[gmx-users] a question on the -sed, trjconv output & g_order

[Mark Abraham]

> h 48 lipids and 1437 waters t=  85.00001
> h 48 lipids and 1437 waters t=  90.00001
> h 48 lipids and 1437 waters t=  95.00001
> h 48 lipids and 1437 waters t= 100.00001
> h 48 lipids and 1437 waters t= 105.00001
> h 48 lipids and 1437 waters t= 110.00001
> h 48 lipids and 1437 waters t= 115.00001
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> h 48 lipids and 1437 waters t= 920.00006
> h 48 lipids and 1437 waters t= 925.00006
> h 48 lipids and 1437 waters t= 930.00006
> h 48 lipids and 1437 waters t= 935.00006
> h 48 lipids and 1437 waters t= 940.00006
> h 48 lipids and 1437 waters t= 945.00006
>
> My doubt is: what are these 1 and 6 in some of the
> time values. I am saving the coordinates after every
> 5ps, why is this not exact, e.g. t= 945.00000 ?

Floating point representations cannot represent all numbers to infinite
precision. In decimal notation, we can't represent 1/3. IIRC in some
floating point representation, we can't represent 1/10. So when you add up
a whole bunch of not-quite-accurate numbers (as you do when accumulating
the time variable) you'll get errors projecting to higher orders of
magnitude.

> I am facing some problem in the g_order analysis tool
> of gromacs version 3.3 which might be because of the
> above-mentioned problem.
>
> When I use -b 0 -e 1000 in the g_order then it gives
> me 10 frames less each time, or in other words 50ps
> missing each time.

I don't understand the sense in which you are getting 10 frames fewer each
time, but probably can't help you here.

> I do not think that mixing the two versions would be a
> good idea, i.e. generating the trr on v3.3 and
> analysing them on v3.2.

If there's a meaningful difference in the .tpr format, it will tell you
and refuse to do something that would be stupid.

Mark