Struct AtomicNeuralNetwork 

pub struct AtomicNeuralNetwork {
    pub networks: HashMap<u8, FeedForwardNet>,
    pub descriptor: BehlerParrinelloDescriptor,
}

Atomic neural network potential (NNP) with one sub-network per element.

Architecture follows Behler (2011): each atom contributes an atomic energy predicted by an element-specific feed-forward network whose input is the Behler-Parrinello descriptor vector.

Fields

networks: HashMap<u8, FeedForwardNet>

Element-specific networks keyed by atomic number.

descriptor: BehlerParrinelloDescriptor

Symmetry-function descriptor shared by all elements.

Implementations

impl AtomicNeuralNetwork

pub fn new(descriptor: BehlerParrinelloDescriptor) -> Self

Create a new AANN with the given descriptor.

pub fn add_element_network(&mut self, atomic_number: u8, net: FeedForwardNet)

Register a sub-network for the given atomic number.

pub fn atomic_energy( &self, atomic_number: u8, descriptor: &[f32], ) -> Option<f32>

Predict the atomic energy for one atom given its descriptor.

Returns None if no network is registered for atomic_number.

pub fn total_energy(&self, positions: &[[f64; 3]], atomic_numbers: &[u8]) -> f64

Sum of atomic energies over all atoms.

Atoms whose element has no registered network contribute 0.

Trait Implementations

impl Debug for AtomicNeuralNetwork

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.