Auto_IPC-RC

An atuoencoder architecture for mining reaction coordinate (or implicit physical characteristic )

This repository contains codes and supplementary data supporting results of this paper:

Evidence for the Generic Existence of Two Local Structures in Liquid Water

ENVIRONMENT REQUIREMENTs

python == 3.11.0
tensorflow-gpu >= 2.4.0
numpy == 1.26.3
scipy == 1.12.0
pandas == 2.1.4
matplotlib == 3.8

INSTALLATION GUIDE: N/A

TRAINING PROCESS

In the ./Linear_$alpha/slop_$phi/ Path (i.e., Auto_IPC-RC/Linear_0.3/slope_455/), the "dp_LDL_simple_fhi47_100_linear.py" is the training main program.

• $alpha and $phi Setting

###  $alpha setting ###
#line 555 
loss_correlation = (tf.reduce_mean((correlation-0.3)**2) )
#line 556  
loss_spearman_cor = (tf.reduce_mean((spearman_cor-0.3)**2) ) 

###  $phi setting ###
aa_loss1 = tf.reduce_mean( (k_cor - tf.math.tan(455.*pi/1000.0))**2 ) #line 564

• Executive command

python3 dp_LDL_simple_fhi47_100_linear.py

• Training output In the ./Linear_$alpha/slop_$phi/log_simple_fhi47_100_linear Path, "test.log", "train.log" and "xe.log" are the outputs. And using "draw.py" can obtain the training process.

TESTING

In the ./Linear_$alpha/slop_$phi/ Path (i.e., Auto_IPC-RC/Linear_0.4/slope_490/), the "dp_LDL_simple_fhi47_100_linear_test_auto.py" is the testing main program.

• Executive command

python dp_LDL_simple_fhi47_100_linear_test_auto.py 1800_188 0

Also can execute the auto_PT.pl or auto_PT2.pl in Auto_IPC-RC, e.g.,

perl auto_PT.pl

• Testing output In the ./Linear_$alpha/slop_$phi/logtest/ Path, "xe.log" or "xe_1800_188.log"/"xe_1800_188.txt" (renamed by auto_PT/auto_PT2.pl) are the outputs.

DATA SET

• DATASET PATH

├── Auto_IPC-RC/
└── dp_LDL/

The training data "dp_LDL" (~8 GB, for tip4p/ice P-T-rho-potential dataset) can be downloaded from the link within the article.

Training-Dataset-PART01 (dp_LDL.z01): Evidence for the Generic Existence of Two Local Structures in Liquid Water

Training-Dataset-PART02 (dp_LDL.z02): Evidence for the Generic Existence of Two Local Structures in Liquid Water

cat dp_LDL.z01 dp_LDL.z02 > dp_LDL.zip
unzip dp_LDL.zip

• DATASET FORMAT

1. In the "new_coord" Path, each $xx.npy (i.e., 1.npy, 2.npy, ...) file represents the $xx step (i.e., 1.npy, 2.npy, ...) during the MD simulation.

$xx.npy shape: (300,30,4), where "300" is the total number of water molecules in a configuration, "30" is the max number of water molecules (oxygen atoms) in a local structure, and "4" refers to (s(r)^{^}, x^{^}, y^{^}, z^{^})-the detail can read End-to-end Symmetry Preserving Inter-atomic Potential Energy Model for Finite and Extended Systems

2. In the "box" Path, each $xx.npy (i.e., 1.npy, 2.npy, ...) file also represents the $xx step (i.e., 1.npy, 2.npy, ...) during the MD simulation. And these files correspond one by one to the files in the coord path.

$xx.npy shape: (9,) - (time, boxx, boxy, boxz, temp, press, rho, pot, ent, pb) for a configuration from $xx step - can read from boxdata.csv. Herein, our model only learn the rho and pot information for a configuration from $xx step.