Please enable JavaScript.
Coggle requires JavaScript to display documents.
MSSC 2024 - Coggle Diagram
MSSC 2024
Basis sets
-
Local Basis sets: Atom-centered, orbitals
-
-
-
Bloch functions
-
Localized
-
-
Gaussian (contraction coef, expo)
-
-
-
-
-
Other phenomenons
Lattice vibration refers to the oscillations of atoms or molecules in a crystal lattice around their equilibrium positions
Spin-orbit coupling is the interaction between an electron's spin magnetic moment and the magnetic field generated by its orbital motion around the nucleus
Vibrational spectra arise from the absorption or scattering of electromagnetic radiation due to transitions between different vibrational energy levels of molecules
Anharmonic effects:
Going beyond the harmonic approximation to include anharmonic terms in the interatomic potential is crucial for modeling phonon-phonon interactions.
-
-
Relativistic effects arise from the need to reconcile quantum mechanics with special relativity for particles moving at speeds comparable to the speed of light
-
Ab initio
Ab initio calculations start from basic, established laws of physics and chemistry, such as quantum mechanics and electrostatics.
-
DFT is a widely used ab initio method that simplifies the many-body problem of interacting electrons by using electron density as the fundamental variable
Ab initio calculations can be computationally intensive, especially for complex systems with many atoms.
-
GNN for materials
Interaction potential
Symmetries ?
-
-
-
A GNN is permutation equivariant if permuting the input nodes results in the same permutation of the output.
-
-
Spherical Harmonics
SE-Conv uses the Clebsch-Gordan tensor product to combine features of different orders while maintaining equivariance.
GNN terms
-
Local interactions: GNNs capture local interactions between atoms by aggregating information from neighboring nodes.
Long-range effects: By stacking multiple GNN layers, information can propagate across longer distances in the molecular graph, allowing the model to capture longer-range interactions and global molecular properties.
Multi-scale interactions: Some GNN architectures explicitly model interactions at different scales, from individual atoms to functional groups or larger molecular fragments.
Attention mechanisms: Graph attention networks can learn to weight the importance of different node interactions (local group/triplets), allowing the model to focus on the most relevant atomic interactions for a given task.
Edge features: Incorporating edge features (e.g., bond types, distances) into the message passing process allows for more detailed modeling of how atoms interfere with each other.
Global readout: Many GNNs use a global readout function to aggregate information from all nodes, capturing how individual atomic interactions contribute to overall molecular properties.
-
Pretraining
Attribute masking and prediction, Edge prediction, Graph reconstruction
-
-
Other materials
-
-
Equivariant NN
-
SE(3)-Transformers: Extend self-attention mechanisms to be equivariant to 3D rotations and translations
E(n)-Equivariant Graph Neural Networks: Achieve equivariance to Euclidean transformations in n-dimensional space
-
-
-
-
-
-
-
-
-
-
-