Schlüssel-Komponenten
We propose a data analysis management platform that allows free and full interaction with the data handling, modelling, simulation, and visualization resources of the PNI centres supported by core routines on the users own computer. This will be achieved through well-defined API which will allow legacy and external high performance resources to be linked in an effective and timely manner.
Levels Involved in Data Analysis Work Flow
Level 1: Data treatment
The first program block comprises fast, efficient, common, generic libraries,
implemented at server or stand-alone level providing the essential initial data
treatment steps, e.g. transformation from angles, detector pixel positions, etc. into
physical space like (Qx,Qy,Qz,E) including normalization, background and sensitivity
corrections.
Level 2: Data visualization
This allows for the interactive presentation of n-dimensional data as projections, cuts
or iso-surfaces. For the challenges presented here, recognition algorithms for
correlations in the data and automated search strategies for clustering and
classifying are of particular significance. This enables the identification of relevant
sections of parameter space. This involves constructing common libraries, but the
final aim is for user tools tailored for the physical problems.
Level 3: Simulations
This offers access to expert software and CPU-intensive applications. The aim is to
perform on-demand and quasi-real time simulations to both obtain first information
within constraints on physical parameters as well as input for further more
sophisticated analysis. Envisaged capabilities include access to Monte Carlo
instrument simulations, DFT and analytical expressions, but also the combination of
these to simulate the total scattering response.
Level 4: Modelling/Analysis
Here, the connection with the physical parameters is made. Comparison is made
between model and data by the use of statistical analysis, e.g. regression or
Bayesian analysis. Tailored plug-in modules from the user community are extremely
important. The challenge is to address comprehensive data sets, so simultaneous
fitting of multi-technique/-multi-probe data is central. Open source and usercontributed
software have successfully been implemented recently, for example in
McSTAS/VITESS by the instrument simulation community, and we will follow the
best practice here.
Level 5: Real space visualization
Especially for the non-expert user, a visualization of the model in easiest terms is of
highest importance, which usually means real space. This step is always an
important one in making the connection between structure and functionality. What is
required here is to translate the determined system parameters into a readily
viewable representation, such as crystal structure or set of vibrational modes. Again,
the most convenient and meaningful representation goes hand in hand with the type
of system being studied, and we take the same approach as before in writing core
libraries linked with tailored tools.