Error Assessment of Computational Models in Chemistry.

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Scientific Modeling

Background:

• Computational models in chemistry utilize approximations, leading to unpredictable effects on results.
• Quantifying the uncertainty of computational results is essential for assessing model suitability.
• Standard performance metrics like mean absolute error fail to differentiate systematic from random errors.

Purpose of the Study:

• To address challenges in performance assessment of computational chemistry models.
• To elucidate sources of uncertainty in computational chemistry.
• To propose solutions for reliable uncertainty quantification.

Main Methods:

• Analysis of errors in computational chemistry models.
• Distinguishing systematic (explainable) from random (unexplainable) error components.
• Examining examples from quantum chemistry literature.

Main Results:

• Common performance statistics are insufficient for accurate uncertainty assessment.
• Identification of key sources contributing to prediction uncertainty.
• A framework for combining individual uncertainty components is discussed.

Conclusions:

• Accurate performance assessment requires distinguishing error types.
• Understanding and quantifying uncertainty is vital for computational chemistry.
• The proposed methods enhance the reliability of computational predictions.