I have a compound with formula A_x B_[(4-x)/3] Vac_[2(1-x)/3]. This ensures charge balance in the material. I tried the following constraints for the lat.in (pasted below), but it doesn’t seem to generate any structures. I wonder if my way of entering constraints is wrong, making it unphysical, so no solutions? If so, any hint on how to modify this?
This should work (with mmaps) but I think you meant to write
3.0B+1.0A<=6.001
3.0B+1.0A>=5.999
Otherwise only a 12-atom supercell at least would have been necessary, which takes time to enumerate for a ternary !
Charge- balanced cluster expansions can be tricky - see https://journals.aps.org/prl/abstract/10 … .98.266101
And https://arxiv.org/abs/1206.5429
Thanks for the references on charge-balanced (neutral) doped-Ceria.
In my formula that pertains to x+(4-x)/3+2(1-x)/3 = 2 sites/f.u.: A_x B_[(4-x)/3] Vac_[2(1-x)/3], if I take the two extreme values of ‘x’ and compute the constraint 3B + 1A, I get:
x=0 i.e. no A-atom present, then 3B + 1A = 3*(4-0)/3 + 1*0 = 4
and
x=1 i.e. no Vac present, 3B + 1A = 3*(4-1)/3 + 1*1 = 4
So, why is the right-hand-side 6.0 in your constraints and not 4.0, like I get above ? Can’t seem to figure out.
Sorry, my rhs of 6.0 didn’t make sense.
The key here is that the compositions (A and B in your equation) are atomic fractions. They sum to 1. A+B+Vac=1 by construction. Your calculation seems to assume A and B are per f.u.
To impose charge balance , simply put the charge of each specie as its coefficient (A, B and Vac).
Yes, it generates structures almost immediately when the constraints aren’t present. But when I put the constraints it doesn’t generate even after days.
My constraints should be:
+1.0A +3.0B <=2.1
+1.0A +3.0B >=1.9
Do you get structures to be generated when you use these constraints?
Now that I came back to this after a while, I got it to work. The trick was to keep the search to the phase space where the charge-balanced structures were more likely to form. This was done with additional constraints on the chemical potential of the A and B species.
Now, according to your paper on doped Ceria, you didn’t fit the total energy to the CE Hamiltonian, but instead the difference wrt the electrostatic energy. This makes sense, because now you get away with a short-ranged cluster expansion. I can do this by computing the electrostatic term for each calculated structure, and subtracting it from the total DFT energy in the file ‘energy’. But how do I put back the electrostatic energy (i.e. Coulomb energy) back in when doing the Monte Carlo using the multi-component emc2 code? OR is there a consistent way of doing all of this in atat internally?
Both maps and emc2 and their multi component versions (mmaps, memc2) accept the option -ks=es to treat electrostatics separately. ( so you actually don’t need to subtract electrostatic energy manually yourself ) you need to provide an input file es.in that gives the charges of each chemical specie, for instance:
Cs=1.0
Cl=-1.0
One atom per line.
Sorry, it could be that the next structures satisfying the constraint are very large…
You can also make your own structures and place them in numbered directories.
(e.g. generate a super cell and move atoms between the subcells)
Hello,
I have a compound with formula AxB1-xC1+xD1-x which ensures it is isoelectronic to known materials. A and B are cations and A,B occupy one kind of lattice sites; C and D are anions and C,D occupy another kind of lattice sites(I have written this in my lat.in). I want to keep the constrain of concentration when I run mmaps.
I tryied different crange.in:
1.0A+1.0B<=0.34
1.0A+1.0B>=0.32
1.0C+1.0D<=0.67
1.0C+1.0D>=0.65
2.0A+1.0B-1.0C<=0.03
2.0A+1.0B-1.0C>=-0.03
or
1.0A+1.0B<=0.34
1.0A+1.0B>=0.32
1.0C+1.0D<=0.67
1.0C+1.0D>=0.65
1.0B-1.0D<=0.03
1.0B-1.0D>=-0.03
But mmaps will generate 8 structures and then stop generating new structures (even after a week). I want to do CE on this "line" because I want to constrain the ratio to find the ground state of these electronic equivalents.
Aha, I see it now. cs is the constituent strain method, and underscores are used to choose any combination of predictors. (Unfortunately I am still getting segfaults when using -ks=es, as posted in the compilation forum.)