Asymmetric Tensors [pdf]
Asymmetric Tensors [pdf] Asymmetric Tensors [pdf]
2D Asymmetric Tensor Field Analysis and Visualization Eugene Zhang Oregon State University
- Page 2 and 3: Introduction • Asymmetric tensors
- Page 4 and 5: Introduction • Existing technique
- Page 6 and 7: Introduction • Given a vector fie
- Page 8 and 9: Introduction • Flow motions and p
- Page 10 and 11: Introduction • However, tensor fi
- Page 12 and 13: Introduction • Questions: - What
- Page 14 and 15: Tensor Decomposition • The set of
- Page 16 and 17: Eigenvector Manifold • Eigenvecto
- Page 18 and 19: Eigenvector Manifold • Eigenvalue
- Page 20 and 21: Eigenvector Manifold • We can foc
- Page 22 and 23: Eigenvector Manifold
- Page 24 and 25: Eigenvector Manifold
- Page 26 and 27: Eigenvector Manifold
- Page 28 and 29: Eigenvector Manifold
- Page 30 and 31: Eigenvector Manifold • Bisectors
- Page 32 and 33: Eigenvector Manifold • Dual-eigen
- Page 34 and 35: Eigenvector Manifold • Degenerate
- Page 36 and 37: Eigenvector Manifold • Visualizat
- Page 38 and 39: Eigenvalue Manifold • Eigenvalues
- Page 40 and 41: Eigenvalue Manifold
- Page 42 and 43: Eigenvalue Manifold
- Page 44 and 45: Eigenvalue Manifold Tensor Magnitud
- Page 46 and 47: Combining Eigenvector and Eigenvalu
- Page 48 and 49: Applications • Sullivan flow (a t
- Page 50 and 51: Applications • Sullivan flow (a t
2D <strong>Asymmetric</strong> Tensor Field<br />
Analysis and Visualization<br />
Eugene Zhang<br />
Oregon State University
Introduction<br />
• <strong>Asymmetric</strong> tensors can model the gradient<br />
of a vector field<br />
– velocity gradient in fluid dynamics<br />
– deformation gradient in solid mechanics
Introduction<br />
• Flow visualization has a wide range of<br />
applications in areo- and hydro-dynamics:<br />
y<br />
– Climatology<br />
– Oceanography and limnology<br />
– Hydraulic engineering<br />
– Aircraft and undersea vehicle design
Introduction<br />
• Existing techniques often focus on velocity<br />
[Laramee et al. 2004, Laramee et al. 2007]<br />
– Good for visualizing particle movement<br />
Trajectories<br />
Vector magnitude<br />
Topology
Introduction<br />
• Basic types of non-translational motions<br />
Rotation (+/-)<br />
Expansion<br />
Pure shear<br />
Contraction
Introduction<br />
• Given a vector field , the local<br />
linearization at is:<br />
Velocity vector field:<br />
translation<br />
Velocity gradient tensor field:<br />
yg<br />
rotation, isotropic scaling (expansion<br />
and contraction), and pure shear.
Introduction<br />
• Rotation:<br />
• Isotropic scaling:<br />
• Anisotropic stretching (aka pure shear):
Introduction<br />
• Flow motions and physical meanings: [Batchelor<br />
1967, Lighthill 1986, Ottino 1989, Sherman 1990]<br />
– Rotation:<br />
• Vorticity<br />
– Isotropic scaling:<br />
• volume change and/or stretching t in the third dimensioni<br />
– Anisotropic stretching:<br />
• rate of angular deformation, related to energy dissipation and rate<br />
of fluid mixing
Introduction<br />
• Velocity gradient tensor has been used in<br />
vector field visualization<br />
– Singularity classification [Helman and Hesselink 1991]<br />
– Periodic orbit extraction [Chen et al. 2007]<br />
– Attachment and separation detection [Kenwright 1998]<br />
– Vortex core identification [Sujudi and Haimes 1995, Jeong and<br />
Hussain 1995, Peikert and Roth 1999, Sadarjoen and Post 2000]
Introduction<br />
• However, tensor field structures were not<br />
investigated and applied to flow visualization<br />
– Velocity gradient is asymmetric<br />
– Past work in tensor field visualization focus on<br />
symmetric tensors [Delmarcelle and Hesselink 1994, Hesselink et<br />
al. 1997, Tricoche et al. 2001, Tricoche et al. 2003, Hotz et al. 2004,<br />
Zheng and Pang 2004, Zheng et al. 2005, Zhang et al. 2007]
Introduction<br />
• Symmetric tensors<br />
– two real eigenvalues<br />
– two mutually perpendicular eigenvectors when not<br />
degenerate<br />
• <strong>Asymmetric</strong> tensors<br />
– can have complex eigenvalues<br />
– eigenvectors not always mutually perpendicularp
Introduction<br />
• Questions:<br />
– What are features in an asymmetric tensor field?<br />
– How to visualize these features?<br />
Wh t l b t th fl f th<br />
– What can we learn about the flow from these<br />
features?
Tensor Decomposition<br />
– Isotropic scaling:<br />
– Rotation:<br />
– Pure shear:
Tensor Decomposition<br />
• The set of 2x2 tensors can be parameterized<br />
by , , , and , such that<br />
– , and<br />
–<br />
• This is a four-dimensional space<br />
• Can we focus on configuration spaces with<br />
lower-dimensions?
Tensor Decomposition<br />
• Eigenvalues only depend on , , and<br />
• Eigenvectors are dependent on , , and<br />
• Define eigenvector and eigenvalue manifolds
Eigenvector Manifold<br />
• Eigenvectors of<br />
are same as<br />
are same as<br />
can be rewritten as
Eigenvector Manifold<br />
Image credit: http://math.etsu.edu/MultiCalc/Chap3/Chap3-4/sphere1.gif
Eigenvector Manifold<br />
• Eigenvalues are constant along any latitude
Eigenvector Manifold<br />
• Real domains and complex domains<br />
• Degenerate curves
Eigenvector Manifold<br />
• We can focus on any longitude and understand<br />
how eigenvectors change
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold
Eigenvector Manifold<br />
• Bisectors never change along any longitude<br />
• They are dual-eigenvectors (Zheng and<br />
Pang 2005)<br />
• Dual-eigenvectors are eigenvectors of
Eigenvector Manifold<br />
• Dual-eigenvectors of
Eigenvector Manifold<br />
• Dual-eigenvectors of
Eigenvector Manifold<br />
• Which side of the Equator matters<br />
• Incorporate the Equator into asymmetric<br />
Incorporate the Equator into asymmetric<br />
tensor topology
Eigenvector Manifold<br />
• Degenerate (circular) points of<br />
– Number, location, index, orientation<br />
Major Eigenvectors of<br />
Symmetric Component<br />
Major Dual-Eigenvectors
Eigenvector Manifold<br />
• Poincaré-Hopf theorem (asymmetric<br />
tensors):<br />
– Given a continuous asymmetric tensor field<br />
defined on a closed surface S such that<br />
has only isolated degenerate points ,<br />
then
Eigenvector Manifold<br />
• Visualization<br />
– Black curves:<br />
– White curves:<br />
– Blue curves:<br />
– Degenerate points<br />
– The Equator<br />
– Degenerate curves
Eigenvector Manifold
Eigenvalue Manifold<br />
• Eigenvalues depend on , , and<br />
• We are interested in relatively strengths<br />
We are interested in relatively strengths<br />
among the three components
Eigenvalue Manifold
Eigenvalue Manifold
Eigenvalue Manifold<br />
Dominant component
Eigenvalue Manifold
Eigenvalue Manifold<br />
All components
Eigenvalue Manifold<br />
Tensor Magnitude
Combining Eigenvector and<br />
Eigenvalue Manifolds<br />
• CCW-dominant region (red) must be in the<br />
northern hemisphere (red)<br />
• CW-dominant region (green) must be in the<br />
southern hemisphere (red)
Combining Eigenvector and<br />
Eigenvalue Manifolds
Applications<br />
• Sullivan flow (a tornado model) [Sullivan:1959]
Applications<br />
• Sullivan flow (a tornado model)<br />
Vector field topology<br />
Dominant eigenvalue +<br />
major eigenvector and<br />
major dual-Eigenvector<br />
Tensor magnitude
Applications<br />
• Sullivan flow (a tornado model)<br />
Vector Field Topology<br />
Dominant Eigenvalue +<br />
Major Eigenvector and<br />
Major Dual-Eigenvector<br />
Tensor Magnitude
Applications<br />
• Sullivan flow (a tornado model)<br />
Vector Field Topology<br />
Dominant Eigenvalue +<br />
Major Eigenvector and<br />
Major Dual-Eigenvector<br />
Tensor Magnitude
Applications<br />
• Sullivan flow (a tornado model)<br />
Vector field topology<br />
Dominant eigenvalue +<br />
major eigenvector and<br />
major dual-eigenvector<br />
Tensor magnitude
Applications<br />
• Diesel engine
Applications<br />
• Diesel engine
Open Questions<br />
• What does tensor index tell us about the<br />
flow, other than zero stretching?<br />
• Can we generate a graph representation for<br />
asymmetric tensors, much like the vector<br />
field topology?
Open Questions<br />
• How do we integrate information from vector<br />
and tensor field analysis?
Open Questions<br />
• How does the analysis carry over to 3D?<br />
– Axis of rotation is not always aligned with any of<br />
the eigenvector directions, which means all of<br />
them could be moving when going from real<br />
domains into complex domains<br />
– How to deal with 3D anisotropic stretching?<br />
– What do the eigenvalue and eigenvector manifolds<br />
look like?
Open Questions<br />
• What is the topology of higher-order tensor<br />
fields, symmetric or not, 2D or 3D, static or<br />
time-varying? And why do we care?
Acknowledgement<br />
• Collaborators<br />
– Dr. Harry Yeh, Professor in fluid mechanics, Oregon<br />
State University<br />
– Darrel Palke, Intel<br />
– Zhongzang Lin, Ph.D. student at Oregon State<br />
University<br />
– Dr. Guoning Chen, postdoctoral researcher at<br />
University of Utah<br />
– Dr. Robert S. Laramee, Lecturer (Assistant Professor)<br />
at Swansea University, UK
Acknowledgement<br />
• Inspirations from:<br />
– Dr. Xiaoqiang Zheng, Nvidia<br />
– Dr. Alex Pang, Professor in computer science, UC<br />
Santa Cruz<br />
– The pioneers in vector and tensor field analysis and<br />
– The pioneers in vector and tensor field analysis and<br />
visualization
References<br />
• Xiaoqiang Zheng and Alex Pang, “2D <strong>Asymmetric</strong> Tensor Analysis”,<br />
IEEE Vis 2005 (http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1532770)<br />
• Eugene Zhang, Harry Yeh, Zhongzang Lin, and Robert S. Laramee,<br />
“<strong>Asymmetric</strong> Tensor Analysis for Flow Visualization”, IEEE Trans. on<br />
Visualization and Computer Graphics<br />
(http://www.computer.org/portal/web/csdl/doi/10.1109/TVCG.2008.68)<br />
org/portal/web/csdl/doi/10 1109/TVCG • Darrel Palke, Guoning Chen, Zhongzang Lin, Harry Yeh, Robert S.<br />
Laramee, and Eugene Zhang, “<strong>Asymmetric</strong> Tensor Visualization with<br />
Glyph and Hyperstreamline Placement on 2D Manifolds”, Tech<br />
Report, Oregon State University<br />
(http://ir.library.oregonstate.edu/jspui/handle/1957/13549)
Questions?