Boundary Elements An Introductory Course | SECOND EDITION C.A. Brabbia and J. Dominguez rove, het WI Teress Computational Mechanics PublicationsBoundary Elements An Introductory Course Second Edition C.A. Brebbia & J. Dominguez WITp RESS Boston, Southampton Computational Mechanics Publications 4CA. Brebita I. Dominguez Wessex Institute of Technology ‘Escuela Tecnica Superior de Ingenieros Industriales Ashurst Lodge University of Seville Ashurst Av. Reina Mercedes, s/n Southampton, S040 7AA, UK 41012, Sevilla, Spain Published by WIT Press/Computational Mechanics Publications Ashurst Lodge, Ashurst, Southampton, S040 7AA, UK Tel: 44(0)1703 293223; Fax: 44(0)1703 292853 E-Mail: witpress@witpress.com http://www. witpress.com For USA, Canada and Mexico Computational Mechanics Inc 25 Bridge Street, Billerica, MA 01821, USA Tel: 978 667 5841; Fax: 978 667 7582 E-Mail: cmina@ix.netcom.com British Library Cataloguing-in-Publication Data A Catalogue record for this book is available from the British Library ISBN 1 85312 160 6 WITPress/Computational Mechanics Publications, Southampton ISBN 1 56252 087 3 Computational Mechanics Inc, Boston Library of Congress Catatog Card Number 91-77179 No responsibility is assumed by the Publisher, the Editors and Authors for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. © WIT Press/Computational Mechanics Publications 1992 Reprinted 1998 Printed and bound in Great Britain by Print in Black Alll rights reserved, No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the Publisher,fo Alexander, Beatriz, Isabel and PelayoDISKETTE A diskette containing all the source codes in this book is available in two formats, 3% inch or 5% inch, for IBM PC, PS2, or compatible machines. Please apply to: Wessex Institute of Technology Ashurst Lodge Ashurst Southampton SO4 2AA UK. Tel: 44(0)703 293223 Fax: 44(0)703 292853 Please send a cheque or money order for £28 (or the equivalent in dollars) and state diskette format required.Table of Contents Why Boundary Elements? .. References Le Chapter 1 Basic Concepts 1.1 Fundamental Concepts 1.2. The Poisson's Equation 1.3 Approximate Solutions :. 14 Weighted Residual Techniques. 1.5 Weak Formulations 16 Boundary and Domain Solutions | 1.7 Concluding Remarks Exercises Le Chapter 2. Potential Problems 2.1 Introduction oe 2.2 Basic Integral Equation | 23° The Boundary Element Method am 24 Computer Code for Potential Problems using ‘Constant Elements (POCONBE) tewaewe B 2.5 -Linear Elements 2.6 Computer Code for Potential Problems sang Linear Elements (POLINBE) RS MER EY 2.7 Discontinuous Elements 2.8 Quadratic and Higher Order 2.9 Computer Code for Potent (POQUABE) 2.10 Computer Code for Muliboundaty Problems (POMCOBE) 2.11 Boundary Elements for Three Dimensional Problems 2.12 Poisson's Equation 2.13 Orthotropy and Anisotropy 2.14 Subregions 2. 2. 2. 2. lements a Problems using Quadratic Elements 15 Helmholtz Equa ion 17 Indirect Formulation 18 Other Approaches for the Treatment of Domain Integrals References ee we ew ew oe Exercises 14 14 7 20 3 37 42 43 45 45 52 37 70 4 87 39 O4 112 18. 123 128 131 133 134 136 137 148 150Chapter 3 153 3.1 Introdu 153, 3.2 Basic Equations of Linear Elasto Se 154 3.3 Fundamental Solutions MYR AS we Sw Bm ae g 160 34 Boundary Integral Formulation =|... NS 3.5 Boundary Element Formulation =... 6 1 1 We eee 172 3.6 Treatment of Domain Integrals and Body Force Terms =... ‘185 4.7 Subregions in Elasticity =. 2. 2... ee 190 38 Indirect Formulations 2... sw ee we Ew 8 193 3.9. Axisymmetric Problems =... tee ee 198 3.10 Anisotropic Elasticity ee ee 203 References . . - Be ee 206 Exercises aE CT SEH ee CK EH B 207 Chapter 4 Two Dimensional Elastostatices. 5 0°. eos 209 4.1 Introduction. . we ee oe 209 4.2 Plate Stretching - Plane St in Problems =. 2 1 1 1 wee 209 4.3, Boundary Element Formulation ee ee 213 44 Constant Element Formulation... se 215 4.3 Elastostatics Code using Constant Elements (ELCONBE) so 220 4.6 Linear Elements. . ary 233 4.7 Quadratic Elements 4, ee ee 238 4% Elastostaties Code using Qua ratie ments (ELQUABE) . . 241 BRR. a Ge we ER Re a we Be 268 Chapter 5 Other Interesting Topics. 5. 6 ee eT $.1 Introduction. Pek ae as 271 5.2 Combination of Boundary and Finite Elements Soe ee 272 $5.3. Approximate Boundary Elements... . ee 276 5.4 Singular Elements for Fracture: Mechanics. tue e we Ye o 278 $5 Steady State Flastodynamics =... 88 References. og 8 8 GFR Be obo: 285 Appendix A: Numerical Integration =. 2 2. ee 87 Appendix B: Diskette Contents. 6. OD Appendix C: References for Further Reading =. . 2... . 294 Appendix D: Answers to Selected Exercises 5. 5. 2... . 307 Subject Index 314Why Boundary Elements? ineers who have been exposed to finite elements may ask themselves why it necessary to produce yet another computational technique. The answer is that finite elements have been proved to be inadequate or inefficient in many engineering applications and what is perhaps more important the method is in many cases cumber- some to use and hence difficult to integrate in Computer Aided Engineering systems. Finite Element Analysis is still a comparatively slow process duc to the need to define or redefine meshes in the piece or domain under study. Boundary elements [1] have emerged as a powerful alternative to finite elements particularly in cases where better accuracy is required due to problems such as stress concentration or where the domain extends to infinity. The most important features of boundary elements however is that it only requires discretization of the surface rather than the volume. Hence boundary element codes are easier to use with existing solid modellers and mesh generators. This advantage is particularly important for designing as the process usually involves a scries of modifications which are more difficult to carry out using finite elements. Meshes can easily be generated and design changes do not require a complete remeshing. This point is illustrated in figure 1 by two views of a turbine blade section, one discretized using a finite element code and the other with boundary elements. Notice the presence of a series of cooling ducts in the blade whose size, position and number have to be reviewed during the design process. Such a variation (es difficulties for finite elements as some cements may easily become distorted or have bad dimension ratios. The boundary element mesh instead is easy to modify. Figure | describes a two dimensional application; these problems are of course compounded for finite clements when working in three dimensions. Boundary clement meshes, especially three dimensional ones can easily be linked to CAE systems as the structure is defined using only the boundary. The jon process is even simpler when using discontinuous elements, which are nol adi ible in finite elements. The mesh shown in figure 2 represents the surface discretization of one eighth of a problem, i.e. a cylinder with a cylindrical perforation across. Notice that the use of elements which sometimes do not meet at corners and are consequently discontinuous in terms of their variables, facilitates the meshing, In addition there is no need to use clements on the planes of symmetry. Figure 3 describes a turbine blade and its base. Notice that discontinuous elements allow for a simple mesh grading. The reason why these clements are possible in boun- dary elements is explained in some of the chapters in this book, From the user’s point of view they offer many advantages in terms of alterations of meshes and general versatility. Figure 4 shows some Von Mises surface stresses concentration at the root of the blade [2].Why Boundary Elements? et. Analysis of a turbine blade using FEM ‘ation in the configuration of cooling clements creates difficulties for the FI code {from a colour original) Figure 2. Cylinder with a cylindrical perforation. The boundary element mesh represents discretization of one eighth of the problemWhy Boundary Elements? 3 More complex three dimensional structures such as the complete crankshaft model shown in figure 5 can be discretized relatively easily using a combination of continuous and discontinuous elements [2]. The model shown in the figure consists of approximately 2,000 surface boundary elements and each throw of the shaft has been represented by a boundary element zone (similar to a finite element substructure or super element) thus making a total of approximately 10,000 degrees of freedom. As only the surface of the shaft has been represented using elements the modelling time is quite rapid and the clement mesh can be auto- matically created from a boundary mode! originating in a CAD system, The solu- tion of problems of this size can nowadays be easily accomplished on the new generation of powerful engincering workstations. (This model was run on an IBM RS6000). This exampie demonstrates that computer time is no longer a primary concern in boundary clement analysis, particularly as it is anticipated further increases in performance for workstations and all other computers within the next few years, The model in figure 5 allows the user to analyse the overall behaviour of the crankshaft under different loadings. Afterwards part of the crankshaft can be Figure 3, Discretization of a turbine blade using discontinuous elements (from a colour original)Why Boundary Elements? Figure 4. Von Mises surface stresses at the root of the blade (from a colour original) Figure 5. Complete crankshaft discretized into boundary elementsWhy Boundary Elements? NOU Figure 7. Wire frame view of crank showing oil passing through the crank6 Why Boundary Elements? studied in more detail to find the effect of oil holes, fillets, etc. on the stress distribution. Figure 6 shows a model for part of a crankshafi discretized with a finer mesh than the one in figure 5. This crank throw has an oil hole as shown in the wire-frame diagram in figure 7. The interesting feature of this problem is that small details can be casily represented without causing a massive increase in the cost of the analysis. Note that the elements describing the oil hole in figure 7 only describe its surface and do not intersect the elements on the exterior part of the model except where the oil hole penetrates the surface. It is evident from these examples that boundary elements are an ideal tool for engin- ering design mainly because it is easy to generate the data required to run a problem and carry out the modifications needed to achieve an optimum design. With computer costs declining while engineers’ time becomes (or should become!) more expensive, the saving in cngincers’ time is of primary importance. (Also, engineers need relief from the dreary task of preparing finite clement data.) More important still, any tool that can shorten the ‘turn around’ time taken by the analysis and design can bring forward the completion date of a project. The future of BEM in enginccring is promising and will continue to be so as long as the developers do not alienate the users by producing codes which arc unreliable or cumbersome to usc. Most of the advantages of BEM are related to its more complex mathematical foundations. This provides a high degree of versatility and accuracy in well-written codes but can have disastrous consequences in the case of poorly written BEM codes. The BEM is more susceptible to crrors when the appropriate numerical techniques arc not uscd and it is then important for developers to understand properly the theory of the method. Although better computational performance is important in BEM, particularly for three dimensional problems, improvements in CPU times should not come at the expense of precision and accuracy. For instance, applying coarse numcrical integration techniques to BEM codes can result in large savings in computer codes and give reasonable results in many cases. For other cases however the solution may be of very poor accuracy or give non-convergent resulis. This makes such codes unreliable. Another important advantage of BEM over FEM is when analysing problems with stress (or flux) concentration. Many such studies have now been carricd out and they tend to demonstrate the high accuracy of boundary clements for problems such as re-cntry corners, stress intensity problems and even fracture mechanics applications. It is not our intention in this introduction to review all these studies but rather to point out the difference in results that can be obtained using one or the other numerical method. As an illustration the finite element solutions found along a line in the neighbourhood of a re-entry corner (figure 8) of a pressure vessel is shown in figure 9. The problem was also analysed, using a photo-clastic model and boundary elements. Besults for a finite clement mesh consisting of approximately 500 degrecs of frccdom (69 clements) and using eight nodes elements are compared against BEM solutions obtained using only 20 elements. It is evident from the figures that while the 69 elements finite element results show lack of equilibrium in the domain as well as on the boundary, reasonably accurate solutions were obtained using boundary elements. It was only8 Why Boundary Elements? » 2 = v s s 20 boundary wlements mosh, 120 degiees of freedom Figure 8 continued when using a very refined finite clement mesh that the FE results were in agree- ment with the boundary clement and photo-elastic model solution i.e for results obtained using 240 elements (and 1,500 degrees of freedom.) Fora full discussion of these results the reader is directed to references [3] to [5]. ‘The development of more powerful hardware specially parallel and vector pro- cessing computers favour the use of BEM. These computers are better suited to deal with the fully populated matrices and the type of operations which are characteristics of boundary elements. The applications shown in figure 10 demonstrate the use of boundary elements for solving a non-linear problem, i.e. the contact analysis of a connecting rod. The model in thi: ‘was gencrated on a CAD system and the geometry automatically passed to the BEASY system and meshed. Notice that only: one quarter of the rod needs to be discretized due to symmetry. The solution of this analysis is shown in figure 11 where contact sur- face gap between the pin and connecting rod is clearly seen. Problems other than stress or temperature analysis can be solved using boundary elements. Typical applications include torsion, diffusion, seepage, Nuid flow and electrostatics. Corrosion engineers have used the method to design better cathodic protection systems for offshore structures, ships and pipelines. Many of these structures are ally three dimensional and the region of interest extends. to infinity. Consequently they could not be cffectively analysed before the develop- ment of boundary clements. Early attempts to use finite differences or finite elements to solve these problems met little success, For these cases the computer model has to represent the potential field around the structure, representing the shiclding effect of the structural geometry aind the effect of the different materials involved. Unlike a structural model the cathodic protection model is concerned with the seawater around the structure and the interface between the seawater structure. Hence the use of FEM to analyse the problem would requi subdivision of the seawater surrounding the structures which is a Herculean task.Why Boundary Elements? ‘uy 241 Bue a ea ss01) iouud wins ‘Rsau)8 jediouyd WrUNURY ~- 841 Buoye sessans dooH ‘Astpunog ns jediouud xen ‘ soon ms aap _ = = a a . . , . i wf g i ad g 2 q or £ OF ‘suowo19 OZ-Waa ~---— unwns Ove Wad sauaware 69-Had eo siuewel@ Oz-wae <== SWaWOIO OFT NBL em -uouis}® 69-3) —o— g candy ui pareotput dul] indino ssans ay) Buoe synsol WAG Pue WIA Jo uosuedWOD “6 ainB_y swuoway9 O2-W3B siuawes ove Wad Om ‘UAW 69-134 ome (089) ssong peoroung wnunneyy,Figure 10. Quarter mode! of a connecting rod Figure 11. Final contact configuration after load application showiThe CONOCO Hutton TLP (Tension Leg Platform) Figure 12.12 Why Boundary Elements? Figure 14. Model of the TLP showing contours of potentials The use of boundary clement method represents the only practical solution for this problem. The advantage of the method is that only the structure needs to be defined as the BEM automatically takes care of the field « i.c. the scawater extending to infinity. Figure 12 shows the first three dimensional BEM cathodic protection application which was the study of the tension leg platform (TLP) built by CONOCO in the Hutton Field in the North Sea. Figure 13 shows the discretization of a quarter of the structure into the boundary elements used in the analysis and figure 14 the results obtained for the potentials on the surface for a particular configuration of the improved code system used. Since then the boundary element method has become the key to the successful and practical analysis of cathodic protection systems and further work has been carried out in this regard particularly at the Computational Mechanics Institute, Southampton, UK. A system is now available which allows the corrosion engineer to evaluate design options, look at problem areas, interpret experimental observations, optimize the design and predict with accuracy and confidence the degree of protection and life expectancy of a cathodie protection system. The advances made in cathodic protection modelling using boundary clements are just one of the apy tions of the technique for systems extending to infinity. The method is nowadays extensively used in other problems with infinite or semi-infinite domains such as those occurring in geomechanii “an engineering, foundations, aerodynamics, flow through porous media and many others. This brief introduction has attempted to point out the advantages of BEM for a wide variety of engincering problems and the reason why the method should be taught on an undergraduate as well as a graduate level. University courses should include the fundamentals of the method and provide workshops on applications while short courses with hands-on applications will help to bring the method to the attention of practising engineers. This book has been written to provide aWhy Boundary Elements? 13 simple ind up to date introduction to the method to help popularize the technique nongst en; The future of BEM hinges on its acceptance by practising engineers, in particular as a design tool. Developers should aim to make the method more accessible to engineers by writing codes which are easy to use and by explaining the fundamentals of the method on the basis of engineering rather than mathematical concepts. This book has been written in a form that can be used as at textbook at undergraduate or graduate level and for the engineer in practice aunts to learn the fundamentals of the technique unaided. Of particular the way in which the mathematics concepts are introduced and immediately applied in simple computer codes. These codes (4 for potential and 2 for elasticity) will facilitate the comprehension of BEM. This book is based on the authors’ many years experience as researchers and hers of boundary elements. It is designed to teach in the most eflective manner the fundamentals of the method rather than to attempt to demonstrate erudition ‘on the subject. Many topics have been deliberately omitted to avoid confusing the reader. The essentials however are all here. It is now left to the reader to build knowledge. References U1] Brebbia, C. A. J. Telles and L. Wrobel, Boundary Element Techniques ~ Theory and Applications in Engineering, Springer-Verlag, Berlin and NY, 1984. [2] Trevelyan, J. BEASY: Self-Teaching Guide. Computational Mechanics Publications, Southampton and Boston, 1991 [3] Floyd, C. G. The Determination of Stress using a Combined Theoretical and Experi- mental Analysis Approach, Computational Methods and Experimental Measurements, Prac. 2ud Int. Conf. Jute Sul 1984 (C. A. Brebbia, Fd.) Springer-Verlag, Berlin and . Southampton and Boston, 2984. lement Progedures— on Mesh Selection, Structures, 21, 257-264, 1985, ©. A. and J. Trevelyan, On the Accuracy and Convergence of Boundary Element Results for the Floyd Pressure Vessel Problem, Technical Note, Computers und Structures, 24, 513 $16, 1986,Chapter I Basic Concepts 1.1 Fundamental concepts Consider a very simple differential cquation applying in a one-dimensional domain x, from x=0 to x= 1, @u N+ b=0 inx a.) dx? w is the function which governs the equation and we usually nced to find it using a numerical technique which gives an approximate solution. 22 is a known positive constant and b is a known function of x. The solution of equation (1.1) can be found by assuming a va n for consisting of a scrics of known shapes (or functions) multiplicd by unknown coefficients. These coefficients can then be found by forcing (1.1) to be a series of points. This is the basis of the collocation (or point colle precise) method and is essentially what one docs when using finite differences. In finite elements instead the solution is found using the concept of distribution of error within the domain. This is somewhat a process of ‘smoothing’ and it is then not surprising that finite clement solutions tend to have tess ‘noise’ than finite difference ones. The concept of distribution or weighting of a differential cquation is not only valid for approximate solutions but it is a fundamental mathematical concept, which can be used in countless enginecring applications. Engineers for instance are very familiar with the principle of virtual work which is usually formulated in terms of work done by internal and external forces. They are usually unaware however that the first ‘demonstration’ of the principle was proposed by Lagrange using the concepts of distributions, applying what are now called the *Lagrangi: multipliers. These concepts are also cssential to study the behaviour of the differential equations, and in particular the type of boundary conditions they require and which arc consistent with them. To understand what these concepts mean before proposing any approximation, one can consider another function wv, arbitrary except for being continuous in the domain x and whose derivatives are continuous up (oa required degree (the degree of continuity will vary with the problem as will be shown shortly). One can now multiply the whole of equation (1.1) by this w function and integrate on the domain x as follows: 1 fat (ee o}was a (1.2)1.1, Fundamental Concepts 15 This operation is called an ‘inner’ product in mathematics and although does not imply any new concepts, allows us to investigate the propertics of the governing equation. This is done by integrating by parts terms with derivatives in the above expression. In this case one can only ‘manipulate’ in this manner the first term, ie. Pufdx?, which gives : ! fea atu tyebas[ v[=0 (ay Notice that the integration by parts has produced two terms. one in the domain with first derivatives of u and w, and the other on the boundaries (which in this case are simply the two points x =0, x= 1). . Furthermore, if the w function has sufficient degree of continuity one can integrate by parts again to obtain Lt ew : 7 } {u mn +20 ow dxt [‘ «[ - [« “lt 0 (14) Expression (1.4) is of course equivalent to (1.3) but here not only has one passed all derivatives to the newly defined w function but the two terms at x =0 and x= 1 give us an insight into the boundary conditions required to solve the problem. In th du or needs to be known at x =0 and x dx 1 (LS) Notice that the w function which in principle was an arbitrary function with a‘certain degree of continuity can be made to satisfy certain boundary conditions if pne wishes to do so. In the principle of virtual displacements for instance, arbitrary functions of this type are defined as virtual displacements but they are assumed to satisfy the homogencous version of the displacement boundary conditions. ic. they are set identically to zero al any points where the displacements arc prescribed. even if those displacements (represented by 1) are not set to zero, ic. w= 0 on the parts of the boundary where w is given. This is done in order to climinate du ecg oe * . terms of the type 1 w | which give rise to a type of work’ one docs not wish to ds have. In general however one can. assume that w and dw/dx. can have values different from zero on the boundaries and this makes expression (1.4) more general. The concept of an arbitrary function w used as a distribution function is retated not only to virtual functions and consequently to virtual work but also to the idea of Lagrangian mullipliers. These are functions of the w type defined in order to satisfy certain cquations. They will be defined better in what follows. Although equation (1.4) gives the user an insight into the type of boundary conditions required to solve the problem, these conditions have not yet been itly incorporated into the problem. In order to do so let us consider that thelo Chapter | Basie Concepts boundary condition: as follows: uit atx=0 tee eet (1.6) a eg atx nd where the derivatives of u are now defined as q and the terms with bars represent known values of the function and its derivatives. [is usual to call the first type of conditions in (1.6) ‘essential and those like q involving derivatives as ‘natural’ Substituting those values into (1.4) gives Lf dw iu wet (u- ont dx + (Law],-1 — (awd o} dx? fer] bet J ln It is now interesting to try to return to the original expression (1.2) by integrating by parts again, but this time passing the derivatives for w to u. The first integration gives, Lf dudw dw -~ + (2 — byw dt if degre" ms} ” +f L, -[¥] + (aw). - Law, [wit +[ ds Jeo ax dea Notice that only the term in fu dwjdx], disappears, Furthermore the following expression results alter carrying out a sccond integration, Lae i {! "4 nH ow} dx -[ | + [‘ | = [« “| a lax dx ey Lae bee Lav deo + Low). =) - [aw] .-0 + [a “l 0 9) Ix Jeo Once again only one term disappears, in this case [gw], - Notice that 4q = dujdx as defined cartier. - Grouping the terms together one now arrives at an interesting expression, different from the original formula (1.4) ic. fd dw if’ "oe ube de la alee + =a | =0 (1.10) dx o dx Jeuo This expression implies that one is trying to enforce not only satisfaction of the1.2. The Poisson's Equation 17 differential equation in x but the two boundary conditions. The w and dw/dx functions can be seen as Lagrangian multipliers. Furthermore nothing has yet been said about approximations; the above expressions are valid for exact solutions as well. In other words the procedure describes a general tool for the investigation of differential equations. 1.2. The Poisson's Equation ‘An important equation in engineering analysis is the so-called Poisson equation which for wo dimensions can be written as ay ay out inQ tt) ext e or Vu=b inQ (1.12) ay where V7()= Ox , is called the Laplace operator. x, and x, are the two np coordinates and b is a known funetion of x1, x2. @ is the domain on which the equation applies and is assumed to be bounded by I. The outward normal to the boundary is defined as a (figure 1.1). The Poisson equation or its homogeneous form (i.e. b = 0) which is the Laplace equation, governs many types of engincering problems, such as seepage and aquifer analysis, heat conduction, diffusion processes, torsion, uid motion and others. Consequently it is a very important equation in engineering analysis. 2 Figure 1.1 Domain under Consideration for Poisson Equation Basic Definitions18 Chapter 1 Basic Concepts Here one can also introduce the idea of multiplying equation (1.12) by an arbitrary w function, continuous up to the second derivative. This gives. § (Vu —bywdQ=0 (1.13) a Integrating by parts the terms in x, and x, gives 1(-2 pee Ot) af wd? =0 (14) a\” ax, dx, Ox, Ox In this case the integration by parts of the two terms produces the derivative of u with respect to the normal, ic. u/én which will later on be called q, i.e. q=dujan. Integrating by parts again, one obtains, ay ‘ "am bw) anf weal fa Mar =o (1.15) . ax} or {for ubw} das fe wal — fui iat = 0 (1.16) a ron Expression (1.16) is equal to (1.13) and hence one can write, {oraw dQ= {nw dns gi warm fu Ba "ur =0 (1.17) s of the where the tcrm in b has been climinated as it appears on the two s equation. Equation (1.17) can a theorem, i.e. so be expressed in the form known as the Green's a aw { {(02upw — (W2wyr} dQ =f (™ wou’ *) dr (118) a 1 \en en Although this theorem is in many cases given as the starting point for many engincering applications, including boundary clement formulations, it is much more illuminating to use the concept of distribution as it illustrates the degree of continuity required of the functions and the importance of the accurate treatment of the boundary conditions. In this regard let us now consider that the F boundary of the 2 domain under study is divided into two paris, [, and P (7 =P, +1) such that, wait on Ty (tg on Ty !1.2. The Poisson's Equation 19 Hence equation (1.16) can now be written as, § {V4 yu ~ hav} dD a + fqwdl +f qwat— falar fu ar=o (1.20) *, te fs aT Once again one can integrate by parts to retrieve the original Laplacian V7u in order to sce how the importance of the boundary conditions affect the equation. Integrating by parts once we have, ow eu ew eu f3- - ~ bw) dQ al ex, dx, Oxy ex, ew af uae + fqwdt sf qwar— fa Warf ar=0 (12H ten ty i Btw a en One can split the first integral on IT into two terms (one on T, and the other on F,), the second of which can be cancelled with the last integral in (1.21). This gives ow w on a + fo" war + § qwat+ § qwar— f ao" ar =o (1.22) nh n Pr ron Integrating again by parts the following expression is obtained { LV — bw} dQ oe —fugart faut f gwar f gwar i fen ff fe ew -ja, ar=0 (1.23) btm The first integral in F can again be written as a summation of two integrals, one on [, and the other on [. The one on F, can be cancelled with the integral on Ty of qw in the above equation, This gives J leva — pw} d= f wg at fdr + f gwar a Pr Tew Tr ow - fa" ar=o0 (24) men20 Chapter | Basic Concepts This f Ja can be written as, J MV2u~ bw} dO f Q—awdl + f wa" ar =0 (1.25) a h 4 en Once again this expression shows that one is trying to satisfy a differential equation in the doma in plus two types of boundary conditions, the “essential” conditions w= ion F, plus the ‘natural’ conditions ¢ = q on T}. This is, very much what been shown in equation (1.10) with the only exception that the sign of the last term is different in both expressions. This is because in (1.10) the derivatives were taken with respect Lo x rather than with respect to the normal, as they are now. 1.3 Approximate Solutions Although the previous sections have introduced the concept of distributions. the formulations apply irrespective of the type of solution one find: they are valid for exact as well as approximate solutions. This section however will investigate what happens when the concept of an approximate solution is introduced in the formulation, In engincering practice the exact solution can only be known in a few simple cases and it is hence important to sce how the solution behaves when onc introduces an approximation. Let us consider now that the function w defines an approximate rather than the exact solution. In this case one ean write for instance, U= 2,9, + 22924... (1.26) where 3, are unknown coefficients and the ¢, are a set of linearly independent functions which are known, x, are generalized coefficients although in some eases they can be associated with nodal values of the variable under consideration. In general in engineering problems, one prefers to use nodal values as they have a clear physical meaning and this is done in finite elements, finite differences or the boundary clement method. In such cases the approximation for w can be written as w=) +d... x L 16; (1.27) where ¢p, are a set of linearly independent functions which are sometimes called interpolation functions. u, are the nodal values of the field variuble or its derivative (or more generally the nodal value of any variable with physical meaning directly related to w or its derivatives). Introducing the upproximation for w into the governing, differential equation one finds that the equition is no longer identically satisfied except for the case in which (1.26) or (1.27) can represent the exact solution. This produces an etror or residual function which will soon be defined.1.3, Approximate Solutions a For instance, introducing an approximate value of w into equation (1.1) one generally finds that Pu iat iu-b#0 in (1.28) ds: The same will generally occur with the boundary conditions corresponding to this equation, ic. u-it40 =0 (1.29) q-a40 0 atxeal One can now introduce the concept of an error function or residual which represents the errors occurring in the domain or on the boundary due to non- satisfaction of the above equations, The error function in the domain is called R and is given by 2 ee (1.30) de and on the boundary one has. Rysu-a and (31) R,=4q-G Although the above case is a particular and relatively simple equation the same occurs for any other problem. (one considers the Poisson’s equation (1.12) for instance, the error function in the domain is R=Vn-b ind (1.32) and the errors for the boundary conditions (equation (1.19)) are defined by Ry=u-H ont, 033) R,=q-§ ont, . The numerical methods used in engineering try to reduce these errors to a minimum by applying different techniques. This reduction is carried out by forcing the errors to be zero at certain points, regions or in a mean sense. This operation can be generally interpreted as distributing these errors. The way in which this distribution is carried out produces different types of error distribution techniques which, in generat, force the integrals of the residuals weighted by a certain function to be zero, Because of this they are called weighted residual techniques.2 Chapter 1 Basic Concepts 1.4 Weighted Residual Techniques The solution of the boundary value problem defined by equations (1.28) and (1.29), (1.32) and (1.33) or similar sets for other problems can be altempted by choosing an approximation for the function u. One can then have three types of method: () If the assumed approximate solution identically satisfies all boundary conditions but not the governing cquations in Q. one has a purely ‘domain’ method. (ii) If the approximate solution satisfies the ficld or governing equations but not the boundary conditions one ‘boundary" method. (iii) If the assumed solution sa neither the field equation nor the boundary conditions, one has a method. Let us first assume that the functions 4, which are defined to approximate a. satisfy all boundary conditions. One then has a residual R function in the domain as the field equations arc gencrally not identically satisfied. The idea is now 10 make R as small as possible by sctling its weighted residual equal to zero for various values of the weighting functions, y,, such that saN “(.44) fRyjdQ=0 ind a These functions have to be linearly independent. Notice that another way of writing (1.34) in a form that is more compact and casy (o operate with, is by defining a new function w, such that w= Bid + Bast Baba = By (1.35) where are arbitrary coefficients. Hence equation (1.34) can now be written in a more compact form as, [XvdQ=0 in 1.36) a Different types of weighting functions y, (or w) will define different approxim: methods. Equation (1.34) or (1.35) will produce a system of algebraic equati from which the unknown values of the or u, cocfficients used in w (equ: (1.26) or (1.27) can be obtained. The approximation can always be improved by increasing the number of N functions used. (N is the number of terms in the approximate solution equal to the number of weighting functions required.) Approximate methods based on equation (1.36) are called weighted residual methods and, given an approximate solution, the method will vary in accordance with the functions used as weighting functions. In what follows a few will be reviewed. ion1.4, Weighted Residual Techniques 23 (i) Subdomain Collocation For this method the domain © is divided in M subdomains and the integral of the error in each of them isset to zero. The weighted functions are simply chosen as. 1 forxeQ, = 1.37) ws { for x40; ( € indicates belonging to and Q, is the F subdomain). Equation (1.34) becomes, PRdx=0, FHbLA.LAN (1,38) a, (ii) Galerkin Method In the case of Galerkin's method the weighting functions are the same as the appoximating functions, i.c. b=; (139) hence equation (1.34) becomes, [Rp dQ=0 j=l... (1.40) a Using the same definition as in (1.35) this can be written as, f RwdQ=0, (4h a with, w= Bibi + Babe +... + Buby (142) This method is the starting point of many finite clement formulations for which the symmetry of ¢,=, coupled to inherently symmetric ficld cquations, lead to mmetric algebraic matrices. : Point Collocation Method: In this case N points x,.x, are chosen in the domain and the residual is set (0 zero at these points This operation can be interpreted as defining weighting functions in terms of Dirac det .c. Wp Mw— xk F= 12, (143)4 Chapter | Basic Concepts A(x —x,) at point x — x; has an infinite value but is such thal its integral gives unity, ic. fAw— x) dos; a aN (1.44) The Dirac function can be interpreted as the limit of a regular function when its base tends to.zero. Hence equation (1.34) can now be written as, {RAtx— x) d=, GH=hQN (1.45) which simply says that the error function is zero at a series of points, ic. Rh 0; teks coe N (1.46) The method consists of setting the residual or error function equal to zero at as many points as there are unknown coefficients in the approximate solution. The distribution of the collocation points is in principle arbitrary, but in practice better results are obtained if they are uniformly distributed. Example 11 As an illustration of how to use weighted residuals, consider the following differential or field equation in the one dimensional domain x (where x varies from Otox=1) ie. gaten0 @) with homogeneous bound:ry conditions, i.e. u=Oatx=Oand x= (b) (Notice that equation (a) is a particular case of equation (1.1) when 2=0 and b=-x) The exact solution of (a) can be found by integration and gives, iC) Moxact = 6 6 6 Let us now attempt to solve (a) using the weighted residual techniques described above, starting by defining an approximate solution which satisfies the boundary conditions and can be written as u= 41), + 2224.06 {d)1.4, Weighted Residual Techniques 25 ‘One can use Hermitian polynomials for , but since only two of them satisfy the homogeneous boundary conditions, only these two will be used, i.e. way + t2p2 (ce) where byex-2te? , © cy The residual or error function in this case is obtained by substituting (e) into equation (a) which gives, ay Riiyad= at x 2, 2, way Og gh O24 dx? dx? = 2, (6x — 4) + 2,(6x —2) +x (g) Let us now. reduce (g) using the various techniques previously described. (i) Subdomain Collocation sider the domain divided into 2 equal parts, one from 0 to 3 and the other } to L In this case one can write, uw 1 J Rdx= J [x\(6x—4) + a2(6x—2)+ x]Jdx=0 @ and ‘ (h) 1 J Rdx= § [ay(6x—4)+ «(6x —2) +x] dx=0 oa Oe which produce the following system of equations ~ 1.21, —0.25a, +0.125 =0 @) 0.25a, + 1.20, +0.375=0 from which one can obtain, @ Substituting (j) into (c) gives the following result xo WR om (k wal = 6 {k) Notice’ that the exact solution (c) has been obtained since the assumed shapes of wate able to represent it.26 Chapter | Basic Concepts (ii) Galerkin Tn this case the weighting functions are, vad baad and the weighted residual expressions are 4 Lan(6x—4) + ap(6x—2) +x1x— 2x? +.x°) dx =0 @ w J fay(6x—4) + a,(6x—2) + x] —x7)dx=0 @ which produces the following algebraic equations in a, and 2. 4a, +0,+1=0 a, — 4a, —1.5=0 m) This also results in a a =o = Here one forces the residual to be zero at a series of points. Consider in this ¢: that R is zero at the two points x =0.25 and x = 0.75. This gives Rhwo.2s = — 10a, — 2a, +1=0 Rhewo.25 = 221 + 104, +3=0 (n) with the same results for a, and a3, ie. m= a=} Notice that this case is rather trivial and the same results have been obtained for all the methods. In general this will not be true when the exact solution canngt be reproduced by the proposed value of w and’ one will find different results depending on the method used. Example 1.2 Let us now study another equation using point collocation such that in this case we will obtain an approximate rather than the exact solution.1.4. Weighted Residual Techniques 2 Consider the equation (1.1), with 2? = | and x = —b, ie. @u ge thts fa) and the homogencous boundary conditions, w=0 al x =0 and x = 1. The exact solution of (a) can be easily obtained by integration and gives aX (b) sin Instead of using (b) we will try to approximate it defining a solution u= 4,9, +4362 + dyhy t.-- (c) where the #, are terms of a polynomial in x, ic. «dy In order to satisfy the boundary conditions exactly, equation (c) has to give, u=Qatx=Oandx=1 te) which implies that, alx=Osu=a,=0 0 atx=l—usa,t+a,tayt.. Hence a, = Oand a, can be expressed in function of the other a, parameter. ic. a, = (ay tag...) (g) Substituting a, =0 and (g) into (c) one can write, = ay(x? — x) + ag? — x) + as(x* — x) +... (h) = x(1—x)(—ay — a4) + x(1 —x)(—ag)x +... Defining now a new set of unknown parameters 2, such that, a= —dy— ay; yy... i) one can write, ua=x(l— xa, +aaxt...) (i)2B Chapter | Basie Concepts This function satisfies the boundary conditions in w and has the degree of continuity required by the derivatives in equation (a), hence i ‘admissible’, We will itlso see that the “distance” between the approxi exact solution decreases when the number of terms in (j) increases and this implies that the approximate formulation « is ‘complete’, i.e. tends to represent the exact solution better and better when the number of terms increases. In order to apply the point collocation technique we will restrict ourselves to two terms in the (j) expression, ic. w= x(1 = x), + 23x) tk) Substituting this function into the governing eq! residual, i.e. ion (a) one finds the following R aay xe(—24 0-22 y+ (2-60 4.2 pay tx mn dx =0 at two points, say x=} and rac delta functions applied at these Collocation can now be interpreted as setting This can also be expressed in terms of two points, i.e. the weighting function is, w= BA = 4) + B2Adle— (m) The weighted residual integrals are represented by 1 JRwdx=0 (n) a or simply, R=0 atx=Sandx (0) Substituting these values of x into (I) one obtains two equations in 2, und 2, They can be written in matrix form as follows, [ : “eNe-th () The solution of this system gives yah ment (a) The approximate value of w ~ equation (k) ~ can now be written as, x(h— x) = (42 + 40% u 27 (42 + 40x) (r)1.4, Weighted Residual Techniques 29 Table 1.1 Results for Point Collocation 0.019078 ~0,009953 0.052258 +0.002027 0.071428 +0.00000 0.065585 0.065806 0.024884 0.030901 0.032350 0.081474 Notice that the error function can now also be fully defined in terms of x, by substituting 2, and a into (I), This gives, R= yh, (—4 + 19x — 2x? — 40x3) (s) These results can be tabulited in table 1.1 where they are compared ay exact solution for u. Notice that the values of R are identically zero at but that this does not mean that the solution for u is exact at those points. Example 13 Let us apply Galerkin’s technique to equation (1,1) for which 2? = | and b= —x with homogeneous boundary conditions t The approximate solution will be the same as in example 1.2., ie. = ayx(l—x) + ax7(1 — x) (a) which cin be written as. w= ay +02 () where , and @2 are the shape functions (, =x(1—x); 62 =x2(1—x)). The residual is the same as previously, i.c. eu +utK dx? () =(-24+x- x2), + (2-6x $07 — pa tx The weighting function w in Galerkin is assumed to have the same shape function as the approximate solution (b), i.c. w= Bid, + Biba (d) The coefficients ff, and fy are arbitrary.30 Chapter | Basic Concepts The weighted residual statement is, , JRwdx=0 (c) a which produces two integral expressions as 8, and fi, are arbitrary, ic. : J RU. + Baba) dx =0 © or simply, fro, dx=0 and Rd: dx=0 (g) ° a Substituting (c) and the functions 4, and ¢3 into (g) gives \ JU —24x—4%)ay + (2— 6x +0? — on + ILE — xy =O ‘t o hy $u-24x-ay + (2-6x +x? = 3a, +x] L2(1 — x] dx =0 a After integration this gives the following system i # iet- {3} w io stlad Ue. Notice that the matrix is symmetric because the equation is of an even arder and the approximate and weighting functions are the same. Solving (i) gives % = 3o5i = ah w Substituting these values into (a) produces the approximate solution for u, ue x(1 — xs + ax) (ky ‘One can also find the residual function R (equation ¢) which is now R= shy (— 16 + 62x — 8x? — 63x4) The results for u and R are given in table 1.2 where they are compared against the exact solution of u. Notice that although the solution is overall more accurate than in the case of using the collocation technique, one now needs to carry out some integrations as shown in formula (h). This operation was not required for the case of point colloc:1.5. Weak Formulations 3t Table 1.2 Results for Galerkin u (exact) u (appr 0.018641 0.018853 — 0.026945 0.051194 0.051162 +0,000485 0.069746 0.069444 +0.013888 0.065582 0.065505 +0.005070 0.030901 0.031146 0.034165 1.5 Weak Formulations The fundamental integral statements of the boundary element and the finite element methods can be interpreted as a combination of a weighted residual statement and a process of integration by parts that reduces or ‘weakens’ the order of the continuity required for the u function. If one returns to equation (1.12) with b =0 for simp! Vu=0 inQ (1.47) one can write formula (1.25) as, . J(v2ujwdQ— f (g—aywdl + f (wa) ar=0 (1.48) a te i én or in terms of residual functions, cw J RwdQ— J Rwdr + [RR ar=0 (1.49) a Vy tm en A special case of this equation is the case for which the function # exactly satisfies the ‘essential’ boundary conditions, u= a on T,, which results in R, =0. In this case equation (1.49) becomes f RwdQ= f Rw (1.50) a re or, few dQ = f (q— Gwar (st) i te A more usual form of this expression can be obtained by integrating by parts once which gives wep (cu Ow | eu aw a )an=—f awa § aw ar (1.52) Ox, AX, f i ox,2 Chapter | Basie Concepts It should be pointed out that equation (1.52) could alse be obt integrating by parts over the domain the weighted residual then introducing the boundary conditions, i.e. starting with I by inent for V2u and S(Vupw dQ=0 (1.53) a one can integrate by parts once to produce the following expression, ou dw Ou a a 7 (i wy ee Jans [war <0 (1.54) a ren ax, xy” ax, ax, Introducing then the corresponding boundary cone in equation (1.52). The last term in equation (1.52) is usually forced 10 be identically equal to zero by the requirement that the w functions have to satisfy the Lagrangian version of the essential boundary conditions, or condition on I, i.e. w=0 on T). This gives a relationship well known in finite elements, i.e. (= ew Cu ew a €x, Oxy Oxy Ary ions in T (P=, +P) results n= § geal (1.55) Py Fquation (1.55) is usually interpreted in terms of virtual work or virtual power, by associating w with a virtual function. Notice that the integral on the left hand side is a measure of the internal virtual work and the one on the right the virtual work done by the external forces q. Equation (1.55) is the starting point of most finite element schemes for Laplacian problems and is usually called a ‘weak’ variational formulation. The ‘weakness’ can be interpreted as due to two reasons, (i) the order of u function continuity has been reduced as its derivatives are now of a lower order (i.c. first rather than second order); (ii) satisfaction of the natural boundary conditions is done in an approximate rather than exact manner, which reduces the accuracy of boundary values of this variable. (Notice that R, is generally different from zero.) The boundary clement formulation can be interpreted as introducing a further formal step in the process of integration by parts on the derivatives of u, and consequently weakening the continuity requirements for u. Ione starts again from equation (1.48) and integrates by parts as before, the more complete expression obtained is as follow: (# ow | Ou Ow a PxyPxy AX, Any, aw \an ~§ qwar—f qwat— fw— ay ar * i ‘ en (1.56) Integrating again in order to eliminate all derivatives in u on the left hand side integral, one finds, ae aw fv dQ =—f quar fqwars fa Mars far 57) a i hy a a1,5, Weak Formulations 33 ‘This is the starting statement for the Boundary Flement formulation of the Laplace equation, The same equation can be obtained starting from the integral of the weighted residual over the domain Q (equation (1.53), integrating by parts twice and thea introducing the boundary conditions. The processes hav@itready been shown from another field equation in formulae (1.13) to (1.16) and then (1.19) and (1.20), the only difference now being that b is zero. Consider now equation (1.1) again to illustrate how a weak formulation can be used and the domain and boundary clement statements are obtained. Let us n (1.10) which was deduced from (1.1) by a process of integrations is and application of boundary conditions, i.e. ifn: wt Bu oe ax —[q—GWens + [« -a Mast dw =0 (1.58) | no ase 0 which can also be expressed in a more compact form in function of residuals, i.e. { rw dv —[Raw]yey +[e, *| 0 (1.59) l.-0 The function « will now be assumed to satisfy exactly the ‘essential’ boundary conditions = fi at x =0. In this case (1.58) becomes, 1 ie we (nye |ac= l@-Dw.-y (1,60) d olds or in terms of (1.59), simply 1 J Rwde = [Rew], (1.61) Integrating by parts equation (1.60) one can write ' , j { a iM +(2u— iow dx = [qv] a0 —[aw]eot (1.62) I the weighting function w is forced to satisfy the homogeneous version of the essential boundary conditions at x =0, equation (1.62) becomes, ' j {i WY atu ot dx= al dxdx [aw].=1 (1.63) which is analogous to equation (1.55) obtained for the Laplace field equation. Notice that equation (1.59) can also be obtained by applying the boundary conditions into statement (1,3) and that this statement was simply obtained by integrating by parts weighted residual expression (1.2). ‘34 Chapter 1 Basic Concepts The Boundary Element type governing statement for the example under discussion is found by carrying out two consecutive integrations by parts of (1.10) and this gives the previously obtained formula (1.7), ie. 7 if a+ aon} d+ {Cawheot — Lawleno! af ot dx’ aw ae = =0 [ =|] on {| oll yo dx. This expression could also have been obtained by carrying out a double integration by parts of the weighted residual equation (1.2) and applying afterwards the boundary conditions. [t is worth noting that both in this one dimensional example and the two dimensional Laplace equations, a Finite Element type statement has been obtained afier the first integration by parts (cquations (1,3) and (1.14)), and Boundary Element type integral equation after the second integration (equations (1.4) (1.15). Example 1.4 In order to understand the effect of weak formulations on the satisfaction of boundary conditions, we will now consider again cquation (1.1) but assume that the boundary conditions are of two types, i.e. atx=O07u=0 (essential® condition) fa atx=l—q == @ ~~ Cratural’ condition) ) Ix The expression previously used for the approximate values of w can not now be applied as the boundary conditions are different. Let us consider again the starting expression, st agxt asx? +... (b) and satisfy exactly the essential condition, at x = aix=0; u=a,=0 ©) but not the natural condition. Hence the approximate solution is now, W=Oyx +aQx? +... (d) where @ =d3,4;=4)...1.5, Weak Formulations 35 The residual will now be different from the one in the previous examples, i.e. du R= dx? tutxaayxta(2+x)tx (e) The weighted residual statement has to include now the natural boundary condition R, residual which is not identically satislied, i.c. 1 J Rwdx =[Row], oy ) ° or in expanded form, Lat {(Sitetx)eds= aah @) 0 \dx? ‘One can now solve equation (g) in its present form or reduce the order of deriv: in the domain and the number of terms on the right hand side by inte parts the du/dx? term. This gives 1 {du di i Fe 4 @ and w= Pid + Baba dex and b= where Substituting these values into (h) one finds, {foie} en ee) (hy +22 + MBG, + pal} dx =(4Bids + BrbaNe=s w or ee {ety + 2aaxMbby + Bax) — (4x tax? + xB LN + fax?) dv = (4. +82) tk)36 Chapter 1 Basic Concepts As the fi, «and fi terms are arbitrary this implies satisfaction of the following two equations, 1 J fla, + 22x) — (ar? tae 4x2 dxe=J a) a 1 J f2ayx + 2atgx?) — (ax? + ant + Yd =| a Integrating the above equations and writing the results in matrix form one finds, fd fash [} teas ad se The values of a, and a are, 4, = 0.9859 + 1,9864q = (n) a = —0.4319 — 0.43229 Notice that an error will now appear when we try to compute the value of q at xelie, [4] = my + 2ee, = 0.1221 + 1.1229 {o) dx fees and hence this value will never be equal to the applied 4, ie. q=0.1221 + 1.12294 q (p) This peculiar result is characteristic of weak formulations such as those used in finite clements, Because of thi i atisfaction of the natural boundary conditions, fe. solutions nd to give poor results for surface fluxes or tractions. The resulting errors in many cases ‘pollute’ the results to such an extent that the finite element solutions are unreliable for many cases of stress or flux concentration except when using very fine meshes. Results for u and R are given in table 1.3 for the case in which G =0. The exact solution sinx Cost (q@) ‘Table 1.3 Results for Weak Formulation and Galerkin Method (exact) u (approximate R 0.084773 0.094271 — 0.669519 0.246953 0.256899 = 0,306901 0.387328 0.384975 0.021175 0.492328 0.478499 031 9 0.549794 0.537471 0.5736711.6. Boundary and Domain Solutions: 7 1.6 Boundary and Domain Solutions In section 4 the weighted residual technique was classified into boundary, domain and mixed methods. Boundary methods were defined as those for which the assumed approximate solution satisfies the governing or field equation in such away that the only unknowns of the problem remain on the boundary. The satisfaction of the field equation may be of its homogencous form or a special form with a singular right hand side. In the process of double integration described earlier one had transferred the derivatives of the approximate solution u to the weighting function w and so the conditions previously imposed on the former apply now to the latter. A boundary method can be obtained by choosing a weighting function w in either of the following two ways, ie. (i) By selecting a function w which satisfies the governing equation in its homogencous form, or (ii) By using special types of functions which satisfy those equations in a way that it is still possible to reduce the problems to the boundary only. The best known of the functions applied as right hand side of the equation in the second method are the Dirac delta functions which give simply a value ata point when integrated over the domain. It is important however to realize that other functions could also be proposed and may be very appropriate for other cases, provided that they can be reduced to the boundary. We will now apply both techniques to our simple equation (1.1), ie. au dx? + Hu —b(x)=0 (1.65) or its weighted residual statement, WT fatwa du}! fF dw]! Pw | — bw | ds re = 66) i[(ee +4 ") | n+ [few al 0 (1.66) The first approuch implies that a solution is known such that ow two (167) without taking into account the actual boundary conditions of the problem. Hence statement (1.66) reduces to ~fowass [tw] fo *|- 0 (1.68) ix” |, [dx This approach is associated with the method called Trefftz,38 Chapter | Basic Concepts The second approach is based on a function w such that ow +h dx? 4, (1.69) where 4, indicates the Dirac function such that age singular at the x, point with f A,dx=1 A, (1.70) =0 at any other point Notice that in this case 2 i[(+ iw) [ox fu, dx = —u, am a where u; represents the value of the function u at the point x,. In this case equation (1.66) becomes, 1 du}! dw]! ~jowans[ | -[«t]-0 (1.72) When the x; point is chosen on the boundary, then equation (1.72) gives a relationship between boundary variables. The secoud approach is the one usually applicd in boundary elements where the function w is called the ‘fundamental’ solution of the governing cquation, or solution of (1.69), Notice that this solution is obtained without taking into con- sideration the boundary conditions of the problem. Domain solutions arc obtained from weighted residual statements when the assumed approximate solutions do not satisfy the governing equations One can return to equation (1.1) which after integration by parts gives the following statement, ‘ : {aa ss si bow a+ [ w[=0 (173) ol dxd. This is a finite element type equation for which the last term can be found to be zero at the boundary points where q = du/dx is unknown, by the requirement that w#0 there, Substituting an approximate solution u in terms of unknown coefficients and known weighting functions leads to a system of equations to solve the problem. Notice that in the case of finite clements the unknown function u is explicitly defined over all the domain. Although the above remarks refer to the starting one dimensional equation (1.1) they also apply for the case of the Poisson cquation (1.12) and the associated weighted residual statements (equations (1.14) and (1.16)). Similar considerations. can be made for many other types of field equations,1.6. Boundary and Domain Solutions 39 Example 1.5 Let us now return to the same equation as defined in Example 1.1 and try to solve it using a weak formulation and considering boundary as well as domain techniques. The field equation is eu SJ+x=0 (a) dx? with boundary conditions, atx =Oand x= (b) (@ Boundary Solution, Homogencous Approach A weighting function which will satisfy the homogencous version of equation (a), i-c. aw =0 tc) dx? is the simple function w=a,x+a, with dw/dx=a, W) Equation (1.68) can be written for the case 4 =0 and 6 = —x as, 1 q fwd tos -[u *| ° © a dx Jo Alter substituting above the boundary conditions (b) and the expressions for w and dw/dx as given by (d) equation (c) becomes ' J xlayx +a) dx + quay + 42)— doa, =O 3 As the above equation has to be satisfied for any arbitrary values of a, and a), it gives the following two expressions 1 q=—Jxtde= ! J (g) 1 4 —do= —Sxdx= a and hence, do=6 (hy These values of g at x =0 and x =1 which are now the problem unknowns, are in this case the exact valucs.40 Chapter 1 Basic Concepts Gi) Boundary Solution, Singular Approach The weighting function in this case is chosen such that ew . at and w A solution of equation (i) regardless of boundary conditions is ia XN e vf j w np RK, Once the boundary conditions are applied, equation (1.72) becomes 1 —u;, + f xwdx + gy, —qoWo (k) 3 Taking into consideration that wg =0 and substituting the other values of w as given by (j) one finds, i ' fave fxd taux, “ or, x; xP u; ax, (m) 2 6 Notice that only one unknown (q,) remains, since one of the boundary stresses disappeared because of the variation of the weighting function w. The value of q, can be determined by tking the coordinate x, = 1, ic. ny =0=5-b4qy (n) Sanh which is the exact value in this case. Any value of u inside the domain can be computed from (I), ie. a= as {o) which is also the exact solution. If instead of the fundamental solution given by (j) one had chosen a funda- mental solution that also satisfies the boundary conditions, then no unknowns would exist either in the domain or at the boundaries and the value of u at any point would be obtained by a single integration. Consider for instance the solution, -{" =x)x x x;1.6. Boundary and Domain Solutions 4t This function satisfies (i) and the boundary conditions w =0 at x=0 and x= 1 {tq = 0). Thus equation (k) gives 1 gf we dx = § (1—x,)x? dx +f x1 — x) dx (@) a a Fa resulting in () which is the exact solution, Fundamental solutions that satisfy the boundary conditions as well as the governing cquations are called Green's functions, (iii) Domain Solution The weighted residual statement used here is the one resulting after one integration by parts has been carried out (equation (1.73), i. 16 dudw du)! a twee dx +] —w] =0 s iezenl-[E-[ ° ‘The proposed approximat boundary conditions, i.e. ¢ solution is the one in Example 1.t which satisfies the u=ayb, +2962 w with gy =x 2x7 43 = ap Va3x?-4x41 ay dd dx 3x? — 2. where ¢, and @, are the Hermitian cubic polynomials. Substituting these values into (s) and using Galerkin, i.e. Wabi tab (a)42 Chapter | Basic Concepts ‘one can write, j [x (3x? — 4x + 1) + ag(3x? — 2x)](3x? — 4x + Idx a 1 =f x(x— 2x? + x) dx (w) 3 ‘ J ey (Bx? — 4x + 1) + 2,(3x? — 2x](3x? — 2x) de a = j x(x? - x?) dx a which after integrating and solving also gives the exact solution, i.e. a=h =—} &) IL is worth noticing that if the approximation used for the ‘weak" formulation is the same as the one used for the original domain weighted residual equation the result will be the same in all cases. The advantage of the ‘weak’ formulation is that the order of the derivatives of u is in this case reduced and hence the order of derivability required by the assumed approximate solution is also reduced. 1.7 Concluding Remarks This chapter has presented the Boundary Element Method as a weighted residual technique. This approach permits relation of the method to other numerical techniques and gives an casy way of introducing boundary elements. For simplicity one dimensional problems have been discussed throughout the chapter to present the relationship between different integral statements and also between approximate techniques. The prescntation was then extended to potential Problems governed by the Laplace or Poisson's equations, which will be used in the next chapter. Some authors prefer to deduce the boundary integral equations: from Green's theorem instead. Notice that this theorém has also been presented here (equation (1.18) where it was shown that it can be derived from Lagrangian multipliers or basic residual type statements. Later on a similar approach will be discussed for elasticity problems as shown in Chapter 3. The beauty of weighted residuals is that they are simple to use and can be applied for a wide range of problems, including some very complex non- linear and time dependent cases which are not discussed in this book.Exercises 43 Exercises 13, M4, 1s, 16. 47. e Solve = +u+x=0 with boundary conditions u(0) = u(1) = 0 using a trial function I of the form w= uy + «4x +4,x? and point collocation for x= 1/2. Plot the solution and compare it with that of example 1.2 of the text and the exact solution given by equation (b) of that example, d Solve Sa = exp(u) from x=0 to x= twith boundary conditions w(0) = u(1)=0 using the same trial function and collocation point of exercise 1.1. Solve V?u=0 in the plane domain 0f=- (2.16) in| Sale! 16 Notice that the surface of the sphere is [,=4ze?. Similarly for the two dimensional case one can define a small circte of radius ¢ and then take the limit when & +0, (2.17) Here the perimeter of the small circle is T, = 2xe.50 Chapter 2 Potential Problems Boundary Integral Equation We have now deduced an equation (2.11) which is valid for any point within the Qdomain. In boundary elements it is usually preferable for computational reasons to apply cquation (2.11) on the boundary and hence we need to find out what happens when the point x' is on I. A simple way to do this is to consider that the point i is on the boundary but the domain itself is augmented by a hemisphere of radius (in 3D) as shown in figure 2.2 (for 2D the same applics but we will consider a semicircle instead). The point x! is considered to be at the centre and then the radius ¢ is taken to zero. The point will then become a boundary point and the resulting expression the specialization of (2.11) for a point on T. At present we will only consider smooth surfaces as represcuted in figure 2.2 and discuss the case of corners in other sections. Boundary surface T (0) Three Dimensional Case. Hemisphere around i Boundary cuve F, Boundary curve Boundary point i (ii) Two Dimensional Case. Semicivcle around i Figure 2.2. Boundary points for two and three dimensional case, augmented by a small hemisphere or semicircle2.2. Basic Integral Equation st It is important at this stage to differentiate between two types of boundary integrals in (2.11) as the fundamental solution and its derivative behave differently. Consider for the sake of simplicity equation (2.11) before any boundary conditions have been applied, i. w+ Sug* dP =futgar (2.18) i r Here P= I, + [and satisfaction of the boundary conditions will be left for later on. Integrals of the type shown on the right hand side of (2.18) are casy to deal with as they present a lower order singularity, i.c. for three dimensional cases the integral around [, gives: Jim {5 que a} ot {5 ‘ ies ar} 2 slim {ox eo (2.19) coo! ane In other words nothing occurs to the right hand side integeal when (2.11) oF (2.18) are taken to the boundary. The left hand side integral however behaves in a different manner. Here we have around T, the following result, tn {fr ar} tn {fat} 2 = tim {-« me =! (2.20) They produce what is called a free term, It is easy to check that the same will ovctir for two dimensional problems in which case the right hand side integral around [; is also identically equal to zero and the left hand side integral becomes tin {fag arf tin ff) ar =lim {-» E}- —hd (2.21) 0 é From (2.19) to (2.21) one can write the following expression for two or three dimensional problems gut Jug? dU =f qu dv (2.22) i r where the integrals arc in the sense of Cauchy Principal Valu This is the boundary integral equation gencrally used as boundary elements. arting point for32 Chapter 2 Potential Problems 2.3 The Boundary Element Method Let us now consider how expression (2.22) can be discretized to find the system of equations from which the boundary values can be found. Assume for simplicity that the body is two dimensional and its boundary is divided into N segments or elements as shown in figure 2.3. The points where the unknown values arc considered are called ‘nodes’ and taken to be in the middle of the element for the so-called ‘constant’ elements (figure 2.3(a)). These are going to be the elements considered in this section, but later on we will also discuss the case of linear elements, i.c those clements for which the nodes are at the extremes or ends (figure 2.3(b)) and curved clements such as the quadratic ones shown in figure 2.3(c) and for which a further mid-element node is required. For the constant elements considered here the boundary is assumed to be divided into N elements. The values of w and q are assumed to be constant over cach element and equal to the value at the mid-clement node. Equation ( Jean be discretized for a given point ‘i’ before applying any boundary conditions, as follows, a y x hut ¥ fugtar = ¥ § quar (2.23) gaat, dairy The point i is one of the boundary nodes. Note that for this type of element (i.e. constant) the boundary is always ‘smooth’ as the node is at the centre of the element, hence the multiplier of w is 4. T; is the boundary of the ‘j" element. The uw and q values can be taken out of the integrals as they are constant over each element. They will be called w and q/ for element 3". Hence i+ i ( ie ar) = ( Sut are (2.24) Notice that there ure now two types of integrals to be curried out over the clements, i.e. those of the following types fqtdl and fwtdl i 6 These integrals relate the ‘i? node where the fundamental solution is acting to any other ‘j° node. Because of this their resulting values are sometimes called influence coefficients, We will call them HY and Gi, i. AYsfqtdt; Gla futar (2.25) Vs Notice that we are assuming throughout that the fundamental solution is applied ula particular ‘i’ node, although this is not explicitly indicated in u*, q* notation to avoid proliferation of indexes. Hence for a particular ‘i’ point one can write, x x Wht Y Ad = ¥ Gig! (2.26) a2.3, The Boundary Element Method 53 element (a) Constant Elements {(b) Linear Elements Nodes 4i\ | deman (©) Quadratic Elements Figure 2.3 Different types of boundary elements If we now assume that the position of f can also vary from 1 to N, ie. we assume that the fundamental solution is applied at each node successively one obiains a system of equations resulting from applying (2.26) to each boundary point in turn.54 Chapter 2 Potential Problems Let us now call if ej We {i when i 2.27) AY 43 when i= j hence equation (2.26) can now be written as x x x aw = ¥ Gq (2.28) i im This set of equations can be expressed in matrix form as HU=GQ (2.29) where H and G are two Nx N matrices and U, Q are vectors of length N. Notice that N, values of u and WN, values of q are known on T, and Tr; respectively (IT, +1, =), hence there are only N unknowns in the system of equations (2.29). To introduce these boundary conditions into (2.29) one has to rearrange the system by moving columns of H and G from one side to the other. Once all unknowns are passed to the left-hand side one can write, AX=F (2.30) where X is a vector of unknowns u's and q's boundary values. F is found by multiplying the corresponding columns by the known values of u's or q's. It is interesting to point out that the unknowns are now a mixture of the potential and its derivative, rather than the potential only as in finite clements. This is a consequence of the boundary clement being a ‘mixed’ formulation and gives an important advantage to the method over finite elements. Equation (2.30) can now be solved and alt the boundary values are then known. Once this is done it is possible to calculate any internal value of u or its derivatives The valucs of u’s are calculated at any internal point ‘i’ using formula (2.11) which can be written as, Jaqu* dP — fag* ar (231) r t Notice that now the fundamentat solution is considered to be acting on an internal point ‘i’ and that all values of u and q are already known. The process is then one of integration (usually numerically). The same discretization is used for the boundary integrals, ic. x % win ¥ Gg ¥ Aw (2.32) int jet The coefficients G'/ and A have been calculated anew for cach different internal point.23, The Boundary Element Method 55 The values of internal fluxes in the two directions, say x, and X3. dx, and q,, = éu/ax,, are calculated by carrying out derivatives on (2.31), aat\! (an. = (: ar wat (se) foe) vO 5 y= (2) 4) Ir (oc) ar (ais) (=) fa ee) Sloe Notice that the derivatives are carried out only on the fundamental solution functions u* and q* as we are computing the variations of lux around the ‘i* point. The boundary integrals are discretized into integrals along the clements, twat = (HJ = (iee)e- 5 1G aru (2.34) wat (sz) = B (U) eB Ee) ‘The kernels to be integrated along the clements are aujex, ‘eu* (#)-3 4 & (canesin 235) and “yt aft (3) -i¢ ( emt rad) =- = (Qr?, — Ue, + rr zrtg] (2.36) Oq* 1 i ae (2r3 = Ing + 2ryram] where r, indicates derivative at the integration point; i.e. oy = or, ax, - and n,, m, are the components of the unit normal. The integration of the expres- sions given in (2.36) is done numerically using a standard Gaussian quadrature. Evaluation of Integrals Integrals like G4 and fin the above expressions can be calculated using numerical integration formulae (such as Gauss quadrature rules) for the case i# j. For the element j= j however the presence on that element of the singularity due to the fundamental solution requires a more accurate integration. For these integrals it56 Chapter 2 Potential Problems 0 a Node @ t want 0 get $12 ——_+}- 72 —___»1 9 Figure 2.4 Element coordinate system is recommended to use higher-order integration rules or a special formula (such as logarithmic and other transformations which will be discussed later on) For the purticular case of constant elements however the * and G" integrals can be computed analytically. The H* terms for instance are identically zero, the normal nnd the element coordinate ire always perpendicular to each other, ie. aut AYsfqture Ve “reo (2.37) n fOr en ‘The integrals in G require special handling. For a two dimensional element for instance they are Giafudre) j ia(!) ar (2.38) i ni; r In order to integrate easily the above expression one can change coordinates to a homogeneous & coordinate over the clement (figure 2.4) such that, (2.39) where ! is the element length. Hence taking into account symmetry (2.35) can be written bk) ha)” “1()im(ea)* = 7 (H[(;.) +f fm( | ) «| (2.40)2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 57 ‘The last integral is equal to L, Le. oe OG.) ‘| ean For more complex cases special weighted formulae are used. The other integrals {i.c. for i # j) can be calculated using simple Gauss quadrature rules. In the two dimensional codes described in this chapter a 4 points rule has been used (see Appendix A). 2.4 Computer Code for Potential Problems using Constant Elements (POCONBE) In what follows the above theory will be employed to produce a simple computer code written in FORTRAN for solving Laplace type problems. The code is valid for isotropic materials and uses constant elements. The program can be run in any IBM/PC type XT or AT or compatibles. Boundary Element codes are substantially different from Finite Element programs, Their internal organization is somewhat simpler as they do not require an assembler. They also produce all the boundary values (u’s and q's) and give generally very precise solutions, Main Program The macro-flow diagram for the POCONBE boundary element code is shown in Figur . The main program defines the maximum dimensions of the system of equations, which in this case is 100 and allocates the input channel 5 and the output channel 6. It calls the following five routines INPUTPC — This routine reads the input to the program GHMATPC - It forms the system matrices H and G and rearranges them in accordance with the boundary conditions into a matrix A. It also creates the right hand side vector F. SLNPD = This is a subroutine for solving systems of equations, with pivoting. INTERPC § - This routine computes the values of potentials and fluxes at internal points. OUTPTPC: Outputs the results. The main routine also reads and opens files for input and output. The general integer variables used by the program are defined as follows. N: Number of clements (equal number of nodes for constant elements). L: Number of internal points where the function is calculated.58 KODE: One dimensional array indicating the type of boundary con: Chapter 2 Potential Problems MAIN PROGRAM INTERPC Figure 2.5 Macro flow diagram ons at the nodes. KODE(J) = 0 means that the value of the potential is known at node J and KODE() = I signifies that the value of q is known at the corresponding boundary node. The following real arrays are used to store data and results. XM: YM: FI: One dimensional array of x, coordinates extreme point of boundary clements. One dimensional array of x, coordinates extreme point of boundary elements, x, coordinates of the nodes. XM(J) contains the x, coordinate of node J. x, coordinates of the nodes. YM(J) contains the x; coordinates ofnode J. - Matrix defined in cquation (2.29). After application of boundary condi- tions the matrix A is stored in the same location. Matrix defined in equation (2.29). Prescribed value of boundary conditions. FI(J) contains the prescribed value of the condition at node J. If KODEW)=0 it means that the potential is prescribed and if KODE(J)=1 that the g is given for the clement associated with the location in those vectors.2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 59 DFI: Right hand side vector in equation (2.30). After solution it contains the values of the unknown u’s and q's. cx: x, coordinate for internal point where the value of u is required. CY: x; coordinate for internal point where the value of w is required. POT: Vector of the potential values for internal points. FLUX1, FLUX2: Vectors of potential derivatives for internal points. The listing of the MAIN program is as follows: PROGRAM POCOMBE PROGRAM 1 ‘THIS PROGRAM SOLVES TWO DIMENSIONAL (PO)TENTIAL PROBLEMS. USING (COM)STANT (R)OUNDARY (E)LEMENTS (CHARACTER®10 FILEIN,FILEOUT © ananaaoe age DIMENSION X( 101), ¥(101) ,X4(100) .¥H( 100) ,#1(100) ,DF24100) DIMENSION KODE( 100) ,CX(20) ,CY(20),POT( 20) ,FLUX1 (20) ,FLUXZ(z0) CONNON/HATG/ G( 100,100) ‘COMMON /HATH/ I 100, 100) ‘COMHON. NL, INP, PR SET MAXIMUN DIMENSION OF THE SYSTEK OF EQUATIONS (Wx) (THIS NUMBER MUST BE EQUAL OR SNALLER THAN THE DIMENSION OF XM, ETC...) wre 100 ASSIGN MUMBERS FOR IMPUT AND OUTPUT FILES 90 fare READ NAMES AND OPEN FILES FOR INPUT AND OUTPUT WRITE(#,* (A) ") ' NAME OF INPUT FILE (MAX. 10 CHART.)* READ(#,? (a) ‘)PILEIN. OPEN (INP, FILESFILEIN, STATUS="0LD"} + OWRETEC,? (AD) "NAME OF OUTDUT FILE (MAX. 10 CKART.)? READ(*,? (a) FILEOUT. (OPEN( IPR, FILE*FILEQUT ,STATUS* "NEW" ) READ DATA CALL, INPUTPG(CH.CY,X,7.RODESFLD COMPUTE H AND G MATRICES AND FORM SYSTEM (AX = F) CALL GHNATPC(X.Y.40,YH,G.H.FI .DF1sKODESNA? SOLVE SYSTEN OF EQUATIONS CALL SLNPD(G,DFI,D,N.NE) COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS CALL INTERPC( FS ,DF1.KODE,CA.OY.X,¥, POT. FLUX? FLUKZ) PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POIKTS CALL OUTPTPC(xX¥, YM,FI DFT CK, CY.POT, FLUX /FLUZ2) 999 9Aa AMA BAA AAA One ‘CLOSE INPUT AND OUTPUT FILES Lose (1P) CLOSE (TPR) STOP. END,60 Chapter 2 Potential Problems Routine INPUTPC ‘All the input required by the program is read in the program INPUTPC and contained in a file whose name is requested by the main program. The file should contain the following input lines (using free FORMAT): (1) Title Line One line containing the title of the problem. (2) Basic Parameter Line One line containing the number of boundary elements and the number of internal points where the function is required. (3) Extreme Points of Boundary Elements Lines The coordinates of the extreme of the elements read in counterclockwise direction for the case shown in figure 2.6(a) and in clockwise direction for 2.6(b). (4) Boundary Conditions Lines As many lines as nodes giving the values of KODE and the value of the potential at the node if KODE =0 or the value of the potential derivative if KODE = 1. (5) Internal Points Coordinates Lines The x,x, coordinates of the internal points arc read in free FORMAT in one or more lincs. This subroutine prints the title, the basic parameters the extreme points of the boundary clements and the boundary conditions. The internal point coordinates are printed in the OUTPTPC routine. Notice that all input is written in free FORMAT. a SUBROUTINE INPUTPC(CX, CY,X,¥,KODE, FT) © PROGRAM 2 CHARACTER*80 TITLE COMMON N, L, INP, IPR DIMENSION CX(1) ,C¥ (2) ,X(2) ,¥(2) ,KODE(1) , FE(2) Ne NUMBER OF BOUNDARY NODES (#NUMBER OF ELEMENTS) L= NUMBER OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED fone WRITE (IPR, 100) 100 FORMAT(” *,79("#")) READ JOB TITLE ano READ(INP,“(A)') TITLE WRITE(IPR,’(A)‘) TITLE READ NUMBER OF BOUNDARY ELEMENTS AND INTERNAL POINTS aan READ(INP, 4)N,L WRITE (IPR, 300)N, 1, 300 FORMAT(//* DATA'//2X, NUMBER OF BOUNDARY ELEMENTS =, 13/2X, ‘NUMBER 1 OF INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED’ =", 13) READ COORDINATES OF EXTREME POINTS OF THE BOUNDARY ELEMENTS IN ARRAYS X AND ¥ aanan WRITE(TPR, 500) 500 FORMAT (//2X, "COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELE IMENTS! , //1X, POINT! , 7%, *X" , 15K, °¥") READ(INP, #) "(X(I),¥(I) ,Tm2,8) DO.10 T#i,W Lo WRITE (IPR, 700) I,X (1) ,¥(1) 700 FORMAT (2X,13,2(2X,E14.5))2.4, Computer Code for Potential Problems using Constant Elements (POCONBE) 61 (8) Numbering direction for extemal surtace. {b) Numbering direction for internal surface - OPEN. CLOSE DOMAIN (anticlockwise) DOMAIN (clockwise) Figure 2.6 Numbering directions for external and internal surfaces READ BOUNDARY CONDITIONS IN FI(I) VECTOR, IF KODE(1)=0 THE FI(I) VALUE IS A KNOWN POTENTTAL;IF KODE(I)=1 THE FI(I) VAWE IS A KNOWN POTENTIAL DERIVATIVE (FLUX). anenn WRITE (TPR, 800) 800 FORMAT(//2X, ‘BOUNDARY CONDITIONS! //2X, ‘NODEY , 6X, CODE’ , 7X, ‘PRESCRT BED VALUE?) DO 20 T=1,N READ(INP, *) KODE(I) , FI(I) 20 WRITE (IPR, 950)I,KODE(I)FI(T) 950 FORMAT(2X,13,9X, 11, 8X, E14.5) c READ COORDINATES OF THE INTERNAL POINTS c IF(L.EQ.0) GO TO 30 READ(INP, 4) (CX(Z) ,C¥ (I), T=2,L) 30 RETURN END Routine GHMATPC The routine GHMATPC forms the G and H matrices of equation (2.29) through its subroutines EXTINPC and LOCINPC, It then rearranges their columns to form A matrix and F vector of (2.30). The subroutines EXTINPC and LOCINPC perform the following functions. EXTINPC: This subroutine computes the H and G matrix elements by means of numerical integration along the boundary elements (using 4 points Gauss quadrature). It calculates all elements except those on the diagonal. LOCINPC: Only calculates the diagonal elements of G matrix, given by equation (2.41). :62 Chapter 2 Potential Problems Notice that as 2, = 0 the diagonal clements of H are simply }. It is also important to point out that as the fundamental solution has been taken as in(+) — without 1 — all terms in G and H are effectively fs (2) multiplied by 1/2) Rearranging the columns of G and H produces the matrix A, which is stored in the original space used by G. The columns of this matrix are the columns of H or G which are multiplied by unknown values of u or q. The right hand side vector F is called DFI in the code and is obtained by multiplying the columas of Gor H by the known values (ic. boundary condition values) of q or u respectively. SUBROUTINE GHMATPC(X,Y,XM,YM,G,,FI,DF1sRODE,NX) PROGRAM 3 THIS SUBROUTINE COMPUTES THE G AND MNATRICES AND FORMS THE SYSTEN OF EQUATIONS AX = F aneeae o COMMON WAL. INP.IPR DIMES TON XC1),¥(1) 4 XNC2) TNC) .FT(2) -RODEC) DIMENSION DFI(1) ,GUWX .NX) , CNX, HX) ‘COMPUTE THE NODAL COORDINATES AND STORE TN ARRAYS XM AND YH eno xOWeL=xC1) Yow) DO 10 Ist, ACT) ( CED OR(0 429) 72 YO YN TDR(Y(Z)o¥(169)/2 COMPUTE THE COEFFICIENTS OF G AND H MATRICES ana Do 30 #1," Do 30 401, Rieder IE (1-3) 20,28,20 20 CALL EXTIMPC(XM(I) YC) »X(J)67(9) -KCRK) -YCRK) C143) G01.) 1,D@1 .DOz, DUI ,DUZ 0} Go 70 30 28 CALL LOCANPC( (2) ¥(2) 4 XK) YORK) ,6(2,9)) HCL,a) 03, 1418925 30 CONTINUE REORDER THE COLUHNS OF THE SYSTEM OF EQUATIONS 1N ACCORDANCE WITH THE BOUNDARY CONDITIONS AND FORM SYSTEM MATRIX A WHICH 15 STORED IW G Do S55 Je1.M IF (KODE( 3} )58,88,40 49 D0 60 181K cHeG(1,3) G(T. 3)e=W(2,3) WO19)e-cH $0 CONTINUE 58 CONTINUE, nee FORM THE RIGHT HAND SIDE VECTOR F WHICH IS STORED IN DFT Do 60 121, DET(T)=0. Do 60 Jenin DFE(1) =DFI(1)4#(1,3)# FICS) 60 CONTINUE RETURN END2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 63 Routine EXTINPC This subroutine computes the values of the off-diagonal coefficient of H and G using a 4-point Gauss integration formula (see Appendix A). It also computes, using the same numerical integration formula, the integrals of the fundamental solution and its derivatives required for the computation of potentials and fluxes at internal points (equations 2.35 and 2.36). Consider now that instead of the system x,—x, we use an x-y system of coordinates. In this case X1, X2, ¥1, Y2 are going to be the coordinates of the extreme points of cach element considering them in clockwise (open domain) or anticlockwise manner (closed domain). . Using numerical integration and changing to a dimensionless system of coordinates the G! and H" terms for cach element and collocation point can be written as, 4 SOT Xa OT V2 on "(cn)" ast $ d (rw Jes Jixi- xp art 22 (Ray we dn 2 = z - jp RDI ETAL + RD2*ETA2) ro xy where RDI=r,= (= XP Nae O- RD2 =r, = —__—_ ? a , is the weighting for each point, XP, YP are the coordinates of the collocation point and ETA!, ETA2 are the components of the unit normal. The values w; and the location of the k points over the element are given in Appendix A. ‘SUBROUTINE EXTINPC(XP,YP,| PROGRAM 4 Y1,42,72, 8,6, D@1 ,D@z,DUL, DI THIS SUBROUTINE COMPUTES THE INTEGRAL OF SEVERAL. NON-SINGULAR FUNCTIONS ALONG THE BOUNDARY ELEMENTS USING A POUR POINTS GAUSS QUADRATURE WNEN KeOs THE OFF DIAGONAL COEFFICIENTS OF THE H AND G MATRICES ARE Kal, ALL THE COEFFICTENTS NEEDED FOR COMPUTATION OF THE POTENTIAL AND FLUXES AY INSERNAL POINTS ARE COMPUTED. (G,i,21,E2,F1,F2) RAs RADIUS = DISTANCE FROM THE COLLOCATION POINT TO THE GAiss INTtoRATLON POINTS ON THE BOUNDARY, ELEMENT EIALETAL NTS OF THE UNIT NORMAL TO THE ELEMENT ROLIRDZSAON © RADIUS DERIVATIVES naanannaanacoose 964 ‘Chapter 2 Potential Problems DINENSION XCO(41,¥CO(4) ,G1(4) .OMECA) DATA G1/0.86113631,-0,86113631 ,0-33998104,-0,93998104/ DATA OWE/G. 34785485 ,0, 34785485 ,0.65214515,0.65214515/ BX OX20X1) 72 AYe(V2-¥1)/2- Bye (y20¥1)/2- SLeSQRTIAX##20ave2) ETAISAY/SL. SAK/SL ov2e0: DQt=0. azo. € COMPUTE G, H, G1, DQ2, DUI AND DUZ CORFFITIENTS SGTCTDOBX YOU) saveGI (1) BY RAS SQRT((XP-XCO(1))#42¢(YP-YCO(T))##2) RDI=(XCO(1)-XP 1/KA RDZ=(YCO(I)-YP)/RA RON=RDI*ETAI+RDZ*ETAZ IF(k) 30,30, 10 10 DU! «DUI +hD1 SONE(T )¢SL/RA DUZEDUZ+RD2¢ONE(T) ¢SL/RA DQL=DQI=1(2.#RDL##2-1.) SETA +2. ¢RDI#RDZ#ETAZ) SOME(T)*SL/RAESZ Dazeba2-((2.#RD2+02-1_)ETAZ#2, #RDI*ROZWETAI } COME I }#SL/RATOZ 30 GeG+ALOG(1/RA)#OME(TIOSL 40 HeN-RDN*OME(T)#SL/RA ‘BETURN END Routine LOCINPC This routine simply computes equation (2.41) to obtain the diagonal elements of G. As we have used throughout the fundamental solution In(1/r), the formula has to be multiplied by 27, i.e. (2.41) becomes, t or=ifin int it PROGRAM 5. IMIS SUBROUTINE COMPUTES THE VALUES OF THE DIAGONAL, FITAENTS OF THE G MATRIX eanoaa AX#(X2°R1)/2- Ga2ssRe{ RETURN END Routine SUNPD ‘This is a standard routine given in reference [10] which can solve the system of equations using pivoting if needed. If the matrix A has a zero in the diagonal it2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 65 will interchange rows, deciding that the system matrix is singular only when no row interchange will produce a non-zero diagonal coefficient. If this happens it will give a message indicating a singularity in that row. After elimination the results are stored in the same right hand side vector DFT. ‘SUBROUTINE SLNPD(A,B,D,N,NX) PROGRAM 6 SOLUTION OF LIWEAR SYSTEWS OF EQUATIONS BY THE GAUSS ELIMINATION METHOD PROVIDING FOR INTERCHANGING ROWS WHEN ENCOUNTERING A ‘ZERO DIAGONAL COEFICTENT SYSTEM MATRIX ORIGINALLY IT CONTAINS THE INDEPENDENT CORFFICIENTS. AFTER SOLUTION IT. CONTAINS THE VALUES OF THE SYSTEN UNKWOWNS Not ACTUAL NUNBER. OF UNKNOWNS: NX? ROW AND COLUMN DIMENSION OF A a 8 DIMENSION BUNK) ,A(HX,NX) TOLe1-F. © 9 aaannansnncnenne CHAU) TEUABS(C)-TOL)1,1,3 1 dO 7 Jex1,e © © TRY TO INTERCHANGE ROWS TO GET MON ZERO DIAGONAL COEFFICIENT © TP(ABS( (ACJ, ))=TOL)7, 7,5 $00 6 LeKWN CACK LD ACK LJ EAC LY 6 ACILLsc, caBiK) BUK)=B(3) Bsc SACK KD ‘co 10's 7 CONTINUE 00 To 8 DIVIDE RO¥ BY DIAGONAL CORFFICTENT 2 CeAUKKD DO4 Sent y 4 ROK, I)4AUE IE DUK} =BeKI Ze ene ELIMINATE UNKNOWN X(K) FHOM ROW DO.10 deKI4N CEA KD DO'S dekr ae 9 ALT, 2)2AU1,3)=C9A(K,3) 10 BEr}eB(1)-CeB(x) 100 CONTINUE ooo COMPUTE LAST UNKNOWN neo IF(ABSC(ACH,N)))=TOL)B, 8, 101 201 BUN) #B(ND/A(N ND APPLY BACKSUBSTITUTION PROCESS TO COMPUTE REWATWING UNKNOWNS. Do 200 Le1.Nd Ken-L xiske DO 200 JeK1,x 200 B(K)=BK)-AUHAI) 9B (I) nee66 Chapter 2 Potential Problems © € COMPUTE VALUE OF DETERMINANT. € Del, 0 380 ret," 250 DsDeAcT. 1) 60 10 300 8 WRITE(? 2) 2 RORMAT(! #288 SINGULARITY IN ROW’, 15) bso. 300 RETURN m0 Routine INTERPC Subroutine INTERPC reorders FI (boundary condition vector) and DFT (unknown vector) in such a way that all the values of the potential arc stored in FI and all the values of the derivatives or fluxcs in DFI. This subroutine also computes the potential values for the internal points using formula (2.32) and the fluxes along x, and x, directions using formula (2.34). Note that because all the H and G terms appear multiplied by 2 the solution for the internal points is also multiplied by 2x. o ‘SUBROUTINE INTERPC(PT,DF2,RODE,CX,CY,X,Y, POT, FLUX ,FLUX2) PROGRAM 7 ‘THIS. SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL AMO THE POTENTIAL DERIVATIVES (FLUXES) AT INTERNAL POINTS onnaee COMMON WL, INP, TPR DIMENSION #1(1) .DP1(1) NOEL 3) ,ox¢ 2) ,OYC1) .XC1) YC) DIMENSION POT(1} , FLUXY (1) ,FLUX2(1) REARRANGE THE FI AND DFI_ARRATS TO STORE ALL THE VALUES OF THE POTENTIAL IN F1 AND ALL THE VALUES OF THE DERIVATIVE IN DF2 pone BO 20 161.8 IF(RODE(T}} 20,20,10 to cHeFL(T) FIC1)spFI(1) DFI(T)=cn 20 CONTINUE COMPUTE THE POTENTIAL AND THE FLUZES AT INTERNAL POINTS one IFUL.FQ.0) GO TO 50 Do 40 Ke1,L. POT(K)=0. FLUZI(K}20. FLUX2(X)=0. DO 30 Js1,K We381 CALL EXTINPC(CK(K) sCVCK) 5X(9) (0) 4XCKR) YORK) (AyD 1,DQy ,DQz,DUI UZ, 1) ‘BOT (Ke) #POT(K} +DFH(3)*B-F1 (3) #4 FLUE (K)*FLUXI (K) ¢DP1(J)®DUI-FI(J)*Q1 30. FLUNZ(K) =FLUN2(K) «DPI C3) *DUZ-FY (3) #0Q2 POT(E)=POT(K)/(2.43. FLUX (x)= 40 FLUxziK, 50 RETURN END Routine QUTPTPC This routine outputs the results. It first lists the coordinates of the boundary nodes and the corresponding values of potential and its derivatives (or Nluxes), It also2.4. Computer Code for Potential Problems using Constant Elements (POCONBE) 67 prints the values of potential and fluxes at internal points if any have been requested. onnene, cinaieont ness SUBROUTEME QUTPFPC(X, YM, FI,DFT 6x )€¥ POT, FLUR) PLAN?) Program © wats sunmoUTINE pRinrs THE VALUES OF THE FOTENTTAL AND 173 NORMAL € e DERIVATIVE AT BOUNDARY NODES. IT ALSO PRINTS THE VALUES OF THE POTENTIAL AND THE FLUXES AT INTERNAL POINTS ‘COMNON ML, INP) TPR DIMENSION HM(1} YC) F1(1) ,DPT(1) ,CK(1) .0Y04) DIMENSION POT(1} ,FLUXI(1),FLUX2( 1) WRITE(IPR; 100) Hoo Formate! },79('#*)//1%, "RESULTS'//2X, "BOUNDARY NODES*//8x,°X" 15 AK,'Y' 919%, }POTENTIAL® , 3X," POTENTIAL DERIVATIVE! /) DO 10"1e1,N 20 WRITE( IPR. 200) XM(1),YH(T),PI(T) ,DFI(T) 200 FORWAT(4(2x,514.5)) IF(L.E@.0) 00 TO 30 WRITECTPR, 300) 300 FORMAT(//,2X, "INTERNAL POINTS" ,//8X, ‘x: 19K, "FLUE 8, 10K, "FLUE Y?/) DO'20 Kel st 20 WRITE(1PR, 400)CR(K) ,CV(K) ,POTLR) , FLUXS (R) FLUKE (R) 400 FORMAT(S(2x,E14.5)), 30 WRITE(IPR, 600) 500 FoRmaT«* *,79('6")) RETURN END 15x, °Y" 13K, "POTENTIAL" Example 2.1 The following example illustrates how the code can be used to analyse a simple potential problem. Consider the ease of a square close domain of the type shown in figure 2.8, where the boundary has been discretized into 12 constant elements with 5 internal points. The input statements are as follows: HEAT FLOW EXAMPLE (DATA) HEAT FLOW EXAMPLE (12 CONSTANT ELEMENTS) 125 0. 0. 2. 0. 4. 0. 6 0. 6. 2. 6. 4. 6. 6. 4. 6. 2. 6. 0. 6. 0. 4. 0. 2. ececo0cee 122 A 8 a ults are printed out as follows.68 Chapter 2 Potential Problems HEAT FLOW EXAMPLE (OUTPUT) eeeatesennnvase: MEAT FLOW EXAMPLE (12 CONSTANT ELENENTS) DATA NUMBER OF BOUNDARY ELEMENTS = 12 NUMBER OF INTERNAL POINTS WHERE THE FUNCTION 15 CALCULATED = & COORDINATES OF THE EXTREWE POINTS OF THE BOUNDARY ELEMENTS Pout x y 1 -900008+00 000008400 2 ‘200008601 ‘ooodor+00 3 ‘40000E¢01 Too000E+00 ‘ ‘e0000r+01 ‘ooo0oe+o0. 5 ‘60000F-01 ‘2o0008+os 6 ‘so000r+o1 40000401, 7 ‘e0000E+01 ‘Go000E+01 8 ‘a0o00n=01 (00005401 9 ‘200008<01 {600005+01 30 ‘a0o00E+00 Seo0005+01 n ‘ooocE+00 Saoc00E+or n ‘00000r-00 ‘z00008+01 BOUNDARY CONDITIONS NODE: cope PRESCRIBED VALVE "000008400 [00000E+00 ‘o00005+00 ‘o0000E+00 ‘ooooor-o0 = Too000k+00 Toooo0R+00 ‘oo00E+00 ‘ooo00E+00. ‘lx00008-03 1300005403 '30000E+03 RESULTS. BOUNDARY NODES x Y POTENTIAL POTENTIAL DERIVATIVE +10000E+01 900008400 252258603 000008400 isvo00e+01 ‘ooonoE+ 00 ‘1s002E+03 Tooo00E+eo [s0000E+01 ‘ooo0ar+00. 1477508402 ‘o000E+00 ie0000+01 [100008+01 1090005+00 2962802 ‘so00E+o1 {30000801 ‘000005+00 487718602 60000801 ‘so00Es 01 ‘oosoor+a0 ‘829625402 Ts00008+01 ‘600005+03 larrsoR+or Toooo0e+00 1300008+01 ‘600005+01 [1s002E+03 ToooooR+o ThooooEso1 ‘eoo00r+01 lesazsz+oa ‘oo000E+00 loo0008+00 ‘s0000k+01 {30000E+03 ‘s2g69E.02 Too000e+00 ‘s00008+03 ‘s00008+03 ‘4s7376002 To0000k+00 [re000E+o1 [300005403 ‘s2969Eeo2 INTERNAL POINTS x 4 Y POTENTIAL, ruux x Fux y +200008+01 +20000E+01 -20028E+03 +80903E+02 —-. 14976000 200008+01 {400006401 {20028E-03 0303802 ‘4ap24Es00 {300008+01 ‘s0000E+01 {as001ze0a © =.$02185+02 403605 -05 ‘40000E¢01 20000801 97 408-02 0308E+02 148648400 Sa00008+01 '400005+01 lesreoz+02 O306E+02 5145645600 Notice the excellent agreement of the results with the exact solution given figure 2.7(a)), when the coarseness of the mesh and the simplicity of the model2.4, Computer Code for Potential Problems using Constant Elements (POCONBE) ao an eo w= 300: _ ~~ os 4-50 a= -50 a0" ——— s —_—1 (2) Definition of the Problem Boundary odes + [+ Intefnat points 1 2 a {b) Discretization into elements and internal nodes 0-0, u=300-7| uo a=0 (©) Boundary conditions Figure 2.7. Simple potential problem eo7” Chapter 2. Potential Problems are considered. On the two vertical sides the fluxes are close to — 50 and 50 as expected and on the horizontal s the value of the potential is ilar to the analytical solution which varies linearly from 300 on the left hand side to 0 on the right. The accuracy of the internal point results is however even more remarkable and this is duc to the way in which these results are computed using formula (2.32), i.c. they are like a weighted average of the boundary values. 2.5 Linear Elements Up to this section we have only considered the casc of constant elements. i.e. those with the values of the variables assumed to be the same all over the clement. Let us now consider a linear variation of u and q for which case the nodes are considered to be at the ends of the clement as shown in figure 2.8. The governing integral statement can now be written as, clu! + fug* dT = f uty dd (2.42) r r Notice that the § coefficient of u! has becn replaced by an unknown ¢ value. This is because c= } applies only for a smooth boundary. The value of for any other boundary can be proved to be, (2.43) 0 Qa where 0 is the internal angle of the ‘corner in radians. This result can be obtained by defining’‘a small spherical or circular region around the corners and then taking the radius of them to zero (similar to what has been shown in section 2.2). Another possibility is (o determine the value of c' implicitly (sce section 2.6) and in this case it is not required to calculate the angle. Alter discretizing the boundary into a series of N elements equation (2.42) can be written x n cult ¥ fag*tdr= Yo f utqdl (2.44) ier ity ‘The integrals in this quation are more difficult to evaluate than those for the constant element as the u's and q's vary lincarly over each T; and hence it is not possible to take them out of the integrals.2.5. Linear Elements r = Noda) vatue ‘Nodst value of ofuorg vorg > (2) Linear Element Definitions m @ element j+2 element j a {b) Etement Intersection Figure 2.8 Linear clement. Basic definitions The values of wand q at any point on the element can be defined in terms of their nodal values and two lincar interpolation functions @, and 3. which are given in terms of the homogeneous coordinate £ as shown in figure 2.8(a), ic. \ wld) = dual + a = woaft} eas i a) = dit! + bur = toaftst is the dimensionless coordinate varying from —1 to + Land the two interpolation functions are W=Ul-Sh da= +8) (2.46) Let us consider the integrals over an clement ‘j*. Those on the left hand side can be written as, fug* at = { (¢1d2]0* aft} a on} (2.477 'n Chapter 2. Potential Problems where for each element ‘/” we have the two terms, hii J Oat dP (2.48) and hij =f dyq* dv (2.49) ‘ Similarly the integrals on the right hand side give oe i! f qut dt = § £6.p2]u* arf \. tnt \ (2.50) i ty Cs a. where gi =f duet al (2.51) iv and gi =§ dot ar (2.52) f ‘Treatment of Corners A domain discretized using boundary elements will present a series of corners which require special attention as the conditions on both sides may not be the same, When the boundary of the domain is discretized into linear elements, node 2 of element ‘j’ is the same point as node 1 of element ‘+ 1° (figure 2.8(b)). Since the potential is unique at any point of the boundary, u? of clement ‘7’ and u! of element 'j+ 1’ are both the same. However, this argument can not be applied as a general rule to the flux, as there are boundary points for which the Mux docs nol have a unique va This takes place at points where the normal to the boundary is not unique (corner points), It may also happen that the flux prescribed along a smooth boundary presents discontinuities at certain particular points. While corners with different values of the flux at both sides exist in many practical problems, discontinuous values of the flux along a smooth boundary are seldom prescribed. To take into account the possibility that the flux of node 2 of an element may be different from the flux of node | of the next clement, the fluxes can be arranged in a 2n array. Substituting equations (2:47) and (2.50) for all ‘j' elements into (2.44) one obtains the following equation for node‘. ut 2 et cA." Y pene? Gay) (2.53) Tad ge2.5, Linear Elements a where 17 is equal to the hi! term of element ‘7° plus the h/~* term of element 1 Menee fo (2.53) rep s the assembled equation for node ‘i. Note the simplicity of this approach. Equation (2.53) can be written as, ¥ an dt YM = ¥ GH (2.54) m im 1s previously shown (equation (2.28)), this formula can be written as x aN YS Miu = ¥ Gigi 2.55) in im and the whole set in matrix form becomes HU=GQ (2.56) where G is now an N x 2N rectangular matrix. Several situations may occur at a boundary node: First that the boundary be smooth at the node, In such a case both fluxes “before” and ‘after’ the node are the same unless they are prescribed as different, but in any case, only one variable will be unknown cither the potenti: ue lux. Second, that the node is at a corner point. In this case four different cases are possible depending on the boundary conditions: (a) Known values: fluxes ‘before’ and ‘after’ the corner. Unknown value: potential (b) Known values: potential, and flux ‘before’ the corner. Unknown value: flux ‘after’ the corner : Known values: potential, and flux ‘after’ the corner. Unknown value: flux ‘before’ the corner (4), Known values: potential. Unknown values: flux ‘before’ and ‘after’ the corner. te There is only one unknown per node for the first three cases, and two unknowns for case (d). As long as there is only one unknown per node, system (2.56) can be reordered in such a way that all the unknowas are taken to the left hand side and obtain the usual system of N x N equations, i.c. AX =F (2.57) where X is the (N) vector of unknowns; A is the (N x N) matrix of coefficients which columns are columns of the matrix H, and columns of the matrix G after a change of sign or sum of two consecutive columns of G with opposite sign when the unknown is the unique value of the flux at the corresponding node. F is the known vector computed by the product of the known boundary conditions and the corresponding coefficients of the G or H matrices.4 Chapter 2 Potential Problems When the number of unknowns at a corner node is two (case (d)), one extra equation is necded for the node. The problem can be solved using the idea of ‘discontinuous’ elements [11] presented in section 2.7. 2.6 Computer Code for Potential Problems using Linear Elements (POLINBE) Although this code has many routines which are similar to those developed for the constant clement case (POCONBE), there are some parts which require modification. Main Program The integer variables have the same meaning as in the constant elements program. ‘The same can be said for the real arrays except for the mid-point coordinates XM and YM that are not needed as now the nodes are at the inter-clement junction, Arrays FI and DFI now have a different meaning. The dimension of FI is (N) while the dimension of DFI is (2N). Prescribed boundary conditions are read in DFI (two per clement). FI is used as the right hand side vector that after solution contains the valucs of the unknowns. Finally both vectors are reordered to put all the values of the potentials in FI and all the values of the fluxcs in DF1 as was done for constant elements. Now, however FI contains onc potential and DFI two fluxes, per node. The program allows for the flux ‘before’ and ‘after’ any node to be different, When two, equal or different, fluxcs are prescribed at a node the potential is computed; if the potential and one flux are prescribed, the other flux is computed; and in the casc that only the potential is prescribed, both fluxes arc considered to be equal. {t should be noticed that in problems with only one uniform region, the case of potential prescribed and two different unknown values of the flux will only take place in a corner where the potential is prescribed along the two clements that join at that corner. This situation is not frequent and since the potential would be known along two different directions emerging from the corner, the potential derivatives along these two directions would.be known and consequently the flux along any direction would also be known. Thus, the three variables would be known at that corner and hence any two of them can be prescribed and the third will be computed. Notice that only for the case of a singularity on the corner would one require replacing the corner node by two different nodes inside cach of the two adjacent clements. The listing is as follow:2.6, Computer Code for Potential Problems using Linear Elements (POLINBE) 15 ce. eeeeeees © PROGRAM POLINBE © © PROGRAM 9 ¢ THIS PROGRAM SOLVES TWO DIMENSIONAL (PO)TENTIAL PROBLEMS © USING (LIN)EAR (BJOUNDARY () LEMENTS © CUARACTER® 10 FILEEN, FILEOUT COMMON /HATG/ (80,180) CORHON/HATH/ (80,80) COMMON WL INP, TPR DIMENSToW X81} ,¥(84) ,FI(80) ,DFI(160) DIMENSION KODE( 160) ,CX(20) .CY{20) ,POT( 20) ,FLUXI(20) ,FLUR2(20) s © SET WAKINUW DINENSION OF THE SYSTEM OF EQUATIONS (Nx) © RE © MAXIMUN NUMBER OF MODES = MAXINUN NUMBER OF ELEMENTS © wx NxLszeNx, ce © ASSIGN NUMBERS FOR INPUT AND OUTPUT FILES © Inpes, IPR c © READ MANES AND OPEN FILES FOR INPUT AND OUTPUT © WRITE(®,* (A) ")_* MAME OF THE INPUT FILE (AX, 10 CHART. |? READ(S,* (A) *) FELEIN OPEN( INP, FIL-E®FILEIN, STATUS=“OLD'} CUAL 0)" NAME OF THE OUTPUT FILE (MAX.10 CHART.1° ad sy pr.eour OPEN( IPR, FILESFILEOUT, STATUS: "NEW" ) © © READ DATA ce CALL INPUTPL(CK,C¥,X,Y,KODE, DPI) © © COMPUTE G AND H MATRICES AND FORM SYSTEK (A X * F) ce CALL GHMATPL(X,¥,G,H,FI,DPY,RODE,WX,NX1) ¢ € SOLVE SYSTEM OF EQUATIONS e ‘CALL, SLNPDCH, FI, DeNAHX) © © COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL POIRTS ce CALL INTERPL(FI,DFI ,KODE,CK,CY,X,Y,POT,FLUXI ,FLUX2) © © PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POINTS ce CALL OUTPTPL(X,Y FI sDFI,CX,CY POT, FLUKI ,FLUXZ) © close (1p) LOSE (IPR) STOP. END Routine INPUTPL The input subroutine is similar to INPUTPC in POCONBE. Only the boundary conditions are prescribed in a different way. Now, two boundary conditions per clement are read in array DFT. Thus. cach node ‘j* may have a different value of the flux. one as the end node of clement ‘j — 1° and the other as the start node of element 3".16 Chapter 2 Potentiat Problem a ja SUBROUTINE INPUTPL(CX, CY, X,Y, KODE, DPT) PROGRAM 10 N= NUMBER OF BOUNDARY ELEMENTS Lm NUMBER OF INTERNAL POINTS naanae CHARACTER®S0 TITLE COMMON NL, INP, TPR DIMENSTON €X(2} ,CY(1) ,X(2),¥(21) KODE (2) , DFE (1) WRITE (TPR, 100) 200 FoRMat(! },79(7«*)) READ JOB TITLE ana READ(INP, (A) ') TITLE WRITE(IPR,'(A)') TITLE READ NUMBER OF ELEMENTS AND INTERNAL POINTS a0 READ(INP, 4) N,L WRETE (TPR, 300)¥,L 300 FORMAT(//* OKTA'//2x, ‘NUMBER OF BOUNDARY ELEMENTS =',13/2X, ’NUMBER 1. OP INTERNAL POINTS WHERE THE FUNCTION IS CALCULATED =’, 13) READ BOUNDARY NODES COORDINATES IN ARRAYS X AND Y ana warre (TPR, 500) 300 FORNAT(//2X, “COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELE IMENTS! //2K, “POINT” 10%, °K! 18%, '¥) READ(INP,*) (X(1),¥(I),I=1,N) 00 10 T=i,8 10 WRITE (IPR, 700) 1, X(1) ,¥ (1) (700 PORMAT(3X,13,2(5X,£14.5)) READ BOUNDARY CONDITIONS IN DFI(I) VECTOR.IF KODE(I)"0 ‘THE DFI(T) VALUE TS A KNOWN POTENTIAL: IF KODE(I)=1 THE DFI(I) VALUE IS A KNOWN POTENTIAL DERIVATIVE (FLUX) ‘TWO BOUNDARY CONDITIONS ARE READ PER SLEMENT. ONE NODE MAY HAVE TWO DIFFERENP VALUES OF THE POTENTIAL DERIVATIVE BUT ONLY ONE VALUE OF THE POTENTIAL eaananaa WRITE(TPR, 800) 800 FORMAT(//2X, ‘BOUNDARY CONDITIONS’ //15X, !~----~~FIRST_NODE-- 19X, '>-=*-SECOND NODE*==-~"/17X, ‘PRESCRIBED’ , 20X, ‘PRESCRIBED’ / 2, 2k, "ELEMENT" , 12%, "VALUE! , 2X, "CODE", 24X, “VALUE? ,7X, “CODE? ) bo 30 T=1,N READ(INP,*)_KODE(2*I~1) , DFE (2*1-2) ,KODE(2*Z) , DPI (241) 20 WRITE (IPR, 950) 1, DFI(2#1"1) ,RODE(2*I~1) , DFI(2#I) , KODE (241) 950 FORMAT (2X, 13,2(10X,£14.7, 5%, 11)) READ COORDINATES OF THE INTERNAL POINTS ona IF(L.£9.0) GO To 30 READ(INP,*) (CX(I),C¥(Z) ,T=1,1) 30 RETURN END Routine GHMATPL Notice that this routine is similar to the one described in POCONBE with the main difference that the g" elements are assembled in a N x 2N matrix instead of an N x N one as was previously the case. This is because two possible values of flux are considered at cach node, one to the left and the other to the right of it. Then the boundary conditions are applied as described earlier to rearrange the system of equations and prepare it for solving.2.6, Computer Code for Potential Problems using Linear Elements (POLINBE) n The diagonal terms in H are computed implicitly. Assuming a constant potential over the whole boundary the flux must be zero and hence HI=0 (2.58) where I is a vector that for all nodes has a unit potential. Since (2.55) has to be satisfied x a H¥=—¥ Hi (forj¥i) (2.59) i which gives the diagonal coefficients in terms of the rest of the terms of the H matrix. The above considerations are strictly valid for close domains. When dealing with infinite or semi-infinite regions, equation (2.56) must be modified. If a’ unit potential is prescribed for a boundless domain the integral f ptar (2.60) " over the external boundary I’, at infinity will not be zero and since p* is due to a unit source, this integral must be {see equation (2.17) when r = e— 00) f ptdD=-1 (2.61) r ‘The diagonal terms for this case are, " H¥=t- Yo HY (for j#i) (2.62) m Notice that as all the terms H’ and G! are multiplied by 27 in the program, because the fundamental solution has been taken as In(1/r), equation (2.62) is also written in the program as H"=2x-Y HY (fori#j) (2.63) int SUBROUTINE GHHATPL(X,Y4G,M.F1,DFI KODE, NE,NX1) PROGRAM 11 ZMHIS SUBROUTINE coNPUTES THE 6 AND Hl WATRICES AND FORNS THE SYSTEN OF EQUATIONS A x AMie"a SQUARE HATEIX (N,K)5°0.15 RECTANGULAR (W,24N) anannna 9 Do io Jei,nn 10 0(1,3)=0. © COMPUTE THE CORFFICIENTS OF G AND H HATRICESci] ‘Chapter 2 panna annne X(Woryax(2) Yow av) Do 100 183m or NS#TaNo? Do 60 I9ENF NS IF (J3-i1)20,30,20 20 Jed3-8 30 7040 30 3033 40 CALL EXTIMPL(X(1) ,¥(1).X(3) YC) .X(991) .Y( 61) oA ARSBI BZ 1,DQIFL .DQ2F1,DQIF2,OqzF2,DULFL,DUZFi ,DULF2,DU2F2,0) TPC dow}42,43,43 42 WL 961) RET, 961 DAT 0 to'44 43: ACT, Tp AMT 1) 0Az 44: (Trobe (Td) OAL GC .ze9~1)5B1 GI1,263)eB2 BO MiE,T)AN(T,T)-A1=A2 Reeien-d RSeLsM DO 96 JJeKF NS IP(I3-¥70,70,60 60 JeJ9- co 7080 10 dead 80 CALL LOCIMPL(X(3) ,¥(3),X(191),Y(J41) ,B1,B2) IF(35-NF}B2,82,83, ez cHeat Bise2 B2-ck 2 G(T, 2ed~2)=B1 95 G(1.283)4B2 ADD ONE TO THE DIAGONAL COEFFICIENTS. FOR EXTERNAL PROBLEMS. TF(M(T,1)) 98,100, 100 pa W(T,1)=6.20318526H(T,3) 100 CONTINUE REORDER THE COLUMNS OF THE SYSTEM OF EQUATIONS IN ACCORDANCE WITH THE BOUNDARY CONDITIONS AND FORM THE SYSTEM NATRIX A WHICH IS STORED IN H Do 155 Te Do 180 Je TF(KODE(2#i-243))110,110, 150 110 IF(TANE.N OR, 3.HE.2) 00 TO 125 IP(RODE(1)} 115,118,119 113 bo 174 Kel 7 aa 118 DO 116 Rea MO) MCR, 1) -GER, 200) 18 GOK. 2eH=0. 00 70 180 428 TR(I-EQ.1 JOR. J.GT.1 .OR. RODE(2#1-2).89.1) GO 70 130 Do 125 Ke,n WORT) HEX T) G(x, 287-1) 329 G(K,201-1)40, 80 40 150 130 DO 132 Kea CueHEK,T-18) WK Loit3)9=G(K, 201-263) 132 G(K,2eT-2e3 }2ec 150 CONTINUE 4158 CONTINUE FORK THE RIGHT HAND SIDE VECTOR F WHICH 1 STORED IM FI DO 160 Tet y¥ FI(I)«0. DO.160 Je14KN FLL) #F1C1}+G(1,3)9DFI (9) 160 CONTIRUE RETURN END Potentiat Problems2.6, Computer Code for Potential Problems using Linear Elements (POLINBE) vn Routine EXTINPL This routine is similar to the one in POCONBE but instead of computing only one value per element for each coefficient as in POCONBE, it now computes two values per element, i.e. the parts of the coefficients corresponding to the adjacent nodes. oo ‘SUBROUTINE. EXTINPL(XP.YP.X1.71.K2,¥24A 4,D@1F1,DQEPL,DGIF2,DQzF2,DUIFI ,DUZF1 ;DUIF2,DURFZ,K) c © PROGRAM 12 c € THIS SUBROUTINE COMPUTES THE INTEGRAL OF SEVERAL NON-SINCULAR © FUNCTIONS ALONG THE BOUNDARY ELEMENTS USING A FOUR POINTS © GAUSS QUADRATURE © WHEN K*0, THE OFF DIAGONAL COEFFICIENTS OF THE H AND G MATRICES © ARE compuTED © WHEN KSI, ALL THE COEFFICIENTS NEEDED FOR COMPUTATION © OF THE POTENTIAL AND FLUXES AT INTERNAL POINTS ARE © COMPUTED (AI,AZ,B1,B2,FA1,P21,F12,F22,E11,E21,E12,E22) € © RAs RADIUS = DISTANCE FROM TRE COLOCATION POINT TO THE © GAUSS INTEGRATION POINTS ON THE BOUNDARY ELEMENTS © ETAL,ETA2 © COMPUNENTS OF THE UNIT NORMAL 70 THE ELEMENT © RDJ,RD2,RDN = RADIUS DERIVATIVES e DIMENSION XCOL4),YCO14) ,G14) MELE? DATA GH70. 08611362) 0-33: 0..33998104/ 52145157 « axe (x3-110/2 BXe (x20x1)72 Aveta nn 72 SLeSanrUateezeayee2) ETAL® AY/SL ETA2S-AK/SL. Aso. Azo. Bio. 2-0 Dairiso. UzF2=0. © © COMPUTE THE TERNS TO BE INCLUDED IN THR G AND H MATRICES © OR THE TERMS NEEDED FOR COMPUTATION OF THE POTENTIAL, € AND FLUXES AT THTERNAL POINTS c po 40 181.4 XCOUT }eAXeGI (1). Yoo! 1} sa¥eat (1). v= SQHT (XP-KCOU I) )##24 (¥P-YCO(1) }##2) RDI=(KCO()-XP)/RA D> YCO(1)=¥PI/RA ‘RDN=RDI*ETAI+RD2*ETA2 Fie(1.-@1(1))/2. F2e(1.90H(1))/2- IF(K)"30,30,10 10 DUL|RDI*OME(T}9SL/RA DUZ=RDZ*ONE(1)8SL/RA DQi=((2-#RD1¢42~1.)#ETAL*2. #RD #RDZ#ETAZ) *OME(1)8SL/RACS? DaZe-((2.*RD2*42-1,]#ETA242, *RD1#RDZ*ETAL )“OME(1)®SL/RAS#Z DQIFieDQiFi+D@iF) DOLF2=pqiF2~DQ)+F2 Dg2Fi=pg2Fi+Daz*Fi DezF2-bazF2+Daz*F2 DUIFI-DUIFI+DUISFi DULF2+DUIF2+DUI+F2 DuzF1«DUZF1+DUZeFI DUZF2eDu2F2+DUDeF?80 Chapter 2 Potential Problems 30 MaRDNEONE(T)3SL/RA GALOG(1/RA) OME 1)#8L, AL=AL-FIOH Routine LOCINPL. This routine computes now the part of the elements of the matrix G corresponding to the integrals along the elements which include the singularity. These integrals are: Bi=f@,In () ar yy , B2= J d.in () ar yy O, (2.64) Using the local system of coordinates in figure 2.9, the integrals can be written as Point (2) 1 1 1 Bl= f =m tn(2) ar=1fa—mm(2) an , Point (1) r ° a, (2.65) 2? n(S)ar Wf n(t) a B2= ~)ar = nf — mad r Ce if3 BI -43 - mo (2.66) a 9. ” 2 a=0 aed t ee | Figure 2.9 Linear element tocal coordinate system2.6. Computer Code for Potential Problems using Linear Elements (POLINBE) ° ‘SUBROUTINE LOCIWPL(X1,¥1 42, ¥2, PROGRAW 13 ‘THIS SUBROUTINE COMPUTES THE PARTS OF THE @ MATRIX COEFFICIENTS CORRESPONDING TO INTEGRALS ALONG AN ELEMENT ‘THAT INCLUDES THE COLLOCATION POINT. SEPeSQRT((X2-x1) #824 (¥2-Y1}0#2) BLeSEPS(1.5-ALOG(SEP) )/2 B2ESEP+ (0. 5-aLOG( SEP) )/2 ‘RETURN EXD aanaana Routine INTERPL This routine replac: 81 s the INTER PC program used in POCONBE. It first arranges all potentials in FI and all their derivatives (or fluxes) in DFT and then computes the values of potentials and fluxes at the internal points if requested. ‘SUBROUTINE INTERPL(PI, DPI ,KODE,CK,CY,X,Y, POT, FLUX! , FLUZ2) c © PROGRAM 14 ¢ © THES SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL G AND THE POTENTIAL DERIVATIVES (FLUXES) AT INTERNAL POINTS c COMMON WL, THP TPR DINENS1OW #1(1} ,DFY (1) , RODE 3) .OX(1),C¥C1),.200),9C1) DIMENSION POT(1},FLUEI{1),FLUKZ(1) c © REARRANGE THE FI AND DF1_ARRAYS TO STORE ALL THE VALUES OF THE © POTENTIAL IN FI AND ALL THE VALUES OF THE DERIVATIVE 1M DFT © DO 155 1etW BO 180 Je1y2 TF(KODE( 241-23) 110,110, 180 130 1F(1.ME.N OR. J.NE.2) @0 TO 125 APCKODE()) 11441245213 113 GHFFLGD FLL) =DET(zeN) DFI(20n) cH 6 To 150 114 DFA(zeny=DFI(1) G0 70 150 125 I(T. Be. DFA(z#1- 150i (201 Go 70 150 130 CHSFI(I~163) FI(1a169)«DF1( 201-263) DPLzei-2e3 ec 150 CONTINUE 185 CONTINUE OR. 3.9.2 OR. KODE(2#1-2).Eq.1) 00 70 130 a COMPUTE THE POTENTIAL AND THE FLUXES AT INTERNAL PDINTS 1F(L,EQ.0) G0 TO $0 O40 KelyL POT(K)=0. FLUKL(K)#0, FLUK2(K) #0. O30. 38148 CALL EXTINPL (CHEK) .CYEK) XC) ¥L2) ,X( Jo) E961) ATLA 43, D917), Da2Fd, DQ1F2,DQ2F2,DULFi ,DUZF1DUIFZ, DUZF2,1) iF(s-n)32,33,33 32 POT(K) =POT(K) +DFI(2#3-1 FLUE (K)=FLUEI (X) ¢DPI(293- 35F1(3)#DQIF1~PI( Jo) DQ1F2 FLUKZ(K)*FLUR2(K) ¢DFL (280. A=FI(9 }9DQ2F1-FY (301) #DQ2F2 ‘a0 TO 30 3,32 DFI( 285) SB2-FI(J)SAI@FI(J41)0A2 /eDUIFA®DFI( 243} eDUIF2 *DUZFI+DFI (265) *DUZFZ82 Chapter 2 Potential Problems 33 POT(K) sPOT(K) #DPI (203-1) 9B1 ¢DFI(203) #02: PLUK (Xk) PLUEI (K) +DFH (203~1) #DUIPIDFI( 3971 (3) SDGIFL“FI(1) #0QLE2 ‘FLUK2(X) #PLUX2(K) ¢DF1 (283-1) *DU2F1¢DFI(2*3) #DU2F2 A-FI(3)9DQ2F1-F1(2) #0q2F2) 418926) SPLUEL(R) /(2.03.1415026) so PLUna(apSPLUEZ(N)/(zc09 1418020) 40 RETURN END 1(3)9A2-FI(L) AAD jeDuiF? Routine OUTPTPL This routine is similar to the onc described in POCONBE but instead of printing the mid-point coordinates it now gives directly the valucs of the coordinates in the X and Y arrays. It also gives two values for the flux at each boundary node: ‘one ‘before’ and the other ‘after’ the node, SUBROUTINE OUTPEPL(X,¥,F1 ,DF1,CX,CY,POT,FLUY1,FLUX2) PROGRAM 15 ‘THIS SUROUTINE PRINTS THE VALUES OF THE POTENTIAL AMD ITS NORWAL DERIVATIVE AT BOUNDARY NODES. IT ALSO PRINTS THE YALUES OF THE POTENTIAL AT INTERNAL POINTS. aeneana COMMON WL, NPS IPR DIMENSION X(1),YE1) -FE(1) ,DFI(1) seXC1),C¥(1) DIMENSION POT(i) FLUXL(1),FLUX2(2) WRITE(IPR,100) 100 FORMAT(* *,18(+8")//2x, *RESULTS*//2X, “BOUNDARY NODES*// 6X, POTENTIAL DERIVATIVE'/ 20x, $x" 18K, "Y", 12%, "POTENTIAL" ,6X, "BEFORE NODE’ ,6X, "AFTER NODE"/) WRITE 1PR,200)"X(1} 44(1) 4F1(1),DFE(2N) .DFL(A) DO 10 Tez.N Yo WRITE( IPR,200) X(1),7(1)41 200 FORMAT(S(2x,E14.5)) IF(1.2@.0) Go To 30 WRITE(TPR,300) 300 FORMAT(//,2X, INTERNAL POINTS" //93 49K, "FLUX X* 10x, "FLUX ¥'/) (1) ,DPI (20-2) DPI ( 201-1) 132K, ‘POTENTIAL, bo" 30 Kew 420 YRITECLTR’400)0R( 1) sCY(R) sPOT(K sFLUR (8) sFLUXZ(K) eb vonvarstateeteesy) So eniTettPn;3i0) 200 onvate ,19(°=")) heron tx Example 2.2 The potential problem solved with POCONBE will now be analysed using linear elements as shown in figure 2.10, The number of elements is still 12 for the lincar (figure 2.10(b)) as well as the constant (figure 2.10(a)}. Although the constant element solution gives reasonable agreement with the known results, the lincar solution is identical to the analytical one within the precision limits of the computer. This result was to be expected since the exact solution varies linearly.2.6. Computer Code for Potential Problems using Lincar Elements (POLINBE) 83 ‘ies {@) Results obtained with the consiant element code 2522 (POCONBE) 12 atements solution 2000 1000 Fi-50.0ti-s0.01 1500 \~50.0) heoo tiooo {-50.0) (-50.0) 2000 (0) Results obtained with the linear element code 3000 (POLINBE) 12 olaments solution Figure 2.10 Results obtained using constant and linear elements for the heat flow caample84 Chapter 2 Potential Problems {c) Results obtained with the linear element code 300 (POLINBE) 4 elements solution Figure 2.10 continued The data corresponding to the 12 element problem is as follows. HEAT FLOW EXAMPLE (DATA) HEAT PLOW EXAMPLE (12 LINEAR ELEMENTS) + 0. 2. 0. 4. 0 6. 0. 6. 2. 6. 4. 6. 6. 2. 6.0. 6. 0, 4. 0. 2. Hupeoouny 300. 0300. 300. 0 300. 300: 0 300. 22 a a a and the output is given by HEAT FLOW EXAMPLE (OUTPUT) NEAT FLOW EXAMPLE (12 LINEAR ELEMENTS) DATA NUMBER OF BOUNDARY ELEMENTS. = 17 NUMBER OF INTERNAL POINTS MERE THE FUNCTION IS CALCULATED = 82.6. Computer Code for Potential Problems using Linear Blements (POLINBE) a5 COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELENENTS point x y 1 -p0ca0E+00 -000008+00 2 ‘20000801, ‘eoaeor+00. 2 ‘400008+01 Toov00E+00 ‘ ‘eo000c+01 Tooo00e+00 5 Teonoor+e1 ‘20000801 6 1000F+01 [400006+01, 7 ‘s0000K+01 ‘eo000801 ° V0000E-01 Te00008+01 ‘2o00eEo1 ‘so000E+01 loocese+00 © = saa00E+01 To0000k +00 ‘ao0008501 ‘nocor+o0. ‘20000601, BOUNDARY CONDITIONS: SECOND. NODE: PRESCRIBED ELEMENT. ‘VALUE cone ‘VALUE cope 1 seoao000n+00 sooon0006+00 2 ‘eonooo0r+00 = 4 3 Toaosooors00 s 4 Toooaacok-00 =o ° 5 Secooo00Es00 =o ° ‘6 Teosoeoors00 =o ° 1 Tovsooogss00 i 3 ‘ooooasoesoo = i ° Tooooooer-es = ‘ooosoo0e«00 = 0 Tso0o0008-03o ‘so00000+03 oo 1 Tavo0000s+03 =o ‘aeoao00E+03 0 RESULTS, BOUNDARY NODES POTENTIAL DERIVATIVE x Y POTENTIAL BEFORE NODE AFTER KODE -000008+00 90000800 200008403 -So000rs02 00000400 Tzovo0rot ToeoonE+00 ‘20000403 ‘o0o008+00 Toooaor+00 Teoooors00 ‘iopoorso3 Tooooer+o0 Tooeaor+00 ‘oooe0E+ 00 ‘o0000E+00 ‘oooccz+oo © -: $0000 +02 ‘eoo0E+00 = :80000E+0z =: $00006+02, ‘oooooesoa = =:So000E+ez © —"s0000E+0z ‘00000800 ‘s00d0E~02 ‘00000800 ‘10000803 Toosoceoo Tooen0e-00 Szooo0808 Tooosoesoo Tocoo0E+00 Sooeoer-00 ‘sooeseso1 ‘Soooceso3 loo000F+00 prt ‘eooo0rs 00 “400008 ‘300008+03. ‘so0008+02 {so0o0rs02 0000E+ 00 [20000801 [s00008+03 ‘800005+02 ‘soocee+02 TERNAL POINTS. x y voTRNTIAL mux x 20000801 +200008¢03 so001E+02 ‘zo000E+01 ‘zooooE+o3 'SO001E+ 02 ‘s0000E+01 ‘iso00ee03 — -:s00008«02 ‘ao0008+01 ‘100008+03. 'S00008+02 DE+03 Example 23 It is interesting to note that in this case even a simple four elements representation can give exact results using double value of the flux at the corners (figure 2.10(c)).86 Chapter 2 Potential Problems ‘The input in this case is HEAT FLOW EXAMPLE (DATA) HEAT FLOW EXAMPLE (4 LINEAR ELEMENTS) 43 0, 0, 6. 0. 6. 6. 0, 6 10.1 0. 00, 00. 10, 10. 0 300. 0 300. 243. 3 a The corresponding output is very accurate taking into consideration the simplicity of the mesh. HEAT FLOW EXAMPLE (OUTPUT) REAT FLOW EXAMPLE (4 LINEAR ELEMENTS) DATA, NUMBER OF BOUNDARY ELEMENTS © 4 NUMBER OF INTERNAL POINTS WHERE THE FUNCTION 1 CALCULATED =» 3, COORDINATES OF THE EXTREME POINTS OF THE BOUNDARY ELENEMTS point x y 1 -800008+00 -000008+00 2 + Teop00Es01 ToooooE+00 3 ‘600008001 ‘soooveso1 ‘ ‘oo000E+00 ‘eeoo0Eso1 BOUNDARY CONDITIONS. FIRST WODE- ‘PRESCRIBED ‘VALUE. +9000000E+00 Teoooo0r +00 ‘evoon00e+00 ‘ovoseoeE+00 Teoceseoe+ 00 RESULTS BOUNDARY NODES. POTENTIAL DERIVATIVE x ¥ POTENTIAL «BEFORE NODE AFTER NODE 000008400 +000008+00 0000807 =000008+00 90008601 Tooa0E+ 90 ‘ooooorsoo =. 8000007 {60000E+0; Teovo0gsar '30000E+02 000006+00. lo00008+00 Teooo0Es01 TS00008+03 ‘eo000e+00 Tso0cuEs02 INTERNAL POINTS g ¥ POTENTIAL FLux 200008401 -400008+01 201388003 -28160F+01, {So000E+01 {30000E+01 80782E-06 34782E+002.6 Computer Code for Potential Problems using Linear Elements (POLINBE) 87 2.7 Discontinuous Elements To avoid the problem of having two unknown fluxes at a corner node (for which only one boundary element equation can be written) the nodes of the two linear elements which mect at the corner can be shifted inside the two elements. The nodes remain as (wo distinet nodes (sce figure 2.11) and one equation can be writ- ten for each node. The potential and the flux are represented by linear functions along the whole elements in terms of their nodal values and both of them are in principle discontinuous at the corncr. Discontinuous elements are also useful for situations in which one of the variables takes an infinite value at the end of the element (for instance al a reentry corner or in fracture mechanics applications). In such cases the value of the variable at the node shifted from the end of the ele- ment is finite and can be computed from the system of equations without numerical difficulties. The values of w and q at any point on a linear element have been defined in terms of their values at the extreme points by equation (2.42). ! MG) = duu! + d.0? = [bibs] {at (2.67) 1 = ba! + 60? = toro {oh Ifthe two nodes of an clement have been shifted from the ends distances a and b respectively as shown in figure 2.11 any of the (wo equations (2.67) can be particularized for the nodes. Nodal vatue of vora Nodal vatue of vorg Figure 2.11, Discontinuous elements88 Chapter 2. Potential Problems wl fouled soi} tot bye Glen) | (2.68) where £, = (2a/l) ~ | and &, = 1 — (2b/f) are the local coordinates for the nodal points. Equation (2.68) can be inserted and after substitution into (2.67) yields the value of w at any point on the element in terms of the nodal values uf (2) = [8:10 { 2.69) where t fl-b -a a-bl-b t-a ‘The same relation can be written for the Mux H)= [631 @ {7} 270) Alter discretizing the boundary into N elements the integral statement for a node “i” can be written as ” x cult ¥ fought dl = ¥ f utg a 7m) ir, fit, ‘The integrals over a discontinuous clement “j" are Jug? ar = Stoo 100° aft oh nea {ie % (2.72) ft a= § C6.d) Qt dt {it = = Lae a) 7 ‘When solving a potential problem continuous and discontinuous elements can be used together in the same mesh. The total number of nodes be equal to the total number af clements plus one additional node per each discontinuous element. The coefficient c’ is equal to 0.5 for the nodes on discontinuous elements. The integrals hi, hi, x and gi along the discontinuous elements given by equation (2.72) can be computed by the usual Gaussian numerical quadrature when the node ‘*’** does not belong to the element. When ‘‘i”* is one of the nodes of the discontinuous element hi = hj =0 and gi and gf can be easily obtained by analytical integration. The element is subdivided into two parts one at each side2.7. Quadratic and Higher Order Elements 89 ‘of the node. The resulting integrals consist of the same basic integrals as those of the regular linear elements given by equation (2.66). 2.8 Quadratic and Higher, Order Elements It is usually more convenient for arbitrary geometries to implement some type of curvilinear elements. The simplest of these are the three noded quadratic elements which require working with transformations. Consider the curved boundary shown in figure 2.12 where I is defined along the boundary and the F position vector is a function of the cartesian system, x, x3. The variables « or q can be written in terms of interpolation functions which are functions of the homogeneous coordinate ie. WE) = yet + pau? + by? =[br bros] 4 we 2.73) a He) = 14! + b247 + da? = [6.02031 54° , where the interpolation functions are I= S45 bs = hE +8) (2.74) 2 Teta) Figure 2.12 Curved boundary90 Chapter 2 Potential Problems oF @ vasianon fot eo gee Retesence system Figure 2.13 Quadratic element These functions are quadratic in ¢ and give the nodal values of the variable w or 4 when specialized for the nodes: i.e. with reference to figure 2.13 (2.75) The integrals along any ‘i element are similar to those computed for the linear element, but there are now three nodal unknowns and the integration requires the use of a Jacobian. Consider for instance an integral for the H type terms, i.e. wu! f uldyq* dP = { [didada]g* at yu? + 4 a w = (HME 0? 2.76) oe where w= foatdr; M=Jdatdry Y= f dygt at 2.7 ty % ty2.8. Quadratic and Higher Order Elements. a1 The evaluation of these terms requires the use of a Jacobian as the ¢, functions are expressed in terms of €, but the integrals are functions of I. For a curve such as given in figure 2. 12, the transformation is simple, {i (ey Cay fas =|G|ag (2.78) where |G| is the Jacobian, Hence one can write, Note We= fold =f d(Ebar|G|ag 2.79) Node Formulae such as (2.66) are generally (oo difficult to integrate analytically and numerical integration must be used in all cases, including those elements with a singularity. Notice that in order to calculate the value of the Jacobian {G| in (2.78) one needs 1 know the variation of the x, and x, coordinates in terms of &. This can be done by defining the geometrical shape of the element in the same way as the variables w and q are defined, i.e. using quadratic imerpolation, X= ONt + ONT + PNT X= GaN} + 2X3 + GaN} (2.80) where the superscript indicates the number of the node. This is a similar concept to the isoparametric elements commonly used in finite elements. cu ic Elements Elements of order higher than quadratic are seldom used in practice, but they may be interesting in some particular applications. Because of this we will briefly describe the case of elements with cubic variation of geometry, and u or q variables. In this case the functions are described by taking four nodes over each element (figure 2.14). u= bul + Pri? 4.60 + bau (2.81) q= dia! + bag? + O39? + bag* and similarly ix} + 2x7 + Gaxt + hart Xa Fie) + G2xT + G3Xt + Gary 2.82) Kp Qix3 + GX) + h3X} + bard92 Chapter 2. Potential Problems a) 8) o @ Reference element Figure 2.14 Cubic elements with four nodes where the interpolation functions are, b= il BLOF HE FD] G2 = sell + EN - 10+ 95? + 1) o3= fell — PML — 38) = el — 7M + 3) (2.83) which can be specialized at the nodes as follows, (2.84)2.8, Computer Code for Potential Problems using Quadratic Elements (POQUABE) 93. Another possibility with cubic elements is to define the variation of w or q in terms of the function and its derivative along the element, at the two end points — ie nodes | and 2 - as shown in figure 2.15. The corresponding function for (same applies for q and x,x,) is then given by ‘ ‘ w= out +6(3*) tut o(‘*) (2.85) with br. =KE-WPEF2 b= —UE-IPE +) 2.86) = HEF IPE 2) bg AEH FE -1) where { is the element length. eat so goat Node 1 Node 2 ke > 12 ——} Reterence element igure 2.15 Cubic clements with only two nodes4 Chapter 2 Potential Problems This last type of cubie clement could be used in cases where we wish to have a correct definition of the derivative along [, for instance to calculate fluxes in that direction, or if we prefer to reduce the number of nodes along the element. In some cases it may still be better to continue defining the geometry with four nodes as it is generally more difficult to have accurate results for the slopes. 2.9 Computer Code for Potential Problems using Quadratic Elements (POQUABE) In what follows a FORTRAN code for potential problems using quadratic elements is described. The program has the same organization as the two previously studied. All variables in the code have the same meaning as for the linear element program (POLINRE). Fl and DFI have a slightly different form. The dimension of FL is (N), N being the number of nodes and that of DFI is (3N£), where NE is the number of elements. The prescribed boundary conditions are read in DFI {three per element). The program allows for the values of flux at both sides of the nodes connecting two elements to be different. Then, (i) when both fluxes are prescribed as different at both sides of the node, the potential is the only unknown; (ii) when the potential and one flux are prescribed, the other flux is the unknown and (iii) if only the potential is prescribed, one value of the flux is the unknown and will be the same on both sides of the node. Thus, the situation at corner nodes of quadratic elements is the same as for linear elements. Main Program The program follows the same structure of the constant and linear potential codes. The listing is as follows. PROGRAN POQUABE PROGRAM 16 THIS PROGRAM SOLVES TWO DIMENSIONAL (PO) TENTIAL PROBLEMS USING (QUAJDRATEC (BJOUNDARY (EDLENENTS, CHARACTER® 10 PILEIR, FILEOUT COMMON /HATG 7G, 100,180} COHMON/WATR/H( 100,100) CORMON NLL, INP, TPR DIWENSION X(101},( 101) DIMENSION DEI(156) ,P1(100) , RODE( 150) DIMENSION CX{20) ,C¥(20) ,POT( 20) ,PLUR3 (20) ,FLUK2(20) ‘SET MAXINUN DIMENSION OF THE SYSTEM OF EQUATIONS (Nx) Xs MAXIMUM NUMBER OF KODES* 2eMAXIWUN NUMBER OF ELEMENTS WXis SeMAXIMUN NUMBER ELEMENTS ennnnne ag enaan is MK 1502.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 95. ASSIGN NUMBERS FOR INPUT AND OUTPUT FILES Ines IPRS READ NAMES AND OPEN FILES FOR INPUT AND OUTPUT WRETE(®," (A) ') * NAME GF INPUT FILE (MAX. 10 CHART.)* READCS,! (a) *)PILEIN, OPEN( INP, FILE®FILEIN, STATUS® "OLD" WRITE(*," (4) _") * NAME OF OUTPUT ILE (MAX. 10 CHART.)* READ(*,* (A) *)FILBOUT. OPEN(1PR,PILE*P3LEOUT ,STATUS="NEW") READ DATA CALL INPUTPQUCK,CY,,¥,KODE,DFI) COMPUTE AND G MATRICES AND PORN SYSTEN (A x © F) CALL ONMATPQUX,Y.6,H,F2, DF KODE N,N) SOLVE SYSTEM OF EQUATIONS ‘CALL SLNPD(H,FI \DoN.HR) COMPUTE POTENTIAL VALUES AT INTERNAL POINTS. (CALL INTERPOIFI ,DFI.KODE.CL,CY.X,Y,POT,FLUXY FLUI2) PRINT RESULTS AT BOUNDARY NODES AND INTERNAL POINTS CALL OUTPTPOLI.Y,F1,DF1,CK,CY.POT FLUE, FLUX2) CLOSE INPUT AND OUTPUT FILES ‘CLOSE (INP) CLOSE (PR) stop ND Routine INPUTPQ This subroutine reads all the input required by the program and requests a file from the user containing the following lines: (i) Title Line Contains the title of the problem (ii) Basic Parameter Line Contains the number of elements and the number of internal points. (iii) Boundary Nodes Coordinates Lines Contains x,x2 coordinates read counter- clock wise for external boundaries and clockwise for internal ones. The lines are organized in free format. (iv) Boundary Conditions Lines As many lines as boundary elements. Thiee values of KODE and the known variables are read for each element, corresponding to the three nodes. In this way a value of the flux may be prescribed for an extreme node as part of one element and a different value as part of the other clement. The potential however must be unique for any node and the flux must also be unique for the mid-node of any element. The known variables are the potential if KODE=0 and the flux if KODE = I. The order of reading is first KODE(/) and then value of the variable (I) for [= 1, 2, 3.9% Chapter 2 Potential Problems (v) Internal Points Coordinates Lines Contain x,x, coordinates of the internal points organized in free format, There will be one or morc lines if necessary. This subroutine first prints the name of the run and the basic parameters. Then the coordinates of the nodes and the boundary conditions for each clement, with codes and prescribed values are printed. The internal points coordinates arc only printed in the subroutine OUTPTPQ. The FORTRAN listing of INPUTPQ is as follows: ‘SUBROUTINE INPUTPOICK,CY,X,7, LODE, DF1) PROGRAM 17 Ee MUNBER OF BOUNDARY ELEXENTS Nos MUNRER OF BOUNDARY NODES = 2 © NE 1 © NUMMER OF INTERNAL POINTS. aannece g CHARACTER®G0 TITLE CCOMNON WL, INP, TPR DIMENSION KODE(1),X(3).713),cR(1) .CVE1) ,DFI(1) WRITE(IPR;100) 00 FoRMAT(* #,790°#*)) READ JOR TITLE READ(INP,"(a)') TITLE WAYTECIPR,"(A)") TITLE [READ MUMBER OF BOUNDARY ELEMENTS AND INTERNAL POINTS READ(INP, #) NEL. WAITE(EPR,210)NE 210 FORMAT //2X, "DATA" /2X,'NUNDER OF BOUNDARY ELEMENTS®* 113/2%,"NUNBER OF INTERNAL, POINTS=",13) Ne2ONE READ BOUNDARY NODES COORDINATES IN ARRAYS X AND Y ana 2X1), 100) TL a) O10 Jei,6 40 WRITE(SPRIZ40) 16x12),YU1) 500 FORKAT(//2X, "BOUNDARY NODES COORDINATES'///4x, APNODE 10K, 5X? 18x, "¥47/) 240 FORMAT(SX,33,215%,E14.7)) READ BOUNDARY CONDITIONS 3N DFI(1) VECTOR, IF KoE(1)=0 THE DFI(1) VALUE 13 A KNOWN POTENTIAL; 2F KODEIT)*1 THE. DPIC) VALUE 3S.A KNOWN POTENTIAL DERIVATIVE (FLUX). ‘THREE BOUNDARY CONDITIONS ARE READ PER ELEMENT. NODES. BETWEEN TWO ELEMENTS. WAY HAVE TWO DIFFERENT VALUES OF THE POTENTIAL DERIVATIVE BUT ONLY ONE VALUE OF THE POTENTIAL. ananaanan WRITE(1PR,800) #00 FORNAT(//3X, BOUNDARY CONDITIONS? //12%, SECOND NODE~=o--" 3X, SRIBED* «14X, D*/1X, "ELEMENT" » BX, "VALUE" CODE” 8k, “VALUE’ ,7X, SCODE" ) (x0De (03-963) ,DFI( 387-905) 9819) 383) ,MODE( 383~-343),901,3) READ COORDINATES OF THE INTERNAL POINTS. JF(L.EG.0) 0 To 30 READUINP,*) (CKUT) sC¥(1) 41 30 RETURN END oan2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 97 Routine GHMATPQ This subsoutine computes the G and H_ system matrices by calling routines EXTINPQ and LOCINPQ. EXTINPQ: Computes the GW and HW (3) submatrices which relate a collocation point with an clement as defined by its three nodes. The collocation point is not any one of the element node: LOCINPQ: Computes the GW (3) submatrix for the case when the collocation point is one of the nodes within the element under consideration (i.e. the singularity is in the same clement). Notice tha: the corre- sponding HW (3) is computed using EXTINPQ because the singularity will occur only on the diagonal term and this is computed later on by adding the off-diagonal terms of the row. The resulting GW and HW submatrices arc assembled in the G and H system matrices. Matrix G is now rectangular since each extreme node of un clement may have different fluxes, i.e. onc ‘before’ and another ‘after’ the node. The diagonal terms in 1 are computed using constant potential considerations, which results in adding row coefficients together. Once the matrices H and G are assembled, the system of equations needs to be reordered in accordance with the boundary conditions to form AX =F where X is.a (N) vector of unknowns, N being the number of nodes; A is a (N x N) matrix whose columns arc 4 combination of columns of H or G depending on the boundary conditions or of two consecutive columns of G when the unknown is the unique value of the tractions at both sides of the extreme node of an element; F is a known vector computed by multiplying the prescribed boundary conditions by the corresponding row terms of G or H. At the end of the subroutine GHMATPQ and after rearranging H contains the matrix A, and FI the F vector. The FORTRAN listing of GHMATPQ is as follows. ‘SUBROUTINE GHHATPO(X,¥.G,H,FI,DFI.RODENA.NX1) PROGRAN 18 Th1S SUBROUTINE COMPUTES THE G AMD H MATRICES AND FORKS ‘TWE SYSTEM OF EQUATIONS AX» F WIS A SQUARE MATRIX (Z8NE,20NE); @ 1S RECTANGULAR (2*NE,3®NE) DIMENSION X€1),¥E1) ,6(00 NRL) MCR 1K) DIMENSION MW( 3} ,Gw(3) ,DQI¥(3) ,Dq2™) 3) ,DULW{ 3) ,DUaM{3) DIMENSION F1(1},0F1(1},RODEC I) COMMON Hy, INP, TPR Ween? D0 20 t91,8 anannaa ¢ 0.1 a1 WUT 0 iz98 Chapter 2 Potential Problems c c © 17 GLT.3)40, 20 CONTINUE Xewetbaxcny Vonet eV) COMPUTE THE OW AND RY MATRICES FOR EACH COLLOCATION POINT AND EACH BOUNDARY ELEMENT 0 40 Lit DO40 Te1,Re1,2 HR(CuboDYA(LL"3-1 ) eC uL—t-2){LE-LoN-2)) 22,2242 21 RODOSLL- 141 TRCCLL-EQ.1).AND.(1.E0.N-1}) NODO=NoDO+ CALL EXTINPQX(LL) 5 Y(LLD (1), VOU) ,XULo1) PCL) XC 142) ,¥CLO2) uM ,W ,DG1W,DQ2u, DULY, DUNO) GALL LOCUNPaCK(I} ,VCL),X(140), 14164) .x(162) 11162) 64, 8ODO? To 34 22 CALL EXTINPQ(X(LL) .YCLL) XL) YUL) oXUE+1) YAT¢0) XU142) .VUA2) '# HW oN, D91N, 092 ,DUIV, DURW,0) PLUG THE GW AND HW MATRICES INTO THE GENERAL G AND H MATRICES. 34 DO 38 Ja1,3 3) or) AF(12N03)"37,38,37 35 1813-3) 37,36,38 36 WELL, 2 )9MLL, thoMWt a) 90 TO 38, 37 HOLL, 1-163 )aH(LL,T-163) 64013) ‘38 CONTINUE 40 CONTINUE COMPUTE THE DIAGONAL COEFFICIENTS OF THE H MATRIX o10 rere BUDO. Do 80 asian IF(T.FQ.3) G0 70 60 MOET PHC TD=HCL SD 60 CONTINUE ADD ONE TO THE DIAGONAL COEFFICIENTS FOR EXTERNAL PROBLEMS. ICEL IY) 68,70,70 65 WC1,1)26, 20918520H(1,2) 70 CONTINUE REORDER THE COLUNNS OF THE SYSTEM OF EQUATIONS IN ACCORDANCE NITH TNE BOUNDARY CONDITIONS AND FORM SYSTEM ATRIX A WHICH 15 STORED IN W o 180 151,NE bo 170 Je1,3 IF(KODE(S#i=3+3)) 110,110,170 110 IF((1-NE).NE.O LOR. J.NE.3) GO TO 126 P(RODE(1)) 115,115,113 113 0-114 Ket CHeHLK I) MAK, 1)5-6(K, 387) eesteo ite) co 70 170 118 Do 116 Ke1w WO, DH 1-60, 901) 136 GUK,301}50. 90 70 170 128 IF(1.EQ.1 JOR. 3.67.1 OR. KODE(3#1-3).E9.1) GO 70 130 Do 120 ker. B(x, 201-1) 2H(x,261-1)-6(R, 381-2) 129 G(x, 301-2) 20 G0 70 170 130 DO 132 Kern CHeMiK, 261-243) W(t, 281-263 )=-G(K 341-363) 192 Gikyded-ded econ 170 continue 180 CONTINUE FORM THE RIGHT MAND SIDE VECTOR F WHICH IS STORED IN FI Do 190 14,4 FIUI)® 0.2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 99 Do 185 Jet. 3NE aes PECL) SFIC1 SGU.) 9DF113) 190 conn RETU END Routine EXTINPQ This subroutine computes using, numerical integration, the (3) submatrices GW and HW that correspond to an element when the collocation point is at a node other than any of those 3 in the clement. The correlation of the collocation points are XP and YP. The integrals are of the type (2.76), i.e. Hw J dat r= T oatla| ag (2.87) ay “1 GW= f gut at= J gulGlae (2.88) um 1 This subroutine also computes the (3) submatrices DULW, DU2W, DQLW and DQ2W which are needed to obtain the x, and x, fluxes at internal points ‘aut! puiw= (6(%) ar (2.89) DU2W = f fo “) ar (2.90) DQIW = j oF ey iT (291) 1 DQIW = f joe “) av (2.92) ‘The Jacobians are calculated by taking derivatives of the expressions for the x, and x, coordinates, which are defined as follows (equation (2.80), X= Oat t+ Oaxt tbat seabird + bax + yd 55) Alter substituting the $, expression (equation (2.74)) the above relationships can be written as, Xy = O(N} ~ NF Ha) + RP — ad) +a (2.94) p= 4G} — 2d 4 ad) + led x tad ee100 Chapter 2 Potential Problems The Jacobian is obtained by substituting (2.94) into (2.78) which gives [Ol = C{G} — 2x} + x})E + AG} — xD}? + (G6) — 2x3 +x) + 4G) — xh P I)? 2.95) A Gauss quadrature formula with ten points has been taken instead of the four points formula applied in the constant and lincar clement cases. The reason is twofold: (i) The variation of potential and flux is quadratic and hence more points should be taken and (ii) the clement geometry is also quadratic. SUDROUTINE EXT2NPQURP, YP, x1, ¥1 ,X2,¥2,23,¥9, 3,0 1, DQ1W, DO2W,DUIW, DUZW, i) © © PROGRAK 19 c ¢ THIS SUBRUOTINE COMPUTES THE NY AND GW MATRICES © MWICN RELATE A NODE (xP.¥P) WITH A BOUNDARY © ELEMENT USING GAUSS QUADRATURE. © 17 ALSO COMPUTES (WHEN KI) THE DQ), DG2¥, DUIV AND DUZW MATRICES © MWICK RELATE AN INTERNAL POINT WITH A BOUNDARY ELEMENT AND ARE. © NEEDED FOR COMPUTATION OF THE INTERNAL, FLUX VALUES. c coma Rapius © RDI.RDZ.RON ¢ RADIUS DERIVATIVES © ETAL,ETAZ © COMPONENTS OF THE UNIT NORMAL TO THE ELEMENT © xc0,¥co INTEGRATION POINT ALONG THE ELEMENT © xual JACOBIAN & INP, TPR BUS) ,qw(9) ,nw¢3) ,Dqrw(3),.Dezw(3) ,DUIW{3) ,DUZW( 9) G1 (18) OWE 10) 1) 1680693666 ,-0.8650629666 2 0.4333883941,-0.4933989041,, / ‘DATA ONE/0..081 -0866713443,0 494513493, '0.2190863625 ,0.2190863675,0, 2602667; (03. 2958242247 ,0.2955262747/ B00 Jels3 Delya-v1)/2 BOAO Te1,10 COMPUTE THE VALUES OF THE SHAPE FUNCTIONS AT THE INTEGRATION POINTS. F(3)eGH(1)8(G1(1)01}40.5 COMPUTE GEOMETRICAL PROPERTIES AT THE INTEGRATION POINTS Xeoex19F(1)4x207(2)ox36FL9) YoOsY1sF(1)«V26F(2)4¥3#F(3) XJATSQRT| (G1 (1)#A0B}##26(G1(1)4CeD) #42) ETAL#(G1(1)8C#D) /294 ETAze=(G1(1) #483) 294 RASSQRT( (XP-X160)##2¢(YP-¥CO) ##2) RD1=(XCO-XP) /RA RD2={YCOMYP) /RA RDNORDI*RTA1#RD228TAZ2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 101 ° © COMPUTE cv, HW DaIN, DOV, DUIY AND DUZY HATRICES Bo 40 301.3 20 Bose a}ebuti pexpveonet #XIASF(IV/RA buzwis - 78 SVSETAT#2 RDI #RD2¢ETA2)* :D2882~ 1. )AETAZSZ-OROIORDIEETAL)® OME TD AXIAR( I) 40 mt )en (3) -RON/HABOMELI@RIAFCD) RETURN END Routine LOCINPQ This subroutine computes using numerical integration, the submatrix GW that corresponds to an element when the collocation point is one of those three in the clement. The integrals are GW = f gutdr (2.96) ‘4 Three cases are considered depending on the position of the collocation point, ic, NODE = I, 2 of 3 (see figure 2.16). (i) Collocation point at Node (1) First a change of coordinates from x,x, to € is defined in the same way that was done in subroutine EXTINPQ. Then, in order to integrate the singularity a new change of variables is carried out, i.c. aft 297 The integral then gives two parts, one with a singular (erm In 1/7 and the other with no singularity. The first part is integrated by means of a special integration formula of the type (sce Appendix A) =f in(! pane ¥ wisn) 2.98) fl, & (in the program 1, GIL(/), w= OMEL(1)). The second part is integrated by the standard Gauss quadrature formula in terms of the variable é (in the program &=GI(N, = OME(/)). The shape functions #,, $2, $3 are given by Fl, F2, F3 in (erms of € and by FLI, FL2, FL3 in terms of 9. XJAL is the Jacobian for the special integration and XJA2 is the Jacobian for the standard Gauss q :102 ‘Chapter 2. Potential Problems (ii) Collocation point at Node (2) In order to integrate the two singularities that appear at both sides of the nodes, the integral is divided into two parts, 2 ® ® Gw = f pur|a]de = f pur|G] de + J our|G| dé (2.99) & a y Then, the first part is changed to the variable (see figure 2.16) 7’= —£ and the second part ot the variable 7 = é. Now cach singular part of these two integrals is computed using the special integration formula, and the two non-singular parts together are integrated using standard 10 points Gauss quadrature. In the program XJAI and XJA11 are the Jacobians for the two special logarthmic integrations and XJA2 is the Jacobian for the standard Gauss quadrature. The functions ¢,. #2 and 5 in terms of the new variables 9 are FLN1, FLN2 and FLN3. (ii) Colfocation point at Node (3) This case is similar to the first one with the logarithmic integration variable being now (2.100) ero grat aot ae oan Coliocation Point Collocation Point Collocation Point man war? ian Figure 2.16 Geometrical coordinates systems for numerical integration SUBROUTINE LOCINPQ(XG1, ¥G1, X62, ¥G2,XG3, ¥3 ,GM, NOOO) PROGRAM 2¢ . THIS SUBROUTINE COMPUTES THE GW MATRIX WHEN THE COLLOCATION POINT IS ONE OF THE NODES OF THE INTEGRATION ELEMENT. ‘THE COEFPICTENTS ARE COMPUTED BY NUMERICAL INTEGRATY aneneo2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 103 © THE NON SINGULAR PART TS COMPUTED USING STANDARD GAUSS QUADRATURE, © THE LOCARITUMSC PART IS COMPUTED USING A SPECIAL QUADRATURE FORMULA. e COMMON N, L, INP, IPR DIMENSION GI (10) ,OME(10) , GIL(10) ,OMEL(10) ,GH(3) DATA FOR THE GAUSS QUADRATURE ono DATA GI/0.9739065265, -0.9739065285,0.8650633666,-0.8650633666 €,0.6794095682, ~0.6794095682, 0. 4333953941, 04333953941, 00. 1488743389, ~0. 1488743389/ DATA OME/0. 0666713443, 0.0666713443,0.1494513491,0.1494513491 @,0.2190863625,0. 2190863625, 0.2692667193, 0. 2692667193, 00.2955242247,0,2955242247/ DATA FOR THE SPECIAL QUADRATURE aaa DATA GIL/0. 0090426309 , 0.0539712662 , 0. 1353118246 ,0,2470524162 8,0. 1802125396, 0.5237923179 ,0.6657752055 , 0. 7941904160, 00.a982610912,0.9688479887/ DATA OMEL/0. 1209551319, 0. 1863695425, 0. 1956608732, 0.1735771421 8,6. 1356956729, 0.0936467585,0.0557877273,0.0271598109, 0.0035151826,0.0016381576/ SET A LOCAL COORDINATES SYSTEM aoe Go T0(2,2,3) ,NO00 2 xa=xe3-kel ya=¥o3~¥61 x2=XG2-XG1 ya=¥o2-¥61. A= (x3-24K2) 40,5, Bi=x2 A2=(¥9-20¥2) 20.5 B2=¥2 Go TO 4 2 Xa=xG3-xo2 yaeva3-va? xaexai-xc2 yaeyo1-¥c2 Alex1ex3 nsx A2eY24¥3 Baava-72 co To 4 3 x25K02-x63 yamva2-v63 Al» (x1-28x2) 40.5 Bi=-x2 Adm (¥1~24¥2) 40.5, Ba=-¥2 4 CONTINUE bo 10 J*1,9 20 GH()=0. bo 250 I=1,10 COMPUTE SHAPE FUNCTIONS FOR NUMERICAL INTEGRATIONS ane o F3=0,54GI (1) #(G¥(1)+1.) F2=1.-GI (I) ##2 Pi90.5¢GE(I) *(GI(T)-1.) FL3=GIL(1) #(2.*G1L(1)-2.) FL2=4. 4G1L(1) *(1.-GIL(1)) FLI=(GIL(4) ~1.)# (2. #GIL() =2.) FLN3=0, 5GIL(I)*(GIL(T) #2.) FLN2=1,~GTL(T) #42 FLN1=0.5*GIL(I) *(GIL(1) -2-)104 Chapter 2 Potential Probleins COMPUTE GEOMETRICAL PROPERTIES Go T0(50,6¢,70) woo © aee 50 XOAL=SORT((4*AL GILT) ~ 2ye2 XIAI=SORT ( (AL*GI (1) #240.54K3) #424 (A2*GT (I) #240.54Y3) #42) XLO=-ALOG (24SQRT ( (G1 (1) «A1+B1) #424 (GI (1) #A24B2) ##2)) ‘S3"FLO*XIAI*OMEL( 1) +P3*XIA2*XLO*OME (1) ‘S2*FL2*XIAI*OMEL( I) +F2#XJA2*XLO*OME (I) SL*PLL*XIA1*OMEL(T) 41 #IA2*XLO*OHE (I) G0 70 200 60 XIAL=SQRT (0.54B1-A2*GIL(I)) ##24 (0.54B2-A24GIL(T) ) #42) HIRLASQRT ( (0, 54B1+A14GIL (I) ) #426 (0. 54D2+A24GIL(T) ) #42) XOA2=SQRT ( (0.5¢B1+A1¢GI (I) ) #424 (0. 5*B2+AZ*GI (1) ) **2) XL0~=0.5#ALOG ( (GZ (I) #41 #0. 54B140.5) #824 (GI (1) #A2#0.5+B240.5) #42) ‘S3=( PL] #XTALSPLN3 ¢XIA11) SOMEL (1) 4F34XJA2 4X10 *OHE (I) S2-FLN2* (XIAL#XIAI1) #OMEL (1) +F2*XIA2*XLO*OHE (1) ‘Ske (FLN3 #XTAL$FLN14XIA11) #OMEL (1) #F1*XIA24XLO*OME (T) Go To 200 70_XIAL=BORT( (24A1~ @2)42 XIAZWSQRT ( (24A1*GI (1) -0.54X1) #424 (2#A2*GI (1) -0.54¥1) ##2) XLO=~ALOG (2 *SORT ( (AL*GI(T)+B1) *#2¢ (A2*GT (I) +B2) #42)) S3“PLLAXTA2 AOMEL (I) 4F3+XIA2+XLO*OME (1) S2=FL2*XTAL*OMEL (1) 4P2*KIA2*XLO*OME(T) S1*FL34X5A1*OMEL (1) #F1+XJA2#XLO*OME (1) "A140.54K3) #424 (44A24CIL(T) -24A240,54Y3) ## "ALAGIL(I) +0. 5X1) #424 (24A2-44A24GIL(I)-0.5¢¥2) #4 c © COMPUTE GW MATRIX c 200 GW(3)= GH(3) #53, GH(2)= cw(2)s82 Ga(1)= oH(1)+s2 e 250 CONTINUE RETURN END Routine INTERPQ This subroutine first rcorders the vectors DFI and FI in such a way that all the boundary fluxes are stored in DFI and all the potentials in FI. It then computes the potentials and fluxes al internal points. The potential at any interior point is given by v= e {i ure arly t {i wear (2.101) no Uy jay The fluxes are given by (2.102)2.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 105 where the integrals along the boundary elements arc computed numerically by calling again the subroutine EXTINPQ. The listing of INTERPQ is as follows: Gia — ‘SUBROUTINE INTERPQ(FI ,DFI ,RODE,CX,C¥,X,¥,POT,FLUKI ,FLUX2) PRogRat 2 € € © THIS SUBROUTINE COMPUTES THE VALUES OF THE POTENTIAL AND THE PLUXES © AT INTERNAL POINTS. € ‘COMMON NL, INP, TPR DIMENSION F1(1} .DF1(1) ,RODE(1) .0x(1) ,c¥(1) DIMENSION ¥(1) .¥(1) ,POPCL) -FLUKI(1) ,FLUXZ(1) DIMENSION HW} ,ow(3) ,DGIWL 3} ,Dq2WE3) ,DUIW( 3) ,DU2W(3) REARRANGE THE FI AMD DFI_ARRAYS TO STORE ALL THE VALUES OF THE POTENTIAL IN F1 AND ALL THE VALUES OF THE DERIVATIVE 16 DFI wEew/2 DO 180 191,85 Do 170 32113, TF(KODE(3#1-345)) 110,110,170 120 TF CI-NE).NE.O OR. JLNE.3) GO TO 125 ER(KODE(1}) 14,114,313 113 ceRSG) FLU )*DEI(3¢1) DFL(3*11=cH, 80.70 170 344 DPLC3er}=DPICED G0 1 170 125 IF(L-EQ.5 OR. 3.67.1 .OR. RODE(3#1-3).29-1) GO To 130 DFII3#1-2)¢DFi (381-3) G0 TO 170 130 CHeFT(241-203) FI( 281-243) aDFX(387-3¢3) DIU 3#1-3+3) CH 179 CONTINUE 180 CONTINUE . © COMPUTE THE VALUES OF THE POTENTIAL AND THE FLUXES AT © ANTERWAL POINTS ce 1F(L.£9.0) Go TO 60 Do 240 Kel, POT(K) 20. FLUE (x) £0, PLUK (K}=0. DO 230 Tel NB CALL EXTIMPQUCK(K) ,CYOK) oX(2#1-1) ,¥¢2#1=1), 202 1, Y( 20161) Hw, Gu, DQiW, DQ2U,DUIW,DUZU, 1) ‘bo 220 30343) TWareei-zed TF(22.6T (20NE)) 1920-2 POF(K)=POT(K) 6ON(3 /@DFE(3#I-3¢9) -WV(3)9FI (192) FLUXY (X )xFLUXG (X) #DU1W(J)*DFI (381-363 )-DQIW(J}#F1(192) 220 FLUX2(K)=FLUXZ (XK) ¢DUZW( 3) #DFI (381-343)-DQ2W( J) 9FLLTI2) 230 CONTINUE, POT(K)=POT(K)/(2.23,1415928536) PURI (X) #FLUXA (x) /(2.03.1415926526) PLUX2(K) #FLUX2(K)/(2. #3.1418926536) 240 CONTINUE ‘30 RETURN EXD STURT) R(ZATY Rout ouTeTea This subroutine prints the results in th following order. (i), Potentials and fluxes at boundary nodes (fluxes ‘before* and ‘after’ cach node are printed. Mid-nodes always have the same flux at both sides). (ii) Internal points potentialsChapter 2 Potential Problems The listing is as follows: ‘SUBROUTINE OUTPTPQ(X,Y,P1 ,DFI,CX,C¥, POT, FLUXI,FLUIZ) ° © PROGRAN 22 e © ‘THIS SUBROUTINE PRINTS THE VALUES OF THF POTENTIAL AMD ITS NORMAL © DERIVATIVE AT BOUNDARY NODES. 1T ALSO PRINTS THE VALUES OF THE © POTENTIAL AT INTERNAL POINTS © COMMON NL INP TPR DIMENSYON XC4) ,YC1) ¢FEC1) ,DET(1) sCXC1) E71) DIMENSION POT( i} ,FLOXI(1),FLUEZUi) © weens2 WRITECIPR, 100) 100 FORMAT(" $,79(°8*)//2%, *RESULTS*//2X, "BOUNDARY NODES'// 156X, "POTENTIAL DERIVATIVE'/ 20x, °K", 15X,"¥" 12K, "POTENTIAL" 6X, °REFORE NODE" 6x," AFTER NODE’ /) ‘WRITECEPR,200) (1) .YC0) ¢F1(1) «DFH (39NEV.DF2(1) WAITECTPR,200) X(2),V(2),F1(2) ,DF1(2) ,DPi (2) 010.142 ,8E WRITE(TPK,200) X(2#1-1) ,¥(201—1) .PI U2» FI 94 L=3) DEI (391-2) 10 WAITECIPR,200) ¥(201) ,V(2e1) ,FL(Z#1) »DFI (301-1) ,DFI(3#1=1) 200 FORMAT(8(24,E14.8)) © IF(L.£9.0) G0 70 30 ‘MRITECIPR, 300) 300 FORMAT //,24, "INTERNAL POINTS" ,//9X,"X", 18K, "'Y" 12%, "POTENTIAL? 19x," FLUX K?, 10K, "FLUX ¥"/) 0'20 Ke1,1. 20 WRITE(IPR,400)CX(K) ,CY(K),POT(E) ,FLURI(K) ,FLUX2(X) 400 FORMAT(S(2X.E18.5)) 30 WRITE(IPR, 500) 500 FORMATE #,79(°#*)) RETURN END Example 2.4 The problem of an elliptical bar under torsion (figure 2.17(a) is analysed using program POQUABE. Under Saint-Venant type torsion the displacements are given by uy = Oxy = Oxyey (a) w= Op where 0 is the torsion angle per unit length and (x, ») is the warping function given by V2p=0 (b) The boundary conditions are us follows. Tractions normal to the boundary are identically zero, hence te)W 2.9, Computer Code for Potential Problems using Quadratic Elements (POQUABE) suUHecueLp AUEpUUY pue WwaIqGo/a sow siuewore sneipeng § (0) sow siuawere onespenD OL (a) 4 9 8 » € tt ay) Jo uONIMYSG “wajqosd uoISIOL {T'z aunBLy ‘suo\upucd AnaWUAS pue suauUjep jeayrowcaD Ce) (900 02 ee oe108 Chapter 2 Potential Problems For the case of an ellipse this becomes, 6) HF on faxttbext The dimensions were assumed to be a = 10 and boundary and at two selected internal points (x, = 2, x, = 2) and (x, = were computed. Because of symmetry ¢ = 0 along the two axes. Thus, only one quarter of the ellipse needs to be discretized. Ten quadratic clements were used here. Two for the short semi-axis, four for the long one and four for one quarter of the ellipse (sce figure 2.17(b)). The data for this case are as follows: (d) Sand values of $ on the x, =3.5) ELLIPTICAL SECTION (10) (DATA) 'LLAPTICAL SECTION UNDER TORSION (10 QUADRATIC ELEMENTS) 0 0. Oy 2.25 0. 2.5 0. 3.75 0. 5. 0. 6.25 0. 7.5 0, 8.75 0. 10. 0. 9.672.273 8.814 2.3617 7.7008 3.1898 6.174 9.933 4.7898 4.3891 3.3044 4.719 1.557 4.939, 0. 5, 0. 3.375 0. 2.5 0, 1.25 00. 0 0. 60, 00,00. 00. 00.00.00 00.00.00. 10.2 -3.379 1 ~4.8334 1 -4.8334°1 =4.9447" 1 ~4,3104 1 423104 1 -314657 1 -2.4412 1 =2:4411 1 -111643 1 0, 00, 00. 00. 00.00.04, 2214. 28. and the output is given by ELLIPTICAL SECTION (10) (OUTPUT) seasesoeens sesenveee: ELLIPTICAL SECTION UNDER TORSION (10 QUADRATIC ELEMENTS) pata NUMBER OF BOUNDARY ELEMENTSs 10 UMBER OF INTERNAL POINTS« 2 BOUNDARY NODES COORDINATES MODE x Y 1 +0000000E+00 +9000000E+00 2 Uizso000+01 ‘oooooooE+oo 3 ‘2800000F+04 Tonooae0r+o0 ‘ 137600006+01, Tooo0000E+o0 5 ‘50000006401 Tovo0000E+00 6 2500008401 ‘ooocoo0E+an2.9. Computer Code for Potential Problems using Quadrat Elements (POQUABE) — 109 +700000E+05 0000000E+00 ‘e7800008+01 ‘enoaooor-00 ‘roeno00<02 Too00000K«00 ‘s6700008+01 2730008401 ‘48140008401 ‘2361700601 {7790800E+01 Taiesa00E+o1 ‘er74000Es01 Tap33g00%-01 [47a98008-01 [43891008-01 ‘33044008603 [1857000801 ‘eooovonE +00 ‘So00000K+03 ‘ovoeco0E+00 ‘3375000E-01 ‘oo0n000E+0o ‘2600000E-01 Teoooo00E «oo 11250000801, BOUNDARY CONDITIONS. -TWIRD NODE: ‘PRESCRISED ELEMENT ‘VALUE cope: “VALUE coe VALUE 1 -o00000E+00 o -o000n00E+00 o -00000005+00 2 InoononsEsco =o Toaooeoers00 8 Tonooaeersoo a Taooao0ors00 =o Iosoooo0Es00 =o Toneagoor-00 : Looooo00e+co 8 Toooeeoces00 =o Tocooo00E«00 5 Tooooooorsco = |= .33790005s01 1 '4833400E+01 ‘ Haaseonrsor 1 Vasea7oorsos 1 310400801 3 STgiesongso1 == S3enst00E-01 1 24411008401 a sihaatioogson = 1 Sa tgaz00Rs03 ‘ooooa00r+a0 ° Too00000E«00 =o lovo0000Es00 «oo [oo00900r+00 10 0000004000 ooo0ovE+0o oo 90000005+00 RESULTS BOUNDARY NODES POTENTIAL DERIVATIVE x y POTENTIAL «BEFORE NODE AFTER NODE 000008+00 900005400 <000008+00 =. 28405E-03 =. 28405E-03 [12600E<01 ‘ooovoE+o0 Tooo00E+00 lrag6ses00 ‘hapesteoo Uzsonorso1 ‘o0000E+00 Tooao0Es00. Iiage6eso1 laaseee-01 ‘00000600 ‘oo000E+00 ‘e2s0ze+03 ‘eoo00E+00 loocoorso0. i2assgeso3 ‘ezso0E+o1 ‘eocone+ao loo000E-00 Uarsa6eso1 larsaeee [Ts000Es01 ‘eoon0E+ 00 Sooo0ce+00. VeasteEso1 Teashancor larsoorso1 ‘oooo0E+a0 Tonaooes00 Uss0a0es01 isho30E+03 ‘hooooEso2 ‘ooo00E+ 00 Seo000E+00 ‘aeaes01 ‘oonear+oo [agro0E.01 Tr2Ta0geo1 745196601 -.33790Es01 S9790E+01 Teaisorso1 ‘23e17Es01 = LAz506Ev02 © = LanaeEso! B34 sD Trzo0aEso1 ‘31meaeso1 T14746eso2 = L4saaTELOL = LanaaTELOS Thes30E601 |< Las76E+02 = =L4alo4EsoL 143104603 [43BB1F6O1 2616E+O2 — -.34G57ECO] «== 24687E FOL ‘Ai rsoEso1 lesesgcor © -LzeTiBeo1 =. 244r1Es01 1158 70E+01 V49380E+01 T480018+01 = 111603Es01 = 1 3643E001 ‘ooo00E+ 00 ‘so000Fs01 000008400 leaaooEs00 i3oisareo1 YennonE+oo Taa7soRot ‘oooo0g+00 Taozz0bs01 2 Tooooer+00. ‘2so008-0: Teonoorsao Treaaaeso1 Tooooer-oo T12s00801 ‘oo000E+00 [14936E+00 Lrasa6E+00 INTERNAL POINTS x y Pun x yuuK Y 12001401 =. 139908407 2E+ol = .24031Es01 Results for some representative boundary nodes and the two internal points arc compared with the known exact solution in the following table.10 Chapter 2 Potential Problems Potential using 10 Enact Boundary node quadratic elements potential solution Example 2.5 This example is the solution of the same elliptical sections as described in figure 2.16(a}) and studied in Example 2.4 but the number of quadratic elements has been reduced to 5. Il is interesting to see how the simple model gives results which are in good agreement with the theory. The input for the five element model (figure 2.17(c)) is as follows: ELLIPTICAL SECTION (5) (DATA) ELLIPPICAL SECTION UNDER TORSTON (5 QUADRATIC ELEMENTS) + 2.5 0. 5. 0. 7.5 0. 10. O. 4°°2.3617) 6.174 3.933 3.3044 4,719 0.5, 0.2.5 00.00. 00 00.00.00, 2 -A,B334 1 6.3206 23104 1 2.4411 1 0, 00. 0 0 3.52.9. Computer Code for Potential Problems using Quadratic Elements (POQUABE) 111 ‘The corresponding output is given below. ELLIPTICAL SECTION (5) (OUTPUT) ELLIPTICAL SECTION UNDER TORSION (9 QUADRATIC ELEMENTS) DATA NUNBER OF BOUNDARY ELEMENTS® 5 NUMBER OF INTERNAL POIHTS* 2 BOUNDARY MODES COORDINATES ODE * Y 1 +90000008+00 00000600 2 ‘eSo00008+03, ToeooneoE<00 3 ‘S0000008-03 ‘ooe00005+00 ‘ ‘rs00000E+03 ‘eoo0000E+00 5 LooooaaEs0z ‘Toooooooeso0 ‘ 88140008401 12361 700E<01 1 [61 740008-01 ‘ae330008-01 a Tas0svo0E+01 14719000E+01 ° ‘ovooco0E+o0 ‘so00000E+01 10 ‘oosoae0E+05 ‘2s000008+01 BOUNDARY CONDITIONS cHcesFIRST NODE---=-- --==-SECOND NODE- THIRD NODE=<==== ‘PRESCRIBED ‘PRESCRIBED PRESCRIBED LEWENT VALUE. cove VALUE. cone VALUE. cove 1 +0000000E+00 0 -ovp0000E+aa 0 -o000000E00 0 2 Tooooooer+on =o Tooooo00g+ao =o Toooovo0rv0n oo 3 Tooonooorsoo = 1 =14a334008s01 = t= sa }0400Ee01 4 slaatoaooreor = Sl2aasioozeor 1 ‘evooo00g-00 3 ‘9000000E+00 0 0000000800 0 ‘90000006000 POTENTIAL DERIVATIVE x Y POTENTIAL EFORE NODE AFTER NODE 00006500 000008+00 +00000E-09 =. 403a88-01 sossaz-o1 ‘25000801 7ODODES 00 ‘oooooeseo Vasaise+o1 UaszasEsot ‘So000Es01 ‘oo000E+00 ‘000005+00 29757E+0 129787201 UrsnopEso1 ‘oonooE+oo ToooooEsoa TassrzEeo1 ‘asn12E-01 Sre000E<02 ‘oooookeco ‘onooors0e Tooocer-00 [apis0Eso1 Te3e7eeo) | =[127T0R002 = 148334Ee01 = 48344EeOL verTAoEs01 Uousa0eeo1 = ; $ ug*\G 48, db, (2.112) ¢ * A whole range of quadrilateral and triangular elements can be defined. For quadrilateral cases one prefers to use Lagrangian type of elements {ie. the ones which have in some cases nodes inside) as such elements give better numerical results when used with point collocation. This is simply because the collocation. points ure better distributed in these cases. Table 2.1 shows some of the clements which can be used, 2.12 Poisson’s Es Many practical applications are governed by the Poisson rather than Laplace equations. fn these cases sources are distributed in the Q domain in accordance with a b(x) function. This produces the following governing equation Vush ind 2.113) where b is a known function of position. Sometimes the above equation can be reduced to Laplace's simply by substituting a particular solution or a change of variables. Care should be taken in these cases also to transform the boundary conditions accordingly. ‘When the function b(x) has a more complex formulation it may be difficult to find any suitable transformation and one needs to start with (2.113) plus the appropriate boundary conditions in order to deduce the basic integral equation, i. { (Vu — byw dQ= f (gq —qhut dl — f (u—ag* at (2.114) a a rh which integrated by parts twice produces, § (V2u* yu dQ— § but dQ= —§ Gut dP — f gu dP a a a A + f ug* d+ § aig* ar (2.115) h fChapter 2 Potential Problems 124 ‘3p = 99 83ibp = 9 p= Te (b= zi? = 820 = FETE = FO TNS =D suopvuny worjod.ani} Log te % ’ 3 ¥ oupaponds CS— DRC + DET = 60 3 4 DE Gd— DUS + DEE =o (RCE + DEE =O ~ WEP = = 86 MED DUIS =O PEN ES = WF EO 21 = EM IDE = suojoung woNDpodsanty — DEH FO FS 4 ENT gto © +9p10 2nvaponbig —_ puoseg to's =D aay 2+ C2 ~ DE = 18262 + WCF + DE = f>mtb— 1d pag Tato pal (2 DCP + DE = FP 2 — DCR DER" suopigun y worwjodiamy suonaung uonmodsanuy L bot z 1 ty ° ty 7 22p1Q . : a0aUry . = Sy away Aly | 1='9 Lato suonysuny uopmjodsamy suoyoung uonnjodsaquy 29P10 3M }SU0.) junisuoy OB, aendueny Tes21epspend, swa|qord Cf 10} QuawiaY BEPNg [eI VELPENd PUE FeINBUEL Jo A19eL TT AGEL2.12. Poisson's Equation 125 Alter substituting the Laplace fundamental solution u* and grouping all boundary terms together (i.e, in T =P, +1), onc obtains cul + f ug* dT + § but dQ = § qué dr (2.116) r a r Notice that although the 6 functions are known and consequently the integrals in Q do not introduce any new unknowns, the problem has changed in character as we need now to carry out integrals in (he domain as well as on the boundary. Regions of integration called cells can now be employed to compute the domain integral in (2.116) (figure 2.21). One usually applies a numerical integration scheme such as Gauss. In this case for each position of the singularity at a boun- dary point i, the integral in (2.116) can be written as, m/e D'={ but d= > (z malo), 2, 2.117) re the different cells (e varies from | to M, where M is the total number of cells describing Q domain), w, are the integration weights, the function (bu*) needs {o be evaluated at r integration points on each cell, Q, is the area of the cell ‘e’. Dlis the result, different for each ‘i° position of the fundamental solution, where i is one of the boundary nodes. Hence equation (2.116) now becomes x ¥ cult Mul + Dis YL G%qt (2.118) a mt or in matrix form HU+D=GQ (2.119) Notice that the domain integrals need to be computed as well when calculating any values of potentials or fluxes at internal points. Hence, N N ¥ Gq ~ ¥ Aw! - di (2.120) ra i where i is now an internal at which the singularity is applied. Concentrated sources are very simple to handle in boundary elements, They are a special case for which the function h at the internal point “I” becomes, b=Q'a' (2.121) where Q! is the magnitude of the source and A! is a Dirac delta function whose integral is equal to 1 at the point f and zero elsewhere. Assuming that a number