The concept of linking a set of already compiled principal components allows to rapidly extend the system by new components. Existing numerical integrators, for example, can be later combined with new visualization techniques without any re-compilation. Comparing different solutions to a specific problem is also very easy as long as the implementation is extracted as a separate component. Currently we can easily compare different dynamical systems, numerical integrators, and visualization techniques.
The design decision to postpone the combination of components to be done after the compilation step has another advantage as well: Multiple developers can work together and extend the system in parallel. Working components of the implementation are made available to the other developers by placing compile code in the DynSys3D file area.
Interactivity is a tough goal to meet. With most of the investigated visualization techniques it is not possible to achieve real time response. The modules allow a variation of the geometric complexity of the results. Therefore at least an interactive inspection is feasible.
DynSys3D provides some mechanisms, which enforce a certain amount of symmetry. Additionally we are trying to increase the amount of symmetry by specifying guidelines, which developers have to follow. For example, we decided to hand over information about the starting location of stream lines by using the AVS data-flow mechanism, although such a seed point also could be specified by the use of UI elements. This clearly does not hold for the specification of the seed locus of a stream surface. Providing the seed input for both techniques (stream line, stream surface) by the same implementation mechanism, more symmetry is achieved.
The separation of interfaces and the implementation of visualization techniques, allows to reuse certain methods within compound visualization techniques. Input data and parameters of such a reused component can be controlled by the module using it and thus it is possible to reduce the parameter set of any AVS module to a meaningful and intuitive extent.
Up to now we were able to omit situations, where we would have had to include some non-trivial techniques from a field we are not experts in, e.g., numerics. For example, we did not implement a general search engine, which reports the characteristic subsets of a dynamical system. Instead we require the dynamical system itself to tell it's characteristic features, e.g., critical points and cycles. Obviously the user runs into troubles if this feature is not implemented for a specific dynamical system he would like to visualize.
See Fig. 8.1 for a sample network showing the concept
of visualization based on DynSys3D. In the upper part three
groups of modules can be distinguished: on the left side a stream
surface is computed and stream arrows are computed. Next to it an
``back-stream'' surface, i.e., a stream surface computed using
backward integration, is computed. On the right side a
representation of the axes of phase space is generated. All three
groups feed their output into the viewer module where the scene is
rendered.
![\framebox[\textwidth]{
\includegraphics[width=.93\textwidth]{pics/net2a.ps}
}](img280.gif)
