quimb.tensor.belief_propagation.diis
====================================

.. py:module:: quimb.tensor.belief_propagation.diis


Classes
-------

.. autoapisummary::

   quimb.tensor.belief_propagation.diis.ArrayInfo
   quimb.tensor.belief_propagation.diis.Vectorizer
   quimb.tensor.belief_propagation.diis.DIIS
   quimb.tensor.belief_propagation.diis.DIISPyscf


Module Contents
---------------

.. py:class:: ArrayInfo(shape, size)

   .. py:attribute:: __slots__
      :value: ('shape', 'size')



   .. py:attribute:: shape


   .. py:attribute:: size


   .. py:method:: __repr__()


.. py:class:: Vectorizer(tree=None, backend=None)

   Object for mapping back and forth between any nested pytree of arrays
   or Tensors and a single flat vector.

   :param tree: Any nested container of arrays, which will be flattened and packed into
                a single vector.
   :type tree: pytree of array, optional


   .. py:attribute:: infos
      :value: None



   .. py:attribute:: d
      :value: None



   .. py:attribute:: ref_tree
      :value: None



   .. py:attribute:: backend
      :value: None



   .. py:attribute:: _concatenate
      :value: None



   .. py:attribute:: _reshape
      :value: None



   .. py:method:: setup(tree)


   .. py:method:: pack(tree)

      Take ``arrays`` and pack their values into attribute `.{name}`, by
      default `.vector`.



   .. py:method:: unpack(vector)

      Turn the single, flat ``vector`` into a sequence of arrays.



   .. py:method:: __repr__()


.. py:class:: DIIS(max_history=6, beta=1.0, rcond=1e-14)

   Direct Inversion in the Iterative Subspace (DIIS) method (AKA Pulay
   mixing) [1] for converging fixed-point iterations.

   [1] P. Pulay, Convergence acceleration of iterative sequences. The case of
   SCF iteration, 1980, Elsevier, https://doi.org/10.1016/0009-2614(80)80396-4.

   :param max_history: Maximum number of previous guesses to use in extrapolation.
   :type max_history: int
   :param beta: Mixing parameter, 0.0 means only use input guesses, 1.0 means only use
                extrapolated guesses (original Pulay mixing). Default is 1.0.
   :type beta: float
   :param rcond: Cutoff for small singular values in the pseudo-inverse of the B matrix.
                 Default is 1e-14.
   :type rcond: float


   .. py:attribute:: max_history
      :value: 6



   .. py:attribute:: beta
      :value: 1.0



   .. py:attribute:: rcond
      :value: 1e-14



   .. py:attribute:: vectorizer


   .. py:attribute:: guesses
      :value: [None, None, None, None, None, None]



   .. py:attribute:: errors
      :value: [None, None, None, None, None, None]



   .. py:attribute:: lambdas
      :value: []



   .. py:attribute:: head
      :value: 5



   .. py:attribute:: backend
      :value: None



   .. py:attribute:: B
      :value: None



   .. py:attribute:: y
      :value: None



   .. py:attribute:: scalar
      :value: None



   .. py:method:: _extrapolate()


   .. py:method:: update(y)

      Given new output `y[i]` (the result of `f(x[i])`), update the
      internal state and return the extrapolated next guess `x[i+1]`.

      :param y: The output of the function `f(x)`. Can be any arbitrary nested
                tree structure with arrays treated at leaves.
      :type y: pytree of array

      :returns: **xnext** -- The next guess `x[i+1]` to pass to the function `f(x)`, with the
                same tree structure as `y`.
      :rtype: pytree of array



.. py:class:: DIISPyscf(max_history=6)

   Thin wrapper around the PySCF DIIS implementation to handle arbitrary
   pytrees of arrays, for testing purposes.


   .. py:attribute:: pdiis


   .. py:attribute:: vectorizer


   .. py:method:: update(y)