Maps ground state checking

I was doing some test runs on a A-B-N system, and met the following problems.

I calculated 89 structures, among which 19 structures have either exceeded the relaxation criterion by checkrelax, or didn’t reach convergence, so I ruled them out, and reran maps -d, and the files were overridden.

A few things I noticed that puzzle me.

  1. The maps.log reads
Maps version 3.16
The internal database of structures extends at least up to 0 atoms/unit cell, see predstr.out
Among structures of known energy, true ground states differ from fitted ground states
New ground states with at most 2 atoms/unit cell predicted , see predstr.out
Concentration range used for ground state checking: [0,1].
Crossvalidation score: 0.0627808

If maps is running in the background all the time while pollmach is generating structures, "The internal database" should reach about 12 atoms/cell, and the new ground states would reach 14 atoms/cell, but apparently the rerun of maps overrides the prestr.out and so on, and these two numbers are somehow reset. I don’t know the implication of it.

  1. The first fig of mapsrep has a few known str underneath the groundstate line. Is it normal?
    [img:3rqe7lyl]https://astro1.panet.utoledo.edu/~terencezl/public/0-1.png[/img:3rqe7lyl]

  2. There seem to be a lot of structures close to the ground state line. Is it a difficult system?

  1. you force maps to restart where it left off using the -g=[nbatom] option. This will first repopulate the internal database with structures with up to nbatom per unit cell.
  2. it is normal, while the calculations are running and not yet converged, to have structures below the convex hull or even have a non-convex "convex hull". These problem should disappear as more structures are calculated.
  3. yes your system a lot of competing gs and that makes convergence more difficult to achieve. One useful shortcut is to focus on a specific composition range (-c0= -c1= options) where the lattice your are considering is known to be stable. The fact that so many structure relax too much is a sign that your considering composition regions where the assumed lattice is mechanically unstable.
    An alternative solution is described in https://dx.doi.org:10.1038/ncomms8559
    Also implemented as the robustrelax_vasp command in ATAT.
  1. What does it mean that "force maps to restart". I was under the impression that, even without the ‘-g’ option, maps will devour whatever directory that has a str.out and an energy file, and include that data point to obtain the ECI as well as future structure generations. The internal database should automatically updates to all the directories it can find.

What if I left off at 12 atoms/cell, but specify that -g=8, or -g=16. I see the results are loyal to the provided -g, but I don’t understand what they mean.

  1. The manual says "[c0,c1] is the concentration range where ground states must be correct." I think I read it somewhere (but can’t find again) that even this range is specified, maps will still go out of range to generate and calculate (with ab initio code) new structures and take those points into consideration.

So is the effect of this specification only ensuring the (ab initio) calculated ground states within the specified range must meet with the predicted values, while if not specified, there might be some compromises for the full [0, 1] range? Is this carried out by giving less weight to structures outside the range? And, what does "correct" mean?

  1. maps does always reload the structures written to disk. But there are many more structures in memory, used for ground state checking for instance. This is why -g I’d useful.

  2. correct gs meAns that the lowest energy structure according to CE and according to an initio calc agree.
    Maps will generate structure only within range specified except if it predicts a 2-phase equilibrium between a structure within the range and outside the range .

I see. So providing [c0, c1] will guide maps to generate the structures to be calculated within the range. What if there are already directories with energy of structures outside of the range. Will maps automatically rule them out?

Answer: No. But you can always rule them out yourself, by checking out the fit.out file and compare concentration.

Indeed!

I should point out that the -c0 and -c1 will not completely prevent generation of structures outside of that range because sometimes the ground state in the range of interest is actually a mixture of two phases that are outside the range of interest and maps checks for that.