A temperature-dependent quantum mechanical/neural net model for vapor pressure.

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Machine Learning

Background:

• Accurate prediction of vapor pressure is crucial for chemical engineering and materials science.
• Existing models may lack accuracy or broad applicability across diverse molecular structures and temperatures.

Purpose of the Study:

• To develop a robust, temperature-dependent model for predicting molecular vapor pressure.
• To leverage machine learning and semiempirical quantum chemistry for enhanced predictive capabilities.

Main Methods:

• Utilized a feed-forward neural network trained on 7681 vapor pressure measurements for 2349 molecules.
• Employed Austin Model 1 (AM1) semiempirical molecular orbital theory for descriptor calculation.
• Implemented a 10-fold cross-validation scheme for rigorous error estimation.

Main Results:

• Achieved high accuracy with a standard deviation of error (s) of 0.322 and a correlation coefficient (R^2) of 0.976 on the training set.
• Validated the model's performance on an unseen set with s = 0.326 and R^2 = 0.976.
• Confirmed the model's physically reasonable temperature-dependence and its ability to predict related properties like boiling point and enthalpy of vaporization.

Conclusions:

• The developed neural network model provides a reliable and accurate method for predicting temperature-dependent vapor pressure.
• The model demonstrates excellent generalization capabilities and physical consistency.
• This approach offers a powerful tool for computational chemistry and related fields requiring precise vapor pressure estimations.