Two saddle points along the CI-NEB images

[lgxyz]

Recent I got a converged CI-NEB results(EDIFFG=-0.05), which has two saddle points like this:
0 0.019203 -156.342942 0.000000
1 0.023043 -156.310846 0.032096
2 0.041036 -156.224943 0.117999
3 0.045308 -156.043714 0.299228
4 0.040988 -155.808322 0.534620
5 0.029823 -155.830571 0.512371
6 0.029259 -155.683854 0.659088
7 0.038446 -155.675139 0.667803
8 0.013002 -155.724566 0.618376
9 0.019982 -155.727274 0.615668
Note that the image 4 and 7 are two saddle points. How can I sovle this kind of question?
Thanks for any suggestions!

[graeme]

Try minimizing image 5 to see if there is an intermediate minimum along your path.

[lgxyz]

Thanks! The image 5 cannot be converged at EDIFFG=0.02 and it oscillates at the forces of 0.06 although I have tried many kinds of optimizers. Further, I have used four images to CI-NEB calculations and it has converged like this:

0 0.019203 -156.342942 0.000000
1 0.038514 -156.241991 0.100951
2 0.046404 -155.889649 0.453293
3 0.048442 -155.774934 0.568008
4 0.015747 -155.687778 0.655164
5 0.019982 -155.727274 0.615668

Along the path, there is only one sandle point (image 4). Could this be reasonable? Thanks for your suggestions!

[graeme]

I suggested minimizing image 5 to see if there is a local minima; I can't imagine any reason why you could not do that with a conservative optimizer, such as IBRION=3, POTIM=0.01 (although, it will be slow).

It looks like you may have a nicely converged saddle in your 4 image band. To check for intermediate minima, you can minimize on either side of that saddle; one will converge quickly to the nearby product state, see if the other converges to the initial minimum.