Version 0.2.0

This package contains a Modern Fortran implementation of the Davidson diagonalization algorithms.
Different schemas are available to compute the correction.

Available correction methods are:

• DPR: Diagonal-Preconditioned-Residue
• GJD: Generalized Jacobi Davidson

Changed

• Improved variables names.
• Removed unnecessary ritz_vectors calculations.
• Optimized the DPR for the free version #34.
• Removed recomputation of the residues #34.

Version 0.1.0

Changed

• Update the whole projection matrix after adding some correction vector. Replace the block update schema.

Fixed

• Fixed several bugs in the matrix-free implementation.

Choose an appropriate initial orthonormal space

New

• Select the initial orthonormal basis set based on the lowest diagonal elements of the matrix

New matrix-free implementation

Version 0.0.4

changed

• For the Matrix-free implementation, the function that computes the target matrix on the fly,
  was replaced by another function representing the action of a matrix-operator over a vector

Reduce complexity of the modules

Version 0.0.3

Change

• split the dense and matrix free into modules
• Moved the lapack calls to their own module
• Moved the array utilities to their own module

Deleted

• Removed the DPR vs GJD benchmark

Generalized eigenvalue and matrix free

Version 0.0.2

Added

• Algorithm for the Generalized eigenvalue problem using the Davidson method.
• Matrix free implementation of the Davidson method using the DPR correction.
• Tests for the lapack wrappers.

Changed

• Append the orthonormalized correction vector to the previous subspace by computing only the
  matrix elements involved with the new correction vector.

• Used as maximum dimension of the subspace 10 times the number of requested eigenvalues.

prototype

Version 0.0.1

Initial prototype

Dependencies

• Fortran compiler >= 6.0 (supporting submodules and move_alloc)
• lapack or MKL.
• Cmake >= 3.0