Grid Generation Methods. 2nd Ed. - [Liseikin]

Grid Generation Methods

Second Edition

Scientific Computation

Editorial BoardJ.-J. Chattot, Davis, CA, USAP. Colella, Berkeley, CA, USAW. E, Princeton, NJ, USAR. Glowinski, Houston, TX, USAY. Hussaini, Tallahassee, FL, USAP. Joly, Le Chesnay, FranceH.B. Keller, Pasadena, CA, USAJ.E. Marsden, Pasadena, CA, USAD.I. Meiron, Pasadena, CA, USAO. Pironneau, Paris, FranceA. Quarteroni, Lausanne, Switzerland

and Politecnico of Milan, Milan, ItalyJ. Rappaz, Lausanne, SwitzerlandR. Rosner, Chicago, IL, USAP. Sagaut, Paris, FranceJ.H. Seinfeld, Pasadena, CA, USAA. Szepessy, Stockholm, SwedenM.F. Wheeler, Austin, TX, USA

For other titles published in this series, go towww.springer.com/series/718

Vladimir D. Liseikin

Grid Generation Methods

Second Edition

Professor Vladimir D. LiseikinRussian Academy of SciencesInstitute of Computational TechnologiesPr. Lavrentjeva 6Novosibirsk [email protected], [email protected]

and

Novosibirsk State UniversityPirogova St. 2Novosibirsk 630090Russia

ISSN 1434-8322ISBN 978-90-481-2911-9 e-ISBN 978-90-481-2912-6DOI 10.1007/978-90-481-2912-6Springer Dordrecht Heidelberg London New York

Library of Congress Control Number: 2009939615

cSpringer Science+Business Media B.V. 2010No part of this work may be reproduced, stored in a retrieval system, or transmitted inany form or by any means, electronic, mechanical, photocopying, microlming, recordingor otherwise, without written permission from the Publisher, with the exception of anymaterial supplied specically for the purpose of being entered and executed on a computersystem, for exclusive use by the purchaser of the work.

Cover design: eStudio Calamar S.L.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Preface to the Second Edition

This second edition is signicantly expanded by new material that discussesrecent advances in grid generation technology based on the numerical solu-tion of Beltrami and diusion equations in control metrics. It gives a moredetailed and practice-oriented description of the control metrics for provid-ing the generation of adaptive, eld-aligned, and balanced numerical grids.Some numerical algorithms are described for generating surface and domaingrids. Applications of the algorithms to the generation of numerical grids withindividual and balanced properties are demonstrated.

Grid generation codes represent an indispensable tool for solving eldproblems in nearly all areas of applied mathematics and computational physics.The use of these grid codes signicantly enhances the productivity and re-liability of the numerical analysis of problems with complex geometry andcomplicated solutions. The science of grid generation is still growing fast; newdevelopments are continually occurring in the elds of grid methods, codes,and practical applications. Therefore there exists an evident need of students,researchers, and practitioners in applied mathematics for new books which co-herently complement the existing ones with a description of new developmentsin grid methods, grid codes, and the concomitant areas of grid technology.

The objective of this book is to give a clear, comprehensive, and easilylearned description of all essential methods of grid generation technology fortwo major classes of grids: structured and unstructured. These classes relyon two somewhat opposite basic concepts. The basic concept of the formerclass is adherence to order and organization, while the latter is prone to theabsence of any restrictions.

The present monograph discusses the current state of the art in methodsof grid generation and describes new directions and new techniques aimed atthe enhancement of the eciency and productivity of the grid process. Theemphasis is put on mathematical formulations, explanations, and examples ofvarious aspects of grid generation.

Special attention is paid to a review of those promising approaches andmethods which have been developed recently and/or have not been suciently

vi Preface to the Second Edition

covered in other monographs. In particular, the book includes a stretchingmethod adjusted to the numerical solution of singularly perturbed equationshaving large scale solution variations, e.g. those modeling high-Reynolds-number ows. A number of functionals related to conformality, orthogonality,energy, and alignment are described. The book includes dierential and varia-tional techniques for generating uniform, conformal, and harmonic coordinatetransformations on hypersurfaces for the development of a comprehensive ap-proach to the construction of both xed and adaptive grids in the interiorand on the boundary of domains in a unied manner. The monograph is alsoconcerned with the description of all essential grid quality measures such asskewness, curvature, torsion, angle and length values, and conformality. Em-phasis is given to a clear style and new angles of consideration where it is notintended to include unnecessary abstractions.

The major area of attention of this book is structured grid techniques.However, the author has also included an elementary introduction to basicunstructured approaches to grid generation. A more detailed description ofunstructured grid techniques can be found in Computational Grids: Adap-tation and Solution Strategies by Carey (1997), Delaunay Triangulation andMeshing by George and Borouchaki (1998), and Mesh Generation Applicationto Finite Elements by Frey and George (2008).

Since grid technology has widespread application to nearly all eld prob-lems, this monograph may have some interest for a broad range of readers,including teachers, students, researchers, and practitioners in applied mathe-matics, mechanics, and physics.

The rst chapter gives a general introduction to the subject of grids. Thereare two fundamental forms of mesh: structured and unstructured. Structuredgrids are commonly obtained by mapping a standard grid into the physicalregion with a transformation from a reference computational domain. Themost popular structured grids are coordinate grids. The cells of such gridsare curvilinear hexahedrons, and the identication of neighboring points isdone by incrementing coordinate indices. Unstructured grids are composed ofcells of arbitrary shape and, therefore, require the generation of a connectivitytable to allow the identication of neighbors. The chapter outlines structured,unstructured, and composite grids and delineates some basic approaches totheir generation. It also includes a description of various types of grid topologyand touches on certain issues of big grid codes.

Chapter 2 deals with some relations, necessary only for grid generation,connected with and derived from the metric tensors of coordinate transforma-tions. As an example of an application of these relations, the chapter presentsa technique aimed at obtaining conservation-law equations in new xed ortime-dependent coordinates. In the procedures described, the deduction ofthe expressions for the transformed equations is based only on the formula fordierentiation of the Jacobian.

Very important issues of grid generation, connected with a description ofgrid quality measures in forms suitable for formulating grid techniques and

Preface to the Second Edition vii

eciently analyzing the necessary mesh properties, are discussed in Chap. 3.The denitions of the grid quality measures are based on the metric tensorsand on the relations between the metric elements considered in Chap. 2. Spe-cial attention is paid to the invariants of the metric tensors, which are thebasic elements for the denition of many important grid quality measures.Clear algebraic and geometric interpretations of the invariants are presented.

Equations with large variations of the solution, such as those modelinghigh-Reynolds-number ows, are one of the most important areas of the ap-plication of adaptive grids and of demonstration of the eciency of grid tech-nology. The numerical analysis of such equations on special grids obtained bya stretching method has a denite advantage in comparison with the classicalanalytic expansion method in that it requires only a minimum knowledge ofthe qualitative properties of the physical solution. The fourth chapter is con-cerned with the description of such stretching methods aimed at the numericalanalysis of equations with singularities.

The rst part of Chap. 4 acquaints the reader with various types of sin-gularity arising in solutions to equations with a small parameter aecting thehigher derivatives. The solutions of these equations undergo large variationsin very small zones, called boundary or interior layers. The chapter gives aconcise description of the qualitative properties of solutions in boundary andinterior layers and an identication of the invariants governing the locationand structure of these layers. Besides the well-known exponential layers, threetypes of power layer which are common to bisingular problems having com-plementary singularities arising from reduced equations, are described. Suchequations are widespread in applications, for example, in gas dynamics. Sim-ple examples of one- and two-dimensional problems which realize dierenttypes of boundary and interior layers are demonstrated, in particular, the ex-otic case where the interior layer approaches innitely close to the boundaryas the parameter tends to zero, so that the interior layer turns out to be aboundary layer of the reduced problem. This interior layer exhibits one morephenomenon: it is composed of layers of two basic types, exponential on oneside of the center of the layer and power-type on the other side.

The second part of Chap. 4 describes a stretching method based on theapplication of special nonuniform stretching coordinates in regions of largevariation of the solution. The use of stretching coordinates is extremly eec-tive for the numerical solution of problems with boundary and interior layers.The method requires only a very basic knowledge of the qualitative proper-ties of the physical solution in the layers. The specication of the stretchingfunctions is given for each type of basic singularity. The functions are denedin such a way that the singularities are automatically smoothed with respectto the new stretching coordinates. The chapter ends with the description ofa procedure to generate intermediate coordinate transformations which aresuitable for smoothing both exponential and power layers. The grids derivedwith such stretching coordinates are often themselves well adapted to the ex-

viii Preface to the Second Edition

pected physical features. Therefore, they make it easier to provide dynamicadaptation by taking part of the adaptive burden on themselves.

The simplest and fastest technique of grid generation is the algebraicmethod based on transnite interpolation. Chapter 5 describes formulas forgeneral unidirectional transnite interpolations. Multidirectional interpola-tion is dened by Boolean summation of unidirectional interpolations. Thegrid lines across block interfaces can be made completely continuous by us-ing Lagrange interpolation or to have slope continuity by using the Hermitetechnique. Of central importance in transnite interpolation are the blendingfunctions (positive univariate quantities depending only on one chosen coordi-nate) which provide the matching of the grid lines at the boundary and interiorsurfaces. Detailed relations between the blending functions and approaches totheir specication are discussed in this chapter. Examples of various types ofblending function are reviewed, in particular, the functions dened throughthe basic stretching coordinate transformations for singular layers describedin Chap. 4. These transformations are dependent on a small parameter so thatthe resulting grid automatically adjusts to the respective physical parameter,e.g. viscosity, Reynolds number, or shell thickness, in practical applications.The chapter ends with a description of a procedure for generating triangular,tetrahedral, or prismatic grids through the method of transnite interpolation.

Chapter 6 is concerned with grid generation techniques based on the nu-merical solution of systems of partial dierential equations. Generation ofgrids from these systems of equations is largely based on the numerical so-lution of elliptic, hyperbolic, and parabolic equations for the coordinates ofgrid lines which are specied on the boundary segments. The elliptic and par-abolic systems reviewed in the chapter provide grid generation within blockswith specied boundary point distributions. These systems are also used tosmooth algebraic, hyperbolic, and unstructured grids. A very important roleis currently played in grid codes by a system of Poisson equations dened asa sum of Laplace equations and control functions. This system was originallyconsidered by Godunov and Prokopov and further generalized, developed,and implemented for practical applications by Thompson, Thames, Mastin,and others. The chapter describes the properties of the Poisson system andspecies expressions for the control functions required to construct nearly or-thogonal coordinates at the boundaries. Hyperbolic systems are useful whenan outer boundary is free of specication. The control of the grid spacing inthe hyperbolic method is largely performed through the specication of vol-ume distribution functions. Special hyperbolic and elliptic systems are pre-sented for generating orthogonal and nearly orthogonal coordinate lines, inparticular, those proposed by Ryskin and Leal. The chapter also reviews someparabolic and high-order systems for the generation of structured grids.

Eective adaptation is one of the most important requirements put ongrid technology. The basic aim of adaptation is to increase the accuracy andproductivity of the numerical solution of partial dierential equations througha redistribution of the grid points and renement of the grid cells. Chapter 7

Preface to the Second Edition ix

describes some basic techniques of dynamic adaptation. The chapter startswith the equidistribution method, rst suggested in dierence form by Boorand further applied and extended by Dwyer, Kee, Sanders, Yanenko, Liseikin,Danaev, and others. In this method, the lengths of the cell edges are denedthrough a weight function modeling some measure of the solution error. Aninteresting fact about the uniform convergence of the numerical solution ofsome singularly perturbed equations on a uniform grid is noted and explained.The chapter also describes adaptation in the elliptic method, performed by thecontrol functions. Features and eects of the control functions are discussedand the specication of the control functions used in practical applicationsis presented. Approaches to the generation of moving grids for the numericalsolution of nonstationary problems are also reviewed. The most importantfeature of a structured grid is the Jacobian of the coordinate transformationfrom which the grid is derived. A method based on the specication of thevalues of the Jacobian to keep it positive, developed by Liao, is presented.

Chapter 8 reviews the developments of variational methods applied to gridgeneration. Variational grid generation relies on functionals related to gridquality: smoothness, orthogonality, regularity, aspect ratio, adaptivity, etc.By the minimization of a combination of these functionals, a user can dene acompromise grid with the desired properties. The chapter discusses the vari-ational approach of error minimization introduced by Morrison and furtherdeveloped by Babuska, Tihonov, Yanenko, Liseikin, and others. Functionalsfor generating uniform, conformal, quasiconformal, orthogonal, and adaptivegrids, suggested by Brackbill, Saltzman, Winslow, Godunov, Prokopov, Ya-nenko, Liseikin, Liao, Steinberg, Knupp, Roache, and others are also pre-sented. A variational approach using functionals dependent on invariants ofthe metric tensor is also considered. The chapter discusses a new variationalapproach for generating harmonic maps through the minimization of energyfunctionals, which was suggested by Dvinsky. Several versions of the function-als from which harmonic maps can be derived are identied.

Methods developed for the generation of grids on curves and surfaces arediscussed in Chap. 9. The chapter describes the development and applicationof hyperbolic, elliptic, and variational techniques for the generation of gridson parametrically dened curves and surfaces. The dierential approaches arebased on the Beltrami equations proposed by Warsi and Thomas, while thevariational methods rely on functionals of surface grid quality measures. Thechapter includes also a description of the approach to constructing conformalmappings on surfaces developed by Khamaysen and Mastin.

Chapter 10 is devoted to the author variant of the implementation of anidea of Eiseman for generating adaptive grids by projecting quasiuniform gridsfrom monitor hypersurfaces. The monitor hypersurface is formed as a surfaceof the values of some vector function over the physical geometry. The vectorfunction can be a solution to the problem of interest, a combination of itscomponents or derivatives, or any other variable quantity that suitably mon-itors the way that the behavior of the solution inuences the eciency of the

x Preface to the Second Edition

calculations. For the purpose of commonality a general approach based on dif-ferential and variational methods for the generation of quasiuniform grids onarbitrary hypersurfaces is considered. The variational method of generatingquasiuniform grids, developed by the author, is grounded on the minimizationof a generalized functional of grid smoothness on hypersurfaces, which was in-troduced for domains by Brackbill and Saltzman. The chapter also includes anexpansion of the method by introducing more general control metrics in thephysical geometry. The control metrics provide ecient and straightforwardlydened conditions for various types of grid adaptation, particularly grid clus-tering according to given function values and/or gradients, grid alignmentwith given vector elds, and combinations thereof. Using this approach, onecan generate both adaptive and xed grids in a unied manner, in arbitrarydomains and on their boundaries. This allows code designers to merge thetwo tasks of surface grid generation and volume grid generation into one taskwhile developing a comprehensive grid generation code.

The subject of unstructured grid generation is discussed in Chap. 11. Un-structured grids may be composed of cells of arbitrary shape, but they aregenerally composed of triangles and tetrahedrons. Tetrahedral grid methodsdescribed in the chapter include Delaunay procedures and the advancing-frontmethod. The Delaunay approach connects neighboring points (of some previ-ously dened set of nodes in the region) to form tetrahedral cells in such a waythat the sphere through the vertices of any tetrahedron does not contain anyother points. In the advancing-front method, the grid is generated by buildingcells one at a time, marching from the boundary into the volume by succes-sively connecting new points to points on the front until all the unmeshedspace is lled with grid cells.

The book ends with a list of references.

Acknowledgement. The author is greatly thankful to G. Liseikin who prepared thetext of the manuscript in LATEX code. Thanks go as well to G. Lukas for correctingthe authors English.

The author is very grateful for the helpful suggestions in the comments forChaps. 1, 3, and 11 made by E. Ivanov, a leading expert in up-to-date codes, gridquality measures, and methods for unstructured grid generation. The author is alsogreatly obliged to the researchers who responded to his requests and sent him theirpapers, namely, T.J. Baker, D.A. Field, E. Ivanov, P. Knupp, G. Liao, M.S. Shep-hard, N.P. Weatherill, and P.P. Zegeling.

The work over the book was supported in part by an Integrated Grant of theSiberian Branch of the Russian Academy of Sciences (20092011): Award No 94, andby the Russian Foundation for Basic Research (RFBR): Award 09-01-12023. Specif-ically, the eorts related to computing gures of grids made by A. Kharitonchick,A. Kofanov, Yu. Likhanova, and I. Vaseva, whom the author thanks very much,were remunerated by payments from these grants.

Novosibirsk Vladimir D. Liseikin

Contents

1 General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 General Concepts Related to Grids . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2.1 Grid Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2.2 Requirements Imposed on Grids . . . . . . . . . . . . . . . . . . . . . 5

1.3 Grid Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101.3.1 Structured Grids Generated by Mapping Approach . . . . 101.3.2 Unstructured Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151.3.3 Block-Structured Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.3.4 Overset Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201.3.5 Hybrid Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.4 Approaches to Grid Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 211.4.1 Methods for Structured Grids . . . . . . . . . . . . . . . . . . . . . . . 221.4.2 Methods for Unstructured Grids . . . . . . . . . . . . . . . . . . . . 23

1.5 Big Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251.5.1 Interactive Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261.5.2 New Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.6 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2 Coordinate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.2 General Notions and Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.2.1 Jacobi Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.2.2 Tangential Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332.2.3 Normal Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.2.4 Representation of Vectors Through the Base Vectors . . . 362.2.5 Metric Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382.2.6 Cross Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.3 Relations Concerning Second Derivatives . . . . . . . . . . . . . . . . . . . 442.3.1 Christoel Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

xii Contents

2.3.2 Dierentiation of the Jacobian . . . . . . . . . . . . . . . . . . . . . . 462.3.3 Basic Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.4 Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492.4.1 Scalar Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . 492.4.2 Vector Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.5 Time-Dependent Transformations . . . . . . . . . . . . . . . . . . . . . . . . . 552.5.1 Reformulation of Time-Dependent Transformations . . . . 552.5.2 Basic Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562.5.3 Equations in the Form of Scalar Conservation Laws . . . . 582.5.4 Equations in the Form of Vector Conservation Laws . . . 62

2.6 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3 Grid Quality Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.2 Curve Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.2.1 Basic Curve Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.2.2 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703.2.3 Torsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.3 Surface Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.3.1 Surface Base Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.3.2 Metric Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733.3.3 Second Fundamental Form . . . . . . . . . . . . . . . . . . . . . . . . . 753.3.4 Surface Curvatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.4 Metric-Tensor Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773.4.1 Algebraic Expressions for the Invariants . . . . . . . . . . . . . . 783.4.2 Geometric Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.5 Characteristics of Grid Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 803.5.1 Sum of Squares of Cell Edge Lengths . . . . . . . . . . . . . . . . 813.5.2 Eccentricity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 813.5.3 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.5.4 Measure of Coordinate Line Torsion . . . . . . . . . . . . . . . . . 85

3.6 Characteristics of Faces of Three-Dimensional Grids . . . . . . . . . 853.6.1 Cell Face Skewness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853.6.2 Face Aspect-Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.6.3 Cell Face Area Squared . . . . . . . . . . . . . . . . . . . . . . . . . . . . 863.6.4 Cell Face Warping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

3.7 Characteristics of Grid Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883.7.1 Cell Aspect-Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893.7.2 Square of Cell Volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893.7.3 Cell Area Squared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893.7.4 Cell Skewness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893.7.5 Characteristics of Nonorthogonality . . . . . . . . . . . . . . . . . . 903.7.6 Grid Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 913.7.7 Characteristics of Deviation from Conformality . . . . . . . . 923.7.8 Grid Eccentricity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

Contents xiii

3.7.9 Measures of Grid Warping and Grid Torsion . . . . . . . . . . 963.7.10 Quality Measures of Simplexes . . . . . . . . . . . . . . . . . . . . . . 97

3.8 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4 Stretching Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1014.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1014.2 Formulation of the Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1024.3 Theoretical Foundation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.3.1 Model Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1054.3.2 Basic Majorants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

4.4 Basic Intermediate Transformations . . . . . . . . . . . . . . . . . . . . . . . . 1164.4.1 Basic Local Stretching Functions . . . . . . . . . . . . . . . . . . . . 1164.4.2 Basic Boundary Contraction Functions . . . . . . . . . . . . . . . 1204.4.3 Other Univariate Transformations . . . . . . . . . . . . . . . . . . . 1254.4.4 Construction of Basic Intermediate Transformations . . . 127

4.5 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

5 Algebraic Grid Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1335.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1335.2 Transnite Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

5.2.1 Unidirectional Interpolation . . . . . . . . . . . . . . . . . . . . . . . . 1345.2.2 Tensor Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1355.2.3 Boolean Summation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

5.3 Algebraic Coordinate Transformations . . . . . . . . . . . . . . . . . . . . . 1395.3.1 Formulation of Algebraic Coordinate Transformation . . 1395.3.2 General Algebraic Transformations . . . . . . . . . . . . . . . . . . 141

5.4 Lagrange and Hermite Interpolations . . . . . . . . . . . . . . . . . . . . . . 1435.4.1 Coordinate Transformations Based on Lagrange

Interpolation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1435.4.2 Transformations Based on Hermite Interpolation . . . . . . 147

5.5 Control Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1505.6 Transnite Interpolation from Triangles and Tetrahedrons . . . . 1515.7 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

6 Grid Generation Through Dierential Systems . . . . . . . . . . . . 1556.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1556.2 Laplace Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

6.2.1 Two-Dimensional Equations . . . . . . . . . . . . . . . . . . . . . . . . 1586.2.2 Three-Dimensional Equations . . . . . . . . . . . . . . . . . . . . . . . 161

6.3 Poisson Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1646.3.1 Formulation of the System . . . . . . . . . . . . . . . . . . . . . . . . . . 1656.3.2 Justication for the Poisson System. . . . . . . . . . . . . . . . . . 1666.3.3 Equivalent Forms of the Poisson System . . . . . . . . . . . . . . 1686.3.4 Orthogonality at Boundaries . . . . . . . . . . . . . . . . . . . . . . . . 1706.3.5 Control of the Angle of Intersection . . . . . . . . . . . . . . . . . . 177

xiv Contents

6.4 Biharmonic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1816.4.1 Formulation of the Approach . . . . . . . . . . . . . . . . . . . . . . . 1816.4.2 Transformed Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

6.5 Orthogonal Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1826.5.1 Derivation from the Condition of Orthogonality . . . . . . . 1836.5.2 Multidimensional Equations . . . . . . . . . . . . . . . . . . . . . . . . 184

6.6 Hyperbolic and Parabolic Systems . . . . . . . . . . . . . . . . . . . . . . . . . 1856.6.1 Specication of Aspect Ratio . . . . . . . . . . . . . . . . . . . . . . . 1866.6.2 Specication of Jacobian . . . . . . . . . . . . . . . . . . . . . . . . . . . 1886.6.3 Parabolic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1916.6.4 Hybrid Grid Generation Scheme . . . . . . . . . . . . . . . . . . . . . 191

6.7 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

7 Dynamic Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1957.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1957.2 One-Dimensional Equidistribution . . . . . . . . . . . . . . . . . . . . . . . . . 196

7.2.1 Example of an Equidistributed Grid . . . . . . . . . . . . . . . . . 1977.2.2 Original Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1997.2.3 Dierential Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2007.2.4 Specication of Weight Functions . . . . . . . . . . . . . . . . . . . . 201

7.3 Equidistribution in Multidimensional Space . . . . . . . . . . . . . . . . . 2097.3.1 One-Directional Equidistribution . . . . . . . . . . . . . . . . . . . . 2097.3.2 Multidirectional Equidistribution . . . . . . . . . . . . . . . . . . . . 2107.3.3 Control of Grid Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2117.3.4 Equidistribution over Cell Volume . . . . . . . . . . . . . . . . . . . 213

7.4 Adaptation Through Control Functions . . . . . . . . . . . . . . . . . . . . 2167.4.1 Specication of the Control Functions in Elliptic

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2167.4.2 Hyperbolic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218

7.5 Grids for Nonstationary Problems . . . . . . . . . . . . . . . . . . . . . . . . . 2187.5.1 Method of Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2197.5.2 Moving-Grid Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . 2197.5.3 Time-Dependent Deformation Method . . . . . . . . . . . . . . . 221

7.6 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

8 Variational Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2278.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2278.2 Calculus of Variations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

8.2.1 General Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2288.2.2 EulerLagrange Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 2298.2.3 Functionals Dependent on Metric Elements . . . . . . . . . . . 2328.2.4 Functionals Dependent on Tensor Invariants . . . . . . . . . . 2338.2.5 Convexity Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

8.3 Integral Grid Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2368.3.1 Dimensionless Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . 236

Contents xv

8.3.2 Dimensionally Heterogeneous Functionals . . . . . . . . . . . . . 2408.3.3 Functionals Dependent on Second Derivatives . . . . . . . . . 242

8.4 Adaptation Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2438.4.1 One-Dimensional Functionals . . . . . . . . . . . . . . . . . . . . . . . 2448.4.2 Multidimensional Approaches . . . . . . . . . . . . . . . . . . . . . . . 245

8.5 Functionals of Attraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2508.5.1 Lagrangian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2508.5.2 Attraction to a Vector Field . . . . . . . . . . . . . . . . . . . . . . . . 2528.5.3 Jacobian-Weighted Functional . . . . . . . . . . . . . . . . . . . . . . 253

8.6 Energy Functionals of Harmonic Function Theory . . . . . . . . . . . 2558.6.1 General Formulation of Harmonic Maps . . . . . . . . . . . . . . 2558.6.2 Application to Grid Generation . . . . . . . . . . . . . . . . . . . . . 2568.6.3 Relation to Other Functionals . . . . . . . . . . . . . . . . . . . . . . . 256

8.7 Combinations of Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2578.7.1 Natural Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . 258

8.8 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

9 Curve and Surface Grid Methods . . . . . . . . . . . . . . . . . . . . . . . . . . 2619.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2619.2 Grids on Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262

9.2.1 Formulation of Grids on Curves . . . . . . . . . . . . . . . . . . . . . 2629.2.2 Grid Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264

9.3 Formulation of Surface Grid Methods . . . . . . . . . . . . . . . . . . . . . . 2669.3.1 Mapping Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2669.3.2 Associated Metric Relations . . . . . . . . . . . . . . . . . . . . . . . . 268

9.4 Beltramian System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2699.4.1 Beltramian Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2699.4.2 Surface Grid System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

9.5 Interpretations of the Beltramian System . . . . . . . . . . . . . . . . . . . 2729.5.1 Variational Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2729.5.2 Harmonic-Mapping Interpretation . . . . . . . . . . . . . . . . . . . 2739.5.3 Formulation Through Invariants . . . . . . . . . . . . . . . . . . . . . 2749.5.4 Formulation Through the Surface Christoel Symbols . . 2759.5.5 Relation to Conformal Mappings . . . . . . . . . . . . . . . . . . . . 2809.5.6 Projection of the Laplace System . . . . . . . . . . . . . . . . . . . . 282

9.6 Control of Surface Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2839.6.1 Control Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2839.6.2 Projection on the Boundary Line . . . . . . . . . . . . . . . . . . . . 2849.6.3 Monitor Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2859.6.4 Control by Variational Methods . . . . . . . . . . . . . . . . . . . . . 2869.6.5 Orthogonal Grid Generation . . . . . . . . . . . . . . . . . . . . . . . . 289

9.7 Hyperbolic Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2909.7.1 Hyperbolic Governing Equations . . . . . . . . . . . . . . . . . . . . 291

9.8 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291

xvi Contents

10 Comprehensive Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29310.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29310.2 Hypersurface Geometry and Grid Formulation . . . . . . . . . . . . . . 295

10.2.1 Hypersurface Grid Formulation . . . . . . . . . . . . . . . . . . . . . 29510.2.2 Monitor Hypersurfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29610.2.3 Metric Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29710.2.4 Christoel Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29810.2.5 Relations Between Metric Elements . . . . . . . . . . . . . . . . . . 300

10.3 Functional of Smoothness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30110.3.1 Formulation of the Functional . . . . . . . . . . . . . . . . . . . . . . . 30110.3.2 Geometric Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . 30210.3.3 Dimensionless Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . 30410.3.4 EulerLagrange Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 30510.3.5 Equivalent Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

10.4 Hypersurface Grid Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30910.4.1 Inverted Beltrami Equations . . . . . . . . . . . . . . . . . . . . . . . . 309

10.5 Formulation of Comprehensive Grid Generator . . . . . . . . . . . . . . 31110.5.1 Energy and Diusion Functionals . . . . . . . . . . . . . . . . . . . . 31110.5.2 Relation to Harmonic Functions . . . . . . . . . . . . . . . . . . . . . 31210.5.3 Beltrami and Diusion Equations . . . . . . . . . . . . . . . . . . . 31310.5.4 Inverted Beltrami and Diusion Equations . . . . . . . . . . . . 315

10.6 Numerical Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31710.6.1 Finite-Dierence Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 31810.6.2 Spectral Element Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 322

10.7 Formulation of Control Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32410.7.1 Specication of Individual Control Metrics . . . . . . . . . . . . 32510.7.2 Control Metrics for Generating Grids with Balanced

Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32910.7.3 Application to Solution of Singularly-Perturbed

Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33010.8 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331

11 Unstructured Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33311.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33311.2 Consistent Grids and Numerical Relations . . . . . . . . . . . . . . . . . . 334

11.2.1 Convex Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33411.2.2 Consistent Grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

11.3 Methods Based on the Delaunay Criterion . . . . . . . . . . . . . . . . . . 33711.3.1 Dirichlet Tessellation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33911.3.2 Incremental Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33911.3.3 Approaches for Insertion of New Points . . . . . . . . . . . . . . 34111.3.4 Two-Dimensional Approaches . . . . . . . . . . . . . . . . . . . . . . . 34211.3.5 Constrained Form of Delaunay Triangulation . . . . . . . . . 34611.3.6 Point Insertion Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . 34811.3.7 Surface Delaunay Triangulation . . . . . . . . . . . . . . . . . . . . . 354

Contents xvii

11.3.8 Three-Dimensional Delaunay Triangulation . . . . . . . . . . . 35411.4 Advancing-Front Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356

11.4.1 Procedure of Advancing-Front Method . . . . . . . . . . . . . . . 35611.4.2 Strategies to Select Out-of-Front Vertices . . . . . . . . . . . . . 35711.4.3 Grid Adaptation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35811.4.4 Advancing-Front Delaunay Triangulation . . . . . . . . . . . . . 35811.4.5 Three-Dimensional Prismatic Grid Generation . . . . . . . . 359

11.5 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387

1General Considerations

1.1 Introduction

An important element of the numerical solution of partial dierential equa-tions by nite-element or nite-dierence methods on general regions is a gridwhich represents the physical domain in a discrete form. In fact, the grid isa preprocessing tool or a foundation on which physical, continuous quantitiesare described by discrete functions and on which the dierential equations areapproximated by algebraic relations for discrete values that are then numeri-cally analyzed by the application of computational codes. The grid techniquealso has the capacity, based on an appropriate distribution of the grid points,to enhance the computational eciency of the numerical solution of complexproblems.

The eciency of a numerical study of a boundary value problem is esti-mated from the accuracy of the computed solution and from the cost and timeof the computation.

The accuracy of the numerical solution in the physical domain depends onboth the error of the solution at the grid points and the error of interpolation.Commonly, the error of the numerical computation at the grid points arisesfrom several distinct sources. First, mathematical models do not representphysical phenomena with absolute accuracy. Second, an error arises at thestage of the numerical approximation of the mathematical model. Third, theerror is inuenced by the size and shape of the grid cells. Fourth, an error iscontributed by the computation of the discrete physical quantities satisfyingthe equations of the numerical approximation. And fth, an error in the solu-tion is caused by the inaccuracy of the process of interpolation of the discretesolution. Of course, the accurate evaluation of the errors due to there sourcesremains a formidable task. It is apparent, however, that the quantitative andqualitative properties of the grid play a signicant role in controlling the in-uence of the third and fth sources of the error in the numerical analysis ofphysical problems.

V.D. Liseikin, Grid Generation Methods, Scientic Computation,DOI 10.1007/978-90-481-2912-6 1, c Springer Science+Business Media B.V. 2010

2 1 General Considerations

Another important characteristic of a numerical algorithm that inuencesits eciency is the cost of the operation of obtaining the solution. From thispoint of view, the process of generating a sophisticated grid may increase thecomputational costs of the numerical solution and encumber the computertools with the requirement of additional memory. On the other hand, theremay be a signicant prot in accuracy which allows one to use a smallernumber of grid points. Any estimation of the contributions of these oppos-ing factors can help in choosing an appropriate grid. In any case, since gridgeneration is an important component of numerical modeling, research in thiseld is aimed at creating techniques which are not too costly but which givea signicant improvement in the accuracy of the solution. The utilization ofthese techniques provides one with the real opportunities to enhance the ef-ciency of the numerical solution of complex problems. Thus grid generationhelps to satisfy the constant demand for enhancement of the eciency of thenumerical analysis of practical problems.

The rst eorts aimed at the development of grid techniques were un-dertaken in the 1960s. Now, a signicant number of advanced methods havebeen created: algebraic, elliptic, hyperbolic, parabolic, variational, Delaunay,advancing-front, etc. The development of these methods has reached a stagewhere calculations in fairly complicated domains and on surfaces that arisewhile analyzing multidimensional problems are possible. Because of its suc-cessful development, the eld of numerical grid generation has already formeda separate mathematical discipline with its own methodology, approaches, andtechnology.

At the end of the 1980s there started a new stage in the development ofgrid generation technique. It is characterized by the creation of comprehensive,multipurpose, three-dimensional grid generation codes which are aimed atproviding a uniform environment for the construction of grids in arbitrarymultidimensional geometries.

The current chapter presents a framework for the subject of grid gener-ation. It outlines the most general concepts and techniques, which will beexpounded in the following chapters in more detail.

1.2 General Concepts Related to Grids

There are two general notions of a grid in an n-dimensional bounded domainor on a surface. One of these considers the grid as a set of algorithmicallyspecied points of the domain or the surface. The points are called the gridnodes. The second considers the grid as an algorithmically described collectionof standard n-dimensional volumes covering the necessary area of the domainor surface. The standard volumes are referred to as the grid cells. The cellsare bounded curvilinear volumes, whose boundaries are divided into a fewsegments which are (n1)-dimensional cells. Therefore they can be formulatedsuccessively from one dimension to higher dimensions. The boundary points

1.2 General Concepts Related to Grids 3

of the one-dimensional cells are called the cell vertices. These vertices are thegrid nodes. Thus the grid nodes are consistent with the grid cells in that theycoincide with the cell vertices.

This section discusses some general concepts related to grids and grid cells.

1.2.1 Grid Cells

For cells in an n-dimensional domain or surface, there are commonly usedn-dimensional volumes of simple standard shapes (see Fig. 1.1 for n = 1, 2, 3).

In one dimension the cell is a closed line or segment, whose boundary iscomposed of two points referred to as the cell vertices.

A general two-dimensional cell is a two-dimensional simply connected do-main, whose boundary is divided into a nite number of one-dimensional cellsreferred to as the edges of the cell. Commonly, the cells of two-dimensionaldomains or surfaces are constructed in the form of triangles or quadrilaterals.The boundary of a triangular cell is composed of three segments, while theboundary of a quadrilateral is represented by four segments. These segmentsare the one-dimensional grid cells.

By a general three-dimensional cell there is meant a simply connectedthree-dimensional polyhedron whose boundary is partitioned into a nite num-ber of two-dimensional cells called its faces. In practical applications, three-dimensional cells typically have the shape of tetrahedrons or hexahedrons.The boundary of a tetrahedral cell is composed of four triangular cells, whilea hexahedron is bounded by six quadrilaterals. Thus a hexahedral cell has sixfaces, twelve edges, and eight vertices. Some applications also use volumes inthe form of prisms as three-dimensional cells. A prism has two triangular andthree quadrilateral faces, nine edges, and six vertices.

Commonly, the edges and the faces of the cells are linear. Linear trian-gles and tetrahedrons are also referred to as two-dimensional simplexes andthree-dimensional simplexes, respectively. The notion of the simplex can beformulated for arbitrary dimensions. Namely, by an n-dimensional simplexthere is meant a domain of n-dimensional space whose nodes are dened bythe equation

x =N+1

i=1

ixi,

where xi, i = 1, . . . , N +1, are some specied points which are the vertices ofthe simplex, and i, i = 1, . . . , N +1, are real numbers satisfying the relations

N+1

i=1

i = 1, i 0.

In this respect a one-dimensional linear cell is the one-dimensional simplex.The boundary of an n-dimensional simplex is composed of n + 1 (n 1)-dimensional simplexes.


Fig. 1.1. Typical grid cells

The selection of the shapes shown in Fig. 1.1 to represent the standardcells is justied, rst, by their geometrical simplicity and, second, becausethe existing procedures for the numerical simulation of physical problems arelargely based on approximations of partial dierential equations using theseelemental volumes. The specic choice of cell shape depends on the geome-try and physics of the particular problem and on the method of solution. Inparticular, tetrahedrons (triangles in two dimensions) are well suited for nite-element methods, while hexahedrons are commonly used for nite-dierencetechniques.

Some applications consider curvilinear cells as well. These grid cells areobtained by deformation of ordinary linear segments, triangles, tetrahedrons,squares, cubes, and prisms.

The major advantage of hexahedral cells (quadrilaterals in two dimensions)is that their faces (or edges) may be aligned with the coordinate surfaces (orcurves). In contrast, no coordinates can be aligned with tetrahedral meshes.However, strictly hexahedral meshes may be ineective near boundaries withsharp corners.

Prismatic cells are generally placed near boundary surfaces which havepreviously been triangulated. The surface triangular cells serve as faces ofprisms, which are grown out from these triangles. Prismatic cells are ecientfor treating boundary layers, since they can be constructed with a high aspectratio in order to resolve the layers, but without small angles, as would be thecase for tetrahedral cells.

Triangular cells are the simplest two-dimensional elements and can beproduced from quadrilateral cells by constructing interior edges. Analogously,tetrahedral cells are the simplest three-dimensional elements and can be de-rived from hexahedrons and prisms by constructing interior faces. The strengthof triangular and tetrahedral cells is in their applicability to virtually any typeof domain conguration. The drawback is that the integration of the physicalequations becomes a few times more expensive with these cells in comparisonwith quadrilateral or hexahedral cells.


The vertices of the cells dene grid points which approximate the physicaldomain. Alternatively, the grid points in the domain may have been generatedpreviously by some other process. In this case the construction of the grid cellsrequires special techniques.

1.2.2 Requirements Imposed on Grids

The grid should discretize the physical domain or surface in such a mannerthat the computation of the physical quantities is carried out as eciently asdesired. The accuracy, which is one of the components of the eciency of thecomputation, is inuenced by a number of grid factors, such as grid size, gridtopology, cell shape and size, and consistency of the grid with the geometryand with the solution. A very general consideration of these grid factors isgiven in this subsection.

1.2.2.1 Grid Size and Cell Size

The grid size is indicated by the number of grid points, while the cell sizeimplies the maximum value of the lengths of the cell edges. Grid generationrequires techniques which possess the intrinsic ability to increase the numberof grid nodes. At the same time the edge lengths of the resulting cells shouldbe reduced in such a manner that they approach zero as the number of nodestends to innity.

An instructive example of a grid on the interval [0, 1] which does not satisfythe requirement of unlimited reduction of the cell sizes when the number ofthe nodes is increased is a grid generated by a rule in which the steps are ina geometrical progression:

hi+1hi

= a, a > 0, a = 1, (1.1)

where hi = xi+1 xi, i = 0, . . . , N 1, are the steps of the grid nodesxi, i = 0, . . . , N, with x0 = 0, xN = 1. The grid nodes xi satisfying (1.1) arecomputed for arbitrary N by the formula

xi =a 1

aN 1 ai

j=1

aj , i = 1, . . . , N,

and consequently we obtain

hi =ai+1(a 1)aN 1 a , i = 0, . . . , N 1.

Thereforelim

N h0 = 1 a if 0 < a < 1,

limN

hN 1 =a 1

aif a > 1,


i.e. the left-hand boundary cell of this grid, if a < 1, or the right-hand bound-ary cell, if a > 1, does not approach zero even though the number of gridpoints tends to innity.

Small cells are necessary to obtain more accurate solutions and to in-vestigate phenomena associated with the physical quantities on small scales,such as transition layers and turbulence. Also, the opportunity to increase thenumber of grid points and to reduce the size of the cells enables one to studythe convergence rate of a numerical code and to improve the accuracy of thesolution by multigrid approaches.

1.2.2.2 Grid Organization

There also is a requirement on grids to have some organization of their nodesand cells, which is aimed at facilitating the procedures for formulating andsolving the algebraic equations substituted for the dierential equations. Thisorganization should identify neighboring points and cells. The grid organiza-tion is especially important for that class of nite-dierence methods whoseprocedures for obtaining the algebraic equations consist of substituting dif-ferences for derivatives. To a lesser degree, this organization is needed fornite-volume methods because of their inherent compatibility with irregularmeshes.

1.2.2.3 Cell and Grid Deformation

The cell deformation characteristics can be formulated as some measures ofthe departure of the cell from a standard, least deformed one. Such standardtriangular and tetrahedral cells are those with edges of equal lengths. The leastdistorted quadrilaterals and hexahedrons are squares and cubes, respectively.The standard prism is evidently the prism with standard linear faces. Cellswith low deformity are preferable from the point of view of simplicity anduniformity of the construction of the algebraic equations approximating thedierential equations.

Typically, cell deformation is characterized through the aspect ratio, theangles between the cell edges, and the volume (area in two dimensions) of thecell.

The major requirement for the grid cells is that they must not be folded ordegenerate at any points or lines, as demonstrated in Fig. 1.2. Unfolded cellsare obtained from standard cells by a one-to-one deformation. Commonly, thevalue of any grid generation method is judged by its ability to yield unfoldedgrids in regions with complex geometry.

The grid deformity is also characterized by the rate of the change of thegeometrical features of contiguous cells. Grids whose neighboring cells do notchange abruptly are referred to as smooth grids.


Fig. 1.2. Normal (left) and badly deformed (center, right) quadrilateral cells

1.2.2.4 Consistency with Geometry

The accuracy of the numerical solution of a partial dierential equation and ofthe interpolation of a discrete function is considerably inuenced by the degreeof compatibility of the mesh with the geometry of the physical domain. First ofall, the grid nodes must adequately approximate the original geometry, that is,the distance between any point of the domain and the nearest grid node mustnot be too large. Moreover, this distance must approach zero when the grid sizetends to innity. This requirement of adequate geometry approximation bythe grid nodes is indispensible for the accurate computation and interpolationof the solution over the whole region.

The second requirement for consistency of the grid with the geometry isconcerned with the approximation of the boundary of the physical domainby the grid, i.e. there is to be a sucient number of nodes which can beconsidered as the boundary ones, so that a set of edges (in two dimensions)and cell faces (in three dimensions) formed by these nodes models ecientlythe boundary. In this case, the boundary conditions may be applied moreeasily and accurately. If these points lie on the boundary of the domain, thenthe grid is referred to as a boundary-tting or boundary-conforming grid.

1.2.2.5 Consistency with Solution

It is evident that distribution of the grid points and the form of the grid cellsshould be dependent on the features of the physical solution. In particular, itis better to generate the cells in the shape of hexahedrons or prisms in bound-ary layers. Often, the grid points are aligned with some preferred directions,e.g. streamlines. Furthermore, a nonuniform variation of the solution requiresclustering of the grid point in regions of high gradients, so that these areas ofthe domain have ner resolution. Local grid clustering is needed because theuniform renement of the entire domain may be very costly for multidimen-sional computations. It is especially true for problems whose solutions havelocalized regions of very rapid variation (layers). Without grid clustering inthe layers, some important features of the solution can be missed, and the ac-curacy of the solution can be degraded. Problems with boundary and interiorlayers occur in many areas of application, for example in uid dynamics, com-bustion, solidication, solid mechanics and wave propagation. Typically the


Fig. 1.3. Boundary layer function for = 102

locations where the high resolution is needed are not known beforehand butare found in the process of computation. Consequently, a suitable mesh, track-ing the necessary features of the physical quantities, as the solution evolves,is required.

A local grid renement is accomplished in two ways: (a) by moving axed number of grid nodes, with clustering of them in zones where this isnecessary, and coarsening outside of these zones, and (b) by inserting newpoints in the zones of the domain where they are needed. Local grid renementin zones of large variation of the solution commonly results in the followingimprovements:

(1) the solution at the grid points is obtained more accurately;(2) the solution is interpolated over the whole region more precisely;(3) oscillations of the solution are eliminated;(4) larger time steps can be taken in the process of computing solutions of

time-dependent problems.

The typical pattern of a solution with large local variation is illustratedby the following univariate monotonic function

u(x) = exp(x/) + x, 0 x 1,

with a positive parameter . This function is a solution to the two-point bound-ary value problem

u + u = 1, 1 > x > 0,u(0) = 1, u(1) = 1 + exp(1/).

When the parameter is very small, then u(x) has a boundary layer of rapidvariation (Fig. 1.3). Namely, in the interval [0, |ln |] the function u(x) changesfrom 1 to + |ln |. For example, if = 105, then |ln | = 5 105 ln 10 0,

with0 < r0 < r1, 0 0 < 1 2.

If 1 = 2 then this function transforms the unit three-dimensional cube intoa space bounded by two cylinders of radii r0 and r1 and by the two planesx3 = 0 and x3 = H. The reference uniform grid in 3 is dened by the nodes

ijk = (ih, jh, kh), 0 i, j, k N, h = 1/N,

where i, j, k and N are positive integers. The cells of this grid are the three-dimensional cubes bounded by the coordinate planes 1i = ih,

2j = jh, and

3k = kh. Corresponding, the structured grid in the domain X3 is determined

by the nodesxijk = x(ijk), 0 i, j, k N.

The cells of the grid in X3 are the curvilinear hexahedrons bounded bythe curvilinear coordinate surfaces derived from the parametrization x()(Fig. 1.4).

1.3.1.1 Realization of Grid Requirements

The notion of using a transformation to generate a mesh is very helpful. Theidea is to choose a computational domain n with a simpler geometry thanthat of the physical domain Xn and then to nd a transformation x() be-tween these domains which eliminates the need for a nonuniform mesh whenapproximating the physical quantities. That is, if the computational area and


the transformation are well chosen, the transformed boundary value prob-lem can be accurately represented by a small number of equally spaced meshpoints. Emphasis is placed on a small number of points, because any trans-formed problem (provided only that the transformation is nonsingular) maybe accurately approximated with a suciently ne, uniform mesh. In practice,there will be a trade-o between the diculty of nding the transformationand the number of uniformly spaced points required to nd the solution to agiven accuracy.

The idea of using mappings to generate grids is extremely appropriatefor nding the conditions that the grid must satisfy for obtaining accuratesolutions of partial dierential equations in the physical domain Xn, becausethese conditions can be readily dened in terms of the transformations. Forexample, the grid requirements described in Sect. 1.2.2 are readily formulatedthrough the transformation concept.

Since a solution which is a linear function is computed accurately at thegrid points and is approximated accurately over the whole region, an attractivepossible method for generating structured grids is to nd a transformationx() such that the solution is linear in n. Though in practice this requirementfor the transformation is not attained even theoretically (except in the case ofstrongly monotonic univariate functions), it is useful in the sense of an idealthat the developers of structured grid generation techniques should bear inmind. One modication of this requirement which can be practically realizedconsists of the requirement of a local linearity of the solution in n.

The requirements imposed on the grid and the cell size are realized by theconstruction of a uniform grid in n and a smooth function x(). The grid cellsare not folded if x() is a one-to-one mapping. Consistency with the geometryis satised with a transformation x() that maps the boundary of n onto theboundary of Xn. Grid concentration in zones of large variation of a functionu(x) is accomplished with a mapping x() which provides variations of thefunction u[x()] in the domain n that are not large.

1.3.1.2 Coordinate Grids

Among structured grids, coordinate grids in which the nodes and cell facesare dened by the intersection of lines and surfaces of a coordinate systemin Xn are very popular in nite-dierence methods. The range of values ofthis system denes a computation region n in which the cells of the uniformgrid are rectangular n-dimensional parallelepipeds, and the coordinate valuesdene the function x() : n Xn.

The simplest of such grids are the Cartesian grids obtained by the in-tersection of the Cartesian coordinates in Xn. The cells of these grids arerectangular parallelepipeds (rectangles in two dimensions). The use of Carte-sian coordinates avoids the need to transform the physical equations. However,the nodes of the Cartesian grid do not coincide with the curvilinear bound-

1.3 Grid Classes 13

ary, which leads to diculties in implementing the boundary conditions withsecond-order accuracy.

1.3.1.3 Boundary-Conforming Grids

An important subdivision of structured grids is the boundary-tted or bounda-ry-conforming grids. These grids are obtained from one-to-one transformationsx() which map the boundary of the domain n onto the boundary of Xn.

The most popular of these, for nite-dierence methods, have become thecoordinate boundary-tted grids whose points are formed by intersection ofthe coordinate lines, while the boundary of Xn is composed of a nite numberof coordinate surfaces (lines in two dimensions) i = i0. Consequently, in thiscase the computation region n is a rectangular domain, the boundaries ofwhich are determined by (n 1)-dimensional coordinate planes in Rn, andthe uniform grid in n is the Cartesian grid. Thus the physical region isrepresented as a deformation of a rectangular domain and the generated gridas a deformed lattice (Fig. 1.5).

These grids give a good approximation to the boundary of the region andare therefore suitable for the numerical solution of problems with boundarysingularities, such as those with boundary layers in which the solution dependsvery much on the accuracy of the approximation of the boundary conditions.

The requirements imposed on boundary-conforming grids are naturallysatised with the coordinate transformations x().

The algorithm for the organization of the nodes of boundary-tted co-ordinate grids consists of the trivial identication of neighboring points byincrementing the coordinate indices, while the cells are curvilinear hexahe-drons. This kind of grid is very suitable for algorithms with parallelization.

Its design makes it easy to increase or change the number of nodes asrequired for multigrid methods or in order to estimate the convergence rateand error, and to improve the accuracy of numerical methods for solvingboundary value problems.

With boundary-conforming grids there is no necessity to interpolate theboundary conditions of the problem, and the boundary values of the region

Fig. 1.5. Boundary-conforming quadrilateral grid


can be considered as input data to the algorithm, so automatic codes for gridgeneration can be designed for a wide class of regions and problems.

In the case of unsteady problems the most direct way to set up a movinggrid is to do it via a coordinate transformation. These grids do not requirea complicated data structure, since they are obtained from uniform grids insimple xed domains such as rectangular ones, where the grid data structureremains intact.

1.3.1.4 Shape of Computational Domains

The idea of the structured approach is to transform a complex physical domainXn to a simpler domain n with the help of the parametrization x(). Theregion n in (1.2), which is called the computational or logical region, can beeither rectangular or of a dierent matching qualitatively the geometry of thephysical domain; in particular, shape it can be triangular for n = 2 (Fig. 1.6)or tetrahedral for n = 3. Using such parametrizations, a numerical solution ofa partial dierential equation in a physical region of arbitrary shape can becarried out in a standard computational domain, and codes can be developedthat require only changes in the input.

The cells of the uniform grid can be rectangular or of a dierent shape.Schematic illustrations of two-dimensional triangular and quadrilateral gridsare presented in Figs. 1.6 and 1.7, respectively. Note that regions in the formof curvilinear triangles, such as that shown in Fig. 1.6, are more suitable for

Fig. 1.6. Boundary-conforming triangular grid

Fig. 1.7. Computational domains adjusted to the physical domains

1.3 Grid Classes 15

gridding in the structured approach by triangular cells than by quadrilateralones. One approach for such gridding is described in Sect. 5.6.

1.3.2 Unstructured Grids

Many eld problems of interest involve very complex geometries that are noteasily amenable to the framework of the pure structured grid concept. Struc-tured grids may lack the required exibility and robustness for handling do-mains with complicated boundaries, or the grid cells may become too skewedand twisted, thus prohibiting an ecient numerical solution. An unstructuredgrid concept is considered as one of the appropriate solutions to the problemof producing grids in regions with complex shapes.

Unstructured grids have irregularly distributed nodes and their cells arenot obliged to have any one standard shape. Besides this, the connectivityof neighboring grid cells is not subject to any restrictions; in particular, thecells can overlap or enclose one another. Thus, unstructured grids provide themost exible tool for the discrete description of a geometry.

These grids are suitable for the discretization of domains with a compli-cated shape, such as regions around aircraft surfaces or turbomachinery bladerows. They also allow one to apply a natural approach to local adaptation, byeither insertion or removal of nodes. Cell renement in an unstructured sys-tem can be accomplished locally by dividing the cells in the appropriate zonesinto a few smaller cells. Unstructured grids also allow excessive resolution tobe removed by deleting grid cells locally over regions in which the solutiondoes not vary appreciably. In practice, the overall time required to generateunstructured grids in complex geometries is much shorter than for structuredor block structured grids.

However, the use of unstructured grids complicates the numerical algo-rithm because of the inherent data management problem, which demands aspecial program to number and order the nodes, edges, faces, and cells of thegrid, and extra memory is required to store information about the connectionsbetween the cells of the mesh. One further disadvantage of unstructured gridsthat causes excessive computational work is associated with increased numbersof cells, cell faces, and edges in comparison with those for hexahedral meshes.For example, a tetrahedral mesh of N points has roughly 6N cells, 12N faces,and 7N edges, while a mesh of hexahedra has roughly N cells, 3N faces, and3N edges. Furthermore, moving boundaries or moving internal surfaces ofphysical domains are dicult to handle with unstructured grids. Besides this,linearized dierence scheme operators on unstructured grids are not usuallyband matrices, which makes it more dicult to use implicit schemes. As aresult, the numerical algorithms based on an unstructured grid topology arethe most costly in terms of operations per time step and memory per gridpoint.

Originally, unstructured grids were mainly used in the theory of elasticityand plasticity, and in numerical algorithms based on nite-element methods.


However, the eld of application of unstructured grids has now expanded con-siderably and includes computational uid dynamics. Some important aspectsof the construction of unstructured grids are considered in Chap. 11.

1.3.3 Block-Structured Grids

In the commonly applied block strategy, the region is divided without holes oroverlaps into a few contiguous subdomains, which may be considered as thecells of a coarse, generally unstructured grid. And then a separate structuredgrid is generated in each block. The union of these local grids constitutesa mesh referred to as a block-structured or multi-block grid. Grids of thiskind can thus be considered as locally structured at the level of an individualblock, but globally unstructured when viewed as a collection of blocks. Thusa common idea in the block-structured grid technique is the use of dierentstructured grids, or coordinate systems, in dierent regions, allowing the mostappropriate grid conguration to be used in each region.

Block-structured grids are considerably more exible in handling com-plex geometries than structured grids. Since these grids retain the simpleregular connectivity pattern of a structured mesh on a local level, theseblock-structured grids maintain, in nearly the same manner as structuredgrids, compatibility with ecient nite-dierence or nite-volume algorithmsused to solve partial dierential equations. However, the generation of block-structured grids may take a fair amount of user interaction and, therefore,requires the implementation of an automation technique to lay out the blocktopology.

The main reasons for using multi-block grids rather than single-block gridsare that

(1) the geometry of the region is complicated, having a multiply connectedboundary, cuts, narrow protuberances, cavities, etc.;

(2) the physical problem is heterogeneous relative to some of the physicalquantities, so that dierent mathematical models are required in dierentzones of the domain to adequately describe the physical phenomena;

(3) the solution of the problem behaves non-uniformly: zones of smooth andrapid variation of dierent scales may exist.

The blocks of locally structured grids in a three-dimensional region arecommonly homeomorphic to a three-dimensional cube, thus having the shapeof a curvilinear hexahedron. However, some domains can be more eectivelypartitioned with the use of cylindrical blocks as well. Cylindrical blocks arecommonly applied to the numerical solution of problems in regions with holesand to the calculation of ows past aircraft or aircraft components (wings,fuselages, etc.). For many problems it is easier to take into account the geom-etry of the region and the structure of the solution by using cylindrical blocks.Also, the total number of blocks and sections might be smaller than when us-ing only blocks homeomorphic to a cube.

1.3 Grid Classes 17

Fig. 1.8. Types of interface between contiguous blocks (a discontinuous; b, c non-smooth; d smooth)

1.3.3.1 Communication of Adjacent Coordinate Lines

The requirement of mutual positioning or communication of adjacent gridblocks can also have a considerable inuence on the construction of locallystructured grids and on the eciency of the numerical calculations. The co-ordinate lines dening the grid nodes of two adjacent blocks need not havepoints in common, and can join smoothly or nonsmoothly (Fig. 1.8). If alladjacent grid blocks join smoothly, interpolation is not required. If the co-ordinate lines do not join, then during the calculation the solution values atthe nodes of one block must be transferred to those of the adjacent block inthe neighborhood of their intersection. This is done by interpolation or (inmechanics) using conservation laws.

The types of interaction between adjacent grid blocks are selected on thebasis of the features of the physical quantities in the region of their intersec-tion. If the gradient of the physical solution is not high in the vicinity of aboundary between two adjacent blocks and interpolation can, therefore, beperformed with high accuracy, the coordinate lines do not need to join. Thisgreatly simplies the algorithm for constructing the grid in a block. If thereare high gradients of the solution near the intersection of two blocks, a smoothmatching is usually performed between the coordinate lines of the two blocks.This kind of conformity poses a serious diculty for structured mesh gener-ation methods. Currently the problem is overcome by an algebraic techniqueusing Hermitian interpolation, or by elliptic methods, involving a choice ofcontrol functions. A combination of Laplace and Poisson equations, yieldingequations of fourth or even sixth order, is also used for this purpose.

1.3.3.2 Topology of the Grid

The correct choice of the topology in a block, depending on the geometry ofthe computational region and the choice of the transformation of the regioninto the block, has a considerable inuence on the quality of the grid. Thereare two ways of specifying the computational region for a block:


Fig. 1.9. Patterns of grid topology

(1) as a complicated polyhedron which maintains the schematic form of theblock subdomain (Fig. 1.7);

(2) simply as a solid cube or a cube with cuts (Fig. 1.9).

With the rst approach, the problem of constructing the coordinate trans-formation x() is simplied, and this method is often used to generate asingle-blocked grid in a complicated domain. The second approach relies on asimplied geometry of the computational domain but requires sophisticatedmethods to derive suitable transformations x().

In a block which is homeomorphic to a cylinder with thick walls, the gridtopology is determined by the topology of the two-dimensional grids in thetransverse sections. In applications, for sections of this kind, which are annularplanes or surfaces with a hole, wide use is made of three basic grid topologies:H, O and C (see Fig. 1.9).

In H-type grids, the computational region is a square with an interior cutwhich is opened by the construction of the coordinate transformation andmapped onto an interior boundary of the region X2. The outer boundary ofthe square is mapped onto the exterior of X2. The interior boundary has twopoints with singularities where one coordinate line splits. H-type grids areused, for instance, when calculating the ow past thin bodies (aircraft wings,turbine blades, etc.).

In O-type grids, the computational region is a solid square. In this case thesystem of coordinates is obtained by bending the square, sticking two oppositesides together and then deforming. The stuck sides determine the cut, calledthe ctive edge, in the block. An example of O-type grid is the nodes and cellsof a polar system of coordinates. The O-type grid can be constructed withoutsingularities when the boundary of the region is smooth. Grids of this kind

1.3 Grid Classes 19

are used when calculating the ow past bulky aircraft components (fuselages,gondolas, etc.) and, in combination with H-type grids, for multilayered blockstructures.

The computational region is also a solid square in a C-type grid, but themapping onto X2 involves the identication of some segments of one of itssides and then deforming it. In the C-type grid, the coordinate lines of onefamily leave the outer boundary, circle the inner boundary and return againto the outer boundary. There is one point on the inner boundary which hasthe same type of singularity as in the H-type grid. The C-type grids arecommonly used in regions with holes and long protuberances.

The O and C-type techniques in fact introduce articial interior cuts inmultiply connected regions to generate single block-structured grids. The cutsare used to join the disconnected components of the domain boundary in orderto reduce their number. Theoretically, this operation can allow one to generatea single coordinate transformation in a multiply connected domain.

The choice of the grid topology in a block depends on the structure of thesolution, the geometry of the domain, and, in the case of continuous or smoothgrid-line communication, on the topology of the grid in the adjacent block aswell. For complicated domains, such as those near aircraft surfaces or turbineswith a large number of blades, it is dicult to choose the grid topology of theblocks, because each component of the system (wing, fuselage, etc.) has its ownnatural type of grid topology, but these topologies are usually incompatiblewith each other.

1.3.3.3 Conditions Imposed on Grids in Blocks

A grid in a block must satisfy the conditions which are required to obtainan acceptable solution. In any specic case, these conditions are determinedby features of the computer, the methods of grid generation available, thetopology and conditions of interaction of the blocks, the numerical algorithms,and the type of data to be obtained.

One of the main requirements imposed on the grid is its adaptation to thesolution. Multidimensional computations are likely to be very costly withoutthe application of adaptive grid techniques. The basic aim of adaptation is toenhance the eciency of numerical algorithms for solving physical problemsby a special nonuniform distribution of grid nodes. The appropriate adaptivedisplacement of the nodes, depending on the physical solution, can increase theaccuracy and rate of convergence and reduce oscillations and the interpolationerror.

In addition to adaptation, the construction of locally structured grids oftenrequires the coordinate lines to cross the boundary of the domain or the sur-face in an orthogonal or nearly orthogonal fashion. The orthogonality at theboundary can greatly simplify the specication of boundary conditions. Also,a more accurate representation of algebraic models of turbulence, the equa-tions of a boundary layer, and parabolic NavierStokes equations is possible in


this case. If for grids of O and C-type the coordinate lines are orthogonal to theboundary of each block and its interior cuts, the global block-structured gridwill be smooth. It is also desirable for the coordinate lines to be orthogonal ornearly orthogonal inside the blocks. This will improve the convergence of thedierence algorithms, and the equations, if written in orthogonal variables,will have a simpler form.

For unsteady gas-dynamics problems, some coordinates in the entire do-main or on the boundary are required to have Lagrange or nearly Lagrangeproperties. With Lagrangian coordinates the computational region remainsxed in time and simpler expressions for the equations can be obtained inthis case.

It is also important that the grid cells do not collapse, the changes in thesteps are not too abrupt, the lengths of the cell sides are not very dierent,and the cells are ner in any domain of high gradient, large error, or slowconvergence. Requirements of this kind are taken into account by introducingquantitative and qualitative characteristics of the grid, both with the help ofcoordinate transformations and by using the sizes of cell edges, faces, angles,and volumes. The characteristics used include the deviation from orthogonal-ity, the Lagrange properties, the values of the transformation Jacobian or cellvolume, and the smoothness and adaptivity of the transformation. For cellfaces, the deviation from a parallelogram, rectangle, or square, as well as theratio of the area of the face to its perimeter, is also used.

1.3.4 Overset Grids

Block-structured grids require the partition of the domain into blocks thatare restricted so as to abut each other. Overset grids are exempt from thisrestriction. With the overset concept the blocks are allowed to overlap, whichsignicantly simplies the problem of the selection of the blocks covering thephysical region. In fact, each block may be a subdomain which is associatedonly with a single geometry or physical feature. The global grid is obtained asan assembly of structured grids which are generated separately in each block.These structured grids are overset on each other, with data communicated byinterpolation in overlapping areas of the blocks (Fig. 1.10).

Fig. 1.10. Fragment of an overset grid

1.4 Approaches to Grid Generation 21

Fig. 1.11. Fragment of a hybrid grid

1.3.5 Hybrid Grids

Hybrid numerical grids are meshes which are obtained by combining bothstructured and unstructured grids. These meshes are widely used for the nu-merical analysis of boundary value problems in regions with a complex geom-etry and with a solution of complicated structure. They are formed by joiningstructured and unstructured grids on dierent parts of the region or surface.Commonly, a structured grid is generated about each chosen boundary seg-ment. These structured grids are required not to overlap. The remainder ofthe domain is lled with the cells of an unstructured grid (Fig. 1.11). Thisconstruction is widely applied for the numerical solution of problems withboundary layers.

1.4 Approaches to Grid Generation

The unique aspect of grid generation on general domains is that grid genera-tion has a high degree of freedom, i.e. mesh techniques are not obliged to haveany specied formulation, so any foundation may be suitable for this purposeif the grid generated is acceptable.

The chief practical diculty facing grid generation techniques is that offormulating satisfactory techniques which can realize the users requirements.Grid generation techniques should develop methods that can help in handlingproblems with multiple variables, each varying over many orders of magni-tude. These methods should be capable of generating grids which are locallycompressed by large factors when compared with uniform grids.

The methods should incorporate specic control tools, with simple andclear relationships between these control tools and characteristics of the gridsuch as mesh spacing, skewness, smoothness, and aspect ratio, in order toprovide a reliable way to inuence the eciency of the computation. Andnally, the methods should be computationally ecient and easy to code.

A number of techniques for grid generation have been developed. Everymethod has its strengths and its weaknesses. Therefore, there is also the ques-tion of how to choose the most ecient method for the solution of any specic


problem, taking into account the geometrical complexity, the computing costfor generating the grid, the grid structure, and other factors.

The goal of the development of these methods is to provide eective andacceptable grid generation systems.

1.4.1 Methods for Structured Grids

The most ecient structured grids are boundary-conforming grids. The gen-eration of these grids can be performed by a number of approaches and tech-niques. Many of these methods are specically oriented to the generation ofgrids for the nite-dierence method.

A boundary-tted coordinate grid in the region Xn is commonly gener-ated rst on the boundary of Xn and then successively extended from theboundary to the interior of Xn. This process is analogous to the interpolationof a function from a boundary or to the solution of a dierential boundaryvalue problem. On this basis there have been developed three basic groups ofmethods of grid generation:

(1) algebraic methods, which use various forms of interpolation or specialfunctions;

(2) dierential methods, based mainly on the solution of elliptic, parabolic,and hyperbolic equations

Documents

Grid Generation Methods. 2nd Ed. - [Liseikin]