Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 6 Various Finite Difference Discretizations for the Advection Equations.

Slides:



Advertisements
Similar presentations
Finite Difference Discretization of Hyperbolic Equations: Linear Problems Lectures 8, 9 and 10.
Advertisements

5.1 Real Vector Spaces.
8.3 Inverse Linear Transformations
Quantum One: Lecture 6. The Initial Value Problem for Free Particles, and the Emergence of Fourier Transforms.
ECE 8443 – Pattern Recognition EE 3512 – Signals: Continuous and Discrete Objectives: Response to a Sinusoidal Input Frequency Analysis of an RC Circuit.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 9 Instructor: Tim Warburton.
Computational Modeling for Engineering MECN 6040
Engineering Mathematics Class #15 Fourier Series, Integrals, and Transforms (Part 3) Sheng-Fang Huang.
CE33500 – Computational Methods in Civil Engineering Differentiation Provided by : Shahab Afshari
Total Recall Math, Part 2 Ordinary diff. equations First order ODE, one boundary/initial condition: Second order ODE.
Atms 4320 Lab 2 Anthony R. Lupo. Lab 2 -Methodologies for evaluating the total derviatives in the fundamental equations of hydrodynamics  Recall that.
Introduction to Numerical Methods I
EE-2027 SaS, L11 1/13 Lecture 11: Discrete Fourier Transform 4 Sampling Discrete-time systems (2 lectures): Sampling theorem, discrete Fourier transform.
19. Series Representation of Stochastic Processes
Subdivision Analysis via JSR We already know the z-transform formulation of schemes: To check if the scheme generates a continuous limit curve ( the scheme.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 11 Instructor: Tim Warburton.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Instructor: Tim Warburton.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 9 Instructor: Tim Warburton.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 2 Instructor: Tim Warburton.
Pseudospectral Methods
Numerical Methods for Partial Differential Equations
1 Chapter 8 The Discrete Fourier Transform 2 Introduction  In Chapters 2 and 3 we discussed the representation of sequences and LTI systems in terms.
Gerschgorin Circle Theorem. Eigenvalues In linear algebra Eigenvalues are defined for a square matrix M. An Eigenvalue for the matrix M is a scalar such.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 8 Instructor: Tim Warburton.
1 Preliminaries Precalculus Review I Precalculus Review II
Motivation Music as a combination of sounds at different frequencies
Introduction to Numerical Methods for ODEs and PDEs Methods of Approximation Lecture 3: finite differences Lecture 4: finite elements.
ME451 Kinematics and Dynamics of Machine Systems Numerical Solution of DAE IVP Newmark Method November 1, 2013 Radu Serban University of Wisconsin-Madison.
Lecture 22 MA471 Fall Advection Equation Recall the 2D advection equation: We will use a Runge-Kutta time integrator and spectral representation.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 5a.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 3 AB2,AB3, Stability, Accuracy Instructor: Tim Warburton.
1 EEE 431 Computational Methods in Electrodynamics Lecture 4 By Dr. Rasime Uyguroglu
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 5 Summary of convergence checks for LMM and one-step time stepping Instructor:
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 8.
6. Introduction to Spectral method. Finite difference method – approximate a function locally using lower order interpolating polynomials. Spectral method.
To Dream the Impossible Scheme Part 1 Approximating Derivatives on Non-Uniform, Skewed and Random Grid Schemes Part 2 Applying Rectangular Finite Difference.
7. Introduction to the numerical integration of PDE. As an example, we consider the following PDE with one variable; Finite difference method is one of.
Engineering Analysis – Computational Fluid Dynamics –
THE LAPLACE TRANSFORM LEARNING GOALS Definition
16.920J/SMA 5212 Numerical Methods for PDEs Lecture 5 Finite Differences: Parabolic Problems B. C. Khoo Thanks to Franklin Tan SMA-HPC ©2002 NUS.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 24.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 23.
Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 12 Instructor: Tim Warburton.
Advanced Engineering Mathematics, 7 th Edition Peter V. O’Neil © 2012 Cengage Learning Engineering. All Rights Reserved. CHAPTER 4 Series Solutions.
Motivation for the Laplace Transform
ECE 576 – Power System Dynamics and Stability Prof. Tom Overbye Dept. of Electrical and Computer Engineering University of Illinois at Urbana-Champaign.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 17.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 13.
Solving Scalar Linear Systems A Little Theory For Jacobi Iteration
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 19.
Topics 1 Specific topics to be covered are: Discrete-time signals Z-transforms Sampling and reconstruction Aliasing and anti-aliasing filters Sampled-data.
The story so far… ATM 562 Fovell Fall, Convergence We will deploy finite difference (FD) approximations to our model partial differential equations.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 25.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 32.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 28.
Runge Kutta schemes Taylor series method Numeric solutions of ordinary differential equations.
Introduction to Symmetry Analysis Brian Cantwell Department of Aeronautics and Astronautics Stanford University Chapter 1 - Introduction to Symmetry.
1 EEE 431 Computational Methods in Electrodynamics Lecture 13 By Dr. Rasime Uyguroglu
Quantum One.
Numerical Methods for Partial Differential Equations
Lecture 4: Numerical Stability
Introduction to Numerical Methods I
Integral Transform Method
Effects of Zeros and Additional Poles
Lecture 19 MA471 Fall 2003.
UNIT II Analysis of Continuous Time signal
Quantum One.
Quantum One.
Presentation transcript:

Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 6 Various Finite Difference Discretizations for the Advection Equations – Phase and Dissipation Error Instructor: Tim Warburton

CAAM 452 Spring 2005 Note on AB3 Stability

CAAM 452 Spring 2005 Review: After Time Stepping… Back to PDE’s We have now proposed two quite distinct but reliable ways to advance an ODE approximately in time. But our original goal was to solve a PDE And not just for one mode.

CAAM 452 Spring 2005 Review: Sample Points in Time Space We chose to sample the ODE at discrete points on the time axis. We can also make the same choice for the spatial representation at each discrete time level. We follow this choice with the dilemma of how to reconstruct an approximation to the derivative of discretely sampled data. Assume c is positive x

CAAM 452 Spring 2005 Review: Choice of Stencil x x x x x And many more combinations

CAAM 452 Spring 2005 Today: Delta Operators We will consider the use of each of these difference formula (and more sophisticated versions) as approximations for the spatial derivatives. We consider stability… by which we mean we will examine the eigenspectrum of the difference operator treated as a matrix. We will use Gershgorin’s theorem frequently:

CAAM 452 Spring 2005 Right Difference Matrix Consider a case with 10 points on a periodic interval: The semi-discrete scheme for the advection equations is: Note: I used indexing from zero to match with classical definitions coming up.

CAAM 452 Spring 2005 Discrete Operator Matrix By Gerschgorin’s theorem we conclude that the matrix has all eigenvalues contained in the circle centered at -c/dx with radius c/dx Thus the eigenvalues are all in the left half plane, with the possible exception of one or more on the origin  good candidate for time-stepping.

CAAM 452 Spring 2005 cont We can be more precise about the location of the eigenvalues of We first introduce the Fourier transform for the data vector: This is simply a linear, map applied to the vector of values with inverse given by:

CAAM 452 Spring 2005 Fourier Transform in Matrix Notation Introducing the following notation for the Fourier transform matrix and a shift operator: We consider the following eigenvalue problem: or

CAAM 452 Spring 2005 Right Shift Matrix We write down explicitly what the right shift matrix is: Where the box around the addition operator indicates that this is addition modulo M This time the delta indicates the Kronecker Delta:

CAAM 452 Spring 2005 cont We perform a similarity transform using the Fourier matrix: Becomes: We now consider the first term: First note (basic properties of the Fourier transform):

CAAM 452 Spring 2005 cont What we discover after some algebra (with slight liberties taken with the notation) is that the Fourier transform matrix diagonalizes the right shift matrix.

CAAM 452 Spring 2005 cont Finally we return to the and can now find its eigenspectrum: Using We find that the matrix is diagonalized by the Fourier matrix and that

CAAM 452 Spring 2005 Important Result: Fourier Transform of Right Shift Matrix i.e. the Fourier (similarity) transform of the right shift matrix is a diagonal matrix

CAAM 452 Spring 2005 Left Difference Matrix Consider a case with 10 points on a periodic interval: The semi-discrete scheme for the advection equations is:

CAAM 452 Spring 2005 cont Applying Gerschgorin this time we find that the eigenvalues are all in a disk centered at +c/dx with radius c/dx. This is bad news !. All the eigenvalues are in the right half plane, apart from the possiblity of a subset which are at the origin. i.e. any non zero mode will be unstable for time stepping operator (and clearly not physical)

CAAM 452 Spring 2005 Detailed Study of the Left Difference Operator Spectrum This time we introduce the left shift operator: This time the eigenvalue problem to consider is: Through a similar analysis (since the Fourier transform also diagonalizes the left shift matrix) we find: This confirms the Gerschgorin estimate.

CAAM 452 Spring 2005 Important Result: Fourier Transform of Left Shift Matrix i.e. the Fourier (similarity) transform of the left shift matrix is a diagonal matrix

CAAM 452 Spring 2005 Polynomials of Shift Matrices We can express most finite difference operators as polynomials of shift operators: We can apply the Fourier similarity transform to diagonalize this: Where is the vector of Fourier coefficents for u. Hence each Fourier coefficient satisfies:

CAAM 452 Spring 2005 Centered Differencing We recall the centered difference operator can be expressed in terms of the left and right difference operators:

CAAM 452 Spring 2005 cont So the operator has purely imaginary eigenvalues As before we must resort to a time-stepping method which will in this case accommodate a portion of the imaginary axis.

CAAM 452 Spring 2005 The 4 th Order Central Difference Formula Recall the fourth order central difference formula: As for the central differencing formula all the eigenvalues are imaginary. We have used the diagonalizing property of the Fourier transform on polynomials of the shift matrices

CAAM 452 Spring 2005 Summary of Mode Evolution Using: We write down the ODE satisfied by each discrete Fourier mode of the transformed data:

CAAM 452 Spring 2005 Interpreting the ODEs We can assert an interpretation of multipliers in the ODE for each discrete Fourier mode. We know that the m’th mode has physical shape where L is the periodic length (L=M*dx) So (reverting back to our approach for turning a PDE into an ODE analytically) we would ideally be solving: Thus we can perform a small theta analysis of the multipliers in each of the cases we just derived.

CAAM 452 Spring 2005 (1) Right Difference The small theta analysis (low wave number) implies the solution will decay

CAAM 452 Spring 2005 (2) Left Difference The small theta analysis shows that the solution propagates and grows exponentially fast

CAAM 452 Spring 2005 (3) Center Difference The small theta analysis shows that the solution propagates and does not grow

CAAM 452 Spring 2005 (4) 4 th Order Center Difference

CAAM 452 Spring 2005 Summary of Small Theta Analysis The dominant remainder term in this analysis relates to a commonly used, physically motivated description of the shortfall of the method: Dissipative Unstable Dispersive

CAAM 452 Spring 2005 Terms The dominant error is dispersive if it causes different modes to travel at different speeds as with To leading order accuracy ctilde is the actual numerical wavespeed, compared with the physical advection speed The dominant error is dissipative if it reduces the wave amplitude over time as with

CAAM 452 Spring 2005 Comparison of Multipliers 6

CAAM 452 Spring 2005 The Imaginary Part We now see that for each of the methods has a dispersive nature (wave speed changes with wave number). Also, the imaginary part of the multiplier is not monotone and linear over [0, pi] or [pi,2*pi] Clearly there will be spurious modes (eigenvalues) which show up in the discrete spectrum as we saw last time. Spurious modes

CAAM 452 Spring 2005 Recap Via a New Example There is a 6 th order central difference scheme: We Fourier transform this:

CAAM 452 Spring 2005 cont We simplify this to: Compare with GKO p83. The eigenvalues are all imaginary Tedious Taylor expansion computations will reveal an error O(theta^7) Next we plot this dispersion relationship against the other examples:

CAAM 452 Spring 2005 Phase Error Plot It should be clear that the 6 th order central difference operator is more accurate over a wider range of theta – but still prone to spurious eigenvalues. Should be (1/6)*dx^2

CAAM 452 Spring 2005 Summary So Far We have established that for the advection equation it is not appropriate to use for the spatial derivative term (non-negative real part for multiplier). The other difference formulae are stable in this context: In this context is referred to as an upwind derivative – it uses a dissipative mechanism to remain stable. The given central difference operators are carefully designed so that their eigenvalues sit on the imaginary axis.

CAAM 452 Spring 2005 cont We have established that in each of the schemes we have replaced the exact ODE multiplier with an approximation, in this lecture denoted by: We found that this approximation is not uniform in wave number – i.e. there is a non monotonic dependence in m (embedded in the error terms). This explains our observation of the spurious eigenvalues in the computer evaluations of the spectrum of the discrete operator.

CAAM 452 Spring 2005 Recall Those Spurious Eigenvalues What should the eigenvalues of the 4 th order derivative operator ideally tend to in the limiting of decreasing dx? Example (10, 20, 80 points with the periodic length 2*pi): >> [dxDeigs, dxD] = test4th(10); >> dx = 2*pi/10; >> sort(dxDeigs/dx) ans = i i i i i i i i i i >> [dxDeigs, dxD] = test4th(20); >> dx = 2*pi/20; >> >> e = sort(dxDeigs/dx); >> e(1:10) ans = i i i i i i i i >> [dxDeigs, dxD] = test4th(80); >> dx = 2*pi/80; >> e = sort(dxDeigs/dx); >> >> e(1:10) ans = i i i i i i i i i i

CAAM 452 Spring 2005 Upwinding Recall: we referred to the differencing operator as the “upwind derivative”. This is due to the physical placement of the sampled data relative to the advection direction: i.e. the differencing only uses data from the upstream (upwind) direction. More on this next time. x Advection direction

CAAM 452 Spring 2005 Next Lecture Introduction of some standard finite-difference methods. Define (for finite difference schemes): 1)Consistency 2)Stability 3)Accuracy 4)Courant-Friedrichs-Lewy condition for stability We will examine convergence: 1)Continuous time+discrete space (as seen today) 2)Discrete time and discrete space 3)Lax-Richtmyer equivalence theory (consistency+stability  convergence)