Dynamic Workflows

Dynamic workflows are created at runtime using OptimizerBuilder. They combine an optimizer, evaluator, and extractor into a single WorkChain class that can be executed with AiiDA run().

Important

Dynamic workflows cannot be submitted to the AiiDA daemon because the generated WorkChain classes are not importable. Use Static Workflows for daemon-managed or long-running tasks.

OptimizerBuilder

OptimizerBuilder is the main factory for assembling workflows dynamically.

Factory Methods

from_problem

Build an optimizer from a simple function-like problem (no structure).

from aiida_reoptimize.base import OptimizerBuilder, BasicExtractor
from aiida_reoptimize.optimizers.convex.GD import AdamOptimizer
from aiida_reoptimize.problems.problems import Sphere

builder = OptimizerBuilder.from_problem(
    optimizer_workchain=AdamOptimizer,
    problem_workchain=Sphere,
    extractor=BasicExtractor(lambda outputs: outputs.value.value),
)
optimizer = builder.get_optimizer()
result = run(optimizer, parameters=Dict({"initial_parameters": [1.0, 1.0]}))

from_ase

Build an optimizer from an ASE Atoms structure and a calculator workchain.

from ase.build import bulk
from aiida_reoptimize.base import OptimizerBuilder, BasicExtractor
from aiida_reoptimize.optimizers.convex.GD import BFGSOptimizer

builder = OptimizerBuilder.from_ase(
    optimizer_workchain=BFGSOptimizer,
    calculator_workchain=FleurScfWorkChain,
    extractor=BasicExtractor(lambda out: out.output_scf_wc_para.total_energy.value),
    calculator_parameters={...},
    bulk=bulk("SrTiO3", "perovskite", a=3.905),
)

from_MPDS

Build an optimizer with a structure retrieved from the MPDS database.

builder = OptimizerBuilder.from_MPDS(
    optimizer_workchain=BFGSOptimizer,
    calculator_workchain=FleurScfWorkChain,
    extractor=extractor,
    calculator_parameters={...},
    mpds_query="WS2/194",
)

Evaluator WorkChains

Two base evaluator WorkChains are used by the builder:

EvalWorkChainProblem: For plain parameter-vector problems. Submits a problem_workchain for each target in a batch.
EvalWorkChainStructureProblem: For structure-based problems. Uses a StructureCalculator to generate builders from parameter vectors.

Extractors

BasicExtractor walks the outputs of each finished AiiDA process using a user-supplied function and returns the objective value. If the process failed or the output is missing, it returns a configurable penalty value instead.