Fig. 2: Machine learning design results and solid-solution method for preparing molecular heterogeneous solids.

a Data-driven computational materials design at the scale of microstructure enabled by supervised machine learning. Training and testing datasets consist of data for each molecular ferroelectric. The machine learning model, typically consisting of a descriptor transformation, a learning algorithm, and an evolutionary inspired optimization aims to discover a statistically correct predictive model from training datasets. b Results of the empirical T_c value (measured data) versus the predicted T_c value of molecular ferroelectrics from the machine learning models. In all cases, the models are accurate, robust, and can be extended to any molecular system. c T_c and processing environment for selected high-temperature molecular magnets. VCr_0.86 (V[Cr(CN)₆]_0.86·2.8H₂O); KVCr (KV[Cr(CN)₆]·2H₂O); VCr_0.87 (V[Cr(CN)₆]_0.87·1.6H₂O); V[TCNE]C (V[TCNE]_x·yCH₂Cl₂); V[TCNE] (V[TCNE]_x); LiCrTHF (Li_0.7[Cr(pyz)₂]Cl_0.7·(THF)); LiCrTHF_0.25 (Li_0.7[Cr(pyz)₂]Cl_0.7·0.25(THF)). d Schematic diagram for the hydrogel-based method for the multiferroic composites. The hydrogel sample with the desired proton, FE ions, and MA nanoparticles can be obtained by directly printing from the precursor. The crystallization process for the IM-VH is performed based on the printed hydrogel which was dried in a nitrogen atmosphere.