Maybe it would have been better to split this into three SO questions, but anyway here's my try:
- This could refer to at least two things, without more context I can't decide which:
- To flatten out the tree structure of a BVH in memory into a simple array which can result in faster memory access / cache utilization.
- It might also refer to the "Fast BVH Construction on GPUs" paper, which introduces the LBVH (linear BVH) structure.
Using buckets is a technique for speeding up SAH evaluation. Basically it helps to cut down the number of candidate splits. I can recommend Pharr & Humphreys' "Physically Based Rendering" book which explains and uses this.
Without more context, I don't know (and with more context as well, probably).