Hessian measures of semi-convex functions and applications to support measures of convex bodies

This paper originates from the investigation of support measures of convex bodies (sets of positive reach), which form a central subject in convex geometry and also represent an important tool in related fields. We show that these measures are absolutely continuous with respect to Hausdorff measures of appropriate dimensions, and we determine the Radon-Nikodym derivatives explicitly on sets of σ-finite Hausdorff measure. The results which we obtain in the setting of the theory of convex bodies (sets of positive reach) are achieved as applications of various new results on Hessian measures of convex (semi-convex) functions. Among these are a Crofton formula, results on the absolute continuity of Hessian measures, and a duality theorem which relates the Hessian measures of a convex function to those of the conjugate function. In particular, it turns out that curvature and surface area measures of a convex body K are the Hessian measures of special functions, namely the distance function and the support function of K. Received: 15 July 1999     

Vertex Cuts

Given a connected graph, in many cases it is possible to construct a structure tree that provides information about the ends of the graph or its connectivity. For example Stallings' theorem on the structure of groups with more than one end can be proved by analyzing the action of the group on a structure tree and Tutte used a structure tree to investigate finite 2‐connected graphs, that are not 3‐connected. Most of these structure tree theories have been based on edge cuts, which are components of the graph obtained by removing finitely many edges. A new axiomatic theory is described here using vertex cuts, components of the graph obtained by removing finitely many vertices. This generalizes Tutte's decomposition of 2‐connected graphs to k‐connected graphs for any k, in finite and infinite graphs. The theory can be applied to nonlocally finite graphs with more than one vertex end, i.e. ends that can be separated by removing a finite number of vertices. This gives a decomposition for a group acting on such a graph, generalizing Stallings' theorem. Further applications include the classification of distance transitive graphs and k‐CS‐transitive graphs.     

On the exterior Dirichlet problem for Hessian quotient equations

In this paper, we establish the existence and uniqueness theorem for solutions of the exterior Dirichlet problem for Hessian quotient equations with prescribed asymptotic behavior at infinity. This extends the previous related results on the Monge–Ampère equations and on the Hessian equations, and rearranges them in a systematic way. Based on the Perron's method, the main ingredient of this paper is to construct some appropriate subsolutions of the Hessian quotient equation, which is realized by introducing some new quantities about the elementary symmetric polynomials and using them to analyze the corresponding ordinary differential equation related to the generalized radially symmetric subsolutions of the original equation.     

Disjunctive Optimization: Critical Point Theory

In this paper, we introduce the concepts of (nondegenerate) stationary points and stationary index for disjunctive optimization problems. Two basic theorems from Morse theory, which imply the validity of the (standard) Morse relations, are proved. The first one is a deformation theorem which applies outside the stationary point set. The second one is a cell-attachment theorem which applies at nondegenerate stationary points. The dimension of the cell to be attached equals the stationary index. Here, the stationary index depends on both the restricted Hessian of the Lagrangian and the set of active inequality constraints. In standard optimization problems, the latter contribution vanishes.     

Lokal-Global-Prinzipien für die Brauergruppe

This article looks at Local-Global-Principles for the Brauer group, modeled after the celebrated theorem of Hasse-Brauer-Noether for the Brauer group of a number field. Pop introduced a property for fields, which holds especially for real closed andp-adically closed fields and yields a Local-Global-Principle for function fields of one variable over such fields. Then he used model theoretical means to generalize these results to arbitrary extensions of transcendental degree one over real closed andp-adically closed fields. This paper achieves this in a more elementary manner. Another result are examples of fields where the Local-Global-Principle is violated.

---

---

This article was processed by the author using the IATEX style filecljour1 from Springer-Verlag.     

Computation of sparse low degree interpolating polynomials and their application to derivative-free optimization

Interpolation-based trust-region methods are an important class of algorithms for Derivative-Free Optimization which rely on locally approximating an objective function by quadratic polynomial interpolation models, frequently built from less points than there are basis components. Often, in practical applications, the contribution of the problem variables to the objective function is such that many pairwise correlations between variables are negligible, implying, in the smooth case, a sparse structure in the Hessian matrix. To be able to exploit Hessian sparsity, existing optimization approaches require the knowledge of the sparsity structure. The goal of this paper is to develop and analyze a method where the sparse models are constructed automatically. The sparse recovery theory developed recently in the field of compressed sensing characterizes conditions under which a sparse vector can be accurately recovered from few random measurements. Such a recovery is achieved by minimizing the ℓ ₁-norm of a vector subject to the measurements constraints. We suggest an approach for building sparse quadratic polynomial interpolation models by minimizing the ℓ ₁-norm of the entries of the model Hessian subject to the interpolation conditions. We show that this procedure recovers accurate models when the function Hessian is sparse, using relatively few randomly selected sample points. Motivated by this result, we developed a practical interpolation-based trust-region method using deterministic sample sets and minimum ℓ ₁-norm quadratic models. Our computational results show that the new approach exhibits a promising numerical performance both in the general case and in the sparse one.     

An extension of Alexandrov?s theorem on second derivatives of convex functions

If f is a function of n variables that is locally L¹ approximable by a sequence of smooth functions satisfying local L¹ bounds on the determinants of the minors of the Hessian, then f admits a second order Taylor expansion almost everywhere. This extends a classical theorem of A.D. Alexandrov, covering the special case in which f is locally convex.     

Fixed Points for Directional Multi-Valued <Emphasis Type="Italic">k</Emphasis>(·)-Contractions

A fixed point theorem for directional multi-valued k(·)-contractions acting m a complete metric space is established which extends similar results both for k(·)-contractions and directional contractions. Such theorem enables to obtain fixed points theorems for the former class of set-valued maps from those valid for the latter one without metrical convexity or proximinality assumptions, thereby contributing to unify the current setting of the theory. Connections with several recent advances on this subject are also examinated.Mathematics Subject Classifications (2000): 47H10, 54H25     

Degenerate values for broyden methods

A Broyden method fails when the free parameter takes on a degenerate value; one such value is well known, but this paper shows that many others exist in general. These values have practical significance if the initial Hessian approximation is indefinite. The BFS formula is special in that it avoids these degenerate values. Properties about how the new degenerate values behave and their relationship to the well-known degenerate value are described. A new result is used about a reduced inverse Hessian method, which is equivalent to a Broyden method but is parameter free and provides a simple proof of Dixon's theorem.     

Coupling of FEM and BEM in Shape Optimization

In the present paper we consider the numerical solution of shape optimization problems which arise from shape functionals of integral type over a compact region of the unknown shape, especially L ²-tracking type functionals. The underlying state equation is assumed to satisfy a Poisson equation with Dirichlet boundary conditions. We proof that the shape Hessian is not strictly H ^1/2-coercive at the optimal domain which implies ill-posedness of the optimization problem under consideration. Since the adjoint state depends directly on the state, we propose a coupling of finite element methods (FEM) and boundary element methods (BEM) to realize an efficient first order shape optimization algorithm. FEM is applied in the compact region while the rest is treated by BEM. The coupling of FEM and BEM essentially retains all the structural and computational advantages of treating the free boundary by boundary integral equations.This research has been carried out when the second author stayed at the Department of Mathematics, Utrecht University, The Netherlands, supported by the EU-IHP project Nonlinear Approximation and Adaptivity: Breaking Complexity in Numerical Modelling and Data Representation     

Finite reduction and Morse index estimates for mechanical systems

A simple version of exact finite dimensional reduction for the variational setting of mechanical systems is presented. It is worked out by means of a thorough global version of the implicit function theorem for monotone operators. Moreover, the Hessian of the reduced function preserves all the relevant information of the original one, by Schur’s complement, which spontaneously appears in this context. Finally, the results are straightforwardly extended to the case of a Dirichlet problem on a bounded domain.     

Compact gradient tracking in shape optimization

In the present paper we consider the minimization of gradient tracking functionals defined on a compact and fixed subdomain of the domain of interest. The underlying state is assumed to satisfy a Poisson equation with Dirichlet boundary conditions. We proof that, in contrast to the situation of gradient tracking on the whole domain, the shape Hessian is not strictly H ^1/2-coercive at the optimal domain which implies ill-posedness of the shape problem under consideration. Shape functional and gradient require only knowledge of the Cauchy data of the state and its adjoint on the boundaries of the domain and the subdomain. These data can be computed by means of boundary integral equations when reformulating the underlying differential equations as transmission problems. Thanks to fast boundary element techniques, we derive an efficient algorithm to solve the problem under consideration.     

Preconditioning the Pressure Tracking in Fluid Dynamics by Shape Hessian Information

Potential flow pressure matching is a classical inverse design aerodynamic problem. The resulting loss of regularity during the optimization poses challenges for shape optimization with normal perturbation of the surface mesh nodes. Smoothness is not enforced by the parameterization but by a proper choice of the scalar product based on the shape Hessian, which is derived in local coordinates for starshaped domains. Significant parts of the Hessian are identified and combined with an aerodynamic panel solver. The resulting shape Hessian preconditioner is shown to lead to superior convergence properties of the resulting optimization method. Additionally, preconditioning gives the potential for level independent convergence.     

A Generalization of the Classical <Emphasis Type="Italic">α</Emphasis>BB Convex Underestimation via Diagonal and Nondiagonal Quadratic Terms

The classical αBB method determines univariate quadratic perturbations that convexify twice continuously differentiable functions. This paper generalizes αBB to additionally consider nondiagonal elements in the perturbation Hessian matrix. These correspond to bilinear terms in the underestimators, where previously all nonlinear terms were separable quadratic terms. An interval extension of Gerschgorin’s circle theorem guarantees convexity of the underestimator. It is shown that underestimation parameters which are optimal, in the sense that the maximal underestimation error is minimized, can be obtained by solving a linear optimization model.     

Shape optimisation analysing the inner structure of sensitivity matrices

This contribution deals with sensitivity analysis in nodal based shape optimisation. Sensitivity analysis is one of the most important parts of a structural optimisation algorithm. The efficiency of the algorithm mainly depends on the obtained sensitivity information. The pseudo load and sensitivity matrices which appear in sensitivity analysis are commonly used to derive and to calculate the gradients and the Hessian matrices of objective functions and of constraints. The aim of this contribution is to show that these matrices contain additional useful information which is not used in structural optimisation until now. We demonstrate the opportunities and capabilities of the new information which are obtained by singular value decomposition (SVD) of the pseudo load and sensitivity matrices and by eigenvalue decomposition of the Hessian matrix. Furthermore, we avoid jagged boundaries in shape optimisation by applying a density filtering technique well-known in topology optimisation. Numerical examples illustrate the advocated theoretical concept. (© 2011 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)     

Shape Hessian for generalized Oseen flow by differentiability of a minimax: A Lagrangian approach

The goal of this paper is to compute the shape Hessian for a generalized Oseen problem with nonhomogeneous Dirichlet boundary condition by the velocity method. The incompressibility will be treated by penalty approach. The structure of the shape gradient and shape Hessian with respect to the shape of the variable domain for a given cost functional are established by an application of the Lagrangian method with function space embedding technique. This work was supported by the National Natural Science Fund of China (No. 10371096) for ZM Gao and YC Ma.     

A Chevalley theorem for difference equations

By a theorem of Chevalley the image of a morphism of varieties is a constructible set. The algebraic version of this fact is usually stated as a result on “extension of specializations” or “lifting of prime ideals”. We present a difference analog of this theorem. The approach is based on the philosophy that occasionally one needs to pass to higher powers of σ, where σ is the endomorphism defining the difference structure. In other words, we consider difference pseudo fields (which are finite direct products of fields) rather than difference fields. We also prove a result on compatibility of pseudo fields and present some applications of the main theorem, e.g. constrained extension and uniqueness of differential Picard–Vessiot rings with a difference parameter.     

The construction of isolated reducibleSU(2)-connections overS 2×S 2

In this paper, we construct a double indexing family of reducibleSU(2)-connctions overS ² ×S ² in a geometric way. By separation of variables, we compute the spectrum of the Hessian of the Yang-Mills functional at these connections. Using the implicit function theorem, we prove that these connections are isolated non-minimal solutions to the Yang-Mills equations onS ² ×S ². This is part of my doctoral thesis at Peking University.