Design of 2D Time-Varying Vector Fields (Supplemental Material)

Guoning Chen, Member, IEEE, Vivek Kwatra, Li-Yi Wei, Charles D. Hansen, Senior Member, IEEE, and Eugene Zhang, Member, IEEE,

1 E XTENSION TO THE D ESIGN OF 2D T ENSOR F IELDS The present framework can be easily extended to the design of 2D time-varying second-order symmetric tensor elds. Tensor elds are currently a popular research subject in both graphics and visualization communities [1, 7, 5, 6, 2, 4]. Tensor elds have rather different characterizations from vector elds. In a tensor eld, there is no denition of pathline since it is oriented-less. Instead, we study the behavior of hyperstreamlines which leads to the tensor eld topology in the static case. Tensor eld topology consists of the degenerate points and their connectivity through separatrices (Note that closed orbits can exist but are barely discussed in the community). Similar to our treatment to a time-varying vector eld, we can create a time-varying tensor eld by modeling its instantaneous topology over time. Therefore, the discussion of the design of orientation elds is still extensible to tensor eld. That is, we can design the degenerate points and their paths, hyperstreamlines, and bifurcations of a time-varying tensor eld in the similar fashion of time-varying vector elds. Therefore, the former approaches for generating a time-varying vector elds can be reformulated for the creation of a time-varying tensor eld. We consider a tensor t , a 2 2 symmetric and traceless matrix, which is of the form cos 2 sin 2 R where R 0 and [0, 2 ). A tensor eld sin 2 cos 2 T is a continuous function that associates every point p = (x, y) R2 with a tensor T (p). p is said to be a degenerate point if T (p) = 0, otherwise, it is regular. More theoretical details can be found in [3]. The 2D time-varying tensor elds can also be designed using the proposed three approaches, i.e. the element-based design, the key frame design, and the eld deformation. For the element-based design, the time-varying design elements adapted from the static case can be dened similarly to those for the time-varying vector elds. For instance, a degenerate point element is dened as D(T , P(t ), M (t )), where T is the tensor that denes the selected type of the degenerate point, P(t ) represents the path of the singular element over time, and M (t ) is the afne transformation matrix (i.e. scaling and rotating) that is exerted on the element along P(t ). Its basis eld at time t is Ti (x; t ) = ed
xpi (t )
2

where Ti (p) is the tensor of the ith element and p = x pi = (x xpi , y ypi ). There are a number of different types of design elements determined by their local tensor, including wedge, trisector, node, center, and saddle [7]. Note that the bifurcation theory for time-dependent tensor elds has not been well established. Thus, the bifurcation element for a timevarying tensor eld is not easy to dene at this moment. Spatial-temporal constrained optimization: In key frame design, the user can specify the instantaneous tensor elds at the desired times. A tensor-valued spatial-temporal constrained optimization can be used to compute the rest of the tensor eld. T (vi ; j ) = kN (i) ik T (vk ; j ) + j, j1 T (vi ; j1 ) + j, j+1 T (vi ; j+1 ) (4)

where T (vi ; j ) represents the average tensor value at position (vi ; j ) in the space time domain. Figure 1 shows a time-varying tensor eld generated using key frame design with the spatiotemporal constrained optimization. This demonstrates the generality of the present frame work to the higherorder eld design (e.g. N-Rosy eld [5]). However, due to rather complicated behaviors in a tensor eld than a vector eld, especially when including time dimension, further comprehensive study is needed.

MiT (t )Ti Mi (t )

x xpi (t ) y ypi (t )

(1)

Similarly, a regular element is dened as (D(t ), P(t )) where D(t ) = (ux (t ), uy (t )) is the direction of the major eigenvector at time t and P(t ) is a prescribed pathline in space and time. As such, the basis eld generated from a regular element at time t is Ti (x; t ) = ed
xpi (t )
2

Mi (t )l (t )

cos 2 (t ) sin 2 (t )

sin 2 (t ) cos 2 (t )

(2)
Fig. 1. A time-varying tensor eld generated using the key frame design and the spatiotemporal constrained optimization. The rst and last elds are the key frames.

and where l (t ) = The total tensor eld is then the sum of these basis elds. T (x; ) = Ti (x; )
i

ux (t )2 + uy (t )2

u (t ) (t ) = arctan( uy (t ) ). x

(3)

R EFERENCES
[1] P. Alliez, D. Cohen-Steiner, O. Devillers, B. L evy, and M. Desbrun. Anisotropic polygonal remeshing. ACM Transactions on Graphics, 22(3):485493, 2003. [2] G. Chen, G. Esch, P. Wonka, P. Mueller, and E. Zhang. Interactive procedural street modeling. ACM Transactions on Graphics, 27(3):103:1103:10, 2008.

[3] T. Delmarcelle and L. Hesselink. The Topology of Symmetric, SecondOrder Tensor Fields. In Proceedings IEEE Visualization 94, pages 140 147, 1994. [4] D. Laidlaw and J. Weickert. Visualization and Processing of Tensor Fields: Advances and Perspectives. Springer Publishing Company, Incorporated, 1st edition, 2009. [5] J. Palacios and E. Zhang. Rotational symmetry eld design on surfaces. ACM Transactions on Graphics, 26(3):56:156:10, 2007. [6] N. Ray, B. Vallet, W.-C. Li, and B. Levy. N-symmetry direction eld design. ACM Transactions on Graphics, 27(2):10:110:13, 2008. [7] E. Zhang, J. Hays, and G. Turk. Interactive tensor eld design and visualization on surfaces. IEEE Transactions on Visualization and Computer Graphics, 13(1):94107, 2007.