CYCLO70: A New Challenging Pericyclic Benchmarking Set for Kinetics and Thermochemistry Evaluation

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Theoretical Chemistry

Background:

• Cycloaddition reactions are crucial in organic synthesis and materials science.
• Existing benchmark datasets may not fully represent the complexity of these reactions.
• Accurate computational modeling is essential for predicting reaction outcomes and designing new materials.

Purpose of the Study:

• Introduce CYCLO70, a novel and challenging benchmark dataset for cycloaddition reactions.
• Evaluate the performance of 93 density functional theory (DFT) functionals on Diels-Alder, dipolar cycloadditions, and rearrangement reactions.
• Analyze the correlation of errors and functional clustering based on performance.

Main Methods:

• Generation of the CYCLO70 dataset using an active learning approach.
• Systematic testing of 93 DFT functionals, from density approximation to double-hybrid levels.
• Principal Component Analysis (PCA) of functional errors across the dataset.
• Validation on Diels-Alder reactions relevant to self-healing polymers.

Main Results:

• CYCLO70 exhibits larger errors compared to existing datasets like BH9.
• The range-separated hybrid ωB97M-V excels in modeling barrier heights and reaction energies.
• PBE-QIDH performs best among double-hybrid functionals; M06-2X and r2SCAN0 are balanced fixed-range hybrids.
• PCA shows high error correlation (98% variance in two components) and clustering of functionals by type.

Conclusions:

• CYCLO70 provides a rigorous test for DFT functionals in cycloaddition chemistry.
• Specific functionals like ωB97M-V and PBE-QIDH demonstrate superior performance.
• Functional performance correlates strongly, and similar functionals cluster together, aiding method selection.
• The dataset's relevance is confirmed by its application to self-healing polymer design.