Frequently Asked Questions

Holy Smoke! Everyone seems to use different symbols (Uij, Bij, ij, ...) throughout the literature. Which ones should I use?
In the misty past, everything started with ij, then came Bij, and eventually Uij popped up. It is a matter of history and conventions, and not the least it depends on which computer programs were and are employed in (chemical) crystallography. One of these pieces of software, dubbed SHELX, has risen to permanent importance, and this has helped to popularize Uij which represents the de facto standard right now.

To get an overview of the different conventions and definitions, the following publication is highly recommended:
Link R. W. Grosse-Kunstleve, P. D. Adams, J. Appl. Cryst. 2002, 35, 477.

As said before, Uij are most often used these days. This is also the convention most frequently encountered in the cif-format.

You are describing a theoretical method here, mainly. What's the expected accuracy?
How accurate is "accurate"? In our first case study, we validated the theory against single-crystal neutron data (the experimental "gold standard") which was extremly encouraging, â?? but nonetheless also revealed problems in some experimental data (Link CrystEngComm 2014, 16, 10907).
Later, we followed up with a study on a specifically selected halogen-bonded molecular crystal (pentachloropyridine) where at 100 K, we saw several dispersion corrections push the agreement with experiment (single-crystal X-ray data) down to below 0.003 ??2. Have a look at our recent work Link CrystEngComm 2015, 17, 7414 for details â?? itâ??s free to access and share.

Where are the limits of the method?
There is no clear distinction line to be drawn (when is this so?) but certain cases may, and will, become problematic. First, the entire prediction is built on DFT results â?? so DFT itself must be up to the job; this is usually not a big deal for purely organic compounds but much more so for, say, compounds of f-block elements. Second, it is important to keep in mind that what we describe here is based on the harmonic approximation â?? at 100 K or below, this works well (and 100 K is a very routine setting nowadays!), however above 200 K one is expected to run into trouble; we have recently explored this in detail (Link CrystEngComm 2015, 17, 7414). Finally, disorder is hard to tackle â?? one needs discrete input positions for the DFT runs.

How long will such a calculation take?
It depends on the size, and especially on the symmetry of your system. For urea, such a computation took us around 3100 CPU hours (this sounds worse than it is, thanks to parallel computing). You'll need to try.

I do not have a VASP licence. Can I still calculate ADPs?
Of course. The phonon code we use (and wholeheartedly recommend), dubbed phonopy, is interfaced to several DFT codes, some of which are freely available; choose what's best for you! Just go to for more information.

I'm an experimentalist but don't have the time to delve into this myself. Can you compute ADPs for me?
We are sorry but we cannot provide such a service. We might want to start a collaboration, though.
last modified: 2016-06-03