quaternion julia sets in the cave

(iway#59)
daniel j. sandin
university of illinois at chicago
electronic visualization laboratory
851 s. morgan st., room 1120
ph)312-996-3002
fax) 312-413-7585
chicago, il 60607-7053
dan@evl.eecs.uic.edu

no web address

collaborators

louis h. kauffman,
yumei dang, math/stat/cs, uic

description

visualizing quaternion julia sets in the cave. participant will be able to interactively change the constants specifying the quaternion julia set, evolving one fractal form from another. real-time rendering will be based on an inverse iteration computed locally, and rendered in a point cloud of small spheres or tetrahedrons. when an interesting view is discovered (or created), the view and constant information will be sent to remote supercomputers and the julia set will be ray-traced in high detail using a distance estimation technique. this ray-tracing technique is fast but not real time, and the networks cannot ship six 1,024 by 768 bit maps in real time either. the user's ability to steer the computation and perform successive refinement of the ray-tracing operation will improve the user's experience.

mathematical description distance estimation algorithm is a delicate matter to establish, even in the case of julia sets over the complex numbers. depends upon the riemann mapping theorem and a number of special results in complex analysis. we have found that the exact analog of this distance estimate works very well (empirically) for quaternionic julia and mandelbrot sets in three and four dimensions. this experimental work demands both new theoretical work and a finer grained approach to the experimentation. on the theoretical side, collaborator yumei dang is pursuing a program to analyze the distance estimate using theorems in analysis over the quaternions. to complement this, she is pursuing the best ways to push the distance estimate algorithm and corresponding rendering processes on good computer systems. since these images at the present time are quite time consuming to create, this limits the flexibility of the experiments. this is the main mathematical reason for wanting the best supercomputer help for this problem. it is to be expected that a good and flexible source of images for higher-dimensional julia and mandelbrot sets will raise many mathematical questions of whose nature we have not yet dreamed.

ray tracing is one of the realistic methods for rendering the julias. it is not a practical method to sample each point at given resolution along each ray. with the use of the distance estimate formula, the amount of sample points per ray is greatly reduced, but it is still time consuming. hence high-performance supercomputing becomes essential to the visualization of julia sets. the ray-tracing algorithm has been tried on a cray y-mp and it is being run on the dec alpha at the pittsburgh supercomputing center, and better results are expected. the alpha supercluster is a layered system comprised of front-end queue controllers and dedicated machines for computations (the alpha system consists of two front-ends, four parallel development nodes, and eight dedicated compute nodes). it can execute a variety of single-threaded codes, such as the ray-tracing algorithm used in this project, with the parallel virtual machine (pvm) software library. the implementation utilizes advanced supercomputing techniques, such as parallelization and vectorization.

non vbns resources needed


machines: cave with associated Onyx and NCSA Power Challenge Array
addnl sites: SC95 facilities

dedicated vbns resources


PVCs

shared vbns resources:

expected peak b/w From Power Challenge Array, about 100,000 bytes/second.
Delays need to be less than .1 sec. between 1MB and 5MB/sec bandwidth. We have a backup plan to use computers at SGI, but so far it has not been confirmed.
protocols: think tcp/ip
workstation time:
test net time:
other:
biographical information:

commodity access issues:

source of financial support


trial before conference

probably fly out to EVL early fall to try cave-cave session. if sgi else where could try a CAVE-to-Indy connection.

entry number: 29


resources

nii/wall (large screen)
cave/immersadesk
vr other
high-speed network

entry number: 59


return to iway menu return to main menu
last updated 20 july 1995
questions or comments to kc@nlanr.net