chivalry123 / maxsat_ising_solver_ginar Goto Github PK

#MAXSAT Ising solver Ginar

This github repo represents the folder /home/key01027/ising_solver/ising_119_some_debug/ in Ginar (which is already shared to all of you).

All the executables are from Wenxuan's directory in Ginar:

/home/key01027/ising_solver/ising_119_some_debug/bin/

You will need to copy the binaries to the destination location with the corresponding input file to make it work.

In the following, I have preset the J_config.in file for different applications. Except the NSITES which simply means how large the supercell size do you want to go into searching for optimal solution. You don't need to set anything else.

The single point solver: Please look into: examples/single_point_solver. Given an ECIs in its 0,1 formulation , solve for the optimal solution up to super cell size of NSITES

To do this, you would need 3 files: PRIM, J_config.in, J_in.in

PRIM indicates the lattice system in the CASM format. J_config.in includes options for the solver including NSITES, please also Check README for more inforamtion. If you are too busy for that, simply use the current settings. J_in.in dictates the ECIs, in its 0/1 formulation.

After requesting a compute node (with qrsh, or qsub), you could use to run the solver

mpiexec.hydra -np 16 ./ising 

The results would be

[key01027@compute-1-83 single_point_solver]$ mpiexec.hydra -np 8 ./ising 
[MASTER] Sent all jobs.
[MASTER] Handled all jobs, killed every process.

 I am still alive at point: sjoqno2hcb02 

printing POSCAROUT

 tell me what is upper bound: -5.957034540000005e-01 what is lower bound: -1.000000000000000e+18

You could check the POSCAROUT for the predicted Ground state up to that supercell size. (For lower bound to run successfully, which we generally do not use in actual practice, we need to set NLOOPS>0 and have the gurobi server correctly run, which needs to be renewed yearly.)

Please look into: examples/binary_system. Perfom binary GS phase diagram search based on CASM output.

CASM files that are needed: eci.out FCLUST PRIM . copy the useful_scripts/casm2wenxuan.py to the running directory (examples/binary_system). Run python casm2wenxuan.py > J_in.in to create J_in.in

Additional GS file needed include J_config.in, mu_in.in, mu_config.in. When you have a different system, you need to change mu_in.in manually. For example, when your PRIM is,

Mg2 Cr4 O8
1.0
4.163840 4.163840 0.000000
4.163840 0.000000 -4.163840
0.000000 4.163840 -4.163840
2 4 8
direct
0.375000 0.875000 0.875000 Mg Va
0.625000 0.125000 0.125000 Mg Va
0.500000 0.500000 0.500000 Cr
0.000000 0.500000 0.500000 Cr
0.000000 0.500000 0.000000 Cr
0.000000 0.000000 0.500000 Cr
0.776100 0.741300 0.741300 O
0.223900 0.258700 0.258700 O
0.241300 0.741300 0.741300 O
0.758700 0.258700 0.258700 O
0.241300 0.741300 0.276100 O
0.758700 0.258700 0.723900 O
0.241300 0.276100 0.741300 O
0.758700 0.723900 0.258700 O

you need to specified the point term about what x is in mu_in.in. Specifically, it would be

Constant 0

Cluster 1
1,1,1,1,1   
J=0

Cluster 2
1,1,1,2,1   
J=0

The two point term 1,1,1,1,1 and 1,1,1,2,1 belongs to the same concentration group. (note that the first 1,1,1 corresponds to the x,y,z, the forth term 1 and 2 corresponds to the sub-lattice sites of the two Mg, Va species. The fifth term, 1 corresponds to the first element, in this case it is the Mg).

mu_config.in only includes only one line MODE_JPLUSMINUS=1

You can then run GS solver with

mpiexec.hydra -np 16 ./ising

you will have the folder called GS_solutions containing the binary ground state phase diagram. The difference between POSCAR_OUT and POSCAR in GS_solutions is that POSCAR_OUT contains the explicit vacancy terms Va or Vac, whereas POSCAR does not. The direcotry tree of GS_solutions looks like:

$tree GS_solutions/
GS_solutions/
|-- hull_debug.txt
|-- hull.txt
|-- x0
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.250000
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.500000
|   |-- POSCAR
|   `-- POSCAR_OUT
`-- x1
    |-- POSCAR
    `-- POSCAR_OUT

more comments below if you want to edit J_config.in or mu_config.in
The difference compared to pointwise solver is MODE_JPLUSMINUS=1 meaning that the J input from CASM is +1/-1 formulation. work_with_mu=1 and scan_chemical_potential=1 to ensure scanning (we are pin-pointing to be specific) of chemical potential for ground states.

The flexible flags include binary_x_min=0,binary_x_max=1,NSITES=2. binary_x_min and binary_x_max is used to indicate the minimum and maximum interested concentration. NSITES denotes the super cell size to search up to.

In terms of what is x, for example, in the PRIM file, 0.375000 0.875000 0.875000 Mg Va. Since Mg corresponds to +1 and Va corresponds to -1, concentration fo Mg is x.

Please look into: examples/ternary_system.

CASM files that are needed: eci.out FBCLUST PRIM . Noted that we would need FBCLUST this time instead of FCLUST for practical reasons. copy the useful_scripts/casm2wenxuan_ternary.py to the running directory (examples/binary_system). Run python casm2wenxuan_ternary.py > J_in_tern_casm.in to create J_in_tern_casm.in

If you are interested in the internal detail of J_in.in, you would see that there is one more index than the previous five. This is due to some very practical issues CASM use 1/0/-1 CE system and the spin variables needs to have square term. So if the sixth index is 1, it correspodns to square term s_i^2, if the sixth index is 0 it corresponds to linear term s_i.

The PRIM looks like (note that we should place all the substitutions terms at front, otherwise it causes a lot more computations due to the current implementation...):

$ cat PRIM 
M2 Ni1 O1
4.104210848570817
0.0 0.5 0.5
0.5 0.0 0.5
0.5 0.5 0.0
1  2  1 
direct
0.000000 0.000000 0.000000 Li Ni Vac 
0.250000 0.250000 0.250000 Li Ni Vac
0.750000 0.750000 0.750000 Li Ni Vac
0.500000 0.500000 0.500000 O

Additional GS file needed include J_config.in, mu_in_tern_casm.in, mu_config.in. The mu_in_tern_casm.in needs to be constructed manually, in our case, this system is

$ cat mu_in_tern_casm.in 	
Constant 0

Cluster 1
Group 1
1,1,1,1,1   
J=0

Cluster 2
Group 1
1,1,1,2,1   
J=0

Cluster 3
Group 1
1,1,1,3,1   
J=0

Cluster 4
Group 2
1,1,1,1,2   
J=0

Cluster 5
Group 2
1,1,1,2,2   
J=0

Cluster 6
Group 2
1,1,1,3,2   
J=0

Group 1 and Group 2 is used to denote the two types of species correspondingly. To enable ternary algorithm, you need to set in J_config.in

work_with_mu=0
scan_chemical_potential=0
ternary_alg=1

The current solver is a bit stupid that if you set work_with_mu=1 and scan_chemical_potential=1. It just activates the binary search algorithm even though you set ternary_alg=1. So, to run everything successfully you need to set work_with_mu=0, scan_chemical_potential=0.

Since in CASM for the specific PRIM Li is +1, Vac is -1, the GS code associate +1 with x and -1 with y autoamtically. If you know the physically interested range of concentration, you could set it here. This could save tons of computation time needed. Since the chemical potential pinpointing is very intense in ternary system:

ternary_x_min=0
ternary_x_max=0.17
ternary_y_min=0
ternary_y_max=1
ternary_z_min=0.166666
ternary_z_max=0.3334

This is also why I do not construct it to solve quaternary system since we could envision that it would be extremely unefficient in terms of chemical potential searching.

After running with:

mpiexec.hydra -np 16 ./ising

You could see the resulted directory tree:

GS_solutions/
|-- hull.txt
|-- x0.000000y0.000000
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.000000y0.166667
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.000000y0.333333
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.000000y0.666667
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.000000y0.833333
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.000000y1.000000
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.166667y0.666667
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.166667y0.833333
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.333333y0.666667
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.666667y0.000000
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.666667y0.166667
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.666667y0.333333
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.833333y0.000000
|   |-- POSCAR
|   `-- POSCAR_OUT
|-- x0.833333y0.166667
|   |-- POSCAR
|   `-- POSCAR_OUT
`-- x1.000000y0.000000
    |-- POSCAR
    `-- POSCAR_OUT

If for some reason, you really need to compile everything somewhere else yourself (not recommended). Here are some important notes

simply use the build.sh to build

remember to put (or remove everything up to it)

/usr/local/bin

on the top of PATH variable, or alternatively remove anaconda and generally locally installed package (from the PATH)

How to install GS solver on new mac: Firstly check brew list, brew uninstall all gcc stuffs, mpich, boost-mpi, ensure the following is set in the environment

alias g++="g++-4.8"
alias gcc="gcc-4.8"

export PATH="/usr/local/sbin:$PATH"
export HOMEBREW_CC=gcc-4.8
export HOMEBREW_CXX=g++-4.8

the key is to do

brew install [email protected]  
brew reinstall mpich --build-from-source  
brew reinstall boost-mpi --build-from-source  

then using build_ising_Mac.sh from /home/key01027/ising_solver/ising_119_some_debug/src/ising/ should work

if there is some hostname error, look at https://stackoverflow.com/questions/23112515/mpich2-gethostbyname-failed

To change your /etc/hosts hostname type the following: sudo nano /etc/hosts

and then add the line 127.0.0.1 my_new_hostname

important! If you use anaconda, you need to temporarily disable Anaconda to use compile GS solver.

Also ensure your PATH looks something like like this in Ginar:

/share/apps/intel/composer_xe_2015.3.187/bin/intel64:/share/apps/intel/composer_xe_2015.3.187/debugger/gdb/intel64_mic/bin:/home/key01027/.rubies/ruby-2.3.1/bin/:/home/key01027/.local/bin/:/home/key01027/ising_solver/boost_build/bin/:/share/apps/bin:/share/apps/intel/bin:/share/apps/intel/advisor_xe_2015.1.10.380555/bin64:/share/apps/intel//impi/5.0.3.048/intel64/bin:/share/apps/intel//itac/9.0.3.051/intel64/bin:/share/apps/intel/composer_xe_2015.3.187/bin/intel64:/share/apps/intel/composer_xe_2015.3.187/debugger/gdb/intel64_mic/bin:/opt/openmpi/bin:/usr/lib64/qt-3.3/bin:/usr/local/bin:/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/sbin:/usr/java/latest/bin:/opt/rocks/bin:/opt/rocks/sbin:/opt/gridengine/bin/linux-x64

Modify the Makefile accordingly to compile (it is not easy and very system dependent). If you use intel compiler this two links should be useful link1, link2