Visualization of datasets that are much to large to be displayed directly at interactive rates requires that the data visualized be somehow decimated before it is displayed. This can often be done in the process of transforming the data into displayable form. For instance, if one is displaying isosurfaces of scalar field, an isocontouring algorithm can tailor the output volume of the geometric information it produces to the capabilities of the display device being used and the communication channel between the platform on which the algorithm is running and the platform used for display. Using such an approach, a scientist can interact effectively with the displayed data, but reanalyzing the data, as for instance requesting a new set of isocontours, can often be very slow to proceed interactively. In order to provide full interrogative visualization capabilities, including reanalysis, it can also be important to be able to do the analysis itself at a resolution that can be tailored to the needs of the scientist. To do this, the input data to be analyzed must itself be preconditioned so that it can be retrieved by the analysis algorithm in a multiscale fashion. This allows the analysis to be done at varying granularities at corresponding speeds, an , ideally, to be done progressively , that is, at ever increasing levels of quality over the time period available for the computation. These considerations motivate the key task outline below for the work proposed here, leveraging from our prior work on each of these tasks.

We will develop schemes for multiresolution organization an compression of large scale mesh and grid datasets, including the preservation of differential/integral properties (e.g. topology preserving). And we will develop algorithms for progressive and adaptive visualization work from progressive, compressed geometric input data streams. These will include accelerated isocontouring, of scalar fields and scalar/vector topology determination (e.g. vortex identification).