Very Genereal FEM/FEA Solution Libraries

[Josh SternModerator]

Also possible is to represent neighborhoods of discretized time series for the entire mesh+parameters & move the sliding window. That could be computationally expensive, but there are tricks to optimize out the cost by moving indices instead of most data at each step.

[Josh SternModerator]

Some other hints about generating optimized exploratory sequences:

Conjugate gradients algorith. Can be generalized in several ways

Space filling curves on grids can be constructed at different scales of detail – give a way to map large integers into grid locations, where near values tend to be near & far values tend to be far – can be used to hash the set of neighborhoods already visited or waiting for a go at next level expansion. Being in a grid site at level I can be used as a basis for rejection sampling filter in the low discrepancy seque

One can look ahead at the next N values of any given discrepancy sequence & sort it in some way – e.g. distance from current point or from highest level of space filling curve or grid where a hit is matched. Again, this can be done with rejection sampling, and it can be used as another virtual dimension for generalized conjugate gradients (go to a less explored area or keep optimizing here?)

[Josh SternModerator]

For algorithmic efficiency, it is often necessary to treat border regions in a special way. Sometimes that involves adding extra border neighborhoods with special values or treatment.

[Author]

Posts