This action might not be possible to undo. Are you sure you want to continue?

**THE GALERKIN GRADIENT LEAST-SQUARES METHOD Leopoldo Penna FRANCA
**

Laborat6rio National de Computaf~to Cienn'flca, 22290 Rio de Janeiro R J, Brazil

**Eduardo Gomes DUTRA DO CARMO
**

Programa de Engenharia Nuclear-COPPE/UFRJ, llha do Fund~o, 21945 Rio de Janeiro R J, Brazil

Received 15 September 1988 Revised manuscript received 14 November 1988

A finite element method is proposed to solve a scalar singular diffusion problem. The method is constructed by adding to the standard Galerkin method a mesh-dependent term obtained from a least-squares form of the gradient of the Euler-Lagrange equation. For the oneodimensional homogeneous problem the method is designed to develop a nodal exact solution. An error estimate shows that the method converges optimally for any value of the singular parameter. Numerical results demonstrate the good stability and accuracy properties of the method.

1. Introduction In most cases, finite element methods have been developed based on the Galerkin method. From the mathematical point of view, the rather general convergence results obtained for the single field elliptic equations created confidence to attempt further extensions in more elaborate models. From engineering experience: however, undesirable pathologies appeared in various situations of practical interest. Spurious oscillations, locking and other undesirable features appeared separately or combined when applying the Galerkin method to thin structures, incompressible media, fluid flows, etc. To avoid these difficulties, intuitive element design, different integration rules and other devices followed. Some of them have been mathematically substantiated. Simplicity has been the driving force behind all these solutions. The desire to maintain the usual interpolation functions allows easy generalizations for more complex nonlinear models. Some specialists have serious questions about this approach and suggest instead strict obedience to the basic stability conditions of the Galerkin method. Their solution requires use of elaborate interpolation functions and has received resistance from code developers. In recent years, the idea of changing the discrete formulation of a given problem has been gaining more momentum. The methodology we are concerned with, consists of adding mesh-dependent terms to the usual Galerkin method. These perturbation terms are functions of the residuals of the Euler-Lagrange equations evaluated elementwise. Since the residuals of the Euler-Lagrange equations are satisfied by the exact solution, consistency is preserved in these methods. The perturbation terms are designed to enhance stability of the original Galerkin formulation without upsetting consistency. Convergence results may be derived for a

0045-7825/89/$3.50 t~ 1989, Elsevier Science Publishers B.V. (North-Holland)

convergence could only be established for the plate configuration (see [35]). Application of these methods to mixed formulations was introduced by Hughes et al. a singular diffusion problem given by or2u . (1) where or2. when combined with the concept of entropy variables [12] and shock capturing ideas [13-18] were successfully applied to the steady compressible Euler and Navier-Stokes equations [19-24]. or SUPG as it is now frequently called. namely. the Galerkin/least-squares method is no panacea. they presented a design of the perturbation term which allows use of any continuous equal-order interpolation for the velocity and pressure variables.42 L. As a major contribution for these mixed formulations.e 2 AU ---f .37]. This simplifies the conception and extension of this idea to various problems. as it is always the case. In particular. Generalizations followed. who referred to it as the streamline upwind/Petrov-Galerkin method. In the general formulation for axisymmetric shell problems.resmethod large family of interpolations..G. allowing combinations of simple finite element polynomials of almost any order including the attractive (from the implementational point of view) equal-order interpolations. Johnson et al. 6 2 and f are given functions and u is the unknown scalar. this method has overcome the difficulties concerning the usual stability conditions (K-ellipticity and Babu~ka-Brezzi condition). Extensions to the incompressible Euler and Navier-Stokes equations were introduced in [8-10]. Direct application of (1) may be found in diffusion problems in semiconductors.35]. demonstrating the inappropriateness of the Galerkin method. Employment of the . Advection-diffusion systems employing space-time elements are analyzed in [25]. This formulation has been generalized and appfied to various mixed formulations (see [27] and the references therein). The standard Galerkin method presents a difficulty for small values of the ratio e2/o-2. the Stokes flow problem [36. Spurious oscillations usually appear in this case. [26].~. a possibility ruled out by key stability requirements within the Galerkin method. This last work [27] and [28-31] are novel with respect to the former works: the perturbation terms are constructed as mesh-dependent least-squares forms of the Euler-Lagrange equations.n gradient least-sq. we analyze a simpler scalar model equation. To simulate the appearance of complex boundary layers in the cylindrical shell problem. Lagrangean elements). Durra do Carm. This makes us think about alternative approaches. the method helps in obtaining better discrete solutions but the analysis indicates that the dependence on a small parameter (proportional to the shell thickness) can not be overcome within the Galerkin/ least-squares method. the axisymmetric shell problem [34. that in general includes the usual finite element polynomials (e. Analysis of the method was provided by Johnson and N~ivert [4]. It was introduced by Hughes and Brooks [1-3]. for example. the Reissner-Mindlin plate [33]..o.g. This approach has its roots on advection-diffusion problems.a. They applied and analyzed a special formulation for the Stokes flow problem (or for the incompressible elasticity problem). Johnson [5]. 38].P. The Oalerkin/least-squares method has been applied and analyzed for the Timoshenko beam problem [31. The Ga!erk. For the cylindrical shell. Franca and E. Hughes and Mallet [11] have extended the method to advection-diffusion systems which. for the scalar linear advection-diffusion equation in steady state. 32]. [6] and N~vert [7] proposed and analyzed extensions to the unsteady case using space-time elements. and the isotropic compressible and incompressible elasticity problem [29.*. However. the arch problem [30].

in Section 5 some conclusions are drawn. let us define the following Hilbert spaces: H . and u. In Section 3 GGLS is generalized to the multi-dimensional equation (1). " . This provides the resulting formulation with a control on derivatives.. (2) where u is the unknown scalar field. t~ . A new concept of a perturbation term is proposed herein. The exact solution is immediate and can be written as u(x) = (e - (4) )-(/ (5) To construct a weak form of this problem.6 2 U . ~ u~ t. r raFsca G. Note that consistency is preserved since the exact solution satisfies the gradient of (1).L ~" . Subsequently the GGLS method is proposed and its parameter is designed so that nodal exactness is achieved for linear shape functions and constant coefficients in the absence of source terms. I) Iu(0) -0. This procedure enhances stability in the H 1 seminorm. independent of e2/or2.. The term has a mesh-dependent parameter and it is evaluated elementwise. In Section 4 numerical experiments are performed which confirm the good stabifity and accuracy of the proposed GGLS method... which is already present in the Galerkin method. since the least-squares form of (1) only contributes to L 2 .. An error analysis is presented which demonstrates for the GGLS method optimal convergence in the H 1 seminorm.. (6) . Dutra do "-t w .{u E HI(0. •R . 2. In Section 2 the one-dimensional model of (1) is studied in detail. We add to the G~erkin method a least-squares form of the gradient of the Enler-Lagrange equation (1). Let us consider the model boundary-value problem which consists of finding u.xx is the second derivative of u with respect to the independent variable x (ffid2u/dx2). satisfying (2). or and 8 are physical parameters. Finally. und E. assumed constant in the subsequent analysis. The Galerkin method employing linear shape functions is analyzed firom the difference equation point of view and its poor stability properties are shown for small e2/or2 values.~ G a l e r k i n gradient least-squares m e t h o d "" . The one-dimensional model problem A singular diffusion problem in one dimension defined on the open unit interval is governed by the following differential equation: 0"2U . The method will be denoted by Galerkin/gradient-least-squares (GGLS). u(1). z z -.s t a b i l i t y . such that u(O)•o.1}.~ Galerkin/least-squares method will not cure the shortcomings either. (3) uO) = 1.0 .

. or (12) (13) (1 . The general difference solution of (13) is of the form ul = cr I • Substituting (14) in (13) implies a second order equation in r. . V~. namely (1 -. 1) is the space of functions possessing square-integrable derivatives having domain equal to the unit interval. To make this statement precise consider a uniform mesh. n~. where x e is the coordinate of the end point of the eth element and n e~ is the total number of elements.a) = 0. The Galerkin gradient least-squares method n~ = (u e H'(0.} CI H .} f3 H~ . .a)u. the Ith equation of (11) is -a(ux_t + 4u n + u~+~) + u~_~ .o~ = 1... (15) •~th roots .(i .2(1 + 2a)u. i.44 L. such that f~ o'2UhVh dx + f j 62Uh._~ .G.a)u. The variational formulation of the problem is: Find u E H.. nel. . The domain of each element e is J2~= (X~_l.xV. k = 1. 1). Let Skh(O. e = 1 . n. . such that f~ O'2UDdx + f~ e2U. (11) Depending on the value of a = ¢r2h2/(6e 2) the Oalerkin formulation may perform very poorly. 06) . 1) = {U. The standard Galerkin method consists o.(0.).2ut + u~+1 = 0 . 1 ) = {u h l ue E P t ( ~ ) . 2(1 + 2 a ) r + (1 . e ffi 1 .. ) " .P. Consider the set of all polynomials of degree not greater than k and denote its restriction t o / ~ by Pk(£~). e ffi1.x dx ffi 0 Vv h E vk(o... and the mesh parameter is denoted by h = max(/~).. (8) To approximate (8) consider a partition of the unit interval 0ffi x o <x~ <. . : ) / ( I . 1). l u~ E Pk(l]*) .e. Franca and E. I) Iu(0) = u(1) =0).x dx = 0 Vv E H 1 . .xDh. Durra do Carmo. where (7) H i ( 0 . + (1 . xe). .a ) r 2 -- (1 ~ . . (9) (10) where u e is the restriction of u h to element e.~ = ( I + 2~ + V(1 + 2 . finding u h E Sk(0.-<x.. and let ut denote the nodal value of the approximate solution at an arbitrary interior node L Employing piecewise linear shape functions.+ 1 = 0 .

+ I = i/. x. However.x~cr2vh..~)u. The higher-order derivatives are zero and the Ith equation of (18) is (1 .G.consider the approximation employing piecewise linear shape functions (k ffi 1) and a uniform mesh._1 -2(1 + 2~)u.P.x e-I • vv. + (1. setting UoffiO and uNffil. Applying the boundary conditions to the difference equation. then the difference solution is given by u.ffi ~ h2 ~. (22) where N is the total number of nodal points minus one. Consider the following Galerkin/gradient-least-squares method (GGLS): Find u h E Ss k(0.e.r~)/(rT. 1). -- E2Uh'xx)'xT(O'2Uh -. Our goal is to design ~"so that the difference solution is nodaily exact.+ I -0.xUh. + (20) (21) ~. for a > 1 both roots are negative and therefore by (14) the difference solution will oscillate from one node to the next! This clearly demonstrates a spurious behavior of the appro~mate solution as compared to the smooth exact solution (5). such that r plel '. which provides the method with better stability properties with respect to the HLseminorm. To overcome this serious drawback in the Galerkin method we present a method with enhanced stability characteristics. The additional term enhances stability through the presence of ¢r2uh. To design the parameter ~. Note that selecting ~"ffi 0 reduces GGLS to the standard Galerkin method. Comparing (23) to the exact . The difference solution is of the form given by (14) and the roots are given by (16) and (17) with a replaced by ~. e v~(o.E2Uh'xx)'x d x ~-0 (18) where ~-is a positive parameter to be designed below.r~). with (19) ¢r2h2 = 6(. I). Dutra do Carmo. Franca and E. dx + Y~ fa ( O ' 2 u h f~ cr2usvh dx + f~ 82Uh.&)u. The Galerkin gradient least-squares method r2 = I/r 45 1.L. i. = (r[ . (17) Note that both roots are positive if a < 1. (23) where r 1 and r 2 are given by (16) and (17) with a replaced by c~.

Franca and E.xx . (29) REMARKS (1) The assumption that the coefficients o. Removing this assumption the definitions of ~"and ~ remain the same as in (21). (2) Addition of a source term.squares method solution.G. The extra terms e!ue to f will be found on the r i o t hand side of (18) as . Durra do Carmo.1 (27) a = cosh(V~'a) + 2 " By (19). a<l.)/((e~h)N_(e ~h)N). we conclude that r1= e ~ . (30) does not change the definition of the parameter in the method.49.2 and • 2 are constant is not essential to the error estimates derived subsequently. i.. 1 ~ a < 8 . (5). a~8.1 .e2u.P.~ " (28) Summarizing.. The Galerkin gradient least. 1 we compare expressions (28) and (29). In Fig.2 and e 2 will depend explicitly on x. written as u(x. (28) or (29). _zh (24) (25) (26) in order to obtain a nodally exact solution. the GGLS method is given by (18).f ._(e ~ h).. However. replacing (1) by o. where o.064a+0.e e .2u .)=((e --~ h). Carrying out the computations in (25) we arrive at _ cosh(V ). or o" h /'2 . In practice we employ an asymptotic expression for ~ given by 1.46 L. cosh(V )+2 1 = cosh(V~'a). ~= 0. with ~" and ~ given by (21) and (28).e. 0. for f ~ 0 we may not expect nodal exactness in general.

G.f in ~. such that .00 o¢..$" G8 --u"u . The Galerkin gradient least-squares method 47 I #1 .aslmMotio O./ ! 8.P.e2vs.Find u E H~(~2).x~). 3. 33) is. Franca and E. The multi-dimensional counterpart of the model problem given in the last section is: Find u(x) satisfying O"211.Q p-! ! | . The multi-dimensional counterpart Let n be an open region in R ". 16. (32) (33) u--0 on F . Optimal and asymptotic values of the stability parameter.x1"(or2vh .00 Fig. where n. (34) The variational formulation of problem (32.d with piecewise smooth boundary F. (31) A numerical example shown in Section 4 demonstrates the excellent performance of GGLS for f # 0 . Durra do Carmo. Consider Ht(~2) to be the Hilbert space of functions possessing square-integrable derivafives with domain •. 1./ / L.8 2 AU ~ .o >~2 is the number of space dimensions. 1 fv s dx + ~ nei fo • e=l f. where f is a given source distribution. Let on r).~dx.00 0.

Ill = = B(v...2h 2 662 .~=v Av.~ .V(o.L(Vh) where VUh ~ V~(~'~) . (36) The GGLS method is: Find u h ~ Vkh(~'~). such that B(~h. Let us define the norm..ll~=w.1 (Stability). n. (44) .~11o~11 = + d-Ilvv~ll = + .and e constants and ~ given by the asymptotic expression (29).£2 AUh) T . l u~ ~ P ~ ( W ) .V(o. ~= cosh(V'6~).1 .20h -.Let v g ( a ) = {u. Durra do Carmo. v~) yv~ ~ v~(~).P. (4o) 1 cosh(VT. where I1" lib is taken elementwise. (35) To approximate (35) we consider a partition of /] into n~ nonoverlapping elements consisting of triangles or quadrilaterals for n~d = 2 and their generalizations for n..1112--. in which we prove convergence: IIIo.G. Franca and E. e = 1..48 L.d > 2..) + 2 o. The set of all polynomials of degree not greater than k restricted to /~" (the element domain) is denoted by Pk (/~').2Uh-. (39) The parameter ~-is defined as a generalization of the one-dimensional case as h 2 .~ ' Ix- (41) (42) For the subsequent error analysis we will consider o. The Galerkin gradient least-squares method fa or2uv dl~ + fa e2VuVv dl] = fa fv da Vv E HI(a) . Uh)". We now establish some preliminary results: (43) LEMMA 3.2l)h -. lily.£2 AUh)d[]. ~" = ~ e. Ill" III..} 1'3n ~ ( a ) ..62 AUh)d/'] e----1 • (3s) L(Vh) = f[~ fVh dO + ~ ell nel f/j • Vfi" . (37) B(Uh' Oh)--'--fn (IT2UhVh @ E2VUh~Oh)dl~ "]" X V(O. II:.

From the defmition of B(. the result follows immediately. 2 O. Let Vl = u~ .). The error in the G G L S method converges as follows. PROOF. Dutra do Carmo.L. The Galerkin gradient least-squares method 49 PROOF. . L E M M A 3. (a) I f a >~l then Illelll "~<~ ~ ¢ ( u ) h ~+~ .711 ~ + . Let u h E V~(I~) denote an interpolant of H~(l~).2 (Orthogonality o f the error). (38).~-v Anll~ o. Then IIInlll ~ .111 ~ + :llv~ll'~ .~ .~=11~11 ~ + : l l v n l l ~-+ ~-II~vn.4. Vv~ ~ V~(~). Then a6e 2 h~llV A. assuming different values in different appearances. IIInlll'. 2 h'll%ll~.G.u denote the interpolation error.dl~. 2 0.r2h 2 -. (b) Let a < 1. We divide the proof in two parts. v~)=0 in which e = u h ./11~ + :11%11 ~= ~ _< ~ 6e 2 I1.. Franca and E. ~ C f u ) h T H E O R E M 3.3 (Interpolation error).211.. (48) . Then vl converges in Ill" Ill as (a) if a >~l.:11~11 ~ + ~ I1%11~ + ~ Ii~:vn 6~ VA~II~ 0.u. (a) Let a ~> 1.~llv~ll ~ ~< .. n 11. (45) L E M M A 3. IIInlll ~ ~< 0"2C(u)h2k*2 (b) if a < l . + ~ ~< ¢2C(u)h ~k+2 .2h2 h2. F'i B(e. (46) IIInlll ~ ~< e2C(u)h 2k .P. (47) where C(u) is a constant only depending on the exact solution u.

eh) (by Lemma 3. (43). which. Durra do Carmo.P. combining the error estimate (48) with the definition of II1"III. In this case. For a < 1 (e2-dominated). convergence in L 2 may be derived employing duality arguments similar to the classical Aubin-Nitsche's trick.1 implies convergence of GGLS. Consistency combined with the enhanced stability properties obtained in Lemma 3.lVeh) d ~ + ~ nel f/. Franca and E.5 (1) The convergence of GGLS is optimal for any value of a.8 2 Aeh) d•[ effil ½111ehlll~ + ½111'1111: Thus II1~111' ~ Ill.B(e h.1) = B(e71. The Galerkin gradient least-squares method (b) If a < 1 then Illelll ~ ~ e 2 C ( u ) h 2k . (3) The Galerkin/least-squares (GLS) method consists in adding to the Galerkin terms the following mesh-dependent term: . (50) The convergence proof goes as follows: Ille~lll e-.G. eh) (by (50)) = _ BOq . (49) Proof.3.50 L. (2) The GGLS method is consistent in the sense that the exact solution u satisfies (37). Let e h = u h .Ill1'.Ill ~ ~ 4111. comparing (49) with (43) indicates optimal rate convergen~.j • v(~- 82 A~)~. Then e=eh +71. For a >~1 (¢2-dominated). ea)l = Ifo (o'2T/eh + e2V. indicates that the fmite element error converges at optimal rate in L 2 and in H 1. eh) (by Lemma 3. when combined with Lemma 3.ill' • In addition we have IIlelll ~ = Ille~ +.V(o:e~ .e in the H 1 seminorm. vI (51) REMARKS 3.2) IB(n.z7 h. concludes the proof.

we plot in Fig. The first experiment is a one-dimensional example. governed by (30).e 2 Avh) d O . The Galerkin gradient least-squares method 5 ~i foe (O'2Uh-. Employing 20 linear shape functions. The boundary conditions are u ( 0 ) = 0 and u(1)=2. 2. 2 the results given by each method as compared to the exact solution. Notice that contrary to the present GGLS method. 82= 10 -s. Recall that accordiug to the hypotheses employed to design ~-there is no guarantee of a nodally exact solution for f ~ 0.L.82 AUh --/)'~(O'20h -. Notice the oscillations of the Galerkin method near the boundary layer.. The GLS method adds stability in L2. f = x. the GLS method does not contribute in stab'fiizing the H 1 seminorm.1667xl04 . which is already under control by the Galerkin method.G. This idea may be applied to. Durra do Carmo. 4. The coincidence of the nodal solution of the GGLS method with the exact solution is remarkable. with or2= 1. Our multidimensional example consists of employing a 20 × 20 mesh of bilinear shape functions on a unit square with boundary conditions: | • 10"8 • 4. i-~he idea of taking the gradient of the Euler-Lagrange equations and adding its least-squares form to the Galerkin method is new and it is the key to the success of the GGLS method for the model equation studied. Numerical results In this section we perform some numerical experiments to compare the Galerkin and the GGLS methods.P. other models in mechanics and it is the subject of our future work. e----I where ~ is a suitable positive parameter. One-dimensionalsimulation employingGalerkin and GGLS methods (f = x).i ® EXACT GALERKIN u*" I O-I 0 X I Fig. Franca and E. .

The resulting GGLS method converges optimally for any values of the coefficients and presents good stability and accuracy properties.52 L. Hughes for his helpful comments and discussions. The Galerkin gradient least-squares method GGLS GALERKIN Fig. In Fig. 3. We take f = 1. This will be the subject of our future work. u(l. y)-u(x. Conclusions To gain control on the H 1 seminorm.G. We believe that a new door is opened to construct finite element methods employing perturbation techniques. y~<l. •2 _ 10-s. combined with the GLS method. a new finite element technique is proposed within a singular diffusion model: addition of a mesh-dependent least-squares form of the gradient of the Euler-Lagrange equation to the standard Galerkin method. . 3 elevation plots are shown for the Galerkin and the GGLS method. y ~ l . Franca and E.R. o. It seems to us that the present method. Durra do Carmo. y)=u(x.P. Heretofore the dominant strategy ~as been to add mesh-dependent least-squares forms of the Euler-Lagrange residuals to the standard Galerkin method.2 = 1. $. 1)=l f o r 0 < ~ x .J. u(O.O)-O for O~<x. Acknowledgment The authors wish to thank Professor T. Two-dimensional elevation plots employing Galerkin and GGLS methods (fffi 1). forms a powerful and simple strategy to simulate successfully the ever present thin plate and shell problems. The spurious oscillations present in the Galerkin solution are eliminated using the GGLS method. The so-called Galerkin/least-squares (GLS) methodology has limited applicability in singular problems as in the model discussed herein and in some thin s~ructural problems. GGLS.

Peri'~ux and B.J. Comput.G. Streamline upwind/Petrov-Galerkin methods for advection dominated flows. COPPE/UFRJ. [8] A.P. Shakib. Mech. Gallagher.T. [2] T. Chichester. The Galerkin gradient least-squares method 53 References [1] A. Stoufflet.P. 32 (1982) 199-259. 1986) 19-43. 1988. New finite element methods for convection-diffusive problems.R. Element and Volume Techniques.J. eds. Finite element method for [211 . Engrg. Hughes. Chalot. Gale~o and E. [4] C. Brooks.C. L. Mizukami. Hughes. in: A. Comput. 58 (1986) 305-328. 68 (1988) 83-95. Methods Appl. Ph.J. L. Mallet. Methods Appl. Methods Appl.R. J. J. in: R. Finite Element Methods for Convection Dominated Flows (ASME.J. Computing Methods in Engineering and Appfied Sciences V (North-Holland. Finite element methods for conveedon-diffusion problems. Glowinski and J. I. Harari. Engrg. Shakib. AMD-78 (ASME.R.capturing operator for multidimensional advective-diffusive systems. [7] U.H. Tezduyar..R. A Petrov-Galerkin finite element method for the compressible Euler and Navier-Stokes equations. [14] E. [17] T. Finite element methods for linear hyperbolic problem. (1980). M. in: T. New York.N. Hughes and M.R. Comput. Brooks and T. Canada. Harari. Methods Appl. Johnson and U. Gale~o. Franca and E. Niivert and J. Streamline diffusion methods for problems in fluid mechanics. Brooks and T. "l~e generalized streamline operator for multidimensional advection-diffusion systems. Engrg.. 47 (1986) 1-18.N. U. Nivert. Amsterdam. Mallet and A. Streamline upwind/Petrov-Galerkin formulations for convection dominated flows with particular emphasis on the incompressible Navier-Stokes equations.J.P. Chichester. G. Ci~nc.E. to appear). F. Methods Appl. New York. Hughes and M. Comput. Hughes and A. [9] C. Franca. Gale~o. Comput.R.C.. Mallet.J.Jre. Gallagher. Johnson. Mec. in: O. Nivert. Spelce. Mallet. G6teborg. po] L. 54 (1986) 223-234. [11] T. eds. eds. Engrg. Mech. 1986) 251-261. 1. Braunschweig. [5] C. A multidimensional upwind scheme with no crosswind diffusion. Feedback Petrov-Galerkin methods for convection dominated problems. submitted. A consistent formulation of the finite element to solve convectivediffuse transport problems..D.P. eds.D. T.G. A finite element method for convection-diffusion problems.J. Johnson and J.R. T. Brooks. Streamline diffusion methods for the incompressible Euler and Navier-Stokes equations. Axelson. Comput.J. Dervieux and B. L.R Hughes. Engrg. An analysis of some finite element methods for advection diffusion problems.pl. Oden and O.J. [lO] C. in: T. [16] A.H. Rev.. eds. F. M. J.J. Engrg. Thesis. A new finite element method for computational fluid dynamics: IIl. A discontinuity. Chalot and T.E. Numerical Methods for Compressible Flows--Finite Difference. [13] E.E Franca and M. A theoretical framework for Petrov-Galerkin methods with discontinuous weighting functions: appfication to the streamline upwind proced. Amsterdam. Harari. Mallet. RJ.C.R. Hughes. Rio de Janeiro. Finite Element in Fluids IV (Wiley. F.. T. T.. Engrg. Hughes. A consistent approximate upwind Petrov-Galerkin method foz convection-dominated problems.E.R. Durra do Carmo. M.F. Methods Appl.T. Saranen. Mallet. Comp. Hughes. Zienkiewicz.G.E Tezduyar and T. [31 T. Mech. [6] C. Mech. 4 (1986) 309-340. Finite Elements in Fluids VI (Wiley. T. Bras. Mallet. Lions. Hughes and A. Dutra do Carmo and A. [19] F. [121 Symmetric forms of the compressible Euler and Navier-Stokes equations and the second law of thermodynamics.R. Dutra do Canno. G. Chalmers University of Technology.J. Mech.F. Comput. Brazil. Johnson. in: Proc. eds. T. M. Thesis. Calculation of two-dimensional Euler flows with a new Petrov-Galerkin finite element method. Shakib and T. 1982) 311-323. Math. ed. A new finite element method for computational fluid dynamics: I. Ph. Frank and A. in: R. Hughes. in: R. L.G. Zienkiewicz. Oden and O. Nuclear Engineering and Energy Planning Department. F. Carey.S. Spelce. Department of Computer Science. Mech. Dutra do Carmo and A. Notes on Numerical Fluid Mechanics (Vieweg. 54 (1986) 341-355.J. Analytical and Numerical Approaches to Asymptotic Problems in Analysis (North-Holland. 58 (1986) 329-336.C. Sweden (1982). Van Der Sluis. 1981) 99-116. Van Leer. Pitkiiranta.L.R. [15] E.. Franca. Franca. Hughes. Comput. Methods A~. Beyond SUPG. 1979) 19-35. Banff. Mech. Dutra do Carmo. Third International Conference on Finite Element Methods in Fluid Flows.G. 1982) 46--65. Mech. Methods Appl. Carey. Hughes.C. 45 (1984) 285-312.R. Engrg.J.L. Johnson. I.N. A new finite element method for computational fluid dynamics: [iS] II.

Comput.R. Miranda. Circumventing the Babuska-Brezzi condition: a stable Petrov-Galerkin formulation of the Stokes problem accomodating equal-order interpolations. Hughes and L. [361 L. Methods Appl. ty based on a mixed Petrov-Galerkin finite element method. Ph.P. [251 T. [29] L. A mixed finite element formulation for Reissner-Mindlin plate theory: uniform convergence of all higher-order spaces. Mech. Franca. [28] L. Stability. Franca. Franca.R. Two classes of mixed finite element methods. Engrg. Mallet. Comput. Miranda.J. A new family of stable elements for the Stokes problem based on a mixed Galerkin/least-squares finite element formulation. Methods Appl. T. to a p ~ r . T.P. Entropy-stable finite element methods for compressible fluids: application to high Mach number flows with schocks. Loula.D. [33] T.J.. Franca. Hughes and L. The Galerkin gradient least. Engrg. Glowinski and J. Miranda. [381 L. Lions. Comput. [27] L. 63 (1987) 281-303. Franca. Loula. Franca and T. Mech. Franca and E. Berlin-Heidelberg. K. Comput. Miranda. convergence and accuracy of a new finite element method for the circular arch problem. CA 94305. [32] L. A new finite element formulation for computational fluid dynamics: V. Loula.G. Comput. Mech. Miranda. Brazil) 581-599.. Mallet and L.R. Durra do Carmo. LNCC. Hughes. Hughes. A new finite element formulation for computational fluid dynamics: VII.. T. RJ 22290.D.P. Applied Mechanics Division. Methods Appl. Mech. Loula and I.z.R. L. Rio de Janeiro. Mech. Franca. Paper 87-0556 (Reno.P.R. Report 013/88. Hughes and I. 1986) 339-360.R. Comput. Methods Appl. Franca.F. A. 1987). [34] A.P.F.S. Methods Appl. The Stokes problem with various well-posed boundary conditions: symmetric formulations that converge for all velocity/pressure spaces.J. Mech. Math. Methods Appl.D.F.P.A. Thesis. A new mixed finite element method for the Timoshenko beam problem. [22] T. Comput. Loula. Brazil (1988). Mech. Hughes. Engrg. New mixed finite element methods.D.J. Report 012/88. Mixed Petrov-Galerkin method for the Timoshenko beam. Numer. Fourth Brazilian Symposium on Piping and Pressure Vessels 2 (Salvador. [371 T.F. (1985). Franca.R. [30] A. I. Loula and I. Mallet and L. Loula.S. T. CA 94305. Engrg. in: Chung. Ph. I.F. 69.D. Methods Appl.R. Mech.J. Stanford University. T. Wunderfich.R.P. [35] A. Hughes and L. Rio de Janeiro.D. Hughes and L. Methods Appl. Franca and A. Engrg. eds. Analysis and finite element approximation of compressible and incompressible linear isotropic elasticity based upon a variational principle. Hughes. Amsterdam. Comput.P.J. Franca and M.F.P. Bergan.squares method high-speed flows: consistent calculation of boundary flux. M. Stanford University. A successful mixed formulation for axisymmetric shell analysis employing discontinuous stress fields of the same order as the displacement field. [23] TJ. eds.P. Finite Element Methods for Nonlinear Problems (Springer. in: P. 72 (1989) 201-231. Franca and M.J. (1989). Engrg. Brazil (1988). Rio de Janeiro RJ 22290.J.D. Hughes.R. Brazil 22290 (1988).D. On mixed finite element methods for axisymmetric shell analysis. A new family of stable elements for nearly incompressible elasti. Computing Methods in Applied Science and Engineering VII (North-Holland.F. A. Report 025/88. (1988) 89-129. T. A new finite element method for computational fluid dynamics: VI.J. ABCM. 67 (1988) 223-240. U. Mallet.P. Balestra. L. Engrg.P. L. A finite element method for computational fluid dynamics. Karr.A.J. LNCC. Methods Appl. 63 (1987) 133-154. Comput.P. Hughes.R. . 59 (1986) 85-99. Franca. 63 (1987) 97-112. AIAA 25th Aerospace Sciences Meeting.J. Franc. Franca and I. Hughes. Engrg. LNCC. 1986) 761-773. Finite Element Analysis in Fluids (University of Alabama in Huntsville Press. RJ. New finite element methods for the compressible Euier and Navier-Stokes equations. Bathe and W. eds.G. Mech. 1989) 1067-1074. L. M. Hughes. in: R. [24] M. Convergence analysis of the generalized SUPG formulation for linear time-dependent multidimensional advective-diffusive systems.R.L. 65 (1987) 85-96.P.P. Thesis. Franca.P. in: Proc. U. Applied Mechanics Division. [31] A. [26] T.J.J.D.R.54 L. Miranda.P. 53 (1988) 123-141. Engrg. Nevada.

- Foi
- a_politica_fiscal_e_a_falsa_crise_da_seguraridade_social_brasileira_analise_financeira_do_periodo_1990_2005.pdf
- Exame_Mecanica_2016.pdf
- Qualidade Da Imagem Radiográfica
- Exame_Mecanica
- Exame Mec II 2016_1
- Itau_Emprestimo
- Newton Da Costa
- Ressonancia Magnetica Videos Importantes
- Sites_02_10_2015
- EnglishSeparate.txt
- 38142076-Apostila-de-Torno-Mecnico.pdf
- Ementa_Usinagem.pdf
- Usinagem tornearia.pdf
- ementa.pdf
- (2) Cadernos.do.Curso.de.Hacker.redes.p
- Curso de Quimica Basica
- 10 Cold-Blooded Professional Killers - Listverse.pdf
- Term Odin a Mica
- Gestao Agropecuaria
- Jogo20
- Experimentos ressonancia
- Entropia Eleicao Inferno
- English Separate
- Dados Tese

Sign up to vote on this title

UsefulNot useful- 1-s2.0-S0005109812001689-main.pdf
- lecture_slides_chapter_1.pdf
- About Ansys 12
- A Geometrical Non-linear Model for Cable
- Introduction to Finite Element Method
- Lecture Notes on Finite Element Methods for Partial Differential Equations
- eth_biwi_00526
- Punch
- Finite Elements Analysis of Structures
- CNC Magazine Vol.09 Issue33
- The Finite Element Method
- CANCAM07-TKant
- Finite Element Analysis of Ship Structuresjan04mr5
- NAS120 WS10 Fairing
- FEM Zabaras 2DFiniteElements
- Article Woo
- Introduction Ansys
- Elmer Finite Element Software for Multiphysical Optimization Problems
- Methodology FEA Analysis for Hull MEG Storage Tank T-5601A&B -Sherwani
- Bia 1
- Efea Tutorial
- FEA on Furniture IKEA
- ajm32
- CE5166_Project1
- Dantzig Lasso
- Folder HFSS
- ANSYS and Cable Stayed Bridge
- 37016489X
- 7 pt scale
- a three-dimensional finite element study to obtain p-y curves for sand.pdf
- Franca Dudu the Galerkin Gradient Least-squares Method