Skip to main content

Simulations, Data Analysis and Algorithms

Computational Astrophysics and Cosmology Cover Image
Figure 9 | Computational Astrophysics and Cosmology

Figure 9

From: A fast multipole method for stellar dynamics

Figure 9

Error-cost relation. Wall-clock time versus relative (top) and scaled (bottom) acceleration error for N= 10 7 particles drawn from a Plummer sphere. The top panel reports runs using multipole-acceptance criterion (16a) with a low-order-FMM estimate for a b , while the bottom panel reports runs using multipole-acceptance criterion (16b) with a low-order-FMM estimate for f b . Each pair of open and filled symbols (of same colour and ordinate) refers to another FMM run with expansion order p as indicated and a different value for parameter ϵ in equation (16). The timings include all phases of the computation, including tree building and low-order estimation of a b or f b - for comparison, the direct-summation calculation for obtaining the ‘true’ accelerations took 25k seconds on the same hardware. The thin dotted and dashed lines are power laws with exponents −0.16 and −0.2, respectively.

Back to article page