Last Modified 17 December 1999

If you like you may want to look at the SKOUT file or standard output, but our main interest will be in the SKENG file. Actually, when all is said and done, we have two SKENG files, depending on the k-point mesh. The file SKENG goes with the r48 mesh, and the file SKENG.r24 goes with the r24. Since the r24 mesh is substantially smaller than the r48 mesh (157 points versus 1202 points), we would like to know if the r24 mesh is sufficiently well converged.

We can plot the results using the r24 mesh (where we used only a few points) and the r48 mesh (where we used lots of points):

This graph has all of the right properties. At q = 0 it reaches the minimum, as it does again at q = 1.5. These points represent the pure fcc lattice, i.e., no stacking fault. The true stacking fault position is at q = ½, where the layers on either side of the fault mimic an hcp lattice. The points q = -½ and q = 1 are where two "C" atoms are on top of one another (see the stacking descriptions). This is obviously an unstable situation which will not be found in nature.

The true anti-stacking fault energy can be seen in the region between q = 0 and q = ½:

Here we see that the anti-stacking fault energy maximum occurs at about q = 0.275 and is about 4 mRy. The true stacking fault energy at q = ½ is about 1.6 mRy, this is consistent with other calculations for stacking fault energy of gold. Note that the stacking fault energy for gold is actually very low compared to other metals. This is consistent with the observed ductility of gold.

Note that on this scale the convergence of the r24 k-point mesh does not look so good, as it gives a stacking fault energy of about 2.3 mRy, about 40% larger than the r48 result. Whether this is enough accuracy depends upon the needs of the individual user. The r24 mesh may be accurate enough for, example, ranking stacking fault energies of different elements. For absolute stacking fault energies, however, it may be necessary to use the larger mesh.

Now that we've actually run the job and looked at the output, it is time to discuss our success at parallelizing the code.

Previous: Running the job

Next: Timings and Parallelization

Look at other examples.

Get other parameters from the Tight-binding periodic table.

Return to the static Reference Manual.