Navigation :

Algorithms

Work in progress!

In Octopus, many operations, such as time propagations, geometry optimization, etc. are implemented in terms of algorithms. An algorithm, in general, contains a set of instructions, which are performed in a well-defined order.

Algorithm container

The algorithm_t class itself, is only an extension of the linked list.

Definition of algorithm_t

  type, extends(linked_list_t), abstract :: algorithm_t
    private
    type(algorithm_iterator_t), public :: iter            !< Iterator for algorithmic operations
    type(algorithmic_operation_t) :: current_ops          !< The current operation

    type(algorithmic_operation_t), public  :: start_operation !< @brief algorithm specific initialization operation;
    !!
    !!                                                           this operation is performed only once before the algorithm loop starts.

    type(algorithmic_operation_t), public  :: final_operation !< @brief algorithm specific finalization operation
    !!
    !!                                                           this operation is performed only once after the algorithm loop ends.

    integer, public :: algo_steps                         !< @brief Number of 'algorithmic steps' per algorithmic iteration
    !!
    !!                                                       This describes the granularity of the iteration. The algorithm iteration
    !!                                                       counter will advance algo_steps per iteration.
    !!
    !!                                                       This also means that the interactions have to be updated at
    !!                                                       algo_steps times per algorithm iteration.

    logical :: iteration_done                             !< Indicate whether the current iteration is done.

    type(iteration_counter_t), public :: iteration        !< Keep track at which iteration this algorithm is.
    real(real64) :: start_time = M_ZERO                          !< Keep the wall clock time when the algorithm started,
    real(real64), public :: elapsed_time = M_ZERO                !< Elapsed wall clock time for printing info

  contains
    procedure :: add_operation => algorithm_add_operation                 !< @copydoc algorithm_add_operation
    procedure :: do_operation => algorithm_do_operation                   !< @copydoc algorithm_do_operation
    procedure :: update_elapsed_time => algorithm_update_elapsed_time     !< @copydoc algorithm_update_elapsed_time
    procedure :: rewind => algorithm_rewind                               !< @copydoc algorithm_rewind
    procedure :: next => algorithm_next                                   !< @copydoc algorithm_next
    procedure :: get_current_operation => algorithm_get_current_operation !< @copydoc algorithm_get_current_operation
    procedure(algorithm_finished), deferred :: finished                   !< @copydoc algorithm_finished
    procedure(algorithm_init_iteration_counters), deferred :: init_iteration_counters             !< @copydoc algorithm_init_iteration_counters
  end type algorithm_t

Algorithmic operations can be added with the function add_operation(). Examples are discussed in the section Propagators.

Placement in the class hierarchy

Algorithm iterator

Definition of algorithm_iterator_t

Algorithmic operations

Definition of algorithmic_operation_t

Some global algorithmic steps are defined in multisystem/propagator.F90:


  ! Known propagation operations
  character(len=ALGO_LABEL_LEN), public, parameter ::   &
    START_SCF_LOOP       = 'START_SCF_LOOP',            &
    END_SCF_LOOP         = 'END_SCF_LOOP',              &
    STORE_CURRENT_STATUS = 'STORE_CURRENT_STATUS'

  type(algorithmic_operation_t), public, parameter :: &
    OP_START_SCF_LOOP       = algorithmic_operation_t(START_SCF_LOOP,       'Starting SCF loop'),         &
    OP_END_SCF_LOOP         = algorithmic_operation_t(END_SCF_LOOP,         'End of SCF iteration'),      &
    OP_STORE_CURRENT_STATUS = algorithmic_operation_t(STORE_CURRENT_STATUS, 'Store current status')

Derived propagators can then add their own steps, such as e.g. in multisystem/propagator_verlet.F90:


  ! Specific verlet propagation operations identifiers
  character(len=30), public, parameter ::      &
    VERLET_START       = 'VERLET_START',       &
    VERLET_FINISH      = 'VERLET_FINISH',      &
    VERLET_UPDATE_POS  = 'VERLET_UPDATE_POS',  &
    VERLET_COMPUTE_ACC = 'VERLET_COMPUTE_ACC', &
    VERLET_COMPUTE_VEL = 'VERLET_COMPUTE_VEL'

  ! Specific verlet propagation operations
  type(algorithmic_operation_t), public, parameter :: &
    OP_VERLET_START       = algorithmic_operation_t(VERLET_START,       'Starting Verlet propagation'),               &
    OP_VERLET_FINISH      = algorithmic_operation_t(VERLET_FINISH,      'Finishing Verlet propagation'),              &
    OP_VERLET_UPDATE_POS  = algorithmic_operation_t(VERLET_UPDATE_POS,  'Propagation step - Updating positions'),     &
    OP_VERLET_COMPUTE_ACC = algorithmic_operation_t(VERLET_COMPUTE_ACC, 'Propagation step - Computing acceleration'), &
    OP_VERLET_COMPUTE_VEL = algorithmic_operation_t(VERLET_COMPUTE_VEL, 'Propagation step - Computing velocity')

It is important to stress here, that one algorithmic step, in general, does not advance any clock. Clocks are only advanced in steps which update the corresponding entity, which could be a system, an exposed quantity or an interaction. It is therefore quite common, that at a specific state of the algorithm, clocks or different entities have different values.