For an example of multiple mpirun's in a single PBS job see: [VASP example](https://code.ornl.gov/CNMS/CNMS_Computing_Resources/blob/master/CADES/VASP/VASP_OneJob_Parallel_Runs)
## Chaining pbs jobs
This is actually a pretty good primer on job chaining techniques.
If you combine the above two you can run many long trajectories or relaxations via backfilling without fragmenting the hpc environment and making highly parallel jobs difficult to launch.