Geometry-Driven Spin-State Switching in Stacked Phenalenyl Dimers

= 1/2 spin state, serves as an appealing building block for molecular magnetic materials. However, synthesizing high-spin-state stacked dimers has proven difficult due to strong tendencies toward antiferromagnetic pairing. Here, we combine density functional theory with convolutional neural networks to investigate how the stacking geometry governs spin alignment in phenalenyl dimers. Our machine-learning-assisted approach enables efficient mapping of the high-dimensional configurational space, revealing that lateral in-plane displacement can dramatically alter the spin state. Notably, even eclipsed α-carbon stackings, which typically favor low-spin states, can be converted into high-spin configurations through sliding. These results uncover a stacking-dependent mechanism of spin-state control and offer a practical strategy for designing high-spin radicals relevant to quantum information science and molecular spintronics.