Given by Geoffrey C. Fox at CPSP713 Case studies in Computational Science on Spring Semester 1996. Foils prepared 15 March 1996
Outside Index
Summary of Material
This describes the structure of Numerical Relativity as a set of differential equations but it does discuss state of the art solvers involving adaptive meshes |
Basic Motivation of General Relativity and its experimental tests |
Metric Tensor, its derivatives and Einstein's equations |
Initial value formulation and structure of elliptic and hyperbolic equations |
Examination of particular finite difference scheme for the Wave equation in three dimensions -- a study to understand large distances issues in solving numerical relativity |
Outside Index Summary of Material
Geoffrey Fox |
NPAC |
111 College Place |
Syracuse NY 13244-4100 |
This describes the structure of Numerical Relativity as a set of differential equations but it does discuss state of the art solvers involving adaptive meshes |
Basic Motivation of General Relativity and its experimental tests |
Metric Tensor, its derivatives and Einstein's equations |
Initial value formulation and structure of elliptic and hyperbolic equations |
Examination of particular finite difference scheme for the Wave equation in three dimensions -- a study to understand large distances issues in solving numerical relativity |
Introductory References:
|
More Technical Background for Discussion In Course:
|
We are familiar with Newton's Laws for the Interactions between Particles: |
"What" Causes this force?
|
Note that gravitational forces cause particles to move in vacuum but we are also familiar with particles moving "downwards" if placed on a nonflat surface |
Einstein's theory of General Relativity is a brilliant generalization of this to describe all gravity in terms of structure of Space-Time |
We do not say say the Earth creates a "Gravitational Force" which causes Apples to fall as noted by Newton |
Rather The Earth distorts space-Time and Apples move in this distorted Space-Time |
In Newton's law, gravitational force of Earth proportional to its Mass |
In Einstein's description, distortion of Space-Time is proportional to Mass of particle |
Everything (including you) distorts Space-Time;
|
We can consider the distortion of Space-Time due to a Heavy Body as analogous to distortion of a FLAT rubber sheet into a DROOPING rubber sheet when a bowling ball is placed in the Middle |
Now place a much lighter body (a marble) near the bowling ball
|
Greek indices m,n .. denote Space-Time and run from 0 (time) to 3. |
Latin indices i,j,k .. denote Space and run from 1 to 3 (x y and z). |
Note indices ALWAYS balance in equations and there is difference between upper and lower indices
|
Again M ij M jk = d i k is Expression of Inverse Condition |
The speed of light c is set to 1.
|
All quantities in General Relativity are a function of the four vector: |
The dynamical variables of the theory are the 10 independent components of the the metric tensor: |
For example, this describes distortion of rubber sheet by bowling ball |
g mn = g nm is SYMMETRIC |
If two Space-Time points are separated by four vector d m |
Square of distance between them is g mn d m d n |
Each of 16 (10 independent) components of 4 by 4 tensor g mn is a function |
of the four vector |
Einstein's equations are nonlinear equations for the metric g and its derivatives usually constructed in terms of so called Christoffel Symbol G |
Newtonian gravity is linear (weak field) limit of General Relativity |
Tests of General Relativity --
|
New experiments will directly observe gravitational waves due to interactions with sensitive detectors on earth
|
Can estimate that Computations of Gravitational Waves from black hole collision would take 100,000 hours on a Cray-YMP
|
Estimate that LIGO will sensitivity to detect black hole gravitational waves around the year 2005 |
Science 256,281-412, 17 April 92 |
Figure 1 shows two components of a gravitational wave versus time
|
Science 256,281-412, 17 April 92 |
Figure 2 shows two components of a gravitational wave versus time
|
Science 256,281-412, 17 April 92 |
Science 256,281-412, 17 April 92
|
(Upper solid curve) and in a more advanced interferometer (lower solid curve) |
The dashed curves show various contributions to the first interferometer's noise. |
Science 256,281-412, 17 April 92 |
Comparison between rms noise h/rms in LIGO's first and advanced detector systems and the characteristic amplitude h/c of gravitational wave bursts from several hypothesized sources
|
Science 256,281-412, 17 April 92 |
Einstein first published equations in 1915 with three predictions: |
Bending of Light as it passed massive objects such as stars
|
There is a small (43 seconds of arc per century -- with 3600 seconds of arc as one degree) shift in Mercury's perihelion not accounted for by Newtonian Gravity
|
"Blue" shift (frequency or energy change) of light as it falls down a gravitational field
|
Observed time delay of radar signals bounced off planets or spacecraft during superior conjunction
|
Observed decrease in period of binary pulsar discovered by Hulse and Taylor(1975)
|
A laser monitors distance between Moon and Earth showing that they both have same acceleration due to sun with a precision of 7 parts per ten trillion(1010)
|
Says that a uniform gravitational field is the same as constant acceleration
|
Alternatively, Equivalence Principle says that the gravitational mass (the mass m1 that appears in Newton's law of gravity)
|
is equal to inertial mass m1 appearing in Newton's law of motion
|
In general relativity, the Equivalence Principle is built in |
We formulate General Relativity in fashion one would expect for solving wave equations in three spatial dimensions
|
The Einstein equations for g mn can now be written in terms of a reduced 3 by 3 symmetric metric matrix q ij |
q ij has 9 components of which 6 are independent |
There are 10 dynamical equations for 6 independent components of q ij
|
When Solving a second order time evolution equation such as |
One typically converts this to a pair of first order in time equations by transformation: |
We perform the analogous step in general relativity, introducing a new symmetric matrix K ij which is called the extrinsic curvature and related to but not exactly equal to single time derivative of q ij |
K is a Symmetric 3 by 3 matrix |
And we now have 14 equations: |
One of the fascinating features of Numerical Relativity is the freedom to choose coordinates
|
The Lapse describes how much time elapses between two spacial surfaces |
The Shift functions N i describe the tangential shift as you move from the t 0 to t 0+ dt spacial surface |
Choose Lapse and Shift to:
|
a and N i are depicted below for two adjacent spacial surfaces t and t + dt |
Define Inverse q ij of 3 by 3 matrix q ij: |
Define the spacial restriction of Christoffel Symbol by:
|
Get full Christoffel Symbol by replacing q ij by g mn and makes all indices Greek running from 0 to 3. |
Note these are very nonlinear and complicated
|
6 First Order in Time Evolution Equations for q ij |
6 First Order in Time Evolution Equations for K ij |
The numerical relativity problem has several key features |
Our first example is based on work of University of Pittsburgh group involving R. Gomez and J.Winicour
|
This studies the wave-like features of Einstein's equations where key numerical problem is reliable extraction of Gravitational waves |
This is nontrivial as gravitational waves can only be extracted after one has evolved a "long way" from the black holes |
It is difficult numerically to reliable evolve oscillating solutions through long distances
|
Key features of Pittsburgh Approach
|
The 3D Wave |
Equation is: |
We will study this as "essential issues" in wave extraction are exhibited
|
Transform to Spherical |
Polar Coordinates |
Introduce characteristic
|
Introduce new computational variables (x,h,z)
|
These three variables are used as well as u (=t-r) replacing time as fourth independent variable |
Replace F by G as dependent variable: |
We must solve: |
In Compact Domain: |
With Boundary Conditions given in diagram: |
Note we assume that you have used other techniques (represntations) to integrate upto: |
This illustrates formalism on previous page |
We take simple equally paced gridpoints in u, x, h, and z. |
We have already seen for convection equation subtly in time discretization of this type of equation. |
A stable explicit method can be obtained by a carefully chosen integral formulation described in Miller's memo. |
This involves integrating over a box PQRS with sides parallel to the characteristic axes u=constant, v=constant |
We find these pictures for box PQRS in (t,r) (u,v) (u,x) spaces |
Alternatively we can view as a brilliant choice of finite difference scheme |
Which is justified a posteriori by the Von Neumann stability analysis given in Pittsburgh memo which shows all eigenvalues have unit modulus as required for
|
This clever choice is: |
Giving: |
Further we similarily: |
Which gives final numerical equations: |