Thermo-ML is a python library for scientists in the field of thermodynamics, who want to tap into the power of machine learning to make accurate predictions. (If you have heard of CALPHAD, ChemSage, FactSage, Thermochem, this project might interest you.)

There are two goals to this project:

1. Make physical & chemical data of atoms & compounds accessible with once click.
2. Develop an AI that accurately predicts unknown properties of elements or even unknown compounds, by learning the hidden relationships between the properties of compounds and its constituent atoms.

• Mar 2021
  • ✅ Started the project
  • ✅ Added “parse.py” module to parse chemical formula into its constituent atoms
• April 2021
  • ✅ Added "database.get_fundamental_constants" module to get major physical/chemical fundamental constants
• May 2021
  • ✅ Refactored “parse.py” module & added test code
  • ✅ Added "database.get_atoms" module to get properties of atoms (e.g. ionization energy, electronegativity, atomic radius, etc).
  • 🛠 Add module to get thermodynamic properties of compounds (e.g. enthalpy, entropy, heat capacity, etc).
    • Idea 1: Convert JANAF database to ML readable format
    • Idea 2: Convert open Thermo-Calc Database Format (TDB) to ML readable format
• June 2021
  • 🛠 Add AI module for predicting enthalpy of formation of oxide compounds just from its chemical formula, using properties of its constituent atoms.
    • Note: The reason why I focus on oxides is; (1) oxides are common, and (2) to keep the charge on cations rather constant. If the charge changes, electronegativity changes, and if electronegativity changes, so does enthalpy. For more info see Kaufman, George B. "Inorganic chemistry: principles of structure and reactivity" (1993) P.184 equation 5.62 and Fig. 5.32.
    • Idea 1: Multilinear regression w/ constraints
    • Idea 2: Quadratic programming w/ constraints
    • Idea 3: Symbolic regression + genetic programming
    • Idea 4: Deep Learning
• July 2021
  • 🛠 Add AI module for predicting electronegativity of elements, including transition metals.
    • Note: Leland C. Allen mentioned that electronegativity of transition metals are difficult to obtain, in his paper "Electronegativity Is the Average One-Electron Energy of the Valence-Shell Electrons in Ground-State Free Atoms".
• Aug 2021
  • 🛠 Add AI module for predicting percentage ionic character of bonds.
    • Note: Linus Pauling said in his book that "We cannot hope to formulate an expression for the partial ionic character of bonds that will be accurate".
• Sept 2021
  • 🛠 Add AI module for predicting entropy of formation of compounds just from its chemical formula, using properties of its constituent atoms.
• Oct 2021
  • Hmm what else can I do...

1. Clone the repository using

git clone https://github.com/soap-tastes-ok/thermo-ml.git

2. Install all required dependencies using

pip install -r /your/directory/thermo-ml/requirements.txt

3. Append library path to system path

import sys
package = '/your/directory/thermo-ml'
if package not in sys.path:
    sys.path.append(package)

To parse a chemical formula into it's constituent atoms, use the ChemParser module.

>>> from thermo_ml import parse
>>>
>>> CP = parse.ChemParser()
>>> CP.atoms("Ca2SiO3(OH)2")

[{'Ca': 2.0, 'Si': 1.0, 'O': 5.0, 'H': 2.0}]

To retrieve atomic properties data, use the database.get_atoms module.

>>> from thermo_ml import database
>>>
>>> atoms = ['H', 'C', 'Ca', 'Si', 'Li']
>>> properties = [
>>>     "Z", "Symbol", "Group",
>>>     "Atomic radii (pm)",
>>>     "Atomic weight (a.m.u.)",
>>>     "Valence electrons"
>>> ]
>>> df = database.get_atoms(atoms, properties)

To retrieve fundamental constants, use the database.get_fundamental_constants module.

>>> from thermo_ml import database
>>> df = database.get_fundamental_constants()

TBD

I’m currently a machine learning engineer in Tokyo, who was previously doing research in computational thermodynamics & developing FactSage @McGill University. (Linkedin)

I will work on this during weekends, so please wait patiently. If you are interested to follow this project, please hit the star to let me know you are there and I’ll try to work faster ;)

To cite Thermo-ML in publications, please use:

Kota Matsuo and Contributors (2021-). Thermo-ML: Thermodynamics powered with Machine learning.
https://github.com/soap-tastes-ok/thermo-ml.git.