deepmr.optim.power_method

Contents

deepmr.optim.power_method#

deepmr.optim.power_method = <function power_method>#

Use power iteration to calculate the spectral norm of a Linop.

From MIRTorch (guanhuaw/MIRTorch)

Parameters:
  • A (Callable | torch.Tensor | np.ndarray) – Linear opeartor

  • x (torch.Tensor | np.ndarray) – Initial guess of singular vector corresponding to max singular value

  • AH (Callable | torch.Tensor | np.ndarray, optional) – Adjoint operator AH = A.H. If not provided, attempt to estimate from A. The default is None.

  • AHA (Callable | torch.Tensor | np.ndarray, optional) – Normal operator AHA = AH * A. If not provided, estimate from A and AH. The default is None.

  • device (str, optional) – Computational device. The default is None (infer from input).

  • niter (int, optional) – Maximum number of iterations. The default is 10.

  • tol (float, optional) – Stopping criterion. The default is None (run until niter).

Returns:

The spectral norm of the operator A.

Return type:

float