<div>Hi,</div>
<div> </div>
<div>I've come across a discussion about the "single X double precision" issue in a NAMD users list e-mail (<a href="http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l/9166.html">http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l/9166.html</a>). </div>
<div>I would like to know what do you think about these 3 specific points:</div>
<div> </div>
<div>1) "there are some places where single precision (FFT, force/energy <br>computation), where single precision can be applied, but in other <br>places (summations) where even double precision may result in <br>artefacts, due to limited numerical accuracy for large enough <br>
systems. there are some applications that show little sensitivity <br>to single/double precision issues (homogeneous bulk lennard-jones <br>systems), but others are notoriously difficult (systems with <br>significant potential drops in one or two dimensions, e.g. metal <br>
surface slabs or lipid bilayers) since you don't have that much <br>error cancellation anymore." <br>
<p>2) "if you store coordinates in single precision, you have to take <br>extra precautions for handling very large systems, since you <br>would store coordinates with different relative precision, <br>depending on how close you are to the origin. e.g., with domain <br>
decomposition, you can define per domain offsets, but that <br>would work only well in case of a large enough number of domains. "<br>
<p>3) "the best way to study the impact of single vs. double would be <br>by running tests with the gromacs code. gromacs _can_ be compiled <br>in single or double precision. for a proper comparison, you'll have <br>
to turn off the assembly innerloops though, since they may use single <br>precision even in double precision mode." </p>
<p>Specifically this last point startled me. In fact, is this true?</p>
<p>Cheers,</p>
<p>Pedro.<br></p></p></div>