The document examines spurious modes that can occur in plane isoparametric elements when reduced or selective integration techniques are used. Eigenvalue and eigenvector analyses are performed on the stiffness matrices of single elements and simple meshes. This reveals the types of mechanisms and near-mechanisms that can develop, such as hourglassing in 4-node elements and Escher modes in 9-node elements. While these modes may lead to singularities in unconstrained meshes, they can still influence the solution when the problem is constrained by introducing spurious low-energy modes. Examples of a cantilever column and simple loading cases demonstrate how the solution can be dominated by excitation of these spurious modes.
The document examines spurious modes that can occur in plane isoparametric elements when reduced or selective integration techniques are used. Eigenvalue and eigenvector analyses are performed on the stiffness matrices of single elements and simple meshes. This reveals the types of mechanisms and near-mechanisms that can develop, such as hourglassing in 4-node elements and Escher modes in 9-node elements. While these modes may lead to singularities in unconstrained meshes, they can still influence the solution when the problem is constrained by introducing spurious low-energy modes. Examples of a cantilever column and simple loading cases demonstrate how the solution can be dominated by excitation of these spurious modes.
The document examines spurious modes that can occur in plane isoparametric elements when reduced or selective integration techniques are used. Eigenvalue and eigenvector analyses are performed on the stiffness matrices of single elements and simple meshes. This reveals the types of mechanisms and near-mechanisms that can develop, such as hourglassing in 4-node elements and Escher modes in 9-node elements. While these modes may lead to singularities in unconstrained meshes, they can still influence the solution when the problem is constrained by introducing spurious low-energy modes. Examples of a cantilever column and simple loading cases demonstrate how the solution can be dominated by excitation of these spurious modes.
The document examines spurious modes that can occur in plane isoparametric elements when reduced or selective integration techniques are used. Eigenvalue and eigenvector analyses are performed on the stiffness matrices of single elements and simple meshes. This reveals the types of mechanisms and near-mechanisms that can develop, such as hourglassing in 4-node elements and Escher modes in 9-node elements. While these modes may lead to singularities in unconstrained meshes, they can still influence the solution when the problem is constrained by introducing spurious low-energy modes. Examples of a cantilever column and simple loading cases demonstrate how the solution can be dominated by excitation of these spurious modes.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.
14, 1545-1557 (1979)
SPURIOUS MODES IN TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS N. BICANICt Institute for Civil Engineering, Zagreb, Yugoslavia E. HINTONS Civil Engineering Department, University College Swansea, U.K. SUMMARY The reasons for the presence of spurious modes in plane isoparametric elements are examined. Eigenvalue and eigenvector analyses of stiffness matrices for single elements and unconstrained and constrained meshes reveal the types of mechanisms and near-mechanisms that can be expected when reduced or selective integration techniques are used with quadrilateral isoparametric elements in the analysis of plane strain problems. INTRODUCTION In limiting-type problems, such as the analysis of nearly incompressible solids or thin Mindlin plates, overstiff solutions frequently occur when displacement based, numerically integrated isoparametric elements are used. To overcome this kind of difficulty, reduced or selective integration techniques have been adopted with considerable success and economy.' Reduced or selective integration schemes are also attractive in transient dynamic problems, especially for nonlinear cases in which the cost of the analysis depends heavily on the integration order used. Unfortunately, the use of low ordewof integration can be extremely dangerous and can lead to elements which can produce not only rigid body response but also extra zero energy modes. These extra modes can combine into either mechanisms, leading to singular stiffness matrices, or into near-mechanisms which are spurious low-energy modes which mask the solution. The object of this paper is to indicate the types of mesh instabilities which can develop when reduced or selective integration techniques are used with linear 4-noded and quadratic 8- and 9-noded quadrilateral isoparametric elements in the analysis of plane strain problems. The strain energy S is split into two parts,2 S1 and S2, where is deviatoric strain energy, and is volumetric strain energy. t Research Assistant. t Lecturer. 0029-5981/79/1014-1545$01.00 0 1979 by J ohn Wiley & Sons, Ltd. Received 12 June 1978 Revised 16 October 1978 and 7 February 1979 1545 1546 N. BICANIC AND E. HINTON In this representation of the strain energy, the usual stiffness matrix K is the sum of the two If an n X n order Gauss-Legendre rule is required to integrate K exactly, we define the matrices K1 and KZ, and x is the vector of nodal displacements. following three integration techniques employed: 1. Full integration: an n X n rule is used for K1 and KZ (exact for parallelograms). 2. Selective integration: an ( n - 1) x (n - 1) rule is used for Kz and an n x n rule is used for K1. 3. Reduced integration: an (n - 1) x (n - 1) rule is used for K1 and K2. For the elements under consideration, n =2 for the 4-noded element and n =3 for the 8- and 9-noded elements. THE CANONI CAL FORM OF THE STIFFNESS MATRI X In the solution of the problem Kx=f (3) where K is symmetric, positive definite stiffness matrix, x is the vector of unknown nodal displacements, and f is the given load vector, we can set X = h (4) Ky =Ay ( 5 ) where Q, is the matrix of eigenvectors of the standard eigenproblem and a is the vector of modal participation factors. Equation (1) is then rewritten as KQ,a =f and after premultiplying with aT we arrive at CPTKQ,a =iDTf (7) cPTKQ, =A (8) As Q, contains the eigenvectors of matrix K, CPTKQ, is in fact the stiffness matrix in its canonical form where A is the diagonal matrix, with the eigenvalues A, on the diagonal. The system of n linear equations can now he decomposed into n independent equations A,&, =@?f i =l , n (9) from which WJ can solve for a, directly and finally for x since x=Q,O! This long-winded procedure clearly provides the solution to (3). It is not a viable approach, however, since it requires the solution to the eigenproblem (9, which is usually much more expensive than the solution of the linear equations. Its use is more demonstrative in nature, and the di can be identified as primary deformation modes of the element or the mesh, the hi as stiffness related to these particular deformation shape, and the ai as participation factors of particular modes in the total solution. TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS 1547 SINGLE-ELEMENT STIFFNESS CHARACTERISTICS The transformation of the stiffness matrix K to its canonical form is now performed for single unconstrained elements. The semidefinite matrix K is rendered positive definite by adding a shift to the diagonal terms of K, leading to the eigenproblem or in which K* is positive definite and the real Ai are obtained from A . = A ? : - P I 1 (14) The results in Figures l(a)-(c) reveal that with both full and selective integration no additional zero modes are present. With reduced integration, however, all three elements under consideration have extra spurious zero energy modes. This suggests that mechanisms or near-mechanisms may develop when individual elements are combined to form a mesh. SIMPLE MESH CHARACTERISTICS Similar procedures are now applied to unconstrained rectangular meshes in order to obtain the number of zero eigenvalues associated with each mesh. The results using reduced integration are shown in Figure 2. Unlike meshes with 4- and 9-noded elements, where there is a tendency for the singularity to remain, the meshes with an 8-noded element show the interesting property that even when only two elements are present no extra zero modes exist. In fact, this can be clearly seen and predicted from the shape of the single 8-noded element eigenvector, which is related to zero energy (Figure 1c)-the shape is not inter-element compatible. On the other hand, the zero energy eigenvectors for 4- and 9-noded elements are inter-element compatible. The eigenvectors and eigenvalues of a 2 X 2 mesh (Figure 3) with the minimum number of displacement constraints required to prevent rigid body motions, show that zero energy modes on the element level combine into the well-known hourglassing or keystoning mode for the 4-noded element and into the so-called Escherf mode4 for the 9-noded element. I t should be noted that these modes are related to zero stiffness and lead to singular system matrices. This means that the problem cannot be solved. (There is an exception in explicit transient dynamic schemes where the stiffness matrix is never assembled and these spurious modes can pollute the solution.) However, it will be shown in the next examples that the elimination of the overall matrix singularity does not suppress the influence of the hourglassing or the Escher modes completely. These spurious modes, which may be related to some finite energy or stiffness, can completely destroy the solution if excited. A CANTILEVER COLUMN The cantilever column example shows clearly the dangers of spurious modes. The mesh with the 8-noded elements behaves well. The lower order of integration serves its purpose and no mesh t M. C. Escher, a famous Dutch graphic artist. 1548 N. BlCANlC AND E. HINTON instability occurs. For 4- and 9-noded element meshes with reduced integration, eigenvalues and eigenvectors again show the presence of hourglassing and Escher modes of deformation, but this time they are related to some finite stiffness. The assembled stiffness matrix is thus not singular and a solution may be obtained. It also means that, unless the applied load vector is orthogonal to the spurious mode, there must be a proportion of that eigenvector in the solution. 4F - 2/ 2 0 0 0 0 999 1 0 1 9 9 2 0 2 0 8F - 3/3 0. 0. 0 . 0.321 0.598 0 . 5 9 8 0 . 9 9 2 1 . 2 2. 21 2. 93 3.60 3. 60 3 . 9 4 4 . 1 7 8 . 8 0 8 . 8 0 9F - 3/3 DaLuLl zJ aDa 0 0 0 0 321 0 561 0 561 0 992 1 2 1 36 1 36 2 21 2 93 3 78 3 78 3 94 4 17 10. 4 1 0 . 4 Figure l(a). Single unconstrained element analysis-full integration (element size 1.0 x 1.0, E =2.0, Y =0.0, G =1.0) TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS 1549 With reference to (10) the modal participation factors are non-zero if qbTf #O and Ai #m (15) From the first six modes, which are illustrated in Figures 4(a)-(c), it is clear that the Ai of the spurious modes are low and consequently any contribution from q5Tf will be significant in the solution. 4s - 2/1 0 0 0 0,777 0.777 1 - 9 9 2 . 0 2 . 0 8s - 3/2 Q 2 13 aDDDoEu 2 93 3 60 3 60 3 94 4 1 8 80 8 80 9s - 3/ 2 0. 0. 0. 0. 222 0, 554 0. 555 0 . 9 9 2 1 . o a 1 13 !3 1 13 13 2 13 D 2 93 E! 3 73 3 73 El 3 94 Ezl 4 1 10. 0 10. 0 Figure l(b). Single unconstrained element analysis-selective integration (element size 1.0 X 1.0, E =2.0, Y =0-0, G =1.0) 1550 N. BICANIC AND E. HINTON 4 R - 1/1 0 El 0 D 2 0 a 0 Q 0 958 Da 0 0 2 . 0 2 . 0 8R - 2/2 0. 0 0 . 0 . 5 9 8 0 . 5 9 8 0 . 9 5 8 0 . 9 5 8 2 . 6 6 3 . 6 0 0 0 3 . 6 0 3. 70 3. 7 0 9R - 2/2 0 0 0 0 . 9 5 8 2 . 0 2.66 3 . 6 0 3 . 6 0 8 . 8 0 8 . 8 0 TITD 0 598 0 598 3 . 7 0 3. 70 8. 80 8.80 Figure l(c). Single unconstrained element analysis-reduced integration (element size 1.0 x 1.0, E =2.0, u =I 0.0, G =1.0) To demonstrate this effect, two simple problems are solved in Figure 5. I t is, we hope, instructive to see that in the case with the horizontal force, where the force vector is orthogonal to the lowest spurious mode, the total solution is very reasonable. In the case with the vertical force, however, the lowest spurious mode dominates the solution completely, making it meaningless. TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS 1551 5 4 3 2 1 z 171 El I 5 I 1 2 3 4 5 / MI / MI Figure 2. Number of zero eigenvalues for unconstrained meshes (mesh size N x M rectangular elements) A CONSTRAINED MESH The examples presented so far contain few or no displacement constraints, thus enabling the mechanisms and near-mechanisms to show clearly. However, if the same procedure is applied to the highly constrained mesh of 9-noded elements using the reduced integration shown in Figure 6, it is seen that Escher modes are again present even for such a coarse mesh. Any forcing vector which is not orthogonal to those eigenvectors must bring a contribution of spurious modes into the solution. This effect will not be found when the 8-noded element is employed. 4 SOR 2.2 R 4 SOR 2.2 R 4 SOR 2.2 R 4 SOR 2.2 R 4 SQR 2.2 R 000 000 000 092 297 020 050 094 272 384 000 000 000 031 0 96 Figure 3. Square mesh 2 X 2 elements with minimum constraints-reduced integration (mesh size 2.0 x 2.0, E =2.4, Y =0.2, G =1.0) 1552 N. BI ~ANI C AND E. HINTON COL 4 FULL COL 8 FULL i ' COL 4 FULL COL 4 FULL COL 4 FULL COL 4 FULL COL 4 FULL COL 4 FULL 377 690 825 376E- 02 803E- 01 971E 01 COL 8 FULL COL 8 FULL COL 8 FULL COL 8 FULL COL 8 FULL COL 8 FULL 136 275 308 1 4 ~0 2 Z~ZE- OI 385~- 01 E E B COL 9 FULL COL 9 FULL COL 9 FULL COL 9 FULL COL 9 FULL COL 9 FULL COL 9 FULL 316E- 01 115 253 264 116E- 02 244E- 01 Figure 4(a). Lowest six eigenvalues and eigenvectors for simple cantilever-full integration (mesh size 2.0 x 6.0, E =2.4, v =0.2, G =1.0, all support nodes fixed) COL 4 SEL 1 COL 8 SEL COL 9 SEL TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS 1553 COL 4 SEL COL 4 SEL COL 4 SEL COL 4 SEL 334E- 02 738E-01 969E-01 35 1 COL 8 SEL COL 8 SEL COL 8 SEL COL 8 SEL 140E-02 291E-01 385E-01 135 COL 9 SEL COL 9 SEL COL 9 SEL COL 9 SEL 116E- 02 244E-01 316E-01 115 COL 4 SEL COL 4 SEl 557 785 COL 8 SEL 266 B COL 244 9 SEL COL 8 SEL 303 1 COL 260 9 SEL Figure 4(b). Lowest six eigenvalues and eigenvectors for simple cantilever-selective integration (mesh size 2.0 x 6.0, E =2 4 , v =0.2, G =1.0, all support nodes fixed) 1554 N. BICANIC AND E. HINTON COL 4 RED COL 4 REDCOL 4 RED COL 4 RED COL 4 RE0 COL 4 RED COL 4 RED 258E 02 258E 02 596E - 01 596E 01 993E 01 993E 01 COL 8 RED COL 8 RED CDL 8 RED COL 8 RED COL 8 RED COL 8 RED COL 8 RED 140E - 02 290E - 01 385E - 01 134 252 293 707E 01 COL 9 RED COL 9 RED COL 9 RED COL 9 RED COL 9 RED COL 9 RED COL 9 RED 116E 02 241E 02 243E 01 328E 01 487E 01 Figure 4(c). Lowest six eigenvectors and eigenvalues for simple cantilever-reduced integration (mesh size 2.0 X 6.0, E =2.4, v =0.2, G =1.0, all support nodes fixed) TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS E = 2 1, v = 0 2 1555 2 ~~ A x &~=~, Q I ZAxk3 1 19750 L572 1 L572 2 637 1 63 10 a ! L53 2 00 0 0 L53 2 t i - 6 L 9 1 102 , L63L 00 00 , L634 0 0 00 L63 L 6 309 1 68 I L 7 0 2 - - X 63 (FRONT SOLVER I 477 8 r; COL 9 FULL 6 EIGENVECTORS VER F ULL 2o ' VER LOAD 10 E = 2 L v =0 2 0 0 0 0 - 10 0 COL 9 RED 6 EIGENVECTORS HOR LOAD 10 HOR RED e:i 00 - 4579 - L62 1 - 4695 0 0 - L695 y63 (FRONT SOLVER) - L z I - IS the mode number a, - is the participation factor for mode I Axk3- is the contribution to the total horizontal displacement at node 63 for mode I xG3 - is the total horizontal displacement at node 63 obtained using front solution Figure 5. Horizontal and vertical point load on simple cantilever (9-noded element-full and reduced integration) 1556 N. BICANIC AND E. HINTON 9 SOR 6.6 FULL 238 9 SOR 6.6 FULL 501 9 SOR 6.6 FULL 238 9 SOR 6.6 FULL 642 9 SOR 6.6 SEL 238 9 SOR 6.6 SEL 507 9 SOR 6.6 SEL 64 1 9 SOR 6-6 FULL 35 1 9 SOR 6.6 FULL 670 9 SOR 6.6 SEL 35 1 9 SOR 6.6 SEC 670 9 SOR 6.6 RED 9 SOR 6.6 RED 9 SOR 6.6 RED 505 597 591 Figure 6. Lowest six eigenvectors and eigenvalues for constrained mesh with 9-noded element (mesh size 6 . 0 ~ 6 . 0 , E =2.4, v =0.2, G =1.0) TWO-DIMENSIONAL ISOPARAMETRIC ELEMENTS 1557 CONCLUSION On the basis of the transformation of the element and system stiffness matrices into their canonical forms, it has been demonstrated that reduced integration, when applied to 4-noded and 9-noded isoparametric elements in plane strain problems, leads to spurious hourglassing and Escher modes, which-if excited-can completely destroy the solution. Further theoreti- cal justifications of these phenomena can be found The discussion here was confined to the stiffness properties of square elements in linear elastic analysis-with general quadrilateral elements, spurious modes appear higher in the spectrum of eigenvectors, but their effect cannot be completely i gn~red. ~ The same procedures for detecting spurious modes, can, of course, be used for a much broader class of elements. [N.B. Some additional useful references on spurious modes and related topics are listed after the main references.] REFERENCES 1. 0. C. Zienkiewicz, R. L. Taylor and J . M. Too, Reduced integration techniques in general analysis of plates and shells, Znt. J. num. Meth. Engng, 3, 275-290 (1971). 2. E. Hinton, E. M. Salonen and N. BikaniC, A study of locking phenomena in isoparametric elements, MAFELAP 78, Conf. Math. Finite Elements and Applications, Brunel University, April 1978, Proceedings to be published by Academic Press. 3. K. J . Bathe and E. L. Wilson. Numerical Methods in Finite Element Analysis, Prentice-Hall, Englewood Cliffs, N.J ., 1976. Swansea, 1978, to be submitted. 4. N. BiCaniC, Nonlinear finite element transient response of concrete structures, Ph.D. Thesis, University College 5. 0. C. Zienkiewicz, The Finite Element Method, 3rd edn, McGraw-Hill, London, 1977. 6. G. Strang and G. J . Fix, An Analysis of the Finite Element Method, Prentice-Hall, Englewood Cliffs, N. J ., 1973. ADDI TI ONAL REFERENCES ON SPURIOUS MODES AND RELATED TOPICS W. P. Doherty, E. L. Wilson and R. L. Taylor, Stress analysis of axisymmetric solids using higher order quadrilateral finite elements, S.E.L. Report No. SESM 69-3, University of California, Berkeley (1969). I. Fried, Numerical integration in the finite element method, Comp. Struct. 4, 921-932 (1974). J . S. Holt and P. S. Hope, Displacement oscillation in plane quadratic isoparametric elements in orthotropic situation, T. J . R. Hughes, R. L. Taylor and W. Kanoknukulchai, A simple and efficient finite element for plate bending, Znt. J. T. J . R. Hughes, W. K. Liu and A. Brooks, Finite element analysis of incompressible viscous flow by the penalty function B. M. Irons, Numerical integration applied to finite element method, Conf. Useof Digital Computers in Structural B. M. Irons, Numerical integration applied to finite element method, C/R/161/71, Dept. of Civil Engineering, B. M. Irons and T. K. Hellen, On reduced integration in solid isoparametric elements when used in shells with B. M. Irons and S. Ahmed, Techniques for Finite Elements. Ellis Horwood, Chichester, U.K., to be published. J . C. Nagtegaal, D. M. Parks and J . R. Rice, On numerically accurate finite element solution in the fully plastic range, R. S. Sandhu and K. J . Singh, Reduced integration for improved accuracy of finite element approximations, Comp. D. Kosloff and G. A. Frazier, Treatment of hourglass patterns in low order finite element codes, Znt. J. Num. Anal. R. L. Taylor, P. J . Beresford and E. L. Wilson, Anon conforming element for stress analysis, Znt. J. num. Meth. Engng, P. Tsipouras, Numerical stability of incompletely integrated shape functions, Comp. Struct. 7, 679-680 (1977). K. J . Willam, Finite element analysis of cellular structures, Ph.D. Thesis, University of California, Berkeley, (1970). Znt. J. num. Meth. Engng, to appear. num. Meth. Engng, 11, 1529-1543 (1977). formulation, J. Comp. Phys, to appear. Engineering, University of Newcastle, U.K. (1966). University College, Swansea, U.K. (1971). membrane modes, Znt. J. num. Meth. Engng, 10, 1179-1183 (1976). Comp. Meth. Appl. Mech. Eng. 4, 153-178 (1974). Meth. Appl . Mech. Engng, 14, 23-37 (1978). Meth Geomech. 2, 57-72 (1978). 10, 1211-1219 (1976).