Polytope-based computation of polynomial ranges

Polynomial ranges are commonly used for numerically solving polynomial systems with interval Newton solvers. Often ranges are computed using the convex hull property of the tensorial Bernstein basis, which is exponential size in the number n of variables. In this paper, we consider methods to compute tight bounds for polynomials in n variables by solving two linear programming problems over a polytope. We formulate a polytope defined as the convex hull of the coefficients with respect to the tensorial Bernstein basis, and we formulate several polytopes based on the Bernstein polynomials of the domain. These Bernstein polytopes can be defined by a polynomial number of halfspaces. We give the number of vertices, the number of hyperfaces, and the volume of each polytope for n=1,2,3,4, and we compare the computed range widths for random n-variate polynomials for n?10. The Bernstein polytope of polynomial size gives only marginally worse range bounds compared to the range bounds obtained with the tensorial Bernstein basis of exponential size.     

Solving nonlinear polynomial systems in the barycentric Bernstein basis

We present a method for solving arbitrary systems of N nonlinear polynomials in n variables over an n-dimensional simplicial domain based on polynomial representation in the barycentric Bernstein basis and subdivision. The roots are approximated to arbitrary precision by iteratively constructing a series of smaller bounding simplices. We use geometric subdivision to isolate multiple roots within a simplex. An algorithm implementing this method in rounded interval arithmetic is described and analyzed. We find that when the total order of polynomials is close to the maximum order of each variable, an iteration of this solver algorithm is asymptotically more efficient than the corresponding step in a similar algorithm which relies on polynomial representation in the tensor product Bernstein basis. We also discuss various implementation issues and identify topics for further study.     

SqFreeEVAL: An (almost) optimal real-root isolation algorithm

Let f be a univariate polynomial with real coefficients, f∈R[X]. Subdivision algorithms based on algebraic techniques (e.g., Sturm or Descartes methods) are widely used for isolating the real roots of f in a given interval. In this paper, we consider a simple subdivision algorithm whose primitives are purely numerical (e.g., function evaluation). The complexity of this algorithm is adaptive because the algorithm makes decisions based on local data. The complexity analysis of adaptive algorithms (and this algorithm in particular) is a new challenge for computer science. In this paper, we compute the size of the subdivision tree for the SqFreeEVAL algorithm.The SqFreeEVAL algorithm is an evaluation-based numerical algorithm which is well-known in several communities. The algorithm itself is simple, but prior attempts to compute its complexity have proven to be quite technical and have yielded sub-optimal results. Our main result is a simple O(d(L+lnd)) bound on the size of the subdivision tree for the SqFreeEVAL algorithm on the benchmark problem of isolating all real roots of an integer polynomial f of degree d and whose coefficients can be written with at most L bits.Our proof uses two amortization-based techniques: first, we use the algebraic amortization technique of the standard Mahler-Davenport root bounds to interpret the integral in terms of d and L. Second, we use a continuous amortization technique based on an integral to bound the size of the subdivision tree. This paper is the first to use the novel analysis technique of continuous amortization to derive state of the art complexity bounds.     

Formalization of Bernstein Polynomials and Applications to Global Optimization

This paper presents a formalization in higher-order logic of a practical representation of multivariate Bernstein polynomials. Using this representation, an algorithm for finding lower and upper bounds of the minimum and maximum values of a polynomial has been formalized and verified correct in the Prototype Verification System (PVS). The algorithm is used in the definition of proof strategies for formally and automatically solving polynomial global optimization problems.     

A New Subdivision Algorithm for the Bernstein Polynomial Approach to Global Optimization

In this paper,an improved algorithm is proposed for unconstrained global optimization to tackle non-convex nonlinear multivariate polynomial programming problems.The proposed algorithm is based on the Bernstein polynomial approach.Novel features of the proposed algorithm are that it uses a new rule for the selection of the subdivision point,modified rules for the selection of the subdivision direction,and a new acceleration device to avoid some unnecessary subdivisions.The performance of the proposed algorithm is numerically tested on a collection of 16 test problems.The results of the tests show the proposed algorithm to be superior to the existing Bernstein algorithm in terms of the chosen performance metrics.     

A new approach to the symbolic factorization of multivariate polynomials

A heuristic factorization scheme that uses learning and other heuristic programming techniques to improve the efficiency of determining the symbolic factorization of multivariate polynomials with integer coefficients and an arbitrary number of variables and terms is described. The learning program, POLYFACT, in which the factorization scheme is implemented is also described. POLYFACT uses learning through the dynamic construction and manipulation of first-order predicate calculus heuristics to reduce the amount of searching for the irreducible factors of a polynomial.Tables containing the results of factoring randomly generated multivariate polynomials are presented: (1) to demonstrate that learning does improve considerably the efficiency of factoring polynomials, and (2) to show that POLYFACT does learn from previous experience.The factorization times of polynomials factored by both the scheme implemented in POLYFACT and Wang's implementation of Berlekamp's algorithm are given. The two algorithms are compared, and two situations where POLYFACT'S algorithm can be used to improve the efficiency of Wang's algorithm are discussed.     

Computation of the Bernstein Coefficients on Subdivided Triangles

We present a procedure for computing the coefficients of the expansion of a bivariate polynomial into Bernstein polynomials over subtriangles. These triangles are generated by partitioning the unit triangle. The coefficients are computed directly from the coefficients on the subdivided triangle from the preceding subdivision level. This allows a recursive computation of the coefficients and facilitates the economical computation of bounds for the range of a bivariate polynomial over a given triangle.     

Subdivision methods for solving polynomial equations

This paper presents a new algorithm for solving a system of polynomials, in a domain of Rⁿ${\mathbb{R}}^n$. It can be seen as an improvement of the Interval Projected Polyhedron algorithm proposed by Sherbrooke and Patrikalakis [Sherbrooke, E.C., Patrikalakis, N.M., 1993. Computation of the solutions of nonlinear polynomial systems. Comput. Aided Geom. Design 10 (5), 379–405]. It uses a powerful reduction strategy based on univariate root finder using Bernstein basis representation and Descarte’s rule  . We analyse the behavior of the method, from a theoretical point of view, shows that for simple roots, it has a local quadratic convergence speed and gives new bounds for the complexity of approximating real roots in a box of Rⁿ${\mathbb{R}}^n$. The improvement of our approach, compared with classical subdivision methods, is illustrated on geometric modeling applications such as computing intersection points of implicit curves, self-intersection points of rational curves, and on the classical parallel robot benchmark problem.     

Hyper-arc consistency of polynomial constraints over finite domains using the modified Bernstein form

This paper describes an algorithm to enforce hyper-arc consistency of polynomial constraints defined over finite domains. First, the paper describes the language of so called polynomial constraints over finite domains, and it introduces a canonical form for such constraints. Then, the canonical form is used to transform the problem of testing the satisfiability of a constraint in a box into the problem of studying the sign of a related polynomial function in the same box, a problem which is effectively solved by using the modified Bernstein form of polynomials. The modified Bernstein form of polynomials is briefly discussed, and the proposed hyper-arc consistency algorithm is finally detailed. The proposed algorithm is a subdivision procedure which, starting from an initial approximation of the domains of variables, removes values from domains to enforce hyper-arc consistency.     

Interrogation of differential geometry properties for design and manufacture

This paper describes a new robust method to decompose a free-form surface into regions with specific range of curvature and provide important tools for surface analysis, tool-path generation, and tool-size selection for numerically controlled machining, tessellation of trimmed patches for surface interrogation and finite-element meshing, and fairing of free-form surfaces. The key element in these techniques is the computation ofall real roots within a finite box of systems of nonlinear equations involving polynomials and square roots of polynomials. The free-form surfaces are bivariate polynomial functions, but the analytical expressions of their principal curvatures involve polynomials and square roots of polynomials. Key components are the reduction of the problems into solutions of systems of polynomial equations of higher dimensionality through the introduction ofauxiliary variables and the use ofrounded interval arithmetic in the context of Bernstein subdivision to enhance the robustness of floating-point implementation. Examples are given that illustrate our techniques.     

Algebraic manipulation in the Bernstein form made simple via convolutions

Traditional methods for algebraic manipulation of polynomials in Bernstein form try to obtain an explicit formula for each coefficient of the result of a given procedure, such us multiplication, arbitrarily high degree elevation, composition, or differentiation of rational functions. Whereas this strategy often furnishes involved expressions, these operations become trivial in terms of convolutions between coefficient lists if we employ the scaled Bernstein basis, which does not include binomial coefficients. We also carry over this scheme from the univariate case to multivariate polynomials, Bézier simplexes of any dimension and B-bases of other functional spaces. Examples of applications in geometry processing are provided, such as conversions between the triangular and tensor-product Bézier forms.     

Algorithms for unconstrained global optimization of nonlinear (polynomial) programming problems: The single and multi-segment polynomial B-spline approach

We investigate the use of the polynomial B-spline form for unconstrained global optimization of multivariate polynomial nonlinear programming problems. We use the B-spline form for higher order approximation of multivariate polynomials. We first propose a basic algorithm for global optimization that uses several accelerating algorithms such as cut-off test and monotonicity test. We then propose an improved algorithm consisting of several additional ingredients, such as a new subdivision point selection rule and a modified subdivision direction selection rule. The performances of the proposed basic and improved algorithms are tested and compared on a set of 14 test problems under two test conditions. The results of the tests show the superiority of the improved algorithm with multi-segment B-spline over that of the single segment B-spline, in terms of the chosen performance metrics. We also compare the quality of the set of all global minimizers found using the proposed algorithms (basic & improved) with those using well-known solvers BARON and Gloptipoly, on a smaller set of four test problems. The problems in the latter set have multiple global minimizers. The results show the superiority of the proposed algorithms, in that they are able to capture all the global minimizers, whereas Gloptipoly and BARON fail to do so in some of the test problems.     

Local hybrid approximation for scattered data fitting with bivariate splines

We suggest a local hybrid approximation scheme based on polynomials and radial basis functions, and use it to improve the scattered data fitting algorithm of (Davydov, O., Zeilfelder, F., 2004. Scattered data fitting by direct extension of local polynomials to bivariate splines. Adv. Comp. Math. 21, 223–271). Similar to that algorithm, the new method has linear computational complexity and is therefore suitable for large real world data. Numerical examples suggest that it can produce high quality artifact-free approximations that are more accurate than those given by the original method where pure polynomial local approximations are used.     

A note on the Bernstein algorithm for bounds for interval polynomials

In computing the range of values of a polynomial over an intervala≤x≤b one may use polynomials of the form $$\left( {\begin{array}{*{20}c} k \\ j \\ \end{array} } \right)\left( {x - a} \right)^j \left( {b - x} \right)^{k - j} $$ called Bernstein polynomials of the degreek. An arbitrary polynomial of degreen may be written as a linear combination of Bernstein polynomials of degreek≥n. The coefficients of this linear combination furnish an upper/lower bound for the range of the polynomial. In this paper a finite differencelike scheme is investigated for this computation. The scheme is then generalized to interval polynomials.     

An identity for multivariate Bernstein polynomials

We prove an identity for multivariate Bernstein polynomials on simplices, which may be considered a pointwise orthogonality relation. Its integrated version provides a new representation for the polynomial dual basis of Bernstein polynomials. An identity for the reproducing kernel is used to define quasi-interpolants of arbitrary order.     

Optimal computation of the Bernstein algorithm for the bound of an interval polynomial

If a polynomial is expanded in terms of Bernstein polynomial over an interval then the coefficients of the expansion may be used to provide upper and lower bounds for the value of the polynomial over the interval. When applying this method to interval polynomials straightforwardly, the coefficients of the expansion are computed with an increase in width due to dependency intervals. In this paper we show that if the computations are rearranged suitably then the Bernstein coefficients can be computed with no increase in width due to dependency intervals.     

Computation of stationary points of distance functions

This paper presents an algorithm for computation of the stationary points of the squared distance functions between two point sets. One point set consists of a single space point, a rational B-spline curve, or a rational B-spline surface. The problem is reformulated in terms of solution of n polynomial equations with n variables expressed in the tensor product Bernstein basis. The solution method is based on subdivision relying on the convex hull property of the n-dimensional Bernstein basis and minimization techniques. We also cover classification of the stationary points of these distance functions, and include a method for tracing curves of stationary points in case the solution set is not zerodimensional. The distance computation problem is shown to be equivalent to the geometrically intuitive problem of computing collinear normal points. Finally, examples illustrate the applicability of the method     

Bounds for an interval polynomial

We discuss the evaluation of the range of values of an interval polynomial over an interval. Several algorithms are proposed and tested on numerical examples. The algorithms are based on ideas by Cargo and Shiska [2] and Rivlin [4]. The one basic algorithm uses Bernstein polynomials. It is shown to converge to the exact bounds and it has furthermore the property that if the maximum respectively the minimum of the polynomials occurs at an endpoint of the interval then the bound is exact. This is a useful property in routines for polynomials zeros. The other basic method is based on the meanvalue theorem and it has the advantage that the degree of approximation required for a certain apriori tolerance is smaller than the degree required in the Bernstein polynomial case. The mean value method is shown to be at least quadratically convergent and the Bernstein polynomial method is shown to be at least linearly convergent.     

Robustness analysis of polynomials with polynomial parameterdependency using Bernstein expansion

This paper considers the robust stability verification of polynomials with coefficients depending polynomially on parameters varying in given intervals. Two algorithms are presented, both rely on the expansion of a multivariate polynomial into Bernstein polynomials. The first one is an improvement of the so-called Bernstein algorithm and checks the Hurwitz determinant for positivity over the parameter set. The second one is based on the analysis of the value set of the family of polynomials and profits from the convex hull property of the Bernstein polynomials. Numerical results to real-world control problems are presented showing the efficiency of both algorithms     

Learning Read-Constant Polynomials of Constant Degree Modulo Composites

Boolean functions that have constant degree polynomial representation over a fixed finite ring form a natural and strict subclass of the complexity class ACC⁰. They are also precisely the functions computable efficiently by programs over fixed and finite nilpotent groups. This class is not known to be learnable in any reasonable learning model. In this paper, we provide a deterministic polynomial time algorithm for learning Boolean functions represented by polynomials of constant degree over arbitrary finite rings from membership queries, with the additional constraint that each variable in the target polynomial appears in a constant number of monomials. Our algorithm extends to superconstant but low degree polynomials and still runs in quasipolynomial time.