refinements “on-the-fly” to quickly confirm the quality of the measurements and to
ascertain whether a chosen set of experimental conditions is giving the desired result.
Such on-the-fly processing requires a fast, reliable auto-processing pipeline. We
venture that this can be best achieved using command-line-based programs that can
be controlled via an external script without the need for user interaction. Automated
processing is particularly essential for making efficient use of the limited experiment
time available at synchrotron and XFEL facilities; processing scripts can easily be
incorporated into workflows so that they are started automatically and can provide
real-time feedback to users to enable fast decision-making.
In a traditional pump-probe experiment, data can be simply accumulated until
enough signal is obtained over an angular rotation, and once a full set of crystal
orientations have been measured, the data can be processed immediately using
standard workflows. For pump-multiprobe experiments, which record data for
several time delays simultaneously, additional preprocessing steps are required to
prepare the data for analysis. The data from individual time delays must be separated, and multiple images collected for a given time delay/rotation may then need to
be summed to obtain sufficient signal-to-noise ratio. Automated routines are
required to prevent this from becoming a bottleneck during an experiment, and
given the huge amount of data involved, this may require access to highperformance computing clusters. After the data has been sorted and preprocessed,
individual datasets are obtained for each time delay, which can be treated as regular
SCXRD datasets and processed as outlined above.
Although ab initio structure solution may be possible with data from individual
time delays, for rapid auto-processing routines, a good-quality starting model for
both the ground state and the photostationary excited state is optimal. These can be
obtained from low-temperature steady-state photocrystallographic measurements
carried out in preparation for the TR-SCXRD study. By using these as reference
models when refining the TR data, the occupations of known excited-state species
can be quickly and automatically refined. Identifying unknown short-lived species is
more complicated, but can be approached by looking for new features in Fourier
electron density difference maps generated between the TR structures and the
ground-state structural model (termed “photodifference maps”). If new short-lived
species are identified, they are likely to be present at a low population level and will
require a new disorder model to be created to refine the data against. At present there
is no substitute for an experienced crystallographer in this situation, and user
intervention is thus still required.
In summary, the aim of automatic data processing pipelines in pump-(multi)probe
SCXRD experiments should be to automate the routine parts of the data processing
and enable users to quickly identify datasets that show changes or new structural
features of interest so that these can be investigated manually by an experienced
crystallographer. Automatic processing pipelines are absolutely critical to the success of TR studies and will only become more important as these methods progress
towards faster data collections at higher time resolution.
Watching Photochemistry Happen: Recent Developments in Dynamic Single-Crystal. . .
233
ascertain whether a chosen set of experimental conditions is giving the desired result.
Such on-the-fly processing requires a fast, reliable auto-processing pipeline. We
venture that this can be best achieved using command-line-based programs that can
be controlled via an external script without the need for user interaction. Automated
processing is particularly essential for making efficient use of the limited experiment
time available at synchrotron and XFEL facilities; processing scripts can easily be
incorporated into workflows so that they are started automatically and can provide
real-time feedback to users to enable fast decision-making.
In a traditional pump-probe experiment, data can be simply accumulated until
enough signal is obtained over an angular rotation, and once a full set of crystal
orientations have been measured, the data can be processed immediately using
standard workflows. For pump-multiprobe experiments, which record data for
several time delays simultaneously, additional preprocessing steps are required to
prepare the data for analysis. The data from individual time delays must be separated, and multiple images collected for a given time delay/rotation may then need to
be summed to obtain sufficient signal-to-noise ratio. Automated routines are
required to prevent this from becoming a bottleneck during an experiment, and
given the huge amount of data involved, this may require access to highperformance computing clusters. After the data has been sorted and preprocessed,
individual datasets are obtained for each time delay, which can be treated as regular
SCXRD datasets and processed as outlined above.
Although ab initio structure solution may be possible with data from individual
time delays, for rapid auto-processing routines, a good-quality starting model for
both the ground state and the photostationary excited state is optimal. These can be
obtained from low-temperature steady-state photocrystallographic measurements
carried out in preparation for the TR-SCXRD study. By using these as reference
models when refining the TR data, the occupations of known excited-state species
can be quickly and automatically refined. Identifying unknown short-lived species is
more complicated, but can be approached by looking for new features in Fourier
electron density difference maps generated between the TR structures and the
ground-state structural model (termed “photodifference maps”). If new short-lived
species are identified, they are likely to be present at a low population level and will
require a new disorder model to be created to refine the data against. At present there
is no substitute for an experienced crystallographer in this situation, and user
intervention is thus still required.
In summary, the aim of automatic data processing pipelines in pump-(multi)probe
SCXRD experiments should be to automate the routine parts of the data processing
and enable users to quickly identify datasets that show changes or new structural
features of interest so that these can be investigated manually by an experienced
crystallographer. Automatic processing pipelines are absolutely critical to the success of TR studies and will only become more important as these methods progress
towards faster data collections at higher time resolution.
Watching Photochemistry Happen: Recent Developments in Dynamic Single-Crystal. . .
233