N^(0.5)??? if that's the computational time complexity, then that's fantastic!! a brute-force all pairs n-body sim's comp time is roughly N^2, which is of course MUCH MUCH worse.TallDave wrote:Rick on simulation challenges:
rnebel wrote:3-D Particle-in-cell is extremely expensive. Resolution goes like (N)**.5 where N is the number of particles. You have multiple timescales and multiple spatial scales to resolve. This means supercomputers.
and then there are methods of approximation like having 1 simulation particle represent thousands of real particles, or just using a simulated tiny wiffleball, and/or really high mag fields, etc.
as far as "super-computers" go, well, there's a law in computer science, and it goes like this: _algorithm triumps hardware_. e.g. if an N^2 algorithm can do 100 particles at a rate of 1 picosecond/second with $x of hardware, to sim 200 particles at that rate would cost $x*(200^2-100^2)/(100^2). which, needless to say, would be _A LOT_ of money, quite regardless of what "x" is.
now if instead you can find, say, an N*log(N) algorithm, well, what an N^2 algorithm can do on a machine that costs $N^2, an N*log(N) algorithm can do on a machine that only costs $N*log(N), and when N is very large (which it usually is), the payoff ratios can become downright astronomical.
so when i hear "that would take a super-computer!" my immediate reaction is: "...or a normal desktop computer and a bit of cleverness."