Not convergent

Greetings Alex and ATAT users,
I have tried the provided example, Al-Cu system using maps. However, it does not converge even more than 120 structures are calculated. Is it normal?

vasp.wrap
[INCAR]
PREC = high
ISMEAR = -1
SIGMA = 0.1
NSW=41
IBRION = 2
ISIF = 3
KPPRA = 100
DOSTATIC

maps.log
The internal database of structures extends at least up to 8 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 9 atoms/unit cell predicted , see predstr.out
Concentration range used for ground state checking: [0,1].
Crossvalidation score: 0.0957489

Thank you very much.
Regards,
Hui

The setting:
KPPRA = 100
is so low that your energies must be essentially random numbers…

Hi Alex,
I have increased the kppra to 8000. However, it still fails to convergent after calculating 70 clusters. Do you have any further suggestions? Thank you very much.
Regards,
Hui

Maps version 3.23
The internal database of structures extends at least up to 7 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 8 atoms/unit cell predicted , see predstr.out
Concentration range used for ground state checking: [0,1].
Crossvalidation score: 0.108548

[INCAR]
PREC = high
ISMEAR = -1
SIGMA = 0.1
NSW=41
IBRION = 2
ISIF = 3
KPPRA = 8000
DOSTATIC

Problem getting the right ground states have been discussed in many other posts. Either reduce the composition range to the minimum you need (-c0 and -c1 options) or if the issue is relaxation to another lattice, use robustrelax_vasp (see https://dx.doi.org/10.1038/ncomms8559 ).

Hi gudutu,

I’m working on a cluster expansion for the Al-Cu system using ATAT in the Al-rich regime. After spending a lot of time on learning how to work with ATAT and doing a lot of DFT and CE-fitting using MAPS, I finally ended up with a well working CE (using 46 input structures), obtaining the correct ground states for Al-rich alloys (GP-zones) and getting good values for transition temperatures during simulated annealing MC-simulations.

On my way I have noticed some important points you maybe should take into account when dealing with Al-Cu:

  • Construction of a reciprocal CE (using the csfit module) improves the convergence of the CE a lot, due to the large mismatch of lattice constants of Al and Cu.
  • Using the checkrelax command and flagging the severely deformed structures with an error is important, because a lot of structures won’t fit the fcc lattice after relaxation. This is due to the large number of non-fcc based intermetallic phases showing up in the Al-Cu phase diagram.
  • For me, using the -c1 options of MAPS was helpful, because I’m only interested in the Al-rich part of the phase diagram. By this I didn’t waste too much time with the calculation of unstable phases.

I think these points will be helpful for getting your work done. Feel free to contact me!

Kind regards,
Tobias

Greetings Alex/Tobi,

I have tried the TiAl system using the following parameters as INCA.
[INCAR]
Encut =500
ISTART = 0
ICHARG = 2
IBRION =2
EDIFF =1E-6
ISMEAR = -1
SIGMA = 0.1
NSW=41
IBRION = 2
ISIF = 3
KPPRA = 8000
DOSTATIC

The parameters are good enough for high precision. However, the result of the fit is very bad. The energy difference of many structures can be more 50%. I attach the fit.out here. Should I exclude all those structures? Would you please help me have a check where the problem is? Thank you very much.
Regards,
Hui

0.000000 0.000000 0.016221 -0.016221 1.000000 0
1.000000 0.000000 0.016959 -0.016959 1.000000 1
0.500000 -0.265234 -0.272962 0.007728 1.000000 2
0.250000 -0.266506 -0.262931 -0.003574 1.000000 3
0.500000 -0.421451 -0.311782 -0.109669 1.000000 4
0.500000 -0.380350 -0.365838 -0.014512 1.000000 5
0.750000 -0.200510 -0.204065 0.003554 1.000000 6
0.250000 -0.319790 -0.271351 -0.048439 1.000000 7
0.500000 -0.360424 -0.355650 -0.004774 1.000000 8
0.750000 -0.201596 -0.212485 0.010889 1.000000 9
0.250000 -0.135906 -0.136791 0.000885 1.000000 10
0.500000 -0.166686 -0.145026 -0.021659 1.000000 11
0.750000 -0.135197 -0.136422 0.001225 1.000000 12
0.333333 -0.309764 -0.301369 -0.008394 1.000000 40
0.666667 -0.294706 -0.290240 -0.004466 1.000000 43
0.333333 -0.162168 -0.227235 0.065067 1.000000 14
0.333333 -0.351740 -0.361803 0.010063 1.000000 16
0.333333 -0.279570 -0.258186 -0.021384 1.000000 18
0.666667 -0.282012 -0.226989 -0.055023 1.000000 20
0.500000 -0.350863 -0.343631 -0.007232 1.000000 21
0.666667 -0.254530 -0.257940 0.003410 1.000000 23
0.666667 -0.308010 -0.283560 -0.024450 1.000000 24
0.333333 -0.226740 -0.253937 0.027197 1.000000 27
0.333333 -0.281238 -0.321221 0.039983 1.000000 29
0.500000 -0.313828 -0.307073 -0.006755 1.000000 30
0.333333 -0.254029 -0.305746 0.051716 1.000000 31
0.666667 -0.236775 -0.253691 0.016917 1.000000 33
0.500000 -0.329680 -0.339383 0.009702 1.000000 34
0.666667 -0.242233 -0.266501 0.024268 1.000000 36
0.666667 -0.304169 -0.281977 -0.022192 1.000000 37
0.166667 -0.219379 -0.206410 -0.012970 1.000000 39
0.333333 -0.329295 -0.280370 -0.048925 1.000000 41
0.500000 -0.322342 -0.303650 -0.018693 1.000000 42
0.666667 -0.212175 -0.207569 -0.004606 1.000000 44
0.833333 -0.169467 -0.161478 -0.007989 1.000000 45
0.166667 -0.086064 -0.200796 0.114732 1.000000 46
0.333333 -0.248880 -0.290142 0.041262 1.000000 47
0.333333 -0.249139 -0.274757 0.025618 1.000000 48
0.500000 -0.283649 -0.292423 0.008774 1.000000 49
0.666667 -0.284577 -0.279014 -0.005563 1.000000 50
0.833333 -0.160850 -0.155864 -0.004986 1.000000 52
0.500000 -0.308138 -0.334653 0.026515 1.000000 145
0.750000 -0.191508 -0.210629 0.019120 1.000000 96
0.125000 -0.192528 -0.163329 -0.029199 1.000000 135
0.500000 -0.214523 -0.247340 0.032817 1.000000 258
0.250000 -0.344148 -0.358482 0.014334 1.000000 251
0.750000 -0.288822 -0.291454 0.002632 1.000000 262
0.875000 -0.112312 -0.135735 0.023423 1.000000 173
0.125000 -0.194543 -0.169618 -0.024925 1.000000 156
0.375000 -0.370260 -0.354359 -0.015901 1.000000 144
0.625000 -0.326594 -0.320844 -0.005749 1.000000 153
0.250000 -0.344759 -0.311396 -0.033364 1.000000 158
0.375000 -0.379355 -0.360647 -0.018708 1.000000 161
0.500000 -0.358508 -0.355650 -0.002858 1.000000 163
0.750000 -0.189630 -0.247357 0.057727 1.000000 169
0.625000 -0.326262 -0.327133 0.000871 1.000000 171
0.250000 -0.338899 -0.336076 -0.002823 1.000000 201
0.375000 -0.361351 -0.350149 -0.011202 1.000000 209
0.625000 -0.313814 -0.316634 0.002821 1.000000 221
0.375000 -0.348310 -0.356437 0.008127 1.000000 252
0.625000 -0.315267 -0.322923 0.007656 1.000000 257
0.500000 -0.375029 -0.360744 -0.014285 1.000000 322
0.083333 -0.118256 -0.107672 -0.010584 1.000000 984

Thank you for your detailed explanations. Indeed, the checkrelax is very useful. But I am still struggling to get a convergent result. Thank you again.
Regards,
Hui

Ti-Al is a typical system where the most stable lattice changes as a function of composition (hcp for Ti-rich, fcc for Al-rich and even bcc around 50%). So, some structures will certainly relax away from the intended lattice.
Possible solutions (use one of them consistently, not a mix!):

  1. limit the composition range with the -c0=… -c1=… options of maps
  2. Use checkrelax to see which structures have relaxed too much and drop them. (touch nn/error)
  3. For structure that relax too much, you could also try to relax ions and volume but not cell shape. See https://dx.doi.org/10.1016/j.calphad.2016.02.005 for how to patch vasp to do this.
  4. Use the inflection detection method (implemented as robustrelax_vasp -id ) see
    https://doi.org/10.1016/j.calphad.2017.05.005 and https://dx.doi.org/10.1103/PhysRevB.95.144113 and https://dx.doi.org/10.1038/ncomms8559.

(Option 1, 3 are the most theoretically justified, especially 3. But 1,2,3 are the quickest.)