Adding structures by manipulating str.out files by hand

Alloy Theoretic Automated Toolkit (ATAT)
1

Hello,

let’s suppose we know of some "desirable superstructures" (in the sense that those should definitely be part of the training set of our CE), so naturally we think about feeding them into the training set of structures manually at some point rather than just waiting for MAPS to "accidentally" propose those, too.
Given several superstructures proposed by MAPS (i.e. their respective str.out files), we’d be interested to know:

Is it "failsafe" to manipulate the atom labels of the sites within those structure files such that they represent our aforementioned "desirable superstructures"?
Here, "failsafe" would ask for:

  • Is MAPS guaranteed to correctly interpret these structures? (Having in mind the "no effort to rotate or scale" warning from the maps -h output)
  • Coherently, will MAPS preclude the same structure when it might "accidentally" propose it later on in the run, maybe within a slightly different str.out file? (Having noticed, to our surprise, some str.out files happen to have, e.g., a negative c-axis - so we are a bit worried whether the used structure enumeration algorithm is robust in that sense)

Thank you very much in advance and best regards!

2

You should be able to add a user-created directory containing a str.out file of your choosing.
It does not need to be called by a number - a mix of letters and numbers is valid.
(You should NOT modify an existing str.out generated by maps, however. Nor should you change directory names of generated structures. That would confuse maps!)
maps should be able to detect that your proposed structure is identical to one it has or would generate and avoids duplicates (provided that the axes of the proposed structure are integer multiples of the unit cell in lat.in), regardless of signs, etc. If you find otherwise - do let me know: it would be a bug.

I slightly object to the term "accidentally propose" :wink: there is some logic to which structure maps proposes… it does try to predict ground states and/or minimize variance of the fit.

3

I do thank you very much for your answer!

I am fully aware of that and just could not come up with a better term on the fly - clearly English is not my first language.
In no way did I intend to claim MAPS just randomly proposes structures. :wink:

Best regards,
christopher

4

No problem :wink: - take care.