ebook img

Illustrating Stability Properties of Numerical Relativity in Electrodynamics PDF

5 Pages·0.37 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Illustrating Stability Properties of Numerical Relativity in Electrodynamics

Illustrating Stability Properties of Numerical Relativity in Electrodynamics A. M. Knapp1, E. J. Walker1 and T. W. Baumgarte1,2 1 Department of Physics and Astronomy, Bowdoin College, Brunswick, ME 04011 and 2 Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL, 61801 WeshowthatareformulationoftheADMequationsingeneralrelativity,whichhasdramatically improved the stability properties of numerical implementations, has a direct analogue in classical electrodynamics. We numerically integrate both the original and the revised versions of Maxwell’s equations,andshowthattheirdistinctnumericalbehaviorreflectsthepropertiesfoundinlinearized general relativity. Our results shed further light on the stability properties of general relativity, illustrate them in a very transparent context, and may provide a useful framework for further improvement of numerical schemes. PACSnumbers: 04.25.Dm,02.60.Lj,95.30.Sf 2 0 Motivated by the prospect of gravitational wave de- and the two evolution equations 0 2 tectionsandtheaccompanyingneedfortheoreticalgrav- itational wave templates, much effort has recently gone dtγij =−2αKij (3) n into the development of numerical relativity algorithms a and J that arecapable ofmodeling the most promisingsources of gravitational radiation, in particular the inspiral and 5 d K =−D D α+α(R −2K Kl +KK −M ). 1 coalescence of binary black holes and neutron stars. In t ij i j ij il j ij ij (4) thepast,progresshasbeenhamperedbynumericalinsta- 1 bilities that arise in straight-forwardimplementations of v the traditional Arnowitt-Deser-Misner (ADM, [1]) 3+1 Hereγij isthespatialmetric,Kij theextrinsiccurvature, 1 decompositionofEinstein’sequations(e.g.[2,3]). These K = γijKij its trace, α and βi are the lapse function 5 andtheshiftvector,andρ,S andM aremattersource instabilities have been associatedwith the mathematical i ij 0 terms. The time derivative is defined as d = ∂ −L , structureofthe ADMequations,andasacureanumber t t β 1 and R 0 ofhyperbolicformulationshavebeensuggested(e.g.[2,4] ij 2 as well as [5] and references therein). Alternatively, Shi- 1 0 bataandNakamura[6]andlaterBaumgarteandShapiro Rij = − γkl γij,kl+γkl,ij −γkj,il −γil,kj (5) 2 c/ [7]suggestedareformulationofthe ADMequationsthat (cid:16) (cid:17) q has been demonstrated to dramatically improve the sta- +γkl ΓmilΓmkj −ΓmijΓmkl , r- bility properties of numerical implementations (e.g. [8]). (cid:16) (cid:17) g Whiletheexactreasonforthisimprovementisstillsome- D andΓi aretheRiccitensor,thecovariantderivative, i jk : whatmysterious(see[9],hereafterAABSS,and[10]),this and the connection coefficients associated with γ . Fi- v ij i newformulation,nowoftencalledtheBSSNformulation, nally, R is the scalar curvature R = γijRij. Equations X is quite widely used. (1) through (4) are commonly refered to as the ADM r InthisBriefReportweshowthattheBSSNreformula- equations [1]. a tionofthe ADM equationshas a directanalogueinclas- The first term in the Ricci tensor (5) is an elliptic op- sicalelectrodynamics(E&M).Wenumericallyimplement erator acting on the components of the spatial metric both the original and the revised versions of Maxwell’s γ . If the Ricci tensor contained only this term, the ij equations, and find that their distinct numerical prop- two evolution equations (3) and (4) could be combined erties reflect those found in general relativity (GR). As to forma waveequationfor γ . This property is spoiled ij suggested by AABSS, these properties can be identified by the appearance of the three other second derivative with the propagation of constraint violating modes. We termsin(5),suggestingthattheseterms maybe respon- presentourfindingsinthehopethat,inadditiontobeing sible forthe appearanceofinstabilities inmany straight- a very transparent illustration of the stability properties forward,three-dimensionalimplementationsoftheADM ofGR,they mayproveusefulforthefuture development equations. Thisproblemcanbeavoidedbyeitherusinga and improvement of numerical algorithms. hyperbolicformulationofGR(e.g.[5]),orbyeliminating In a 3+1 decomposition of GR, Einstein’s equations themixedsecondderivativesasintheBSSNformulation. split into the two constraint equations In this formulation,the conformally related metric γ¯ ij is defined as γ¯ = e−4φγ , where the conformal factor ij ij R−KijKij +K2 =2ρ (1) eφ is chosen so that the determinant of γ¯ij is unity, γ¯ = 1. The conformal exponent φ as well as the trace of and the extrinsic curvature, K, are evolved as independent variables. Their evolution equations can be found from D Kj −D K =S (2) j i i i the traces of equations (3) and (4), while the trace-free 2 parts of those equations form evolution equations for γ¯ In terms of a vector potential A , Maxwell’s equations ij i and the trace-free part of the extrinsic curvature, A¯ . can be written as the evolution equations ij The latter equation still contains the Ricci tensor R¯ ij associated with γ¯ij, which contains all the mixed second ∂tAi = −Ei−Diψ (10) derivativesof(5). Thecrucialstepistorealizethatthese ∂ E = −DjD A +DjD A −4πj , (11) t i j i i j i second derivatives can be absorbed in a first derivative and the constraint equation of the “conformalconnection functions” D Ei =4πρ . (12) Γ¯i ≡γ¯jkΓ¯i =−γ¯ij , (6) i e jk ,j Here Ei is the electrical field, ρ the charge density, j e i where the last equality holds because γ¯ = 1. Here and the flux, and ψ the scalar gauge potential. Identifying in the following all barred quantities are associated with A with γ , E with K , and ψ with βi, we see that γ¯ . In terms of Γ¯i, the Ricci tensor can be written i ij i ij ij the structure of the evolution equations (10) and (11) is very similar to that of equations (3) and (4), and that 1 R¯ = − γ¯lmγ¯ +γ¯ ∂ Γ¯k+Γ¯kΓ¯ + (7) the constraint equation (12) can be similarly identified ij 2 ij,lm k(i j) (ij)k withthemomentumconstraint(2). InanalogywithGR, γ¯lm 2Γ¯k Γ¯ +Γ¯k Γ¯ we refer to equations (10) through (12) as System I. l(i j)km im klj In further analogy,we willeliminate the mixed second (cid:16) (cid:17) derivative in (11) by introducing a new variable Γ (compare [11]). Evidently, the only remaining second derivativetermistheellipticoperatorγ¯lmγ¯ ,iftheΓ¯i ij,lm Γ≡D Ai (13) are considered as independent functions. For that pur- i pose, an evolution equation is derived by permuting a (compare (6)), in terms of which (11) becomes time and space derivative in (6) ∂ E =−DjD A +D Γ−4πj . (14) t i j i i i 2 ∂tΓ¯i =−∂j 2αA¯ij −2γ¯m(jβi),m+ 3γ¯ijβl,l+βlγ¯ij,l . As in (7), the mixed second derivatives have been ab- (cid:16) (cid:17)(8) sorbed in a first derivative of the new variable. An evo- lution equationfor Γ can againbe derived by permuting The divergence of the extrinsic curvature can now be a time and space derivative in the definition of the new eliminated with the help of the momentum constraint variable (13) (2), which yields the evolution equation ∂ Γ=D ∂ Ai =−D Ei−DiD ψ =−4πρ −DiD ψ. t i t i i e i ∂ Γ¯i =−2A¯ij∂ α (9) (15) t j 2 +2α Γ¯i A¯kj − γ¯ij∂ K−γ¯ijS +6A¯ij∂ φ Here we have used the constraint (12), which, as in GR, jk 3 j j j will turn out to be crucial (see (21) below). Equations +βj∂(cid:16)¯ Γi−Γ¯j∂ βi+ 2Γ¯i∂ βj + 1γ¯liβj +γ¯(cid:17)ljβi . (10), (14), (15) form the evolution equations of what we j j 3 j 3 ,jl ,lj call System II. The definition (13) together with (12) form the constraint equations of System II. As in[7]we willreferto the ADM equations(1)through For vanishing scalar potential ψ = 0, an analytical (4)asSystemI,andtothenewBSSNsystemofequations vaccuum solution to Maxwell’s equations is given by the as System II. A complete listing and derivation of the purely toriodal dipole field latter can be found in [7]. Inanefforttobetter understandthe improvednumer- e−λv2 −e−λu2 ve−λv2 +ue−λu2 ical behavior of System II, AABSS linearized Systems I Aφˆ=Asinθ −2λ . r2 r ! andII andidentifiedtheir characteristicstructure. They (16) found that SystemI has constraintviolating modes with a characteristic speed of zero. In System II, the charac- Here A is the amplitude, λ parameterizes the size of the teristicspeedofthesemodeschangestothespeedoflight. wavepackage,anduandv aretheretardedandadvanced AABSSfurther demonstratedthatina non-linearmodel time u≡t+r and v ≡t−r. According to (10), Eφˆ can problem the existence of non-propagating,constraint vi- befoundbytakingatimederivativeof(16). Sinceρ =0 olating modes may lead to numerical instabilities, as en- e and D Ai = 0 in this solution, ψ = 0 is consistent with counteredinimplementationsofSystemI.Theiranalysis i both the Coulomb gauge and the Lorentz gauge alsodemonstratedthatthe usageofthe momentumcon- straint in the derivation of (9) is crucial for the charac- ∂ ψ =−D Ai−4πρ =−Γ−4πρ (17) t i e e teristicspeedfortheconstraint-violatingmodetochange to a non-zero value, and hence for the stability of the (where the second equality applies for System II), if ap- system. In the following we will show that very similar propriateboundaryandinitialconditionsarechosen. All properties can be found in E&M. results shown below were obtained with Lorentz gauge. 3 FIG. 1: The integrated constraint violation kCk≡( C2dV)1/2 for evolutions usingLorentz gauge with resolution 323 and 643 with outer boundaries at 4 (OB4) and 8 (OB8). R Asinitialdataforourdynamicalsimulationsweadopt time t=9.375, C is larger in System II than in System the analytical solution (16) at t=0 I, as one expects from Fig. 1. At a much later time (t=200), however, System I has settled down to a con- Aφˆ=0, Eφˆ =8Arsinθλ2e−λr2 (18) stant shape, while C in System II has almost completely dissipated. with A = λ = 1 and transformed into cartesian coordi- These numerical results demonstrate that, as in GR, nates. the constraintviolationC behaves very differently in the We numerically implement System I and II following two systems. In E&M, this different behavior can be the algorithm of [7] as closely as possible. In particular, understood very easily from an analytic argument. wewrotethecodeinthreespatialdimensionsusingcarte- For System I, it is easy to show that a time derivative sian coordinates. We use an iterative Crank-Nicholson of the constraint violation (19) vanishes scheme [12] to update the evolution equations, and im- pose outgoing wave boundary conditions on Ei, Ai and ∂tC =Di∂tEi−4π∂tρe =−4π(Diji+∂tρe)=0, (20) Γ as in [7]. We verified that the numerical solution con- wherewehaveusedthecontinuityequationD ji+∂ ρ = verges to the analytical solution to second order as long i t e 0. ThisexplainswhyatlatetimestheprofileofCremains as the solution is not affected by the outer boundaries unchanged in System I. This property is the analogue of (which are first order accurate). AABSS’sfindingthatthelinearizedADMequationshave We compare the performance of System I and II by non-propagating,constraint violating modes. monitoring the constraint violation ForSystemII,ontheotherhand,itcanbeshownthat C ≡D Ei−4πρ . (19) C satisfies a wave equation i e ∂2C = ∂ Di∂ E −4π∂2ρ In Fig. 1, we show integrated values kCk ≡ ( C2dV)1/2 t t t i t e for Systems I and II for two different gridsizes (323 and = ∂tDi(−DjDjAi+DiΓ−4πji)−4π∂t2ρe 643) and two different locations of the outerRboundary = −Di(D Dj∂ A −D ∂ Γ)−4π∂ (Dij +∂ ρ ) j t i i t t i t e (at 4 (OB4) and 8 (OB8)). = Di D Dj(E +D ψ)−D (4πρ +DjD ψ) At early times, System I violates the constraint (12) j i i i e j to a lesser degree than System II. After about a light- = D (cid:16)Dj(DiE −4πρ )=D DjC, ((cid:17)21) j i e j crossing time, when the electro-magnetic wave has left thenumericalgrid,kCksettlesdowntoanearlyconstant which explains why the constraint violations propagate value,whichprimarilydependsonthegrid-resolution. In awayinSystemII.ThisistheanalogueofAABSS’sfind- System II, kCkis also largestafter about a light-crossing ing that in the relativistic System II (the BSSN equa- time, but after that kCk decreases exponentially. As one tions)the constraintviolatingmodespropagatewith the might expect, the decay time of this exponential fall-off speed of light. Moreover, we now realize why the usage scales with the location of the outer boundary. of the constraint (12) in (9) was crucial – without hav- To further compare these Systems, we compare snap- ing made this substitution the terms on the last line of shotsoftheconstraintviolationC forthe323 OB4evolu- (21) would have cancelled, ∂2C = 0, leading to a non- t tion in Fig. 2. With identical initial data, both Systems propagatingconstraint violation as in System I. The ad- have identical values for C at t = 0. At an intermediate dition of constraint equations to the evolution equations 4 0.03 0.03 0.02 0.02 0.01 0.01 0 0 −0.01 −0.01 −0.02 −0.02 −0.03 −0.03 4 4 3 4 3 4 2 3 2 3 2 2 1 1 1 1 0 0 0 0 0.03 0.03 0.02 0.02 0.01 0.01 0 0 −0.01 −0.01 −0.02 −0.02 −0.03 −0.03 4 4 3 4 3 4 2 3 2 3 2 2 1 1 1 1 0 0 0 0 0.03 0.03 0.02 0.02 0.01 0.01 0 0 −0.01 −0.01 −0.02 −0.02 −0.03 −0.03 4 4 3 4 3 4 2 3 2 3 2 2 1 1 1 1 0 0 0 0 FIG. 2: C in the z = .0625 plane (the first of our grid points) for System I (left column) and System II (right column) for a 323 OB4 evolution in Lorentz gauge. We show results for t=0 (top row), t=9.375 (middle row) and t=100 (bottom row). C settles down toa constant profile in System I,but dissipates away in System II. hasbeenfoundtobecrucialinseveralotherformulations and the boundary. Similar problems are likely to occur of Einstein’s equations as well (e.g. [4, 13]). in GR, but might be easier to analyze in the simpler To summarize, we see that the numerical stability framework of E&M. properties of two formulations of GR are beautifully re- Acknowledgments flected in similar formulations of E&M. Maxwell’s equa- tions therefore provide a verytransparentframeworkfor analyzing these properties, which may be useful for fu- This paper was supported in part by NSF Grant ture algorithm development. For example, Fig. 2 shows PHY 99-02833 to the University of Illinois at Urbana- that the outer boundaries produce much more noise in Champaign and Bowdoin College as a subrecipient. System I than in System II, which points to an incon- AMK gratefully acknowledges support from the Mellon sistency between the treatment of the interior equations Foundation. [1] R. Arnowitt, S. Deser & C. W. Misner, in Gravitation: 75, 600 (1995). AnIntroductiontoCurrentResearch,editedbyL.Witten [3] A.M.Abrahamset.al.,Phys.Rev.Lett.80,1812(1998). (Wiley,New York,1962). [4] A.Anderson&J.W.York,Jr.,Phys.Rev.Lett.82,4384 [2] C.Bona,J.Mass´o, E.Seidel&J.Stela,Phys.Rev.Lett. (1999). 5 [5] O.Reula, Living Rev.Rel. 1, 3 (1998). M. Suen,Phys.Rev. D 62, 124011 (2000) (AABSS). [6] M. Shibata & T. Nakamura, Phys. Rev. D 52, 5428 [10] S. Frittelli & O. Reula, J. Math. Phys. 40, 5143 (1999); (1995). Miller, M., gr-qc/0008017 (2000). [7] T. W. Baumgarte & S. L. Shapiro, Phys. Rev. D 59, [11] T. De Donder, La gravifique einsteinienne (Gauthier- 024007 (1999). Villars,Paris,1921);C.Lanczos,Phys.Z.23,537(1922); [8] M.Alcubierreet.al.,Phys.Rev.D62,044034(2000);L. [12] S. A. Teukolsky,Phys. Rev.D 61, 087501 (2000). Lehner,M.Huq,&D.Garrison,Phys.Rev.D62,084016 [13] B. Kelly et. al., Phys. Rev. D 64, 084013 (2001); G. (2000);M.Alcubierre&B.Bru¨gmann,Phys.Rev.D63, Yoneda & H.Shinkai, Phys.Rev.D 63 124019 (2001). 104006 (2001). [9] M.Alcubierre,G.Allen,B. Bru¨gmann, E.Seidel,&W.-

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.