[gmx-users] B-Splines

[Erik Lindahl]

On Friday, Mar 28, 2003, at 10:33 US/Pacific, Lianqing Zheng wrote:

> Thanks, Erik! I'm reading that paper. However, it seems that the codes
> didn't directly use the recursive definition of Cardinal B-Splines to
> evaluate them. Actually, I tried this way in my own SPME code and  
> found it
> was very time-consuming due to the recursions. Instead, the Gromacs and
> Darden codes are very efficient to get M_n(u-j). What mathematics is
> behind this efficient evaluation of B-Splines?

Hi,

We do perform the recursion in the routine make_bsplines(); there is a  
loop over the order, and for each order a loop over lower-order  
bsplines to calculate the current one.

Both we and Darden store as much constants as possible outside this  
calculation of course, and if you define it as recursive function calls  
the overhead would kill your performance.

Cheers,

Erik

------------------------------------------------------------------------ 
-----
Erik Lindahl, MSc, PhD     <lindahl at stanford.edu>
D109, Fairchild Building
Dept. Structural Biology, Stanford University School of Medicine
Tel. 650-7250754    Fax. 650-7238464