### Refine

#### Document Type

- ZIB-Report (6)
- Article (2)

#### Keywords

- Automatic Differentiation (3)
- Lipschitz Continuity (2)
- Nonsmooth (2)
- Piecewise Linearization (2)
- Trapezoidal Rule (2)
- ADOL-C (1)
- Adjoint Equations (1)
- Algorithmic Differentiation (1)
- Automatic differentiation (1)
- Chain Rule (1)

#### Institute

Most nonlinear computations require the evaluation of first and higher derivatives of vector functions defined by computer programs. It is shown here how vectors of such partial derivatives can be obtained automatically and efficiently if the computer language allows overloading (as is or will be the case for C++, PASCAL-XSC, FORTRAN90, and other modern languages). Here, overloading facilitates the extension of arithmetic operations and univariate functions from real or complex arguments to truncated Taylor-series (or other user- defined types), and it generates instructions for the subsequent evaluation of adjoints. Similar effects can be achieved by precompilation of FORTRAN77 programs. The proposed differentiation algorithm yields gradients and higher derivatives at a small multiple of the run-time and RAM requirement of the original function evaluation program. {\bf Keywords:} Automatic Differentiation, Chain Rule, Overloading, Taylor Coefficients, Gradients, Hessians, Reverse Accumulation, Adjoint Equations. {\bf Abbreviated title:} Automatic Differentiation by Overloading.

It is shown how piecewise differentiable functions F : IR^n -> IR^m that are defined by evaluation programmes can be approximated locally by a piecewise linear model based on a pair of sample points \check x and \hat x. We show that the discrepancy between function and model at any point x is of the bilinear order O(||x - \check x||*||x - \hat x||). As an application of the piecewise linearization procedure we devise a generalized Newton's method based on successive piecewise linearization and prove for it sufficient conditions for convergence and convergence rates equalling those of semismooth Newton. We conclude with the derivation of formulas for the numerically stable implementation of the aforedeveloped piecewise linearization methods.

It is shown how piecewise differentiable functions \(F: R^n → R^m\) that are defined by evaluation programs can be approximated locally by a piecewise linear model based on a pair of sample points x̌ and x̂. We show that the discrepancy between function and model at any point x is of the bilinear order O(||x − x̌|| ||x − x̂||). This is a little surprising since x ∈ R^n may vary over the whole Euclidean space, and we utilize only two function samples F̌ = F(x̌) and F̂ = F(x̂), as well as the intermediates computed during their evaluation. As an application of the piecewise linearization procedure we devise a generalized Newton’s method based on successive piecewise linearization and prove for it sufficient conditions for convergence and convergence rates equaling those of semismooth Newton. We conclude with the derivation of formulas for the numerically stable implementation of the aforedeveloped piecewise linearization methods.

Recent research has shown that piecewise smooth (PS) functions can be approximated by piecewise linear functions with second order error in the distance to
a given reference point. A semismooth Newton type algorithm based on successive application of these piecewise linearizations was subsequently developed
for the solution of PS equation systems. For local bijectivity of the linearization
at a root, a radius of quadratic convergence was explicitly calculated in terms
of local Lipschitz constants of the underlying PS function. In the present work
we relax the criterium of local bijectivity of the linearization to local openness.
For this purpose a weak implicit function theorem is proved via local mapping
degree theory. It is shown that there exist PS functions f:IR^2 --> IR^2 satisfying the weaker
criterium where every neighborhood of the root of f contains a point x such that
all elements of the Clarke Jacobian at x are singular. In such neighborhoods
the steps of classical semismooth Newton are not defined, which establishes
the new method as an independent algorithm. To further clarify the relation between a PS function and its piecewise linearization,
several statements about structure correspondences between the two are proved.
Moreover, the influence of the specific representation of the local piecewise linear models
on the robustness of our method is studied.
An example application from cardiovascular mathematics is given.

In this article we analyze a generalized trapezoidal rule for initial value problems with piecewise smooth right hand side F:IR^n -> IR^n.
When applied to such a problem, the classical trapezoidal rule suffers from a loss of accuracy if the solution trajectory intersects a non-differentiability of F. In such a situation the investigated generalized trapezoidal rule achieves a higher convergence order than the classical method. While the asymptotic behavior of the generalized method was investigated in a previous work, in the present article we develop the algorithmic structure for efficient implementation strategies
and estimate the actual computational cost of the latter.
Moreover, energy preservation of the generalized trapezoidal rule is proved for Hamiltonian systems with piecewise linear right hand side.

We present an extension of Taylor’s theorem towards nonsmooth evalua-
tion procedures incorporating absolute value operaions. Evaluations procedures are
computer programs of mathematical functions in closed form expression and al-
low a different treatment of smooth operations and calls to the absolute value value
function. The well known classical Theorem of Taylor defines polynomial approx-
imation of sufficiently smooth functions and is widely used for the derivation and
analysis of numerical integrators for systems of ordinary differential or differential
algebraic equations, for the construction of solvers for the continuous nonlinear op-
timization of finite dimensional objective functions and for root solving of nonlinear
systems of equations. The herein provided proof is construtive and allow efficiently
designed algorithms for the execution and computation of generalized piecewise
polynomial expansions. As a demonstration we will derive a k-step method on the
basis of polynomial interpolation and the proposed generalized expansions.

In this article we analyze a generalized trapezoidal rule for initial value problems with piecewise smooth right hand side \(F:R^n \to R^n\) based on a generalization of algorithmic differentiation. When applied to such a problem, the classical trapezoidal rule suffers from a loss of accuracy if the solution trajectory intersects a nondifferentiability of \(F\). The advantage of the proposed generalized trapezoidal rule is threefold: Firstly, we can achieve a higher convergence order than with the classical method. Moreover, the method is energy preserving for piecewise linear Hamiltonian systems. Finally, in analogy to the classical case we derive a third order interpolation polynomial for the numerical trajectory. In the smooth case the generalized rule reduces to the classical one. Hence, it is a proper extension of the classical theory. An error estimator is given and numerical results are presented.

In this article we analyse a generalized trapezoidal rule for initial value problems with piecewise smooth right-hand side F : IR^n -> IR^n based on a generalization of algorithmic differentiation. When applied to such a problem, the classical trapezoidal rule suffers from a loss of accuracy if the solution trajectory intersects a nondifferentiability of F. The advantage of the proposed generalized trapezoidal rule is threefold: Firstly, we can achieve a higher convergence order than with the classical method. Moreover, the method is energy preserving for piecewise linear Hamiltonian systems. Finally, in analogy to the classical case we derive a third-order interpolation polynomial for the numerical trajectory. In the smooth case, the generalized rule reduces to the classical one. Hence, it is a proper extension of the classical theory. An error estimator is given and numerical results are presented.