29.11.2012 Views

Tretz-Discontinuous Galerkin Methods for Time-Harmonic Maxwells

Tretz-Discontinuous Galerkin Methods for Time-Harmonic Maxwells

Tretz-Discontinuous Galerkin Methods for Time-Harmonic Maxwells

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

ESCO 2012, PLZEŇ — 25–29TH JUNE 2012<br />

Trefftz-DG methods<br />

<strong>for</strong> time-harmonic<br />

Maxwell’s equations<br />

Andrea Moiola<br />

DEPARTMENT OF MATHEMATICS AND STATISTICS, UNIVERSITY OF READING<br />

Joint work with R. Hiptmair (ETH Zürich) & I. Perugia (Università di Pavia)


<strong>Time</strong>-harmonic PDEs<br />

Propagation, scattering and interaction of time-harmonic<br />

acoustic and electromagnetic waves are governed by:<br />

the Helmholtz equation<br />

−∆u −ω 2 u = 0<br />

and the Maxwell equations<br />

in Ω ⊂ R N , N = 2, 3,<br />

∇×(∇×E)−ω 2 E = 0 in Ω ⊂ R 3 .<br />

ω > 0 is the wavenumber.<br />

We supplement them with impedance (Robin) boundary<br />

conditions in bounded polygonal/polyhedral domains.


Oscillations<br />

<strong>Time</strong>-harmonic solutions are inherently oscillatory: a lot of DOFs<br />

needed <strong>for</strong> any polynomial discretization!<br />

[Helmholtz BVP, picture by T. Betcke]<br />

Wavenumber ω = 2π/λ is the crucial parameter.


Pollution effect<br />

Big issue in FEM solution <strong>for</strong> high wavenumbers: pollution effect<br />

��<br />

��<br />

��<br />

��<br />

��<strong>Galerkin</strong><br />

error��<br />

��<br />

��<br />

��<br />

��<br />

≥ C ω<br />

��best<br />

approximation error��<br />

a<br />

a > 0, ω → ∞.<br />

It affects every (low order) method in h: [BABUŠKA, SAUTER 2000].<br />

⇓<br />

Oscillating solutions + pollution effect<br />

= standard FEM are too expensive at high frequencies!<br />

Special schemes required, p-version preferred (hp even better!).


Trefftz methods<br />

How to deal with these phenomena?<br />

Trefftz methods incorporate the wavelength in the FEM space:<br />

trial and test functions are solutions of Helmholtz/Maxwell’s<br />

equations in each mesh element, e.g.:<br />

Vp ⊂ T(Th) = � w ∈ L 2 (Ω) 3 : ∃s > 0 s.t. w |K ∈ H 1/2+s (K),<br />

∇×(∇×w |K)−ω 2 �<br />

w |K = 0 in each K ∈ Th .<br />

Hope: reduce number of DOFs needed.<br />

For instance:<br />

vector plane waves x ↦→ ae iωx·d , d ∈ S 2 , a·d = 0;<br />

spherical waves (vector Fourier–Bessel functions);<br />

singular solutions;<br />

. . .


Trefftz methods<br />

How to deal with these phenomena?<br />

Trefftz methods incorporate the wavelength in the FEM space:<br />

trial and test functions are solutions of Helmholtz/Maxwell’s<br />

equations in each mesh element, e.g.:<br />

Vp ⊂ T(Th) = � w ∈ L 2 (Ω) 3 : ∃s > 0 s.t. w |K ∈ H 1/2+s (K),<br />

∇×(∇×w |K)−ω 2 �<br />

w |K = 0 in each K ∈ Th .<br />

Hope: reduce number of DOFs needed.<br />

For instance:<br />

vector plane waves x ↦→ ae iωx·d , d ∈ S 2 , a·d = 0;<br />

spherical waves (vector Fourier–Bessel functions);<br />

singular solutions;<br />

. . .


Wave-based methods<br />

How to “match” traces across interelement boundaries?<br />

Many techniques are available <strong>for</strong> Helmholtz:<br />

Least squares schemes: Method of fundamental solutions<br />

(MFS);<br />

Lagrange multipliers: <strong>Discontinuous</strong> enrichment method<br />

(DEM);<br />

Partition of unity method (PUM/PUFEM), non-Trefftz;<br />

<strong>Discontinuous</strong> <strong>Galerkin</strong> (DG): Ultraweak variational<br />

<strong>for</strong>mulation (UWVF).<br />

For Maxwell: UWVF by Cessenat and Després 1996-2003,<br />

For Maxwell: UWVF by Huttunen, Malinen and Monk 2007, . . .<br />

Here: extension to Maxwell of a family of Trefftz-discontinuous<br />

<strong>Galerkin</strong> (TDG) methods that includes the UWVF.<br />

Main focus: p-version. Ideas from simpler model: Helmholtz.


Model problem<br />

Homogeneous time-harmonic Maxwell equations with<br />

impedance boundary conditions:<br />

�<br />

∇×(µ −1 ∇×E)−ω 2 ǫ E = 0 in Ω,<br />

µ −1 (∇×E)×n−iωϑ(n×E)×n = g on ∂Ω,<br />

Ω ⊂ R 3 , Lipschitz (star-shaped) bounded polyhedral<br />

domain,<br />

g ∈ L 2 T (∂Ω),<br />

ω > 0 wavenumber,<br />

ǫ,µ > 0 (piecewise) constant (here: ǫ ≡ µ ≡ 1),<br />

0 �= ϑ ∈ R.


TDG: derivation — I<br />

1 Introduce a mesh Th on Ω;<br />

2 multiply the Maxwell equations with a test field ξ and<br />

integrate by parts on a single element K ∈ Th:<br />

�<br />

�<br />

� � 2<br />

(∇×E)·(∇×ξ)−ω E·ξ dV + n×(∇×E)·ξ dS = 0;<br />

3<br />

�<br />

K<br />

K<br />

∂K<br />

integrate by parts again: ultraweak step<br />

E· � ∇×∇×ξ−ω 2 ξ � �<br />

� �<br />

dV+ (n×E)·(∇×ξ)+n×(∇×E)·ξ dS = 0;<br />

∂K


TDG: derivation — I<br />

1 Introduce a mesh Th on Ω;<br />

2 multiply the Maxwell equations with a test field ξ and<br />

integrate by parts on a single element K ∈ Th:<br />

�<br />

�<br />

� � 2<br />

(∇×E)·(∇×ξ)−ω E·ξ dV + n×(∇×E)·ξ dS = 0;<br />

3<br />

�<br />

K<br />

K<br />

∂K<br />

integrate by parts again: ultraweak step<br />

E· � ∇×∇×ξ−ω 2 ξ � �<br />

� �<br />

dV+ (n×E)·(∇×ξ)+n×(∇×E)·ξ dS = 0;<br />

∂K


TDG: derivation — I<br />

1 Introduce a mesh Th on Ω;<br />

2 multiply the Maxwell equations with a test field ξ and<br />

integrate by parts on a single element K ∈ Th:<br />

�<br />

�<br />

� � 2<br />

(∇×E)·(∇×ξ)−ω E·ξ dV + n×(∇×E)·ξ dS = 0;<br />

3<br />

�<br />

K<br />

K<br />

∂K<br />

integrate by parts again: ultraweak step<br />

E· � ∇×∇×ξ−ω 2 ξ � �<br />

� �<br />

dV+ (n×E)·(∇×ξ)+n×(∇×E)·ξ dS = 0;<br />

∂K


TDG: derivation — II<br />

4 choose a discrete Trefftz space Vp(K) and replace traces<br />

on ∂K with numerical fluxes �Ep and �Hp:<br />

E → Ep (discrete solution) in K ,<br />

E → �Ep ,<br />

∇×E<br />

iω → �Hp<br />

on ∂K ;<br />

5 use the Trefftz property: ∀ ξ p ∈ Vp(K)<br />

�<br />

K<br />

Ep· � ∇×∇×ξ p −ω 2 �<br />

ξp dV<br />

� �� �<br />

=0<br />

�<br />

� �<br />

+ (n× �Ep)·(∇×ξ p)+iω(n× �Hp)·ξ p dS = 0.<br />

∂K<br />

� �� �<br />

TDG eq. on 1 element<br />

Two things to set:<br />

discrete space Vp and numerical fluxes � Ep, � Hp.


TDG: derivation — II<br />

4 choose a discrete Trefftz space Vp(K) and replace traces<br />

on ∂K with numerical fluxes �Ep and �Hp:<br />

E → Ep (discrete solution) in K ,<br />

E → �Ep ,<br />

∇×E<br />

iω → �Hp<br />

on ∂K ;<br />

5 use the Trefftz property: ∀ ξ p ∈ Vp(K)<br />

�<br />

K<br />

Ep· � ∇×∇×ξ p −ω 2 �<br />

ξp dV<br />

� �� �<br />

=0<br />

�<br />

� �<br />

+ (n× �Ep)·(∇×ξ p)+iω(n× �Hp)·ξ p dS = 0.<br />

∂K<br />

� �� �<br />

TDG eq. on 1 element<br />

Two things to set:<br />

discrete space Vp and numerical fluxes � Ep, � Hp.


TDG: the space Vp<br />

The abstract error analysis works <strong>for</strong> every discrete Trefftz space!<br />

Possible choice: plane waves space (aℓ, dℓ ∈ S 2 , aℓ · dℓ = 0)<br />

Vp(Th) =<br />

�<br />

v ∈ L 2 (Ω) 3 : v|K(x) =<br />

p�<br />

ℓ=1<br />

α K 1,ℓaℓe iωd ℓ·x +α K 2,ℓ(aℓ × dℓ)e iωd ℓ·x ,<br />

α K ν,ℓ ∈ C, ∀K ∈ Th<br />

2p = number of basis plane waves (DOFs) in each element.<br />

�<br />

.


Numerical fluxes<br />

We choose the numerical fluxes, on interior faces:<br />

�Ep : = {Ep}− β<br />

iω [∇×Ep]T ,<br />

�Hp : = 1<br />

iω {∇×Ep}+α [Ep]T ,<br />

and on ∂Ω:<br />

�Ep : = Ep −δϑ −1<br />

�<br />

n×(∇×Ep)<br />

iω<br />

�Hp : = 1<br />

iω ∇×Ep<br />

�<br />

∇×Ep<br />

−(1−δ)<br />

iω<br />

+ϑ(n×Ep)×n+ g<br />

�<br />

iω<br />

−ϑ(n×Ep)− n×g<br />

iω<br />

{·} = averages, [·]T = tangential jumps on the interfaces,<br />

α, β > 0, δ ∈ (0, 1<br />

2 ] parameters at our disposal (in L∞ (Fh)),<br />

independent of h, p, ω (other choices possible).<br />

,<br />

�<br />

,


Variational <strong>for</strong>mulation of the TDG<br />

With this fluxes, summing over the elements K ∈ Th, the TDG<br />

method reads: find Ep ∈ Vp(Th) s.t.<br />

�<br />

Ah(Ep,ξ p ) =<br />

δ<br />

iωϑ (n×g)·(∇×ξ �<br />

p ) dS+ (1−δ)(n×g)·(n ×ξ p ) dS<br />

∂Ω<br />

∀ ξp ∈ Vp(Th), where (FI �<br />

h = interior skeleton)<br />

Ah(Ep,ξ p ) = −{Ep}· [∇×ξ p ]T − {∇×Ep}· [ξp ]T dS<br />

F I h<br />

�<br />

+ (1−δ)(n×Ep)·(∇×ξ)−δ(∇×Ep)·(n×ξ) dS<br />

∂Ω<br />

�<br />

− i<br />

F I h<br />

ω −1 β [∇×Ep]T · [∇×ξ p ]T +ωα [Ep]T · [ξ p ]T dS<br />

�<br />

δ<br />

− i<br />

∂Ω ωϑ n×(∇×Ep)·n×(∇×ξ p)+ω(1−δ)ϑ(n×Ep)·(n×ξ p) dS.<br />

Ep ↦→ (−Im Ah(Ep,Ep)) 1<br />

2 is a norm on the Trefftz space ⇒ ∃!Ep.<br />

∂Ω


Variational <strong>for</strong>mulation of the TDG<br />

With this fluxes, summing over the elements K ∈ Th, the TDG<br />

method reads: find Ep ∈ Vp(Th) s.t.<br />

�<br />

Ah(Ep,ξ p ) =<br />

δ<br />

iωϑ (n×g)·(∇×ξ �<br />

p ) dS+ (1−δ)(n×g)·(n ×ξ p ) dS<br />

∂Ω<br />

∀ ξp ∈ Vp(Th), where (FI �<br />

h = interior skeleton)<br />

Ah(Ep,ξ p ) = −{Ep}· [∇×ξ p ]T − {∇×Ep}· [ξp ]T dS<br />

F I h<br />

�<br />

+ (1−δ)(n×Ep)·(∇×ξ)−δ(∇×Ep)·(n×ξ) dS<br />

∂Ω<br />

�<br />

− i<br />

F I h<br />

ω −1 β [∇×Ep]T · [∇×ξ p ]T +ωα [Ep]T · [ξ p ]T dS<br />

�<br />

δ<br />

− i<br />

∂Ω ωϑ n×(∇×Ep)·n×(∇×ξ p)+ω(1−δ)ϑ(n×Ep)·(n×ξ p) dS.<br />

Ep ↦→ (−Im Ah(Ep,Ep)) 1<br />

2 is a norm on the Trefftz space ⇒ ∃!Ep.<br />

∂Ω


TDG norms<br />

We define two DG norms on the Trefftz space T(Th):<br />

|||w||| 2 F h := ω −1<br />

�<br />

�<br />

�β 1/2 �<br />

�<br />

[∇×w]T<br />

+ω −1<br />

�<br />

�<br />

�(δ/ϑ) 1/2 �<br />

�<br />

n×(∇×w)<br />

|||w||| 2<br />

F +<br />

h<br />

+ω −1<br />

∀ w, v ∈ T(Th):<br />

:= |||w||| 2 F h +ω<br />

�<br />

�<br />

�α −1/2 {(∇ × w)T }<br />

− Im Ah(w,w) = |||w||| 2 Fh<br />

|Ah(v,w)| ≤ 2|||v||| F +<br />

h<br />

� 2<br />

0,FI h<br />

� 2<br />

0,∂Ω<br />

�<br />

�<br />

�β −1/2 �<br />

�<br />

{wT }<br />

�<br />

�<br />

� 2<br />

0,FI h<br />

|||w|||Fh<br />

�<br />

�<br />

+ω �α 1/2 �<br />

�<br />

[w]T<br />

� 2<br />

0,F I h<br />

�<br />

�<br />

+ω �(1−δ) 1/2 ϑ 1/2 �<br />

�<br />

(n×w)<br />

� 2<br />

0,FI h<br />

�<br />

�<br />

+ω<br />

�<br />

�<br />

�<br />

�δ −1/2 ϑ 1/2 (n×w) � 2<br />

0,∂Ω<br />

⇒<br />

.<br />

� 2<br />

0,∂Ω<br />

Existence & uniqueness Ep,<br />

quasi-optimality<br />

in Fh −F +<br />

h norms.<br />

,


Quasi-optimality in mesh-independent norms<br />

We have quasi-optimality / error estimates in DG norm:<br />

|||E−Ep|||Fh<br />

≤ 3 inf<br />

ξp∈Vp(Th) |||E−ξ + p||| F ,<br />

h<br />

we want them in a more explicit mesh-independent norm<br />

(e.g., L 2 (Ω)).<br />

Tools:<br />

Helmholtz decomposition;<br />

special duality argument from MONK, WANG 1999;<br />

new wavenumeber-explicit stability bounds <strong>for</strong> the<br />

BVP (using Rellich identities <strong>for</strong> Maxwell);<br />

new regularity estimates (in H 1/2+ǫ (curl;Ω)).<br />

We obtain (<strong>for</strong> some ǫ > 0):<br />

�<br />

(E−Ep)·v dV Ω �E−Ep� H(div,Ω) ′ := sup �v�H(div;Ω) v∈H(div;Ω)<br />

≤ C � h −1/2 1 3 −<br />

(ω 2 −<br />

+ω 2)+h ǫ (ω 1 3<br />

2 −<br />

+ω 2) � |||E−Ep|||Fh .


Quasi-optimality in mesh-independent norms<br />

We have quasi-optimality / error estimates in DG norm:<br />

|||E−Ep|||Fh<br />

≤ 3 inf<br />

ξp∈Vp(Th) |||E−ξ + p||| F ,<br />

h<br />

we want them in a more explicit mesh-independent norm<br />

(e.g., L 2 (Ω)).<br />

Tools:<br />

Helmholtz decomposition;<br />

special duality argument from MONK, WANG 1999;<br />

new wavenumeber-explicit stability bounds <strong>for</strong> the<br />

BVP (using Rellich identities <strong>for</strong> Maxwell);<br />

new regularity estimates (in H 1/2+ǫ (curl;Ω)).<br />

We obtain (<strong>for</strong> some ǫ > 0):<br />

�<br />

(E−Ep)·v dV Ω �E−Ep� H(div,Ω) ′ := sup �v�H(div;Ω) v∈H(div;Ω)<br />

≤ C � h −1/2 1 3 −<br />

(ω 2 −<br />

+ω 2)+h ǫ (ω 1 3<br />

2 −<br />

+ω 2) � |||E−Ep|||Fh .


Quasi-optimality in mesh-independent norms<br />

We have quasi-optimality / error estimates in DG norm:<br />

|||E−Ep|||Fh<br />

≤ 3 inf<br />

ξp∈Vp(Th) |||E−ξ + p||| F ,<br />

h<br />

we want them in a more explicit mesh-independent norm<br />

(e.g., L 2 (Ω)).<br />

Tools:<br />

Helmholtz decomposition;<br />

special duality argument from MONK, WANG 1999;<br />

new wavenumeber-explicit stability bounds <strong>for</strong> the<br />

BVP (using Rellich identities <strong>for</strong> Maxwell);<br />

new regularity estimates (in H 1/2+ǫ (curl;Ω)).<br />

We obtain (<strong>for</strong> some ǫ > 0):<br />

�<br />

(E−Ep)·v dV Ω �E−Ep� H(div,Ω) ′ := sup �v�H(div;Ω) v∈H(div;Ω)<br />

≤ C � h −1/2 1 3 −<br />

(ω 2 −<br />

+ω 2)+h ǫ (ω 1 3<br />

2 −<br />

+ω 2) � |||E−Ep|||Fh .


Orders of convergence<br />

Last missing ingredient: best approximation estimates <strong>for</strong> vector<br />

plane/spherical waves.<br />

�<br />

aℓeiωx·dℓ , aℓ × dℓeiωx·dℓ �<br />

ℓ=1,...,(q+1) 2<br />

|aℓ| = |dℓ| = 1, dℓ · aℓ = 0.<br />

We show them by proving analogous results <strong>for</strong> the Helmholtz<br />

equation; tools: Vekua’s theory, harmonic polynomial<br />

approximation, spherical waves, “inversion” of Jacobi–Anger<br />

expansion. . .<br />

Ak x dk<br />

Algebraic orders in h and q, using 2(q+1) 2 spherical/plane<br />

waves per element:<br />

|||E−Ep|||Fh ≤ C (ωh)ω −5/2<br />

�<br />

h<br />

qλ �k−3/2 �∇×E� k+1,ω,Ω ,<br />

�E−Ep� H(div;Ω) ′ ≤ C (ωh)(ω −5/2 +ω −4 ) hk−2<br />

q λ(k−3/2) �∇×E� k+1,ω,Ω .<br />

Ak<br />

dk


Orders of convergence<br />

Last missing ingredient: best approximation estimates <strong>for</strong> vector<br />

plane/spherical waves.<br />

�<br />

aℓeiωx·dℓ , aℓ × dℓeiωx·dℓ �<br />

ℓ=1,...,(q+1) 2<br />

|aℓ| = |dℓ| = 1, dℓ · aℓ = 0.<br />

We show them by proving analogous results <strong>for</strong> the Helmholtz<br />

equation; tools: Vekua’s theory, harmonic polynomial<br />

approximation, spherical waves, “inversion” of Jacobi–Anger<br />

expansion. . .<br />

Ak x dk<br />

Algebraic orders in h and q, using 2(q+1) 2 spherical/plane<br />

waves per element:<br />

|||E−Ep|||Fh ≤ C (ωh)ω −5/2<br />

�<br />

h<br />

qλ �k−3/2 �∇×E� k+1,ω,Ω ,<br />

�E−Ep� H(div;Ω) ′ ≤ C (ωh)(ω −5/2 +ω −4 ) hk−2<br />

q λ(k−3/2) �∇×E� k+1,ω,Ω .<br />

Ak<br />

dk


Summary and open problems<br />

TDG method <strong>for</strong> Maxwell:<br />

<strong>for</strong>mulation, well-posedness,<br />

h- and p-convergence in energy and H(div;Ω) ′ norms,<br />

approximation estimates <strong>for</strong> plane and spherical waves.<br />

Not discussed: strong ill-conditioning <strong>for</strong> (moderately) high p.<br />

Open problems:<br />

hp-version,<br />

non-constant coefficients ǫ(x), µ(x)<br />

and non-homogeneous eqs.,<br />

adaptivity on PW directions, ongoing work<br />

with T. Betcke and J. Phillips (UCL),<br />

non star-shaped domains,<br />

better approximation estimates,<br />

other PDEs,<br />

. . .


THANK YOU!

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!