Computer Science

While there has been some discussion on how Symbolic Computation could be used for AI there is little literature on applications in the other direction. However, recent results for quantifier elimination suggest that, given enough example problems, there is scope for machine learning tools like Support Vector Machines to improve the performance of Computer Algebra Systems. We survey the authors own work and similar applications for other mathematical software. It may seem that the inherently probabilistic nature of machine learning tools would invalidate the exact results prized by mathematical software. However, algorithms and implementations often come with a range of choices which have no effect on the mathematical correctness of the end result but a great effect on the resources required to find it, and thus here, machine learning can have a significant impact.

In this paper we present the first-ever computer formalization of the theory of Gröbner bases in reduction rings, which is an important theory in computational commutative algebra, in Theorema. Not only the formalization, but also the formal verification of all results has already been fully completed by now; this, in particular, includes the generic implementation and correctness proof of Buchberger's algorithm in reduction rings. Thanks to the seamless integration of proving and computing in Theorema, this implementation can now be used to compute Gröbner bases in various different domains directly within the system. Moreover, a substantial part of our formalization is made up solely by "elementary theories" such as sets, numbers and tuples that are themselves independent of reduction rings and may therefore be used as the foundations of future theory explorations in Theorema. In addition, we also report on two general-purpose Theorema tools we developed for an efficient and convenient exploration of mathematical theories: an interactive proving strategy and a "theory analyzer" that already proved extremely useful when creating large structured knowledge bases.

We consider exact matrix decomposition by Gauss-Bareiss reduction. We investigate two aspects of the process: common row and column factors and the influence of pivoting strategies. We identify two types of common factors: systematic and statistical. Systematic factors depend on the process, while statistical factors depend on the specific data. We show that existing fraction-free QR (Gram-Schmidt) algorithms create a common factor in the last column of Q. We relate the existence of row factors in LU decomposition to factors appearing in the Smith normal form of the matrix. For statistical factors, we identify mechanisms and give estimates of the frequency. Our conclusions are tested by experimental data. For pivoting strategies, we compare the sizes of output factors obtained by different strategies. We also comment on timing differences.

A matrix is called Bohemian if its entries are sampled from a finite set of integers. We determine the maximum absolute determinant of upper Hessenberg Bohemian Matrices for which the subdiagonal entries are fixed to be 1 and upper triangular entries are sampled from {0,1,⋯,n} , extending previous results for n=1 and n=2 and proving a recent conjecture of Fasi & Negri Porzio [8]. Furthermore, we generalize the problem to non-integer-valued entries.

A paper by Bruno Salvy and the author introduced measured multiseries and gave an algorithm to compute these for a large class of elementary functions, modulo a zero-equivalence method for constants. This gave a theoretical background for the implementation that Salvy was developing at that time. The main result of the present article is an algorithm to calculate measured multiseries for integrals of functions of the form h*sin G, where h and G belong to a Hardy field. The process can reiterated with the resulting algebra, and also applied to solutions of a second order differential equation of a particular form.

Recently, it has been shown constructively how a finite set of hypergeometric products, multibasic hypergeometric products or their mixed versions can be modeled properly in the setting of formal difference rings. Here special emphasis is put on robust constructions: whenever further products have to be considered, one can reuse --up to some mild modifications-- the already existing difference ring. In this article we relax this robustness criteria and seek for another form of optimality. We will elaborate a general framework to represent a finite set of products in a formal difference ring where the number of transcendental product generators is minimal. As a bonus we are able to describe explicitly all relations among the given input products.

In this paper we introduce the notion of an Open Non-uniform Cylindrical Algebraic Decomposition (NuCAD), and present an efficient model-based algorithm for constructing an Open NuCAD from an input formula. A NuCAD is a generalization of Cylindrical Algebraic Decomposition (CAD) as defined by Collins in his seminal work from the early 1970s, and as extended in concepts like Hong's partial CAD. A NuCAD, like a CAD, is a decomposition of n-dimensional real space into cylindrical cells. But unlike a CAD, the cells in a NuCAD need not be arranged cylindrically. It is in this sense that NuCADs are not uniformly cylindrical. However, NuCADs--- like CADs --- carry a tree-like structure that relates different cells. It is a very different tree but, as with the CAD tree structure, it allows some operations to be performed efficiently, for example locating the containing cell for an arbitrary input point.

We present a new modular proof method of termination for second-order computation, and report its implementation SOL. The proof method is useful for proving termination of higher-order foundational calculi. To establish the method, we use a variation of semantic labelling translation and Blanqui's General Schema: a syntactic criterion of strong normalisation. As an application, we apply this method to show termination of a variant of call-by-push-value calculus with algebraic effects and effect handlers. We also show that our tool SOL is effective to solve higher-order termination problems.

We describe, study, and experiment with an algorithm for finding all solutions of systems of polynomial equations using homotopy continuation and monodromy. This algorithm follows a framework developed in previous work and can operate in the presence of a large number of failures of the homotopy continuation subroutine. We give special attention to parallelization and probabilistic analysis of a model adapted to parallelization and failures. Apart from theoretical results, we developed a simulator that allows us to run a large number of experiments without recomputing the outcomes of the continuation subroutine.

In recent years, the approximate basis computation of vanishing ideals has been studied extensively and adopted both in computer algebra and data-driven applications such as machine learning. However, symbolic computation and the dependency on monomial ordering remain as essential gaps between the two above-mentioned fields. In this paper, we propose the first efficient monomial-agnostic approximate basis computation of vanishing ideals, where polynomials are manipulated without any information of monomials; this can be implemented in a fully numerical manner and is thus desirable for data-driven applications. In particular, we propose gradient normalization, which achieves not only the first efficient and monomial-agnostic normalization of polynomials but also provides significant advantages such as consistency in translation and scaling of data points, which cannot be realized by existing basis computation algorithms. During the basis computation, the gradients of polynomials at the given points are proven to be efficiently and exactly obtained without performing differentiation. By exploiting the gradient information, we further propose a basis reduction method to remove redundant polynomials in a monomial-agnostic manner. Finally, we also propose a regularization method using gradients to avoiding overfitting of the basis for the given perturbed points.