Results File#

A SAFIRE run writes every measurement it takes into a single HDF5 file at the path <output_name>.results.h5 relative to where the executable was run. The output_name defaults to the name of the job file without extension, and can also be set explicitly in the job file.

See also

The safiretools.Results class provides tools to quickly extract measurements from the results file and perform statistical postprocessing on them.

File Layout#

<output_name>.results.h5

  • /

    • 🏷️ format_version 1

    • Measurements/

      • Stage<s>/

        • <observable>/

          • bins \((N_\mathrm{bins}, \dots)\)

        • MixedEstimator/

          • <observable>/

            • bins \((N_\mathrm{bins}, \dots)\)

          • ...

        • BackPropEstimator/

          • Steps=<m>/

            • <observable>/

              • bins \((N_\mathrm{bins}, \dots)\)

            • ...

Entry

Description

format_version

The version of the results file layout.

Stage<s>

The measurements of one execution stage, numbered \(s = 0, 1, \dots\) in the order of the execute blocks. See Multiple Execution Stages.

Stage<s>/<observable>

An observable of the energy estimator. See Energy Estimator.

MixedEstimator/<observable>

An observable of the mixed estimator. See Observables.

BackPropEstimator/Steps=<m>/<observable>

An observable of the back-propagation estimator with a back-propagation length of \(m\) steps, one group per entry of propagation_steps. See Back-Propagation Estimator.

bins

The binned measurements of the observable. The leading axis is the bin, followed by the axes of a single measurement, indexed as in the definition of the observable. For complex observables, the array is stored in the TRIQS/h5 interleaved format with a trailing dimension of 2 and a __complex__ attribute.

The group of an observable can itself be nested, as for the pair correlation functions PairCorr/<α>_<β>/bins.

The results file of a previous run with the same output_name is deleted when the run starts. The measurements of an execution stage are written once, when the stage has finished, so a run that is aborted keeps only the stages it completed.

Each bin holds the average of binsize consecutive measurements, set in the execute block of the job file. A bin is written out only once it is full, so the measurements of a trailing partial bin do not appear in the file.

Every measurement is stored as defined in Estimators, so the weighted averages are already divided by the total walker weight at the time of the measurement, and that denominator is not stored. Averaging an observable over its bins is therefore an ordinary unweighted mean.