Title: AVOIDING GAUGE AND CONSTRAINT SHOCKS IN NUMERICAL RELATIVITY
1AVOIDING GAUGE AND CONSTRAINTSHOCKSIN
NUMERICAL RELATIVITY
Bernd Reimann
ICN, Mexico City AEI, Golm
AEI, 30/09/2004
2Aim and Topics
- The aim of this talk is to give a brief summary
of joint work done together with M. Alcubierre,
D. Núñez and J.A. González. - It is work in progress, and a paper is in
preparation. The topics are - FORMATION and AVOIDANCE of SHOCKS
- What kind of shocks arise by what mechanisms for
systems of PDE's? - gradient catastrophe gt (indirect)
linear degeneracy - blow-up in finite time gt source
criteria - Numerical examples (equations of wave-type)
- EINSTEINS EQUATIONS in SPHERICAL SYMMETRY
- What shocks can form in the particular case of
Einsteins equations? - gauge shocks gt
shock-avoiding lapse - constraint shocks gt adjustments
using constraints - Numerical examples (robust stability test,
evolutions of Gaussian perturbations on top of
Minkowski)
3Formation of Shocks for Wavelike Equations
- We use mathematical theory concerning
shock-/singularity-formation for hyperbolic
systems of PDEs with wavelike character - Shock formation mechanisms
S. Alinhac, Blowup
for Nonlinear Hyperbolic Equations, 1995 - Gradient catastrophe, linear degeneracy
P. Lax, Hyperbolic
Systems of Conservation Laws and the Mathematical
Theory of Shock Waves, 1973 - Blowup in finite time, analysis of source terms
F. John, Nonlinear Wave
Equations, Formation of Singularities, 1979 F.
John, Formation of Singularities in
One-Dimensional Nonlinear Wave Propagation, Comm.
Pure Appl. Math., vol. 27, p. 377-405, 1974 - Gauge shocks in Einsteins equations
M. Alcubierre,
The Appearance of Coordinate Shocks in Hyperbolic
Formulations of General Relativity, Phys. Rev. D
55, p. 5981-5991, 1997 - This theory typically states that for
perturbations of order
shocks form
on a timescale of order , whereas
in shock- avoiding cases one can obtain
timescales
4Evolution Equations
- We are interested in scalar PDEs like
- and their generalization being u-v-evolution
systems with PDEs of the - form together with
- Here and are n- and m-dimensional
vectors, and A is a strongly hyperbolic mxm
matrix depending on the us only. Its real
eigenvalues and its linearly independent
eigenvectors are - and
- Using matrices R and R-1 consisting of the
eigenvectors, the matrix A can be diagonalized by
introducing the eigenfields ,
us lapse, metric components vs partial
derivatives of the us w.r.t. x and t (denoted by
Ds and Ks)
5Geometric Blow-Up and Linear Degeneracy
- The blow-up mechanism is due to focusing of
characteristics. Here derivatives of u become
infinite whereas u itself remains finite. - Example Burgers equation
yields - Here particles move with zero acceleration.
Shocks form as particles with high velocity
collide with others ahead having lower velocity. - This gradient catastrophe occurs at time
- The corresponding criteria for shock-avoidance
is linear degeneracy, which demands
and generalizes for systems of PDEs to - By replacing , M. Alcubierre
in the 90ies developed for the u-v-system of
PDEs the condition of indirect linear degeneracy
,
6ODE-Mechanism and the Source Criteria
- Due to self-increase in its influence domain u
blows up in finite time. - Example ODEs of the form
have solutions
- which become infinite at the time
- For the u-v-system of PDEs when writing the
evolution of the main system in terms of
eigenfields, one finds - In particular, for the time derivative along a
characteristic follows - Hence, for the source criteria we demand in
general - or, if we identify in a particular case more
dangerous terms ,
7Equations of Wave-Type
- As a first example we study equations of
wave-type having the form - which by introducing and
we write as first order system - Eigenvalues
Eigenfields - For indirect linear degeneracy from
- follows that c is a constant (which we set to 1).
- For the source criteria from
- follows that q has to be free of terms
(i.e. no term, and only quadratic
terms
are allowed).
8Numerical Examples (1)
Convergence factors
INDIRECT LINEAR DEGENERACY
NO
YES
vs
Gradient catastrophe (1) c u, q 0
Standard wave equation c 1, q 0
Y E S
S O U R C E
?x u
Gradient catastrophe (2) c u, q 2uD2-K2/u
Wavelike equation c 1, q 2D2-K2
?t u
Time derivative of Burgers equation c u, q
2uD2
Blow-up in finite time c 1, q 2D2
N O
Initial perturbation
9Numerical Examples (2)
INDIRECT LINEAR DEGENERACY
NO
YES
vs
Standard wave equation c 1, q 0
Gradient catastrophe (1) c u, q 0
Y E S
S O U R C E
Gradient catastrophe (2) c u, q 2uD2-K2/u
Wavelike equation c 1, q 2D2-K2
Time derivative of Burgers equation c u, q
2uD2
Blow-up in finite time c 1, q 2D2
N O
Wavelike, shock-free behavior!
10Numerical Examples (3)
INDIRECT LINEAR DEGENERACY
NO
YES
vs
Standard wave equation c 1, q 0
Gradient catastrophe (1) c u, q 0
Y E S
S O U R C E
Gradient catastrophe (2) c u, q 2uD2-K2/u
Wavelike equation c 1, q 2D2-K2
Time derivative of Burgers equation c u, q
2uD2
Blow-up in finite time c 1, q 2D2
N O
Blow-up in u, code crashes at t 7!
11Numerical Examples (4)
INDIRECT LINEAR DEGENERACY
NO
YES
vs
Standard wave equation c 1, q 0
Gradient catastrophe (1) c u, q 0
Y E S
S O U R C E
Gradient catastrophe (2) c u, q 2uD2-K2/u
Wavelike equation c 1, q 2D2-K2
Time derivative of Burgers equation c u, q
2uD2
Blow-up in finite time c 1, q 2D2
N O
Spikes form in D and K, run with highest
resolution crashes at t 12!
12Numerical Examples (5)
INDIRECT LINEAR DEGENERACY
NO
YES
vs
Standard wave equation c 1, q 0
Gradient catastrophe (1) c u, q 0
Y E S
S O U R C E
Gradient catastrophe (2) c u, q 2uD2-K2/u
Wavelike equation c 1, q 2D2-K2
Time derivative of Burgers equation c u, q
2uD2
Blow-up in finite time c 1, q 2D2
N O
Spikes form in D and K, code crashes at t 13!
13Numerical Examples (6)
INDIRECT LINEAR DEGENERACY
NO
YES
vs
Standard wave equation c 1, q 0
Gradient catastrophe (1) c u, q 0
Y E S
S O U R C E
Gradient catastrophe (2) c u, q 2uD2-K2/u
Wavelike equation c 1, q 2D2-K2
Time derivative of Burgers equation c u, q
2uD2
Blow-up in finite time c 1, q 2D2
N O
Gradient in u and hence D and K become infinite,
crash at t 7!
14Einsteins Equations us and vs
- For Einsteins equations in spherical symmetry,
we adopt the Bona-Maso evolution equation for the
lapse, - where f(?) is an arbitrary function. We consider
no shift, and work with the metric components - These variables are combined in the vector
- As vs we introduce spatial and time derivatives
of these quantities - However, rather than working with DA and KA, we
introduce for convenience D DA - 2DB and K KA
2KB - These derivatives build the vector
15Constraint and Evolution Equations
- In terms of these variables we obtain the
Hamiltonian constraint - and the momentum constraint
- The PDEs for the evolution of the us and vs
are given by - and, when adding constraints to the
ADM-equations,
hD? ?/A1/2 mD? ?
hD ?/A1/2 (-8mD) ? hDB ?/A1/2
(2mDB) ? (2hK) ?/A
mK ?/A1/2 (1/2hKB) ?/A mKB ?/A1/2
sources quadratic in D and K
16Eigenvalues and Eigenfields
- Demanding integrability for the us, , the
6 adjustments in the evolution equations of the
Ds have to vanish hDs mDs 0 - In addition we set the adjustments in the
evolution equation for KB to
and - in order to obtain the following eigenvalues and
eigenfields - At this point, only the adjustments hK and mK
and the free parameter ? (scaling the eigenspeed
of ) remain undetermined. - In the following we fix these 3 free parameters
demanding a strongly hyperbolic principal part
(restricting ) and indirect
linear degeneracy and/or the source criteria.
This also follows from the source criteria!
17Gauge Shocks
- For the eigenpair traveling along the time
lines, no shocks are found as - For the gauge eigenpair indirect
linear degeneracy yields - and the source criteria gives
- Hence, both conditions agree in pointing out the
formation of gauge shocks unless f(?) satisfies - Harmonic slicing, , is a member of
this shock-avoiding family. Furthermore, M.
Alcubierre pointed out that 1log-slicing,
is an approximate solution often
more useful in NR.
Identical gauge shocks occur in toy 11
relativity!
18Constraint Shocks
- For the constraint eigenpair,
indirect linear degeneracy gives - This result hence depends on the gauge choice,
but the adjustments .
together with hK -2 and mK 0 seem to be
preferred. - The source criteria yields
- Here we find
and arbitrary ?. - Alternatively, demanding for particular cases
the coefficient of the mixed term to
vanish, the adjustments found for indirect linear
degeneracy, hK -2 and mK 0, are favored. - Hence, both conditions agree in detecting gauge
shocks unless f(?) satisfies
Constraint shocks do not exist in 11!
19Robust Stability Test ?-family (1)
exponential growth for ADM and for ? ? 0 or
? ? 1
essentially no error growth for 0 ltlt ? ltlt
1 and ? gtgt 1
crossing times
20Robust Stability Test ?-family (2)
eigenspeeds for wf? and w0 become similar
as ?f? ? 0
ADM (no adjustments, not a member of the
?-family)
eigenspeeds for wf? and wc? become similar
as ?f? ? 1
initial average error
?
21Robust Stability Test hK (1)
exponential growth for hK ? -2 (e.g. for ADM)
essentially no error growth for hK ? -2
crossing times
22Robust Stability Test hK (2)
ADM (no adjustments, i.e. hK 0)
adjustment suggested by indirect linear
degeneracy or by eliminating mixed terms in the
sources
initial average error
hK
23hK-hKB Parameter Search
Initial data Minkowski Gaussian perturbation
in KB Evolution Harmonic slicing (f?1) 2D
parameter search Output for the 2-norm of the
Hamiltonian and momentum constraints is shown at
times t 10, 20 and 30 as a
function of the adjustments hK and hKB (black
crash
dark-gray exp. growth.. light-gray linear
growth white no growth)
exponentially growing error for ADM
shock-avoiding valley for the ?-family
(source criteria)
shock-avoiding point at hK -2 (indirect
linear degeneracy)
24Conclusions and Outlook
- Mechanisms for shock formation have been
identified (gradient catastrophe blow-up in
finite time), and shock-avoiding conditions
(indirect linear degeneracy source criteria)
have been derived and tested numerically for
equations of wave-type. - For Einsteins equations in 11 dimensions gauge
shocks, and in spherical symmetry in addition
constraint shocks have been found. Here gauge
shocks can be avoided by choosing for the
Bona-Maso lapse the free function as f 1
const/alpha2, and the constraint shocks by adding
the constraints in a suitable way to the
evolution equations. By doing so,
shock-avoiding evolution systems with only linear
growth in the constraints are obtained. - Our recommendation for evolution systems in NR
hence is - strong hyperbolic principal part
shock-avoiding source terms - In future work we want to analyze in spherical
symmetry different systems (ADM, BSSN, KST)
and/or generalize this analysis to 31.