4. RESEARCH PROJECTS


The research projects have been organized in the SCAT (Scientific Computing and Applications in Technology) program. The SCAT program consists of projects in five main research areas. In the year 1993 these areas have been

Within each research area a reasonable level of synergy between different projects is generally provided either by use of similar methods or by study of similar problems with different types of methods.

The different research areas have a common mathematical background which makes the communication possible also between projects in different areas. Moreover, much effort has been invested in the common computational environment which on the one hand reduces the routine work in the projects and on the other hand makes it easier for the projects to communicate with each other.

4.1. FOUNDATIONS OF COMPUTATIONAL METHODS

Although many of the projects deal with fundamental questions, the main emphasis is in most of them in some special problems. That is why only relatively few projects can be considered to belong to this category.

Finite Element Methods and Applications (MICHEL KRÍZEK, RAINO A.E. MÄKINEN, PEKKA NEITTAANMÄKI, VALERY RIVKIND and TIMO TIIHONEN)

The project deals with finite element analysis of some important, mostly nonlinear, problems. The goal is to give a rigorous analysis for the theoretical background of these questions. Of course, it is not possible to cover all technically interesting cases. Nevertheless, several typical techniques are introduced for solving nonlinear problems which can then be modified for other problems. There is no general numerical method capable of solving an arbitrary nonlinear problem and, therefore, every class of problems has to be investigated individually. The investigation of nonlinear problems consists of several steps. The problem is split into a series of subproblems, including, as a rule, the proof of the existence (and, if applicable, also uniqueness) of the true solution, the construction of concrete finite elements, the treatment of convergence questions together with the numerical solution of the approximate problem. Among others, Maxwell equations, contact problems, free boundary problems, Stefan-like problems, crystal growth problems and problems in elasticity and plasticity have been handled.

Research collaboration with researchers from Czech Republic, Germany, USA, France, England, Italy, Poland, Russia and other universities in Finland.

Optimization with Applications to Optimal Control (MARKO M. MÄKELÄ and KAISA MIETTINEN)

This project consists of two parts. The first part deals with nonsmooth analysis and optimization. Subdifferential calculus has been applied to generalize the optimization theory and especially the optimality conditions of nondifferentiable functions. This theoretical basis has made it possible to develop effective bundletype optimization methods for nonsmooth and nonconvex objective functions subject to linear, nonlinear, nonsmooth and nonconvex constraints. The new variant, called proximal bundle method has proved to be very efflcient and reliable in solving nonsmooth and also smooth optimization problems. The methods have been implemented as a subroutine library and tested with standard test examples from the literature in different environments (microcomputers, workstations, mainframe and super computers).

The second part of the project deals with multiobjective optimization. The aim has been to generate interactive optimization methods for complicated problems. Two different kinds of interactive procedures have been developed for nonsmooth multiobjective optimization. Some optimal control problems arising from optimal shape design and continuous casting processes have been solved as applications.

4.2. APPLICATION ORIENTED PROJECTS

4.2.1 Computational Mechanics and Physics

Modelling and Simulation of Paper Making Process (RICHARD DU CROO DE JONGH, JARI HÄMÄLÄINEN, KAI HILTUNEN, MATTI KURKI, TIMO TIIHONEN and REIJO PIETIKÄINEN)

The goal of the project is to construct simulation models for some key parts of a paper machine as well as to study the properties of paper.

Homogenization and dimensional reduction have been applied in order to obtain a computationally feasible model for fluid flow in the headbox of a paper machine. The model has been published in a PhD thesis and in some conference articles. Also some numerical methods for predicting the shape of the free jet at the outlet of the headbox have been developed.

Modelling of the so-called former section of a paper machine (where fluidlike suspension of water and fibres gradually solidifies to a porous fibre net) has been started. A part of this work has been undertaken at the Technical University of Delft. Mathematical properties of the model are studied and numerical methods are being developed.

In addition, a model has been derived for moisture induced deformation of copying paper. The behaviour of paper sheets has been found to be strongly nonlinear and to exhibit bifurcation phenomena. The model has also been tested experimentally.

Free Boundary Problems (JAROSLAV HASLINGER, JARI JÄRVINEN, PEKKA NEITTAANMÄKI, VALERY RIVKIND and TIMO TIIHONEN)

Free boundary problems form one of the main research fields of the laboratory. In the past the work on free boundary problems and related control problems initiated the research in nonsmooth optimization and in state constrained control problems. Currently the laboratory takes part in the ESF programme Mathematical Treatment of Free Boundary Problems.

The theoretical work on free boundary problems concentrates now on modelling the evaporation process. The goal is to find adequate free boundary conditions for a coupled system of Navier-Stokes and Stefan equations as well as to develop suitable techniques for mathematical analysis and numerical solution of such problems.

Another main line in the research is the use of techniques developed for shape optimization in solving free boundary problems and analysing the solution methods. Often the free boundary conditions can be formulated as optimality conditions of a shape optimization problem. The main emphasis has been on a method which is based on a suitable relaxation of such a shape optimization problem.

There is also work going on in numerical simulation of an industrial process for producing silicon crystals. A numerical model has been constructed which takes into account all major mechanisms of heat transfer including solidification and heat radiation.

Inverse Problems in Electromagnetism (PEKKA NEITTAANMÄKI, MAREK RUDNICKI and ANTONIO SAVINI)

In electromagnetism the goal has been to introduce efflcient and reliable methods for the synthesis of magnetic and electric fields. On a more concrete level the research effort has been focussed on the forth-coming monograph "Inverse Problems and Optimal Design in Electricity and Magnetism" (submitted for publication at the Oxford University Press). It collects material from optimization, optimal design and inverse problems in a form that is applicable in the analysis and development of electromagnetic systems.

The Bounding Approach of the Timing Simulation (CORNELIU A. MARINOV and JUKKA-PEKKA SANTANEN)

The main performance criterion for digital circuits is the speed of the signal propagation which can be expressed by a delay time. The calculation of the exact transient solution is a very time consuming task especially in the early stages of the design. Due to this, upper and lower bounds are used instead of the exact solution in the so-called time simulators. The upper and the lower bound estimates for the solution can be calculated by using the parameters of the circuit, but the achieved bounds may not be tight enough. Another possible way of bounding the solution is to use the Waveform Relaxation method which gives tightening monotonic bounds for the solution when some special initial bounds for the solution have been defined.

4.2.2 Control and Identification

Optimal Control of Nonlinear Systems and Applications (TIMO MÄNNIKKÖ, PEKKA NEITTAANMÄKI, TUOMO RÄISÄNEN, DAN TIBA and TIMO TIIHONEN)

The purpose is to discuss optimal control problems, especially systems arising from free boundary problems. The main interest has been in optimal control problems governed by nonlinear parabolic systems including, among others, parabolic variational inequalities and systems with phase transitions. The aim is twofold: firstly to give a theoretical approach to the subject and secondly to present detailed algorithms (with convergence proofs) necessary in computerizing the optimal control processes. Several practical examples are being worked out in detail in order to demonstrate the usefulness of the proposed methods. The results have been collected in the monograph "Optimal Control of Nonlinear Parabolic Systems: Theory, Algorithms and Applications" (Marcel Dekker, Inc. to appear in 1994).

Collaboration with researchers from France, Germany, USA, Japan and Romania.

Optimization of Distributed Systems with Applications (TIMO MÄNNIKKÖ)

The main subject of this project is the theory of sentinels and their applications. Sentinels are "filters" which are sensitive to the variations of some specific parameters, and at the same time insensitive to the variations of all the other unknown parameters appearing in the system.

As an application, sentinels are used to identify unknown source terms in a parabolic partial differential equation with incompletely defined boundary and/or initial conditions.

This project is done in international co-operation with researchers from Compiègne University of Technology, France.

Hemivariational Inequalities and Optimal Control Problems (JAROSLAV HASLINGER and MARKKU MIETTINEN)

In this project hemivariational inequalities and optimal control problems governed by hemivariational inequalities are studied. An approximation of hemivariational inequalities has been developed. It can be realized numerically by using nonsmooth optimization methods.

Also sensitivity analysis of optimal control problems is studied by using a regularization method and set-valued analysis.

Distributed Parameter Identification (TOMMI KÄRKKÄINEN, PEKKA NEITTAANMÄKI and ANTTI NIEMISTÖ)

The aim is to develop methods for the identification of a functional coefflcient. It is assumed, that one has a distributed observation of the solution of an elliptic or parabolic partial differential equation. These observations are used to determine the unknown coefficient in the equation. In physical systems this parameter can represent the heat conductivity or the diffusion coefficient.

Several different discretization schemes based on variants of finite element, spectral and multigrid methods are considered. In addition to new efflcient numerical algorithms one of the purposes is the improvement of techniques for convergence analysis of different approaches. At the moment there exist few error estimates for parameter identification problems and only for elliptic equations. One of the goals is also the treatment of quasilinear and nonlinear problems with some linearization techniques.

General Regularization Methods for Solutions of Ill-Posed Problems (TIMO TASKINEN)

A problem is called well-posed if it is uniquely solvable for each data and if the solution depends continuously on the data. Otherwise the problem is called ill-posed. One way to overcome the ill-posedness is to use regularization methods. In this project new regularization schemes are developed and modified to be in their most suitable form for some particular problems. Error bounds obtainable and the speed of convergence are also matters of interest.

This work is done in co-operation with professor Gennadi Vainikko at the University of Tartu in Estonia.

Parallel Processing and Process Control (ERKKI LAITINEN and PAULI VIRTA NEN )

Nowadays, simulation models are widely used in process control. In the dynamic simulation the computing time and accuracy are important. In this project a parallelized simulation program is developed for controlling the cooling of a continuous casting machine. The model has been installed in a casting machine which casts about 500 000 tons steel in a year. The complicated model which consumes a lot of computer resources has been parallelized for several transputers inside a microcomputer.

I addition, a control program for controlling the fibre orientation of paper in a paper making process is developed. This application is based on Fourier analysis, optimization and simple analysis of fluid flow. The control model uses a computer network and the knowledge base of the paper factory.

The programming tools used are C++ and Windows. Collaboration with Rautaruukki steel works in Raahe and Kymi paper mills in Kuusankoski.

Parameter Estimation for Continuous Casting Process (ERKKI LAITINEN and MARKKU MALINEN)

The estimation of the secondary cooling parameters is crucial for the correct simulation of the heat transfer in continuous casting. Based on the measurements of the water flow rate and the surface temperature in a real casting process one can calculate the cooling parameters. In this project the aim is to find the optimal heat transfer coefficients for the simulation and to develop an easy-to-use interface to execute that searching in the MS-Windows environment.

Collaboration with Rautaruukki steel works in Raahe.

Optimal Motions of Articulated Systems (MIKKO TARKIAINEN)

In this research the time optimal motion planning of articulated systems with any number of degrees of freedom is studied. The method is applied to robotic motions. Full nonlinear and coupled rigid body dynamics including any state dependent control and state constraints are considered. The method has been extended to handle higher order dynamics, e.g. actuator dynamics and joint and link flexibilities. The method makes it possible to minimize the motion time, to study the effect of system parameters on the performance, etc.

The international collaboration has been established with prof. Z. Shiller at the University of California in Los Angeles. Results have been presented in the 1993 IEEE International Conference on Robotics and Automation, Atlanta, USA. An application to model biomechanical processes, i.e. optimal human motions, is planned in collaboration with prof. P. Komi at the Department of Biology of Physical Activity.

4.2.3 Computational Methods for Large-Scale Problems

Fast Direct and Fictitious Domain Methods for Partial Differential Equations (ERKKI HEIKKOLA, YURI A. KUZNETSOV, TUOMO ROSSI and JARI TOIVANEN)

The fast direct methods are considered for solving separable linear systems arising from finite difference or finite element approximations of partial differential equations on rectangular domains. Well known methods like FFT-methods, cyclic reduction, marching technique and FACR(l) methods have been analyzed. This means that exact arithmetical complexity estimates have been derived, furthermore, a new fast direct method has been considered and the applications of the different methods have been investigated. Also, new combinations of the different methods have been introduced.

The fictitious domain methods are developed for solving linear systems of equations arising from the finite element discretization of second order partial differential equations. The fictitious domain methods are preconditioned iterative processes to solve extended linear systems. This means that the original problem is extended to a rectangular domain in such a way that the fast direct solvers can be used in preconditioning.

The fictitious domain methods are almost optimal in the sense of the arithmetical cost. Furthermore, the memory requirements during the iteration can be reduced significantly by using a so-called partial solution technique. Another advantage is that the mesh generation is simple due to the nearly rectangular structure of the mesh.

This methodology has been applied in solving problems arising from fluid dynamics and electromagnetism. Especially, the solution of the potential flow problem, the Stokes equation and the wave Helmholtz equation have been considered.

Parallel Algorithms for Partial Differential Equations (YURI A. KUZNETSOV, RAINO A.E. MÄKINEN and PASI TARVAINEN)

The most cost efficient hardware configuration for large-scale scientific computing is a distributed memory system consisting of a cluster of workstations connected with a fast local network. However, developing efflcient general purpose software for a distributed memory multiprocessor is quite a challenging problem.

This project has two objectives. The first one is to develop a general purpose finite element package that can be run on a distributed memory multiprocessor. The second one is to study domain decomposition methods, a class of methods that is naturally suited for parallel computing.

Overlapping domain decomposition methods are applied to the numerical solution of free boundary problems, especially obstacle problems. The approach is important for nonlinear boundary value problems for two reasons. It provides a possibility of using parallel processing and, what is perhaps more important, means for isolating a neighbourhood of the free boundary for a special treatment.

Moreover, the investigation of the convergence properties of nonoverlapping do- main decomposition algorithms reliable to parallel computers has been continued. The main task has been to investigate the effect of preconditioners on the speed of convergence of the domain decomposition method in the case of elliptic boundary value problems.

The project is supported by the Academy of Finland.

Numerical Methods for Shape Optimization (JAROSLAV HASLINGER, RAINO A.E. MÄKINEN, JACQUES PERIAUX and JARI TOIVANEN)

In shape optimization, the optimization variable is the shape of the domain on which a partial differential equation (the state equation) is posed. The research has been oriented towards finding efflcient numerical methods for general optimal shape design problems. In addition, shape optimization has been applied to solve shape design problems in fluid dynamics.

The most crucial step in shape optimization is sensitivity analysis, i.e. calculation of the gradient of the cost function. There are basically two ways to perform the sensitivity analysis. The first approach uses the discretized model and carries out design sensitivity analysis by differentiating the algebraic equations. The second approach uses the material derivative of continuum mechanics to analyse the changes in the shape of the domain.

The algebraic approach has been applied to solve shape optimization problems for a transonic flow. In co-operation with Dr. J. Chleboun (Czech Republic) a primal hybrid variational formulation has been applied to solve shape optimization problems. The formulation has the advantage of giving an accurate approximation of boundary fluxes or stresses. This is important when boundary formulae are used for the sensitivity analysis. Shape optimization problems governed by Navier- Stokes equations with non-standard boundary conditions have also been studied.

In a joint project with Dassault Aviation (France), the fictitious domain method is applied to shape optimization of an airfoil profile. The objective of this study is to compare the performance of the fictitious domain and the traditional moving mesh approaches.

4.2.4. Pilot Projects

Cellular Radio Networks (VEIKKO HARA, ERKKI LAITINEN, ALI LATTUNEN and LIINA NENONEN)

The research has been concentrated on two applications which model the operation of cellular radio networks. The first application is a model in which optimal base station parameters are estimated by using a simulation model and optimization methods. The parameters to be optimized are the handover limits, i.e. the values of the signal strength and the C/I-ratio, where the mobile station has to switch from one base station to another. The model has been implemented by using SLAM II and Fortran in the VAX/VMS environment.

The aim of the other application is to be able to replace a part of manual network planning. The application makes use of numerical maps, propagation models and basic radio system parameters in order to predict the transmission loss and the field strength at the given distance from the base station. The implementation of this network planning tool has been done by using Borland C++ for Windows. Collaboration with the Telecommunication Laboratory of the Technical Research Centre of Finland.

Neural Networks in Telecommunication Planning (VEIKKO HARA and ALEXANDRU MURGU)

The aim of the project is to adapt some basic deterrninistic and stochastic methods of dynamic programming onto neural networks (feedback-recurrent and feedforward). The actual goal is to get a powerful tool for handling large scale optimization problems arising from telecommunication planning. The main topics have been sequential decision processes (like Markov decision problems and other approaches which can be finally set as Markovian decision problems) with a special emphasis on telecommunication network flow control and routing. The theory of diffusion processes and brownian control are used to be able to describe more accurately the stochastic nature of the arrival and flow processes of the customers in small and medium sized communication networks. The intention is to construct a soft computing tool for planning.

Collaboration with the Telecommunication Laboratory of the Technical Research Centre of Finland.

Multimedia and Telematic Control of Computers (ERKKI LAITINEN and MIKA REKKILÄ)

Voice identification and the telematic control of computers are of great interest nowadays. In this pilot project the possibility of sending voice messages with computers and the remote control of a computer by telephone are studied. The pilot application is to build up a control program which controls alarm messages in an alarm centre. The main objective of the program is to translate digital alarm messages into voice messages and send them forward. On the other hand, the program must be able to handle incoming telecommunication (voice or digital) messages.

Collaboration with Hedegren security.

Hypermedia in Personnel Education (LEENA LALLUKKA)

A steel factory called Rautaruukki Co in Raahe has its own hypermedia application which contains metallurgical knowledge about steel industry, production processes and products. A group of engineers has been developing this knowledge system since the year 1989.

The project concentrates on exploiting this metallurgical knowledge system to personnel education in Rautaruukki Co. This exploiting is based on theories of different learning styles in adult education, hypermedia technology and hypermedia in education. One part of this project is to develop useful applications to support the knowledge system. This project includes also one or more educational periods, where personnel education is based on this system and these different theories.

4.3. COMPUTATIONAL ENVIRONMENT

Program development has a major role in many of the projects. In this respect many projects have shared interests and needs. This has led to the development of program libraries common to the whole laboratory. Locally developed software packages are called FEMPAK and NSOLIB.

FEMPAK (RAINO A.E. MäKINEN and JARI TOIVANEN)

FEMPAK 1.0 is a portable software package for the numerical solution of twodimensional partial differential equations. It is based on the finite element method. The package consists of a set of low-level subroutines for the assembly and solution of the discrete finite element equations, a set of example driver programs and software to visualize the results. The graphical devices supported are X-windows and PostScript.

NSOLIB (MARKO M. MÄKELÄ)

NSOLIB is a Fortran subroutine library for nonsmooth optimization problems. There exist algorithms for single and multiple objective optimization subject to different kinds of constraints. The methods are able to handle either simple bounds for variables, linear, nonlinear or nonsmooth constraints, or all of them at the same time. NSOLIB subroutines are implementations of the proximal bundle method, which is at the moment the most efflcient and reliable class of methods for nonsmooth optimization. The need of NSOLIB was noticed since the commercial subroutine packages do not contain any codes for nonsmooth optimization.

Commercial stlbroutine libraries mainly used are NAG and DISSPLA.

Computer Facilities

The Laboratory of Scientific Computing and the Department of Mathematics have 30 HP-9000 series workstations and X-terminals and 20 486-level PCs for researchers and students. In addition, the facilities at the Finnish National Supercomputer Centre (Cray XMP EA/432 and Convex C3840) and the Computer Centre of the University of Jyväskylä (VAX 8650 and SUN 4/260 computers connected to terminals and PCs with a local area network) are available.


Chapter III Contents Chapter V
webmaster@mit.jyu.fi -- Created 14.6.1994 -- Last Change 21.6.1994