dxtb.calculators

dxtb.calculators#

Calculators: Overview#

The dxtb.Calculator object is the center-piece of dxtb, providing a simple interface to compute energies, forces and other properties of molecules.

The calculator is instantiated with the atomic numbers and a parametrization.

import torch
from dxtb import Calculator
from dxtb import GFN1_XTB

numbers = torch.tensor([1, 1])
calc = Calculator(numbers, GFN1_XTB)

It is recommended to always pass the dtype and device to the calculator.

import torch
import dxtb
from dxtb.typing import DD

dd: DD = {"device": torch.device("cpu"), "dtype": torch.double}
numbers = torch.tensor([1, 1])

calc = dxtb.Calculator(numbers, dxtb.GFN1_XTB, **dd)
print(calc.device)  # Expected output: cpu
print(calc.dtype)  # Expected output: torch.float64

The dtype and device can be conveniently changed after instantiation in the same way as for any other PyTorch tensor.

import torch
import dxtb

numbers = torch.tensor([1, 1])
calc = dxtb.Calculator(numbers, dxtb.GFN1_XTB, **dd)
calc = calc.type(torch.float32)
print(calc.dtype)  # Expected output: torch.float32

To configure settings, a dictionary of settings can be passed to the calculator.

import torch
import dxtb

numbers = torch.tensor([1, 1])
settings = {"maxiter": 100}
calc = dxtb.Calculator(numbers, dxtb.GFN1_XTB, opts=settings)

Additional tight binding components can also be added.

import torch
import dxtb

dd = {"device": torch.device("cpu"), "dtype": torch.double}
numbers = torch.tensor([1, 1])

ef = dxtb.components.field.new_efield(torch.tensor([0.0, 0.0, 0.0], **dd))
calc = dxtb.Calculator(numbers, dxtb.GFN1_XTB, interaction=[ef], **dd)

Calculators: Get Properties#

The calculator can be used to compute energies, forces, dipole moments and other properties. The properties are computed by calling the respective get_ method, just as in ASE.

Depending on which calculator you choose, the properties are calculated using analytical, autograd, or numerical derivatives. The default uses automatic differentation. For details, see the calculator types.

Calculators: Caching#

All properties can be cached. However, caching is not enabled by default. To enable caching, pass {"cache_enabled": True} to the calculator options.

Warning

Caching may lead to side effects if automatic differentiation is used multiple times. If you encounter any issues, try running reset(). If this does not help, disable caching or report the issue.

Classes

AnalyticalCalculator

Parametrized calculator defining the extended tight-binding model.

AutogradCalculator

Parametrized calculator defining the extended tight-binding model.

EnergyCalculator

Parametrized calculator defining the extended tight-binding model.

GFN1Calculator

Calculator for the GFN1-xTB method.

GFN2Calculator

Calculator for the GFN2-xTB method.

NumericalCalculator

Parametrized calculator defining the extended tight-binding model.
Contents