Local correlation: clusters in molecules (CIM)
By partitioning a large molecule into strongly-interacting domains, fragment-based
local correlation approximations like the cluster in molecule (CIM) method overcome
the steep polynomial scaling of CCSD(T) and other many-body methods. As system
size increases, fragment sizes remain constant; the total cost of a computation
increases only linearly with the number of fragments:
For many systems, we observe huge savings in time-to-solution with very little
loss in accuracy. CIM methods often recover nearly 100% of the canonical correlation
energy: