Power bundle adjustment (as linear solver) -- experience / "user study"

[Christoph Vogel]

Hello, I saw the power bundle adjustment being implemented in Ceres.
I am amazed how quickly new ideas are integrated and exposed to the test of time. You are doing an incredible job with ceres.
I am interested in the "power bundle-adjustment" as linear solver to solve the Schur complement, mostly for the application of bundle-adjustment or other relevant vision tasks.

I want to ask the community, if anyone did experience data (ideally also publicly available) where the method does not perform well? On the evaluated BAL data it works quite well in my experience.
This should be cases where the spectral radius is still (relatively) larger, even after applying the 'pre-conditioning'.

I would be grateful for anyone that can share the experience with me, or even show me data where it does not perform so well.

Thanks, Christoph