Automatic generation of reaction energy databases from highly accurate atomization energy benchmark sets

Area of Science:

• Computational chemistry
• Quantum chemistry
• Chemical benchmarking

Background:

• Reaction energy benchmarks are crucial for evaluating computational chemistry methods.
• Existing methods for creating benchmarks can be resource-intensive.
• Atomization energy databases offer a potential source for generating reaction energy data.

Purpose of the Study:

• To demonstrate a method for automatically generating reaction energy benchmark sets from arbitrary atomization energy databases.
• To analyze the performance of electronic structure methods on a large, newly created reaction energy benchmark set.
• To investigate the correlation between performance on atomization and reaction energy benchmarks.

Main Methods:

• Utilized the W4-11 atomization energy database to generate over 11,000 reaction energies.
• Employed standard quantum chemical calculations for benchmark creation.
• Analyzed the performance of various electronic structure methods against the generated reaction energy data.

Main Results:

• Successfully generated a large reaction energy benchmark set (>11,000 energies) with minimal computational overhead compared to atomization energy calculations.
• Identified atomization energy as a weak predictor of a method's accuracy for reaction energies.
• Evaluated the performance of density functional approximations based on their parameterization.

Conclusions:

• Automatic generation of reaction energy benchmarks from atomization data is feasible and efficient.
• Method performance on atomization energies does not reliably predict performance on reaction energies.
• Further analysis is needed to understand the relationship between method design and performance on diverse chemical benchmarks.