Unusual energy behavior for one image during NEB calculation (B-NEB)

[Branislav Milovanović]

Dear CP2K developers/users,

I am running a NEB calculation in CP2K and I observe an unusual behavior for one replica (structure/image 3). In the attached energy profiles, all images evolve smoothly except image 3, which shows a noticeable deviation in the first couple iterations

The system is in triplet state and the calculation uses a hybrid meta-GGA functional (M06-2X) with GAPW.

I have also attached the relevant DFT and BAND input sections.

What could cause one specific NEB image to behave differently during the optimization, while all neighboring images remain smooth? 

Could this be related to small spring constant (K_SPRING 0.02) or use of DIIS optimizer in NEB? Or I am just using NEB in a wrong way?