Prediction of self-diffusion of polar/non-polar, spherical/non-spherical, and hydrogen-bonding molecules in liquids, compressed gases or supercritical fluids.

The machine learning models were trained with a database of 5551 experimental data points from over 216 systems.

The AARD achieved by the models ML5-D11 and ML8-D11 in the test set were 9.06 % and 7.14 %, respectively.

1. Download this repository by clicking Code > Download ZIP. Unzip the folder.

2. Install Python. We recommend a installing the Anaconda Distribution or Miniconda.

3. Open the Anaconda Prompt and change the directory to where you extracted the repository files: cd path/to/folder.

4. Create a conda virtual environment using the provided environment.yml file: conda env create -f environment.yml

5. Activate the environment with: conda activate ml.

6. You can now use the models following instructions bellow either in a .py script file or in a Jupyter Notebook (already provided in the environment by running jupyter lab).

To use the ML5-D11 model: Call the program and provide the properties:

1. Temperature (K)
2. Critical volume (cm3/mol)
3. Critical temperature (K)
4. Density (g/cm3)
5. Acentric factor

To use the ML8-D11 model: Call the program and provide the properties:

1. Temperature (K)
2. Critical volume (cm3/mol)
3. Critical temperature (K)
4. Density (g/cm3)
5. Acentric factor
6. Pressure (bar)
7. Substance identifier (SMILES)

To use the ML5-D11 model:

from ml_D11 import ML5_D11

model=ML5_D11()

model.predict(temp=[112.25], crit_vol=[99.2], crit_temp=[190.4], dens=[0.4222], acent_fact=[0.011])
# Output: array([4.7065659e-05])

To use the ML-D11 model:

from ml_D11 import ML8_D11

model=ML8_D11()

model.predict(temp=[112.25], crit_vol=[99.2], crit_temp=[190.4], dens=[0.4222], acent_fact=[0.011], press=[8.61], smiles=['C'])

# Output: array([4.71234228e-05])

The outputed D11 values are in cm2/s.

More usage examples are provided in the examples.ipynb file.