1120  62.5 10     1        10       0.10
1    0.6808  2.1137  63.8 10     1        20       0.10