triclinic box skew too large in restart file

Dear all,

I run an NPT simulation with a triclinic box. At the end of the run a restart file is generated. If I try to read that restart file to start another run, it gives the error: triclinic box skew is too large. I thought that if the skew got too large during a simulation, lammps would flip the box so that it was no longer too large. Why is this happening?
Also, this does not always happen; i have had success reading different restart files in the past. With this specific system though, this usually happens.
My version of lammps is 24Apr13

Thanks, Steven

Fix npt with triclinic boxes does not flip when the tilt factor
exceeds +/- 0.5, like fix deform does. Rather it allows
an extra delta in the tilt to avoid the box sitting right
at 0.5 and flipping back and forth. So it lets the tilt
be as big as 0.6 before flipping to -0.4 and vice versa.

If you saved a restart when the tilt exceeded 0.5, you
can use the “box tilt large” command to allow
it to be read back in. Assuming you are not going
to deform it more dramatically in the subsequent run,
that should be fine.