Introduction Discretisation Constrained evolution schemes Summary and Outlook
Symplectic time integrators for numerical General
Relativity
Ronny Richter, Christian Lubich
Semiannual Meeting of the SFB/TR7Garching
September 25 2007
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Motivation
Symplectic integrators preserve energy and improve stabilitye.g. in simulations of the outer solar system.
Can these integration methods be used to improve simulationsin General Relativity?
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Motivation
Symplectic integrators preserve energy and improve stabilitye.g. in simulations of the outer solar system.
Can these integration methods be used to improve simulationsin General Relativity?
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Contents
1 IntroductionCanonical Formulation of ODENumerical Integration Methods
2 DiscretisationDerivation of the discrete ADM HamiltonianFree evolution
3 Constrained evolution schemesThe momentum constraints as hidden constraintsNumerical tests
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Contents
1 IntroductionCanonical Formulation of ODENumerical Integration Methods
2 DiscretisationDerivation of the discrete ADM HamiltonianFree evolution
3 Constrained evolution schemesThe momentum constraints as hidden constraintsNumerical tests
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Canonical Formulation of ODE
Hamiltonian Systems
If a system of ordinary di�erential equations can be written as
pi = −∂qiH(p, q), qi = ∂pi
H(p, q) (1)
then it is a Hamiltonian system.
The function H(p, q) is called Hamiltonian.
If (p(t), q(t)) is a solution of (1) then
H(p(t0), q(t0)) = H(p(t1), q(t1)) ∀t0, t1. (2)
The Hamiltonian is a �rst integral.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Canonical Formulation of ODE
Hamiltonian Systems
If a system of ordinary di�erential equations can be written as
pi = −∂qiH(p, q), qi = ∂pi
H(p, q) (1)
then it is a Hamiltonian system.
The function H(p, q) is called Hamiltonian.
If (p(t), q(t)) is a solution of (1) then
H(p(t0), q(t0)) = H(p(t1), q(t1)) ∀t0, t1. (2)
The Hamiltonian is a �rst integral.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Canonical Formulation of ODE
Hamiltonian Systems
If a system of ordinary di�erential equations can be written as
pi = −∂qiH(p, q), qi = ∂pi
H(p, q) (1)
then it is a Hamiltonian system.
The function H(p, q) is called Hamiltonian.
If (p(t), q(t)) is a solution of (1) then
H(p(t0), q(t0)) = H(p(t1), q(t1)) ∀t0, t1. (2)
The Hamiltonian is a �rst integral.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Canonical Formulation of ODE
Canonical and symplectic transformations
Canonical transformations preserve the Hamiltonian character:Let p′ = ψp(p, q) and q′ = ψq(p, q).The map ψ is called canonical transformation if
p′i = −∂q′
iH ′(p′, q′), q′
i = ∂p′iH ′(p′, q′),
with H ′(ψp(p, q), ψq(p, q)) = H(p, q).
Symplectic maps preserve the so-called symplectic 2-form (anotion of area in phase space).
Theorem (Jacobi, 1836):A map ψ is symplectic, if and only if it is canonical.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Canonical Formulation of ODE
Canonical and symplectic transformations
Canonical transformations preserve the Hamiltonian character:Let p′ = ψp(p, q) and q′ = ψq(p, q).The map ψ is called canonical transformation if
p′i = −∂q′
iH ′(p′, q′), q′
i = ∂p′iH ′(p′, q′),
with H ′(ψp(p, q), ψq(p, q)) = H(p, q).
Symplectic maps preserve the so-called symplectic 2-form (anotion of area in phase space).
Theorem (Jacobi, 1836):A map ψ is symplectic, if and only if it is canonical.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Canonical Formulation of ODE
Canonical and symplectic transformations
Canonical transformations preserve the Hamiltonian character:Let p′ = ψp(p, q) and q′ = ψq(p, q).The map ψ is called canonical transformation if
p′i = −∂q′
iH ′(p′, q′), q′
i = ∂p′iH ′(p′, q′),
with H ′(ψp(p, q), ψq(p, q)) = H(p, q).
Symplectic maps preserve the so-called symplectic 2-form (anotion of area in phase space).
Theorem (Jacobi, 1836):A map ψ is symplectic, if and only if it is canonical.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical Integration Methods
Symplectic Integrators
De�nition:A numerical one-step method is called symplectic if theone-step map yn+1 = Φ∆t(yn) is symplectic whenever themethod is applied to a smooth Hamiltonian system.
The value of the Hamiltonian (i.e. the energy) is preserved.
The symplectic 2-form is preserved.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical Integration Methods
Symplectic Integrators
De�nition:A numerical one-step method is called symplectic if theone-step map yn+1 = Φ∆t(yn) is symplectic whenever themethod is applied to a smooth Hamiltonian system.
The value of the Hamiltonian (i.e. the energy) is preserved.
The symplectic 2-form is preserved.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical Integration Methods
Examples for symplectic integrators
Often symplectic integrators are implicit methods (dependingon the structure of the Hamiltonian and the method itself).
Integrators for free Hamiltonian systems
Symplectic Euler method
Symplectic Runge-Kutta methods
Störmer-Verlet method
...
Integrators for Hamiltonian systems with holonomicconstraints
Symplectic Euler method with constraints
Partitioned Lobatto IIIA - IIIB pair
Rattle method
...
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical Integration Methods
Examples for symplectic integrators
Often symplectic integrators are implicit methods (dependingon the structure of the Hamiltonian and the method itself).
Integrators for free Hamiltonian systems
Symplectic Euler method
Symplectic Runge-Kutta methods
Störmer-Verlet method
...
Integrators for Hamiltonian systems with holonomicconstraints
Symplectic Euler method with constraints
Partitioned Lobatto IIIA - IIIB pair
Rattle method
...
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical Integration Methods
Examples for symplectic integrators
Often symplectic integrators are implicit methods (dependingon the structure of the Hamiltonian and the method itself).
Integrators for free Hamiltonian systems
Symplectic Euler method
Symplectic Runge-Kutta methods
Störmer-Verlet method
...
Integrators for Hamiltonian systems with holonomicconstraints
Symplectic Euler method with constraints
Partitioned Lobatto IIIA - IIIB pair
Rattle method
...
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Contents
1 IntroductionCanonical Formulation of ODENumerical Integration Methods
2 DiscretisationDerivation of the discrete ADM HamiltonianFree evolution
3 Constrained evolution schemesThe momentum constraints as hidden constraintsNumerical tests
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Canonical Formulation of GR
A Hamiltonian that describes General Relativity has beendeveloped since the late �fties.
The ADM Hamiltonian is1
HADM =
∫d3x
(α
(πijπij −
1
2πi iπ
jj
)− αhR
+ 2πijhikDjβk
), (3)
where
πij =√h
(Khij − K ij
). (4)
1We replaced the lapse function N by the slicing density α := N/√h.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Canonical Formulation of GR
A Hamiltonian that describes General Relativity has beendeveloped since the late �fties.
The ADM Hamiltonian is1
HADM =
∫d3x
(α
(πijπij −
1
2πi iπ
jj
)− αhR
+ 2πijhikDjβk
), (3)
where
πij =√h
(Khij − K ij
). (4)
1We replaced the lapse function N by the slicing density α := N/√h.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
First simple examples
For �rst tests we consider only those solutions where the 3-metrichij ful�ls the following properties
hij = 0 for i 6= j
hij(x , y , z) ≡ hij(x)
hyy = hzz
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
First simple examples
For �rst tests we consider only those solutions where the 3-metrichij ful�ls the following properties
hij = 0 for i 6= j
hij(x , y , z) ≡ hij(x)
hyy = hzz
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
First simple examples
For �rst tests we consider only those solutions where the 3-metrichij ful�ls the following properties
hij = 0 for i 6= j
hij(x , y , z) ≡ hij(x)
hyy = hzz
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
A simpli�ed Hamiltonian
These systems are described by a simpli�ed ADM Hamiltonian
H =
∫dx
[α
(1
2πxxπxxhxxhxx − πxxπyyhxxhyy
)− α
(1
2∂xhyy∂xhyy − 2hyy∂
2
xhyy + ∂xhyy∂x log(hxx)
)+ 2πxxhxx∂xβ
x + πxxβx∂xhxx + πyyβx∂xhyy
]. (5)
The 3-metric h provides the position variables (q in (1)).
The corresponding canonical momentum is π (p in (1)).
For the moment we take α and βx to be given in advance.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
A simpli�ed Hamiltonian
These systems are described by a simpli�ed ADM Hamiltonian
H =
∫dx
[α
(1
2πxxπxxhxxhxx − πxxπyyhxxhyy
)− α
(1
2∂xhyy∂xhyy − 2hyy∂
2
xhyy + ∂xhyy∂x log(hxx)
)+ 2πxxhxx∂xβ
x + πxxβx∂xhxx + πyyβx∂xhyy
]. (5)
The 3-metric h provides the position variables (q in (1)).
The corresponding canonical momentum is π (p in (1)).
For the moment we take α and βx to be given in advance.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
A simpli�ed Hamiltonian
These systems are described by a simpli�ed ADM Hamiltonian
H =
∫dx
[α
(1
2πxxπxxhxxhxx − πxxπyyhxxhyy
)− α
(1
2∂xhyy∂xhyy − 2hyy∂
2
xhyy + ∂xhyy∂x log(hxx)
)+ 2πxxhxx∂xβ
x + πxxβx∂xhxx + πyyβx∂xhyy
]. (5)
The 3-metric h provides the position variables (q in (1)).
The corresponding canonical momentum is π (p in (1)).
For the moment we take α and βx to be given in advance.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
A simpli�ed Hamiltonian
These systems are described by a simpli�ed ADM Hamiltonian
H =
∫dx
[α
(1
2πxxπxxhxxhxx − πxxπyyhxxhyy
)− α
(1
2∂xhyy∂xhyy − 2hyy∂
2
xhyy + ∂xhyy∂x log(hxx)
)+ 2πxxhxx∂xβ
x + πxxβx∂xhxx + πyyβx∂xhyy
]. (5)
The 3-metric h provides the position variables (q in (1)).
The corresponding canonical momentum is π (p in (1)).
For the moment we take α and βx to be given in advance.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Method of lines
Derive a discrete Hamiltonian:
Approximation by piecewise linear functions
Centered 2nd order �nite di�erences for the spacial derivatives
Integration of the product of piecewise linear functions
The discrete Hamiltonian only depends on the functionalvalues at the grid points.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Method of lines
Derive a discrete Hamiltonian:
Approximation by piecewise linear functions
Centered 2nd order �nite di�erences for the spacial derivatives
Integration of the product of piecewise linear functions
The discrete Hamiltonian only depends on the functionalvalues at the grid points.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Method of lines
Derive a discrete Hamiltonian:
Approximation by piecewise linear functions
Centered 2nd order �nite di�erences for the spacial derivatives
Integration of the product of piecewise linear functions
The discrete Hamiltonian only depends on the functionalvalues at the grid points.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Method of lines
Derive a discrete Hamiltonian:
Approximation by piecewise linear functions
Centered 2nd order �nite di�erences for the spacial derivatives
Integration of the product of piecewise linear functions
The discrete Hamiltonian only depends on the functionalvalues at the grid points.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Derivation of the discrete ADM Hamiltonian
Method of lines
Derive a discrete Hamiltonian:
Approximation by piecewise linear functions
Centered 2nd order �nite di�erences for the spacial derivatives
Integration of the product of piecewise linear functions
The discrete Hamiltonian only depends on the functionalvalues at the grid points.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Application of free evolution schemes
From the discrete Hamiltonian derive the ordinary di�erentialequations for the grid variables.
The discrete Hamiltonian system is then a system of �rst orderODE in 4N variables (N: number of grid points).
Given initial data it is straight forward to apply well knownintegration methods (symplectic as well as non-symplectic).
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Application of free evolution schemes
From the discrete Hamiltonian derive the ordinary di�erentialequations for the grid variables.
The discrete Hamiltonian system is then a system of �rst orderODE in 4N variables (N: number of grid points).
Given initial data it is straight forward to apply well knownintegration methods (symplectic as well as non-symplectic).
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Application of free evolution schemes
From the discrete Hamiltonian derive the ordinary di�erentialequations for the grid variables.
The discrete Hamiltonian system is then a system of �rst orderODE in 4N variables (N: number of grid points).
Given initial data it is straight forward to apply well knownintegration methods (symplectic as well as non-symplectic).
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Gauge wave testbed
The method was tested by simulating the gauge wave testbed.
The analytical solution is
g = (1 + H)(dt2 − dx2) + dy2 + dz2, (6)
where H = A sin(2π(x − t)).
After the 3 + 1-decomposition w.r.t. t = const slicing oneobtains
hxx = 1 + H, hyy = 1 = hzz , hij = 0 for i 6= j . (7)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Gauge wave testbed
The method was tested by simulating the gauge wave testbed.
The analytical solution is
g = (1 + H)(dt2 − dx2) + dy2 + dz2, (6)
where H = A sin(2π(x − t)).
After the 3 + 1-decomposition w.r.t. t = const slicing oneobtains
hxx = 1 + H, hyy = 1 = hzz , hij = 0 for i 6= j . (7)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Gauge wave testbed
The method was tested by simulating the gauge wave testbed.
The analytical solution is
g = (1 + H)(dt2 − dx2) + dy2 + dz2, (6)
where H = A sin(2π(x − t)).
After the 3 + 1-decomposition w.r.t. t = const slicing oneobtains
hxx = 1 + H, hyy = 1 = hzz , hij = 0 for i 6= j . (7)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Results
Results of simulations with A = 0.01 and A = 0.1 for 50 and 200grid points:
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Stability of the Hamiltonian system
In the high amplitude simulations (A = 0.1) the highfrequency modes grow and the code crashes.
It turns out that the linearisation of the system correspondingto (5) is not well-posed (the norm of the solution at a time t
cannot be bounded by the norm of the initial data).
If the continuous problem is not well-posed one cannot expectthat the discrete evolution is stable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Stability of the Hamiltonian system
In the high amplitude simulations (A = 0.1) the highfrequency modes grow and the code crashes.
It turns out that the linearisation of the system correspondingto (5) is not well-posed (the norm of the solution at a time t
cannot be bounded by the norm of the initial data).
If the continuous problem is not well-posed one cannot expectthat the discrete evolution is stable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Free evolution
Stability of the Hamiltonian system
In the high amplitude simulations (A = 0.1) the highfrequency modes grow and the code crashes.
It turns out that the linearisation of the system correspondingto (5) is not well-posed (the norm of the solution at a time t
cannot be bounded by the norm of the initial data).
If the continuous problem is not well-posed one cannot expectthat the discrete evolution is stable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Contents
1 IntroductionCanonical Formulation of ODENumerical Integration Methods
2 DiscretisationDerivation of the discrete ADM HamiltonianFree evolution
3 Constrained evolution schemesThe momentum constraints as hidden constraintsNumerical tests
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Type of constraints
There are symplectic constrained evolutionschemes for holonomic constraints, i.e. those thatdepend on the position variables only(e.g. the Rattle method).
To impose these constraints additionalforces of constraints
and corresponding Lagrange multipliers are added:
p = −∂qH(p, q)− Fg (q, λ), q = ∂qH(p, q), g(q) = 0.
The forces of constraints are Fg = λ∂qg(q).
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Type of constraints
����������
��������
����
pfree
There are symplectic constrained evolutionschemes for holonomic constraints, i.e. those thatdepend on the position variables only(e.g. the Rattle method).
To impose these constraints additionalforces of constraints
and corresponding Lagrange multipliers are added:
p = −∂qH(p, q)− Fg (q, λ), q = ∂qH(p, q), g(q) = 0.
The forces of constraints are Fg = λ∂qg(q).
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Type of constraints
����������
��������
����
���
���
pfree
Fg
There are symplectic constrained evolutionschemes for holonomic constraints, i.e. those thatdepend on the position variables only(e.g. the Rattle method).
To impose these constraints additionalforces of constraints
and corresponding Lagrange multipliers are added:
p = −∂qH(p, q)− Fg (q, λ), q = ∂qH(p, q), g(q) = 0.
The forces of constraints are Fg = λ∂qg(q).
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Modi�cation of the Poisson structure
Formally one may impose the momentum constraints in thatway.
Interpret the shift β as the canonical momentum of a positionvariable ρβ that does not appear in the Hamiltonian.
The Hamiltonian equations of motion for β and ρβ are then
β = −∂ρβH = 0, ρβ = ∂βH = −2C, (8)
where C is the momentum constraint.
If the discrete holonomic constraint ρβ = 0 is imposed in theconstrained evolution scheme then automatically a discretemomentum constraint C = 0 is satis�ed.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Modi�cation of the Poisson structure
Formally one may impose the momentum constraints in thatway.
Interpret the shift β as the canonical momentum of a positionvariable ρβ that does not appear in the Hamiltonian.
The Hamiltonian equations of motion for β and ρβ are then
β = −∂ρβH = 0, ρβ = ∂βH = −2C, (8)
where C is the momentum constraint.
If the discrete holonomic constraint ρβ = 0 is imposed in theconstrained evolution scheme then automatically a discretemomentum constraint C = 0 is satis�ed.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Modi�cation of the Poisson structure
Formally one may impose the momentum constraints in thatway.
Interpret the shift β as the canonical momentum of a positionvariable ρβ that does not appear in the Hamiltonian.
The Hamiltonian equations of motion for β and ρβ are then
β = −∂ρβH = 0, ρβ = ∂βH = −2C, (8)
where C is the momentum constraint.
If the discrete holonomic constraint ρβ = 0 is imposed in theconstrained evolution scheme then automatically a discretemomentum constraint C = 0 is satis�ed.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Modi�cation of the Poisson structure
Formally one may impose the momentum constraints in thatway.
Interpret the shift β as the canonical momentum of a positionvariable ρβ that does not appear in the Hamiltonian.
The Hamiltonian equations of motion for β and ρβ are then
β = −∂ρβH = 0, ρβ = ∂βH = −2C, (8)
where C is the momentum constraint.
If the discrete holonomic constraint ρβ = 0 is imposed in theconstrained evolution scheme then automatically a discretemomentum constraint C = 0 is satis�ed.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Invertibility of the forces of constraints
Imposing the constraint ρβ = 0 leads to a force of constraintFβ that in�uences only β:
β = −∂ρβH − Fβ. (9)
There is no β in the equation ρβ = ∂βH = −2C.It is thus not possible to adjust this force of constraint (i.e.the corresponding Lagrange multiplier) such that ρβ = 0 isalways satis�ed.
The way out:
Impose another holonomic constraints for which thecorresponding force has in�uence on πij .
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Invertibility of the forces of constraints
Imposing the constraint ρβ = 0 leads to a force of constraintFβ that in�uences only β:
β = −∂ρβH − Fβ. (9)
There is no β in the equation ρβ = ∂βH = −2C.It is thus not possible to adjust this force of constraint (i.e.the corresponding Lagrange multiplier) such that ρβ = 0 isalways satis�ed.
The way out:
Impose another holonomic constraints for which thecorresponding force has in�uence on πij .
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Invertibility of the forces of constraints
Imposing the constraint ρβ = 0 leads to a force of constraintFβ that in�uences only β:
β = −∂ρβH − Fβ. (9)
There is no β in the equation ρβ = ∂βH = −2C.It is thus not possible to adjust this force of constraint (i.e.the corresponding Lagrange multiplier) such that ρβ = 0 isalways satis�ed.
The way out:
Impose another holonomic constraints for which thecorresponding force has in�uence on πij .
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Invertibility of the forces of constraints
Imposing the constraint ρβ = 0 leads to a force of constraintFβ that in�uences only β:
β = −∂ρβH − Fβ. (9)
There is no β in the equation ρβ = ∂βH = −2C.It is thus not possible to adjust this force of constraint (i.e.the corresponding Lagrange multiplier) such that ρβ = 0 isalways satis�ed.
The way out:
Impose another holonomic constraints for which thecorresponding force has in�uence on πij .
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Choosing a gauge
This constraint must depend on the metric hij .
The freedom of choosing a spacial coordinate system was notused yet.
Choose the gauge such that some function g of the metric andits spacial derivatives vanishes.
One obtains
β = −∂ρβH − Fβ = −Fβ, ρβ = ∂βH = −2C,
πij = −∂hijH − F
ijg , hij = ∂πijH,
ρβ = 0, g(hij) = 0. (10)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Choosing a gauge
This constraint must depend on the metric hij .
The freedom of choosing a spacial coordinate system was notused yet.
Choose the gauge such that some function g of the metric andits spacial derivatives vanishes.
One obtains
β = −∂ρβH − Fβ = −Fβ, ρβ = ∂βH = −2C,
πij = −∂hijH − F
ijg , hij = ∂πijH,
ρβ = 0, g(hij) = 0. (10)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Choosing a gauge
This constraint must depend on the metric hij .
The freedom of choosing a spacial coordinate system was notused yet.
Choose the gauge such that some function g of the metric andits spacial derivatives vanishes.
One obtains
β = −∂ρβH − Fβ = −Fβ, ρβ = ∂βH = −2C,
πij = −∂hijH − F
ijg , hij = ∂πijH,
ρβ = 0, g(hij) = 0. (10)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
The momentum constraints as hidden constraints
Choosing a gauge
This constraint must depend on the metric hij .
The freedom of choosing a spacial coordinate system was notused yet.
Choose the gauge such that some function g of the metric andits spacial derivatives vanishes.
One obtains
β = −∂ρβH − Fβ = −Fβ, ρβ = ∂βH = −2C,
πij = −∂hijH − F
ijg , hij = ∂πijH,
ρβ = 0, g(hij) = 0. (10)
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Gauge choice for the gauge wave
For the gauge wave testbed the function hxx should havesinusoidal character.
Calculate the Fourier transform of hxx and require that thehigher frequency modes vanish → g(hij).
Use the shift to satisfy the gauge constraint and the force ofconstraint Fijg to satisfy ρβ = 0.
Possible problem:
For yet unknown reasons it is not possible to require thehighest frequency mode of hxx to vanish → we ignore thisconstraint, too.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Gauge choice for the gauge wave
For the gauge wave testbed the function hxx should havesinusoidal character.
Calculate the Fourier transform of hxx and require that thehigher frequency modes vanish → g(hij).
Use the shift to satisfy the gauge constraint and the force ofconstraint Fijg to satisfy ρβ = 0.
Possible problem:
For yet unknown reasons it is not possible to require thehighest frequency mode of hxx to vanish → we ignore thisconstraint, too.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Gauge choice for the gauge wave
For the gauge wave testbed the function hxx should havesinusoidal character.
Calculate the Fourier transform of hxx and require that thehigher frequency modes vanish → g(hij).
Use the shift to satisfy the gauge constraint and the force ofconstraint Fijg to satisfy ρβ = 0.
Possible problem:
For yet unknown reasons it is not possible to require thehighest frequency mode of hxx to vanish → we ignore thisconstraint, too.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Gauge choice for the gauge wave
For the gauge wave testbed the function hxx should havesinusoidal character.
Calculate the Fourier transform of hxx and require that thehigher frequency modes vanish → g(hij).
Use the shift to satisfy the gauge constraint and the force ofconstraint Fijg to satisfy ρβ = 0.
Possible problem:
For yet unknown reasons it is not possible to require thehighest frequency mode of hxx to vanish → we ignore thisconstraint, too.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Possible problems
The number of forces Fg equals the number of constraintsg(hij) = 0 (i.e. N − 4 here).
The number of constraints ρβ = 0 is N.
It is possible that the freedom in choosing the Lagrangemultipliers in the forces Fijg is not su�cient to satisfy ρβ = 0.
However, the operator that assigns the momentum constraintsto the momenta πij 7→ C clearly has a kernel.
It might be that the missing four forces of constraints onlymove the momentum within this kernel.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Possible problems
The number of forces Fg equals the number of constraintsg(hij) = 0 (i.e. N − 4 here).
The number of constraints ρβ = 0 is N.
It is possible that the freedom in choosing the Lagrangemultipliers in the forces Fijg is not su�cient to satisfy ρβ = 0.
However, the operator that assigns the momentum constraintsto the momenta πij 7→ C clearly has a kernel.
It might be that the missing four forces of constraints onlymove the momentum within this kernel.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Possible problems
The number of forces Fg equals the number of constraintsg(hij) = 0 (i.e. N − 4 here).
The number of constraints ρβ = 0 is N.
It is possible that the freedom in choosing the Lagrangemultipliers in the forces Fijg is not su�cient to satisfy ρβ = 0.
However, the operator that assigns the momentum constraintsto the momenta πij 7→ C clearly has a kernel.
It might be that the missing four forces of constraints onlymove the momentum within this kernel.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Possible problems
The number of forces Fg equals the number of constraintsg(hij) = 0 (i.e. N − 4 here).
The number of constraints ρβ = 0 is N.
It is possible that the freedom in choosing the Lagrangemultipliers in the forces Fijg is not su�cient to satisfy ρβ = 0.
However, the operator that assigns the momentum constraintsto the momenta πij 7→ C clearly has a kernel.
It might be that the missing four forces of constraints onlymove the momentum within this kernel.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Possible problems
The number of forces Fg equals the number of constraintsg(hij) = 0 (i.e. N − 4 here).
The number of constraints ρβ = 0 is N.
It is possible that the freedom in choosing the Lagrangemultipliers in the forces Fijg is not su�cient to satisfy ρβ = 0.
However, the operator that assigns the momentum constraintsto the momenta πij 7→ C clearly has a kernel.
It might be that the missing four forces of constraints onlymove the momentum within this kernel.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Numerical tests
Results of Rattle
Results of the constrained evolution using Rattle for A = 0.1 and50 grid points:
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Summary
For small amplitude in the gauge wave testbed the freeevolution schemes seem to be quite stable.
For higher amplitudes one obtains growing high frequencymodes.
The origin of this instability is presumably the nonwell-posedness of the continuous equations.
Through a modi�cation of the ADM Hamiltonian one mayapply constrained evolution schemes, too.
The scheme investigated here is still unstable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Summary
For small amplitude in the gauge wave testbed the freeevolution schemes seem to be quite stable.
For higher amplitudes one obtains growing high frequencymodes.
The origin of this instability is presumably the nonwell-posedness of the continuous equations.
Through a modi�cation of the ADM Hamiltonian one mayapply constrained evolution schemes, too.
The scheme investigated here is still unstable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Summary
For small amplitude in the gauge wave testbed the freeevolution schemes seem to be quite stable.
For higher amplitudes one obtains growing high frequencymodes.
The origin of this instability is presumably the nonwell-posedness of the continuous equations.
Through a modi�cation of the ADM Hamiltonian one mayapply constrained evolution schemes, too.
The scheme investigated here is still unstable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Summary
For small amplitude in the gauge wave testbed the freeevolution schemes seem to be quite stable.
For higher amplitudes one obtains growing high frequencymodes.
The origin of this instability is presumably the nonwell-posedness of the continuous equations.
Through a modi�cation of the ADM Hamiltonian one mayapply constrained evolution schemes, too.
The scheme investigated here is still unstable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Summary
For small amplitude in the gauge wave testbed the freeevolution schemes seem to be quite stable.
For higher amplitudes one obtains growing high frequencymodes.
The origin of this instability is presumably the nonwell-posedness of the continuous equations.
Through a modi�cation of the ADM Hamiltonian one mayapply constrained evolution schemes, too.
The scheme investigated here is still unstable.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Outlook
Try to �nd Hamiltonian formulations of General Relativity thatare well-posed.
Investigate the constrained evolution schemes further.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity
Introduction Discretisation Constrained evolution schemes Summary and Outlook
Outlook
Try to �nd Hamiltonian formulations of General Relativity thatare well-posed.
Investigate the constrained evolution schemes further.
Ronny Richter University of Tübingen
Symplectic time integrators for numerical General Relativity