第一原理分子動力学プログラム STATE Senri Wiki

• バックアップ一覧
• 差分 を表示
• ソース を表示
• バックアップ を表示
• AENET_tutorial_e/example へ行く。
  • 1 (2021-01-21 (木) 00:50:09)
  • 2 (2021-01-21 (木) 00:54:29)
  • 3 (2021-01-21 (木) 13:08:42)

• 追加された行はこの色です。
• 削除された行はこの色です。

running
* Running the TiO2 example [#p04fc2aa]
Having compiled AENET executables (*.x), let us run them by using the example provided at the AENET web page (see the [[download>http://ann.atomistic.net/download/]] section).

** Prep [#s68ea619]
In the AENET directory (~/AENET), create the example directory as
 $ cd ~/AENET
 $ mkdir examples
and change the directory
 $ cd examples
Download the example file
 $ wget http://ann.atomistic.net/files/aenet-example-01-TiO2.tar.bz2
and expand it
 $ tar jxvf aenet-example-01-TiO2.tar.bz2
to get 
 aenet-example-01-TiO2/
In this directory, we can find the following directories:

 01-generate/  02-train/  03-predict/  04-MD-ASE/  LICENSE  README  xsf/

I prepared sample job scripts for AENET to be used in our smith cluster system, which are located in
 ~ikutaro/AENET/example/scripts
and those for generation, training, and prediction are respectively,
- qsub_generate.sh
- qsub_train.sh
- qsub_predict.sh

** Training set generation [#f3902a38]

Go to the 01-generate/ directory and execute

 $ ~/AENET/src/aenet-2.0.3/bin/generate.x-2.0.3-ifort_intelmpi generate.in > generate.out

NOTE: generate.x is not parallelized with MPI or OpenMP, and the number of cores used should be 1 when submitting the job using the queueing system.

Or using a job script:
 $ qsub qsub_generate.sh

In this example, the file "TiO2.train" is generated, and will be used in the training.

** Potential training [#x338bcd1]

Go to the 02-train/ directory and execute the following (for example, depending on your system)

 $ mpirun -np 8 ~/AENET/src/aenet-2.0.3/bin/train.x-2.0.3-ifort_intelmpi train.in > train.out

Using a job script:
 $ qsub qsub_train.sh

When the training is finished, we obtain the following potentials

- O.10tw-10tw.ann
- Ti.10tw-10tw.ann

which will be used in the prediction.

** Using the potential to predict energy and forces [#wa7bbddf]

Go to the 03-predict/ directory and execute the following (for example, depending on your system)

 $ mpirun -np 8 ~/AENET/src/aenet-2.0.3/bin/predict.x-2.0.3-ifort_intelmpi predict.in > predict.out

Make sure that there are ANN potentials generated in the previous step.
Using a job script:
 $ qsub qsub_predict.sh