Thank you for your suggestions, I have read Voter’s paper ( J Chem Phys, 112, 9599(2000)) again, I try to simulate the example about diffusion on the surface Ag(100). But I got a strange result, as is shown here,

STEP | CPU | N | M | STATUS | Barrier | Margin | t_lo | delt_lo |

• | - | - | - | - | - | - | - | - |

311550 | 6574.862 | 1 | 15 | DF | 0 | 0.843 | 9109070514 | 216452.729 |
311550 | 6574.862 | 2 | 15 | E | 0 | 0 | 9109286966 | 216452.729 |
316550 | 6606.801 | 2 | 0 | DF | 0 | 0 | 9109286966 | 5000 |
316550 | 6606.801 | 3 | 0 | E | 0 | 0 | 9109291966 | 5000 |
431200 | 7184.133 | 3 | 0 | DF | 0.478 | 0 | 9109291966 | 11975905845 |
431450 | 7187.26 | 3 | 1 | D | 0.478 | 0.117 | 9109291966 | 12002599872 |

I am very stange about the the barrier is zero. I don’t know why there is a zero barrier, is it wrong in NEB or something else?

I check these results in the log.neb again, I find that the EBF is zero, but when I calculate it as the Manual say( the forward energy barrier is the potential energy of the highest replica minus the energy of the first replica) the result is not zero. It’s very strange, I don’t know whether the algorithm is wrong or something else. Here is the part of results in log.neb file:

I don’t know. Maybe Aidan can comment.
TAD uses NEB as a kernel once it finds an “event”.
If the event is bogus, i.e. the 2 end points relax into
the same configuration, then NEB will find no barrier.
So it’s possible some setting for fix event is not tuned
for your system and it is finding bogus events.

Steve

Your NEB is screwed up, because the first replica has the highest energy, the second replica is lowest. In other words, your NEB path is U shaped instead of humped. You should relax your initial states better and use more NEB replicas. This is why TAD is hard. There is a lot of experimentation required to make it work well for a particular system. I suggest you do some experiments with the silicon vacancy example first.

Aidan