Clarification

[sitela]

Hello,

this is not a bug report, but rather a clarification request. I would like to use the code for an MD simulation. One thing I can't understand, though, is how I can obtain the value for the energy after atomic coordinates are changed by the algorithm that implements the dynamics. I've read all the docs but only single-point calculations are described, where atomic positions are input just once upon creation of the calculator object.

In the ASE implementation the energy was a property of the molecule object, so whenever positions were changed, the energy could be recalculated accordingly. Here, however, the calculator does not seem to depend on the molecule object. Thanks!

[awvwgk]

The details are somewhat hidden in the implementation, the xtb C-API (in xtb.libxtb) provides four objects: environment, molecule, calculator and results.

While this is a good design for C and Fortran, it does not really match the Python philosophy, therefore the Python API is implementing the wrapper classes containing all relevant API objects. A Calculator from xtb.interface contains beside the calculator also a molecule and an environment, this allows to bundle everything in one Python object and hide the details of the C-API. This means the Python Calculator can also use an update to change the cartesian coordinates.

In ASE the concept is inverted, an Atoms object has its (ASE) Calculator. Therefore, the Python Calculator for xtb is wrapped in an ASE Calculator and the necessary checks for cache invalidation and updating of coordinates are hidden inside from the user, as they should happen automatically as the Atoms is changed and new properties are requested.

So it largely depends on which layer you are looking at, C-API, Python-API or ASE-API, they focus on different concepts, abstract different tasks and target different audiences. Depending on how you want to implement the molecular dynamics a certain abstraction layer might be more suitable for your problem than the other, but the details depend on your problem.

[awvwgk]

Without the code snippet I can't tell for sure, but it seems like you requested a property from the results object which is not present. Make sure to actually perform a single point calculation before querying the results object.

[sitela]

I used the code snippet you pointed to on this page: https://github.com/grimme-lab/xtb/blob/77cf01e/include/xtb.h#L126-L131.

[awvwgk]

Please have a detailed look at the example.

The error printout you found is generated on purpose for the example, by requesting a property from the results object before performing the calculation:

https://github.com/grimme-lab/xtb/blob/af6e157/TESTSUITE/c_api_example.c#L47-L51

Of course you have to create a suitable example for your application first.

[sitela]

Thanks, it now works. I have one more problem, tough: the coordinates in molecule are declared as constant double of fixed length. How can my program modify them and re-pass them to xtb_updateMolecule?

[awvwgk]

For your application you will probably have a struct containing the coordinate array, if you have a 3 by n contiguous array of the coordinates in Bohr you can just pass it by reference to the xtb_updateMolecule procedure. In any other case you have to gather it in a 3 by n contiguous array first, convert it to Bohr and than call the API.