Topologically Guaranteed Bivariate Solutions of Under-constrained Multivariate Polynomial Systems Yonathan Mizrahi Prof. Gershon Elber Department of Mathematics Technion Department of Computer Science Technion 1 Outline Topology – basic concepts: General intro Homeomorphisms Manifolds Classification theorems solvers – a quick review? Bivariate solutions with topological guarantee Conclusion Subdivision Center for Graphics and Geometric Computing, Technion 2 Topology – basic concepts 3 Topology The study of properties of space that are invariant under continuous mappings, for example: Closed/Open sets Compactness Connectedness Interior/boundary points Center for Graphics and Geometric Computing, Technion 4 Topology A topological space is a set 𝑋 equipped with a family of subsets 𝜏 ⊂ 2 𝑋 called “open”, that must satisfy a list of requirements Topological spaces are considered equivalent if there’s a continuous bijection with a continuous inverse between them. Such a function is called a homeomorphism. Center for Graphics and Geometric Computing, Technion 5 Equivalent spaces - examples open intervals in ℝ An open interval and the entire line Two open discs in the plane A sphere and the boundary of a cube Two Center for Graphics and Geometric Computing, Technion 6 Manifolds Topological spaces* that are locally homeomorphic to ℝ𝑛 are called 𝑛-manifolds A curve with no self intersections is a manifold of dimension one A two dimensional manifold is also called a regular surface A sub-manifold is a manifold w.r.t to the induced topology (defined by intersection with open sets of the embedding space) Center for Graphics and Geometric Computing, Technion 7 Classification theorems Connected compact manifolds: Dimension zero: a point Dimension one: a closed interval or a closed loop Dimension two: determined by the genus and the number of boundary loops (orientable case) Center for Graphics and Geometric Computing, Technion 8 Subdivision solvers A quick review? 9 Algebraic constraints Problems (often…) reduce to: 𝐹 𝑥 =0 The Stewart Platform (Kinematics) http://www.physikinstrumente.com Bisectors (CAGD) Center for Graphics and Geometric Computing, Technion 10 The general problem Let 𝐹: 𝐷 ⊂ ℝ𝑛 → ℝ𝑘 𝑘 ≤ 𝑛 , where 𝐷 is the hyper-box: 𝐷 = 𝑎1 , 𝑏1 × ⋯ × 𝑎𝑛 , 𝑏𝑛 . Consider the problem of finding: 𝑘 𝑍 = 𝑥 ∈ 𝐷: 𝐹 𝑥 = 0 ∈ ℝ Center for Graphics and Geometric Computing, Technion 11 Basic assumptions 1. The underlying function 𝐹 is piecewise 𝐶 1 smooth 2. The gradients of the scalar components of 𝐹 = 𝑓1 , … , 𝑓𝑘 are linearly independent at solution points (i.e. a maximal rank differential) Center for Graphics and Geometric Computing, Technion 12 Basic solution set properties Under the regularity assumption, the solution set: 𝑍 = 𝑥 ∈ 𝐷: 𝐹 𝑥 = 0 ∈ ℝ𝑘 Is an 𝑛 − 𝑘 dimensional sub-manifold of 𝑛 ℝ (Inequalities: further trim, post-process) Center for Graphics and Geometric Computing, Technion 13 Function representation Useful properties of the Bézier (or Bspline) representation: Convex hull property Convergence of coefficients via domain subdivision 𝐹 𝑥1 , … , 𝑥𝑛 = 𝑓1 , … , 𝑓𝑘 = … 𝑖1 𝑃𝑖1…𝑖𝑛 𝐵𝑖1 ,𝑚1 𝑥1 … 𝐵𝑖𝑛 ,𝑚𝑛 𝑥𝑛 . 𝑖𝑛 Center for Graphics and Geometric Computing, Technion 14 Recursive subdivision solvers The key idea! (extended next to 2DOF systems) 𝐷0 𝐷1,2 𝐷1,1 Positivity/Negativity test – purge? Topological test – terminate subdivision and trace numerically? Subdivide (tighter coefficients!) Solve and unite solutions Center for Graphics and Geometric Computing, Technion 15 Motivation – the 2D case surfaces in ℝ3 Bisector surfaces Sweeping (implicit) surfaces 2-DOF kinematic systems Medical iso-surfaces Minkowski sums Implicit Center for Graphics and Geometric Computing, Technion 16 Related Work I Fully determined systems (0-DOF) Lane-Reisenfeld 1981 Sherbrooke-Patrikalakis 1993 Elber-Kim 2001 Hanniel-Elber 2007 Wei-Feng-Lin 2011 Center for Graphics and Geometric Computing, Technion 17 Related Work II Under determined systems: 1-DOF Sederberg et-al 1988, 1996, 2007… Grandine-Klein 1997 Mourrain-Pavone 2008 Burr-Choi-Galehouse-Yap 2008 Bartoň-Elber-Hanniel 2011 Center for Graphics and Geometric Computing, Technion 18 Related Work III Under determined systems: 2-DOF Lorensen-Cline 1987 (“Marching cubes”) Plantinga-Vegter 2004 Bloomenthal et al 1988, 1995 Stadner-Hart 2005 Lin-Yap-Yu 2013 Center for Graphics and Geometric Computing, Technion 19 Related Work IV Previous results frequently used (0D): Cone Test (CT): root isolation in fully determined systems. (Figure) Elber and Kim, 2001: Geometric Constraint Solver using Multivariate Rational Spline Functions Center for Graphics and Geometric Computing, Technion 20 Related Work V Previous results frequently used (1D): No Loop Test (NLT) Single Component Test (SCT) (Figure) Barton, Hanniel & Elber, 2011: Topologically Guaranteed Solution of Underconstrained Polynomial Systems via No-loop and Single Component Tests. Center for Graphics and Geometric Computing, Technion 21 Our Contribution Topologically guaranteed subdivision termination criteria for ℝ𝑛 , 𝑛 > 3: Terminate when the unknown solution is homeomorphic to a closed disk. A triangulation method in ℝ𝑛 : Yielding a mesh with disc topology Constrained to a known boundary loop Vertices on the solution surface. Center for Graphics and Geometric Computing, Technion 22 Topologically Guaranteed Subdivision Termination Step 1: Solve the boundary problem, chain and count loops (subdivide if two or more) Step 2: Attempt to ensure existence of a 2D planar homeomorphic image of the unknown solution. If 2D planar homeomorphic image is possible: Empty boundary ⇒ empty solution Single loop ⇒ disc topology (terminate) Center for Graphics and Geometric Computing, Technion 23 The Single Loop Boundary Test Recursion invariant: boundary problem is solved (chain all curves on faces) More than one loop - subdivide One (or no) loop: Exactly one (or no) component with boundary Other components if exist: must be closed surfaces (resolved in next phase test) Center for Graphics and Geometric Computing, Technion 24 The Injective Projection Test Check sufficient conditions for projecting the (unknown, perhaps empty) solution on a 2D plane Such a mapping (if exists) is a homeomorphism onto its image. Center for Graphics and Geometric Computing, Technion 25 The Injective Projection Test If such an injective projection exists, the solution cannot contain any closed surfaces. If possible – the solution cannot contain closed surfaces! Previous ambiguity resolved: Zero loops – empty solution One loop – homeomorphic to a disk Center for Graphics and Geometric Computing, Technion 26 The actual IPT procedure 1 𝐹 𝑥 = 0 ∈ ℝ𝑛−2 𝑥𝑘 = 𝑢 ∈ [𝑎𝑘 , 𝑏𝑘 ] 𝑥𝑙 = 𝑣 ∈ [𝑎𝑙 , 𝑏𝑙 ] Theorem: An injective projection is possible on coordinates 𝑘, 𝑙 if Eq. (1) has at most a single solution for any 𝑢, 𝑣 ∈ 𝑎𝑘 , 𝑏𝑘 × [𝑎𝑙 , 𝑏𝑙 ]. Center for Graphics and Geometric Computing, Technion 27 The actual IPT procedure 1 𝐹 𝑥 = 0 ∈ ℝ𝑛−2 𝑥𝑘 = 𝑢 ∈ [𝑎𝑘 , 𝑏𝑘 ] 𝑥𝑙 = 𝑣 ∈ [𝑎𝑙 , 𝑏𝑙 ] The CT (Cone Test) for uniqueness of 𝑛 × 𝑛 systems uses gradients only Independent of 𝑢, 𝑣 The normal cones are the original 𝑛 − 2 along with 𝑒𝑘 and 𝑒𝑙 (the standard basis elements) Center for Graphics and Geometric Computing, Technion 28 The actual IPT procedure An example for 𝑛 = 3: Center for Graphics and Geometric Computing, Technion 29 Tessellation (The Numeric Step) The back-projection idea: 3. Vertices in ℝ𝑛 with the same 𝑘, 𝑙 coordinates: 2. The 𝑛 − 2 dim’ normal space where the back projection problem is solved 1. Vertices in ℝ2 : Center for Graphics and Geometric Computing, Technion 30 Test Results 𝑥 2 + 𝑦 2 + 𝑧 2 − 0.52 𝑧 − 0.75 = 0 Center for Graphics and Geometric Computing, Technion 31 Test Results 6 2 2 𝑖 2 2 𝑥 + 𝑦 + 𝑧 − 0.8 × 0.5 =0 𝑖=0 (Degree 14 with seven disjoint components) Center for Graphics and Geometric Computing, Technion 32 Test Results 𝑥 2 𝑦 2 + 𝑦 2 𝑧 2 + 𝑧 2 𝑥 2 + 0.32 𝑥𝑦𝑧 = 0 Subdivision tolerance: 0.05 (left) and 0.005 (middle and right). Polyline loops only (left and middle) and tessellation (right). 33 Center for Graphics and Geometric Computing, Technion Test Results Let 𝑆1 𝑢, 𝑣 , 𝑆2 𝑟, 𝑠 be parametric surfaces. The bisector surface is the locus of points for which the (locally) shortest distance to each of the input surfaces is equal. Two equations and four unknowns (solved in 𝑢, 𝑣, 𝑟, 𝑠 space and mapped to 𝑥, 𝑦, 𝑧 : 𝑆1 𝑢, 𝑣 − 𝑆2 𝑟, 𝑠 × 𝑛1 𝑢, 𝑣 , 𝑛2 𝑟, 𝑠 = 0 𝑛2 𝑟, 𝑠 2 𝑆1 𝑢, 𝑣 − 𝑆2 𝑟, 𝑠 , 𝑛1 𝑢, 𝑣 2 − 𝑛1 𝑢, 𝑣 2 𝑆1 𝑢, 𝑣 − 𝑆2 𝑟, 𝑠 , 𝑛2 𝑟, 𝑠 2 = 0 Center for Graphics and Geometric Computing, Technion 34 Test Results Bisector examples: Center for Graphics and Geometric Computing, Technion 35 Test Results Bisector examples (contd’) Center for Graphics and Geometric Computing, Technion 36 Test Results Sweep surfaces: A section curve 𝐶 𝑢 is swept along a trajectory curve 𝑇 𝑣 The surface is defined by: 𝑆 𝑢, 𝑣 = 𝑇 𝑣 + 𝑀 𝑣 𝐶 𝑢 Here consider 𝑀 𝑣 = 𝐼 but can be adopted to any matrix (used for rotating and non-uniform scaling – a different problem) Center for Graphics and Geometric Computing, Technion 37 Test Results Problem - what if the section curve is given implicitly: 𝑓1 𝑥, 𝑦, 𝑧 = 0 𝑓2 𝑥, 𝑦, 𝑧 = 0 Solve (in 𝑥, 𝑦, 𝑧, 𝑣 space and project): 𝑓1 𝑥, 𝑦, 𝑧 − 𝑇 𝑣 =0 𝑓2 𝑥, 𝑦, 𝑧 − 𝑇 𝑣 =0 Center for Graphics and Geometric Computing, Technion 38 Test Results Implicit sweep examples: planar sections of a torus (“Oval of Casini”) along cubic & quadratic trajectories): Center for Graphics and Geometric Computing, Technion 39 Test Results Implicit sweep examples (sphere-torus intersection curve swept along a cubic trajectory): Center for Graphics and Geometric Computing, Technion 40 Test Results Minkowski sums: 𝑆1 ⨁𝑆2 = 𝑎 + 𝑏: 𝑎 ∈ 𝑆1 , 𝑏 ∈ 𝑆2 Envelope points satisfy the matched normal vectors constraint: 𝜕𝑆2 𝑟,𝑠 𝑁1 𝑢, 𝑣 , 𝜕𝑟 =0 𝑁1 𝑢, 𝑣 𝜕𝑆2 𝑟,𝑠 , 𝜕𝑠 =0 Where: 𝑁1 = 𝜕𝑆1 𝜕𝑆1 × 𝜕𝑢 𝜕𝑣 Center for Graphics and Geometric Computing, Technion 41 Test results Center for Graphics and Geometric Computing, Technion 42 Test results Center for Graphics and Geometric Computing, Technion 43 Test results Center for Graphics and Geometric Computing, Technion 44 Conclusion and Future Work Topological guarantee: Can be achieved assuming regularity Up to subdivision tolerance Singular sub-domains can be contained Framework can theoretically be extended to higher dimensions with the same topological arguments Center for Graphics and Geometric Computing, Technion 45 Thank you The research leading to this research has received partial funding under the European Union's Seventh Framework Programme FP7/2007-2013/ under REA grant agreement PIAPGA-2011-286426 - the GEMS project http://www.geometrie.tuwien.ac.at/ig/gems 46
© Copyright 2026 Paperzz