Which algorithm does cartographer use to find pose in pure localization?

[yuzy2019]

In demo pure localization, I see cartographer search the whole map to find the optimal initial pose. I am wondering which algorithms are used in this part.

Is branch-and-bound used just like loop-closure? I think this algorithms is used to accelerate correlative scan matching by minimizing the search window. But in pure localization, how can cartographer find the estimate to initialize the search window?

[shamengjun]

in pure localization, firstly start to build one new trajectory, use this new trajectory data to scan match with originally loaded .pbstream that contain trajectory_id 0 's slam state, and the new trajectory keeps 3 submaps at least, I think it use the 3 submaps scan match all submaps in trajectory_id 0, so the process is a liitle slow in my case at least, I also think some methods to fast this process.