SchNet is a deep learning architecture designed to model atomistic systems, particularly for molecules and materials. It uses continuous-filter convolutional layers to capture quantum-mechanical interactions and enables accurate predictions of properties across chemical space. The model learns chemically plausible embeddings of atom types, allowing it to predict potential-energy surfaces and energy-conserving force fields for molecular dynamics simulations. SchNet is a variant of Deep Tensor Neural Networks (DTNNs) and incorporates rotational, translational, and permutation invariances, as well as periodic boundary conditions. It is capable of efficiently incorporating chemical knowledge and constraints through filter-generating neural networks. SchNet has been demonstrated to accurately predict properties of molecules and materials, including formation energies of bulk crystals and quantum-mechanical properties of C20-fullerene. The model's ability to learn local chemical potentials provides insights into quantum-mechanical observables. SchNet has been applied to perform path-integral molecular dynamics simulations, significantly accelerating the simulation process compared to traditional ab initio methods. The architecture is scalable and efficient, enabling the prediction of properties with high accuracy and allowing for the exploration of chemical space. SchNet's design allows for the incorporation of fundamental symmetries and invariances, making it a powerful tool for quantum chemistry and materials science.SchNet is a deep learning architecture designed to model atomistic systems, particularly for molecules and materials. It uses continuous-filter convolutional layers to capture quantum-mechanical interactions and enables accurate predictions of properties across chemical space. The model learns chemically plausible embeddings of atom types, allowing it to predict potential-energy surfaces and energy-conserving force fields for molecular dynamics simulations. SchNet is a variant of Deep Tensor Neural Networks (DTNNs) and incorporates rotational, translational, and permutation invariances, as well as periodic boundary conditions. It is capable of efficiently incorporating chemical knowledge and constraints through filter-generating neural networks. SchNet has been demonstrated to accurately predict properties of molecules and materials, including formation energies of bulk crystals and quantum-mechanical properties of C20-fullerene. The model's ability to learn local chemical potentials provides insights into quantum-mechanical observables. SchNet has been applied to perform path-integral molecular dynamics simulations, significantly accelerating the simulation process compared to traditional ab initio methods. The architecture is scalable and efficient, enabling the prediction of properties with high accuracy and allowing for the exploration of chemical space. SchNet's design allows for the incorporation of fundamental symmetries and invariances, making it a powerful tool for quantum chemistry and materials science.