Numerical Methods for Partial Differential Equations
description
Transcript of Numerical Methods for Partial Differential Equations
Numerical Methods for Partial Differential Equations
CAAM 452Spring 2005
Lecture 2Instructor: Tim Warburton
CAAM 452 Spring 2005
Note on textbook for finite difference methods
• Due to the difficulty some students have experienced in obtaining Gustafasson-Kreiss-Oliger I will try to find appropriate and equivalent sections in the online finite-difference book by Trefethen:
http://web.comlab.ox.ac.uk/oucl/work/nick.trefethen/pdetext.html
CAAM 452 Spring 2005
Recall Last Lecture
• We considered the model advection PDE:
• defined on the periodic interval [0,2pi)
• We recalled that any 2pi periodic, C1 function could be represented as a uniformly convergent Fourier series, so we considered the evolution of the PDE with a single Fourier mode as initial condition. This converted the above PDE into a simpler ODE for the time-dependent coefficient (i.e. amplitude) of a Fourier mode:
0u uc
t x
ˆ0ˆ
dui cu
dt
CAAM 452 Spring 2005
Recall: the Advection Equation
• We wills start with a specific Fourier mode as the initial condition:
• We try to find a solution of the same type:
1) Find 2 -periodic , such that 0,2 , 0,
0
given
1 ˆ( ,0) = 0,22
where is a smooth 2 -periodic function of one frequency
i x
u x t x t T
u uc
t x
u x f x e f x
f
1, ,ˆ
2i xu x t e u t
CAAM 452 Spring 2005
cont
• Substituting in this type of solution the PDE:
• Becomes an ODE:
• With initial condition
0u uc
t x
1 1 ˆ,ˆ ˆ
2 2ˆ
0ˆ
i x i xu u duc c e u t e i cu
t x t x dt
dui cu
dt
ˆ,0u f
CAAM 452 Spring 2005
cont
• We have Fourier transformed the PDE into an ODE.• We can solve the ODE:
• And it follows that the PDE solution is:
ˆ0ˆ ˆ, ,0ˆ ˆ
ˆ,0ˆ
i ct i ct
dui cu
u t e u e fdtu f
1: , ,ˆ
21ˆ ˆsolution : , ,ˆ2
1 ˆinitial condition: 2
i x
i x cti ct
i x
ansatz u x t e u t
u t e f u x t e f f x ct
f x e f
CAAM 452 Spring 2005
Note on Fourier Modes
• Note that since the function should be 2pi periodic we are able to deduce:
• We can also use the superposition principle for the more general case when the initial condition contains multiple Fourier modes:
1 ˆ2
1 ˆ,2
i x
i x ct
f x e f
u x t e f f x t
CAAM 452 Spring 2005
cont
• Let’s back up a minute – the crucial part was when we reduced the PDE to an ODE:
• The advantage is: we know how to solve ODE’s both analytically and numerically (more about this later on).
0u uc
t x
ˆ0ˆ
dui cu
dt
CAAM 452 Spring 2005
Add Diffusion Back In
• So we have a good handle on the advection equation, let’s reintroduce the diffusion term:
• Again, let’s assume 2-pi periodicity and assume the same ansatz:
• This time:
2
2
u u uc d
t x x
1,ˆ
2i xu e u t
2
2
u u uc d
t x x
2ˆˆ ˆ
dui cu d u
dt 2ˆ
ˆdu
ic d udt
CAAM 452 Spring 2005
cont
• Again, we can solve this trivial ODE:
2ˆˆ
dui c d u
dt
2
,0ˆ ˆic d t
u u e
2
,0ˆ i ct x d tu u e e
CAAM 452 Spring 2005
cont
• The solution tells a story:
• The original profile travels in the direction of decreasing x (first term)
• As the profile travels it decreases in amplitude (second exponential term)
2
,0ˆ i ct x d tu u e e
2d tu f ct x e
CAAM 452 Spring 2005
What Did Diffusion Do??
• Advection:
• Diffusion:
• Adding the diffusion term shifted the multiplier on the right hand side of the Fourier transformed PDE (i.e. the ODE) into the left half plane.
• We summarize the role of the multiplier…
0u uc
t x
2
2
u u uc d
t x x
ˆ0ˆ
dui cu
dt
2ˆˆ
dui c d u
dt
CAAM 452 Spring 2005
Categorizing a Linear ODE
Re
Im
Exponential growth Exponential decay
Incr
easi
ngly
osc
illat
ory
In
crea
sing
ly
osc
illat
ory
Here we plot the dependence of the solution to the top left ODE on mu’s position in the complex plane
ˆ 0ˆ ˆ ˆ tduu u u e
dt
CAAM 452 Spring 2005
Categorizing a Linear ODE
Re
Im
Here we plot the behavior of the solution for 5 different choices of mu
CAAM 452 Spring 2005
Summary
• When the real part of mu is negative the solution decays exponentially fast in time (rate determined by the magnitude of the real part of mu).
• When the real part of mu is positive the solution grows exponentially fast in time (rate determined by the magnitude of the real part of mu).
• If the imaginary part of mu is non-zero the solution oscillates in time.
• The larger the imaginary part of mu is, the faster the solution oscillates in time.
Reˆ 0 0 cos Im sin Imˆ ˆ ˆ ˆ ttduu u u e u e t i t
dt
CAAM 452 Spring 2005
Solving the Scalar ODE Numerically
• We know the solution to the scalar ODE
• However, it is also reasonable to ask if we can solve it approximately.
• We have now simplified as far as possible.
• Once we can solve this model problem numerically, we will apply this technique using the method of lines to approximate the solution of the PDE.
ˆˆ
duu
dt
CAAM 452 Spring 2005
Stage 1: Discretizing the Time Axis
• It is natural to divide the time interval [0,T] into shorted subintervals, with width often referred to as:
• We start with the initial value of the solution u(0) (and possibly u(-dt),u(-2dt),..) and build a recurrence relation which approximates u(dt) in terms of u(0) and early values of u.
, or dt t k
duf u
dt
CAAM 452 Spring 2005
Example
• Using the following approximation to the time derivative:
• We write down an approximation to the ODE:
u t dt u tdudt dt
00
u dt uf u
dt
du
f udt
CAAM 452 Spring 2005
Example cont
• Rearranging:
• We introduce notation for the approximate solution after the n’th time step:
• Our intention is to compute • We convert the above equation into a scheme to
compute an approximate solution:
00
u dt uf u
dt
0 0u dt u dtf u
nu
0
1 0 0
0u u
u u dtf u
nu u ndt
CAAM 452 Spring 2005
Example cont
• We can repeat the following step from t=dt to t=2dt and so on:
• This is commonly known as the:– Euler-forward time-stepping method
– or Euler’s method
0
1 0 0
2 1 1
1
0
n n n
u u
u u dtf u
u u dtf u
u u dtf u
CAAM 452 Spring 2005
Euler-Forward Time Stepping
• It is natural to ask the following questions about this time-stepping method:
• Does the answer get better if dt is reduced? (i.e. we take more time-steps between t=0 and t=T)
• Does the numerical solution behave in the same way as the exact solution for general f?(for the case of f(u) = mu*u does the numerical solution decay and/or oscillate as the exact solution)
• How close to the exact solution is the numerical solution?
• As we decrease dt does the end iterate converge to the solution at t=T
0
1
0
n n n
u u
u u dtf u
CAAM 452 Spring 2005
Experiments
• Before we try this analytically, we can code it up and see what happens.
• This is some matlab code for Euler forward applied to:
0
1
0
n n n
u u
u u dt u
http://www.caam.rice.edu/~caam452/CodeSnippets/EulerForwardODE.m
CAAM 452 Spring 2005
What dt can we use?
• With dt=0.1, T=3, uo=1, mu=-1-2*i
The error is quite small.
Matlab
CAAM 452 Spring 2005
Larger dt=0.5
• With dt=0.5, T=3, uo=1, mu=-1-2*i
The error is visible but the trend is not wildly wrong.
CAAM 452 Spring 2005
Even Larger dt=1
• With dt=1, T=3, uo=1, mu=-1-2*i
The solution is not remotely correct – but is at least bounded.
CAAM 452 Spring 2005
dt=2
• With T=30, dt=2, uo=1, mu=-1-i
Boom – the approximate solution grows exponentially fast, while the true solution decays!
CAAM 452 Spring 2005
Observations
• When dt is small enough we are able to nicely approximate the solution with this simple scheme.
• As dt grew the solution became less accurate
• When dt=1 we saw that the approximate solution did not resemble the true solution, but was at least bounded.
• When dt=2 we saw the approximate solution grew exponentially fast in time.
CAAM 452 Spring 2005
cont
• Our observations indicate two qualities of time stepping we should be interested in:
– stability: i.e. is the solution bounded in a similar way to the actual solution?
– accuracy: can we choose dt small enough for the error to be below some threshold?
CAAM 452 Spring 2005
ndtf udt
nu
1nt tn
1nu
Geometric Interpretation:
• We can interpret Euler-forward as a shooting method.• We suppose that is the actual solution, compute the
actual slope and shift the approximate solution by
• Note:– the blue line is not the exact solution, but rather the same
ODE started from the last approximate value of u computed.– in this case we badly estimated the behavior of even the
approximately started solution in the interval dt two kinds of error can accumulate!.
ndtf unu
nf u
0
1
0
n n n
u u
u u dtf u
CAAM 452 Spring 2005
Interpolation Interpretation:
• We start with the ODE:
• Integrate both sides from over a dt interval:
• Use the fundamental theorem of calculus:
• Finally, replace f with a constant which interpolates f at the beginning of the interval…
duf u
dt
1 1n n
n n
t t
t t
dudt f u dt
dt
1
1
1
n
n
n
n
tt
n ntt
dudt u u u
dt
0
1
0
n n n
u u
u u dtf u
CAAM 452 Spring 2005
cont
1
1
1 1 1
1
1
n
n
n
n
n n n
n n n
tt
n ntt
t t t
n n n
t t t
n n n
dudt u u t u t
dt
f u t dt f u t dt f u t dt f u t dt
u t u t f u t dt
Again, we have recovered Euler’s method.
CAAM 452 Spring 2005
Let’s choose f(u) = mu*u
We can solve this immediately:
Next suppose Re(mu)<=0 then we expect the actualsolution to be bounded in time by u(0).For this to be true of the approximate solution werequire:
Stability
0
1
0
1n n n n
u u
u u dt u dt u
0
1
1
0
1 1 0n
n n
u u
u dt u dt u
1 1dt
CAAM 452 Spring 2005
Stability Condition for Euler Forward
• Since mu is fixed we are left with a condition which must be met by dt
• which is true if and only if:
• The region of the complex plane which satisfies this condition is the interior of the unit circle centered at -1+0*i
1 1dt
2 21 Re Im 1dt dt
x Re
Im
CAAM 452 Spring 2005
cont
• i.e. will not blow up if it is located inside the unit disk:
• Notice: the exact solutions corresponding to Re(mu)<0 all decay. However, only the numerical solutions corresponding to the interior of the yellow circle decay.
dt
Re
Im
x-1+0i
CAAM 452 Spring 2005
Disaster for Advection!!!
• Now we should be worried. The stability region:
• only includes one point on the imaginary axis (the origin) but our advection equation for the periodic interval has mu which are purely imaginary!!!.
• Conclusion – the Euler-Forward scheme applied to the Fourier transform of the advection equation will generate exponentially growing solutions.
Re
Im
x-1+0i
ˆ0ˆ
dui cu i c
dt
CAAM 452 Spring 2005
Nearly a Disaster for Advection-Diffusion!!!
• This time the stability region must contain:
• Because the eigenvalues are shifted into the left half part of the complex plane, we will be able to choose dt small enough to force the mu*dt into the stability region.
• We can estimate how small dt has to be in this case
Re
Im
x-1+0i
2ˆˆ
dui c d u
dt
2dt i c d dt
CAAM 452 Spring 2005
cont (estimate of dt for advection-diffusion)
• For stability we require:
2
2 22
2 22 2
2 2 2
1 1
1 1
1 1
2 0
0 or
0 or
2
dt
i c d dt
dt d cdt
dt d dt d cdt
dt
dt d c d
CAAM 452 Spring 2005
cont
• The only interesting condition for stability is:
• What does this mean?, volunteer?• i.e. how do influence the maximum stable dt?• Hint: consider
i) d small, c large
ii) d large, c small
2 2 2
2
ddt
d c
, ,d c
CAAM 452 Spring 2005
Multistep Scheme (AB2)
• Given the failure of Euler-Forward for our goal equation, we will consider using the solution from 2 previous time steps.
• i.e. we consider
• Recall the integral based interpretation of Euler-Forward:
1 1n n n nu u dt af u bf u
1
1
n
n
t
n n n
t
u t u t f u t dt f u t dt
CAAM 452 Spring 2005
cont
• We interpolate through f(un-1)and f(un) then integrate between tn and tn+1
Ifn+1
f(un-1)
tn+1tntn-1
Approximate integral
fLinear interpolant of the f values
f(un)
CAAM 452 Spring 2005
cont
• This time we choose to integrate under the interpolant of f which agrees with f at tn and tn-1
• The unique linear interpolant in this case is:
• We need to compute the following integral of the interpolant..:
11 1
1 1
n nn n
n n n n
t t t tI f f u f u
t t t t
1
n
n
t
t
I fdt
CAAM 452 Spring 2005
cont
Last step: homework exercise
1 1
1 1
1 1
11 1
1 1
11
2 21
1
1
2 2
3 12 2
n n
n n
n n
n n
n n
n n
t t
n nn n
n n n nt t
t t
n nn n
t t
t t
n nn n
t t
n n
t t t tI fdt f u f u dt
t t t t
f u f ut t dt t t dt
dt dt
f u f ut tt t t t
dt dt
dt f u f u
CAAM 452 Spring 2005
Adams-Bashford 2
• The resulting AB2 scheme requires the solution at two levels to compute the update:
1
1 1
1
3 12 2
n
n
t
n n
t
n n n
u u If dt
u f u f u
CAAM 452 Spring 2005
Next Time
• Stability region for AB2• Generalization to AB3, AB4 (use more historical data to compute
update)
• Accuracy/consistency of Euler-Forward, AB2,AB3,AB4• Convergence.
• Runge-Kutta schemes (if time permits)
• READING:
http://web.comlab.ox.ac.uk/oucl/work/nick.trefethen/1all.pdf
• p10-p55 (do not do exercises)