The data in this folder were used in the paper "Empirical Classification of Trajectory Data: An Opportunity for the Use of Machine Learning in Molecular Dynamics." Carpenter, B. K.; Ezra, G. S.; Farantos, S. C.; Kramer, Z. C.; Wiggins, S. J. Phys. Chem. B, 2017, 121, in press, doi: 10.1021/acs.jpcb.7b08707.

There are two files, cirque_Cart_train.txt and cirque_Cart_test.txt.  Each is a plain text file consisting of six columns of numbers, five reals and one integer per row.  The real numbers specify the initial conditions for a trajectory on the potential described in the paper.  The first two numbers are respectively the x and y Cartesian coordinates of the starting point, while the third and fourth numbers specify respectively the initial x and y components of the momentum.  The fifth real number is the initial angular momentum, with respect to an origin at the origin of the potential.  The integer in the sixth column specifies which connectivity class the trajectory belongs to, according to the following code:

1	AA
2	AB
3	AC
4	BB
5	BC
6	CC

The meanings of the two-letter classifications are provided in the paper.

Some of the data in the files have a classification code of 0.  These correspond to trajectories that failed to satisfy one of the termination criteria after 1000 time units.  Subsequent to the creation of the data sets, it was recognized that all such instances correspond to trapped trajectories inside one of the Type 3 periodic orbits (see paper for definition).  They rightfully belong, therefore, to either the AA or BB classification.  The choice is made simply from the sign of x.  If x<0 then a trajectory with classification 0 should be assigned to AA, but if x>0 it should be assigned to BB.

The two files correspond to data used in both the manual analysis and the neural network analysis described in the paper.  In each case, parameter values were determined from the data in the file cirque_Cart_train.txt and subsequently tested against the data in the file cirque_Cart_test.txt.