Some comments on DL_POLY input directives

Trajectory directive

Format: traj nstraj istraj keytrj.

Meaning of the parameters:

nstraj

start dumping to HISTORY file at this time step
So this number is the number specified in the equilibration steps directive $\mathbf{+1}$

istraj

timestep interval between two successive dumps

keytrj

what to write to HISTORY file:

0	coordinates only
1	coordinates + velocities
2	coordinates + velocities + forces

The total number of dumps ndumps is given by:

$$\begin{eqnarray*} \texttt{ndumps}&=&\frac{\texttt{N}-\texttt{nstraj}}{\texttt{istraj}}+1 \end{eqnarray*}$$

where N is the total number of steps (the value of the steps directive).

So, to dump ndumps configs:

$$\begin{eqnarray*} \texttt{N}&=&\texttt{nstraj} + \texttt{istraj}*(\texttt{ndumps} - 1) \end{eqnarray*}$$

Rolling averages: stack directive

The meaning of the stack directive is as follows: when DL_POLY prints instantaneous data to the OUTPUT file (this depends upon the print directive, or is a default time interval), it will also print the rolling averages evaluated on as many simulation steps as specified by the stack directive. In other words, at step $N$, having specified a value of $S$ in the stack directive, the rolling average of a quantity $A$ is computed as :

$$\begin{eqnarray*} \mathit{rolling\ average}&=&\frac{A_N+A_{(N-1)}+\cdots+A_{(N-S+1)}}{S} \end{eqnarray*}$$

So, the rolling averages in the OUTPUT file, are NOT accumulating averages!

Job and close time

job time is the total time the program will run: after that, the program will stop. close time is the time needed by the program to gracefully stop itself (e.g. terminating the current cycle). When:

$$\texttt{CPU time}>(\texttt{job time}-\texttt{close time})$$

the program will start closing operations and will stop.

NOTE: for reasons which I have not deeply investigated, it may happen that DL_POLY and the machine which it is running on have different ideas about time. For instance, on t3e.cineca.it the time limit for each single PE is 10800s, but I've found empirically that this corresponds to a limit of about 1000s for the DL_POLY job time directive.