![Page 1: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/1.jpg)
The Finite Difference Method
Heiner Igel
Department of Earth and Environmental SciencesLudwig-Maximilians-University Munich
Heiner Igel Computational Seismology 1 / 32
![Page 2: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/2.jpg)
Outline
1 IntroductionMotivationHistoryFinite Differences in a Nutshell
2 Finite Differences and Taylor SeriesFinite Difference DefinitionHigher DerivativesHigh-Order Operators
3 Finite-Difference Approximation of Wave Equations
Heiner Igel Computational Seismology 2 / 32
![Page 3: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/3.jpg)
Introduction Motivation
Motivation
Simple conceptRobustEasy to parallelizeRegular gridsExplicit method
Heiner Igel Computational Seismology 3 / 32
![Page 4: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/4.jpg)
Introduction History
History
Several Pioneers of solving PDEs with finite-difference method(Lewis Fry Richardson, Richard Southwell, Richard Courant, KurtFriedrichs, Hans Lewy, Peter Lax and John von Neumann)First application to elastic wave propagation (Alterman and Karal,1968)Simulating Love waves and was the frst showing snapshots ofseimsic wave fields (Boore, 1970)Concept of staggered-grids by solving the problem of rupturepropagation (Madariaga, 1976 and Virieux and Madariaga, 1982)
Heiner Igel Computational Seismology 4 / 32
![Page 5: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/5.jpg)
Introduction History
History
Several Pioneers of solving PDEs with finite-difference method(Lewis Fry Richardson, Richard Southwell, Richard Courant, KurtFriedrichs, Hans Lewy, Peter Lax and John von Neumann)First application to elastic wave propagation (Alterman and Karal,1968)Simulating Love waves and was the frst showing snapshots ofseimsic wave fields (Boore, 1970)Concept of staggered-grids by solving the problem of rupturepropagation (Madariaga, 1976 and Virieux and Madariaga, 1982)
Heiner Igel Computational Seismology 4 / 32
![Page 6: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/6.jpg)
Introduction History
History
Several Pioneers of solving PDEs with finite-difference method(Lewis Fry Richardson, Richard Southwell, Richard Courant, KurtFriedrichs, Hans Lewy, Peter Lax and John von Neumann)First application to elastic wave propagation (Alterman and Karal,1968)Simulating Love waves and was the frst showing snapshots ofseimsic wave fields (Boore, 1970)Concept of staggered-grids by solving the problem of rupturepropagation (Madariaga, 1976 and Virieux and Madariaga, 1982)
Heiner Igel Computational Seismology 4 / 32
![Page 7: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/7.jpg)
Introduction History
History
Several Pioneers of solving PDEs with finite-difference method(Lewis Fry Richardson, Richard Southwell, Richard Courant, KurtFriedrichs, Hans Lewy, Peter Lax and John von Neumann)First application to elastic wave propagation (Alterman and Karal,1968)Simulating Love waves and was the frst showing snapshots ofseimsic wave fields (Boore, 1970)Concept of staggered-grids by solving the problem of rupturepropagation (Madariaga, 1976 and Virieux and Madariaga, 1982)
Heiner Igel Computational Seismology 4 / 32
![Page 8: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/8.jpg)
Introduction History
History
Extension to 3D because of parallel computations (Frankel andVidale, 1992; Olsen and Archuleta, 1996; etc.)Application to spherical geometry by Igel and Weber, 1995;Chaljub and Tarantola, 1997 and 3D spherical sections by Igel etal., 2002Incorporation in the first full waveform inversion schemes initiallyin 2D, e.g. (Crase et al., 1990) and later in 3D (Chen et al., 2007)
Heiner Igel Computational Seismology 5 / 32
![Page 9: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/9.jpg)
Introduction History
History
Extension to 3D because of parallel computations (Frankel andVidale, 1992; Olsen and Archuleta, 1996; etc.)Application to spherical geometry by Igel and Weber, 1995;Chaljub and Tarantola, 1997 and 3D spherical sections by Igel etal., 2002Incorporation in the first full waveform inversion schemes initiallyin 2D, e.g. (Crase et al., 1990) and later in 3D (Chen et al., 2007)
Heiner Igel Computational Seismology 5 / 32
![Page 10: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/10.jpg)
Introduction History
History
Extension to 3D because of parallel computations (Frankel andVidale, 1992; Olsen and Archuleta, 1996; etc.)Application to spherical geometry by Igel and Weber, 1995;Chaljub and Tarantola, 1997 and 3D spherical sections by Igel etal., 2002Incorporation in the first full waveform inversion schemes initiallyin 2D, e.g. (Crase et al., 1990) and later in 3D (Chen et al., 2007)
Heiner Igel Computational Seismology 5 / 32
![Page 11: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/11.jpg)
Introduction Finite Differences in a Nutshell
Finite Differences in a Nutshell
Snapshot in space of thepressure field pZoom into the wave fieldwith grid points indicatedby +Exact interpolate usingTaylor series
Heiner Igel Computational Seismology 6 / 32
![Page 12: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/12.jpg)
Introduction Finite Differences in a Nutshell
1D acoustic wave equation
p(x , t) = c(x)2 ∂2x p(x , t) + s(x , t)
p pressurec acoustic velocitys source term
Approximation with a difference formula
p(x , t) ≈ p(x , t + dt)− 2p(x , t) + p(x , t − dt)dt2
and equivalently for the space derivative
Heiner Igel Computational Seismology 7 / 32
![Page 13: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/13.jpg)
Finite Differences and Taylor Series Finite Difference Definition
Finite Differences and Taylor Series
Heiner Igel Computational Seismology 8 / 32
![Page 14: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/14.jpg)
Finite Differences and Taylor Series Finite Difference Definition
Finite Differences
Forward derivative
dx f (x) = limdx → 0
f (x + dx)− f (x)dx
Centered derivative
dx f (x) = limdx → 0
f (x + dx)− f (x − dx)2dx
Backward derivative
dx f (x) = limdx → 0
f (x)− f (x − dx)dx
Heiner Igel Computational Seismology 9 / 32
![Page 15: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/15.jpg)
Finite Differences and Taylor Series Finite Difference Definition
Finite Differences
Forward derivative
dx f+ ≈ f (x + dx)− f (x)dx
Centered derivative
dx f c ≈ f (x + dx)− f (x − dx)2dx
Backward derivative
dx f− ≈ f (x)− f (x − dx)dx
Heiner Igel Computational Seismology 10 / 32
![Page 16: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/16.jpg)
Finite Differences and Taylor Series Finite Difference Definition
Finite Differences and Taylor Series
The approximate sign is important here as the derivatives at pointx are not exact. Understanding the accuracy by looking at thedefinition of Taylor Series:
f (x + dx) = f (x) + f ′(x) dx + 12! f ′′(x) dx2 + O(dx3)
Subtraction with f(x) and division by dx leads to the definition ofthe forward derivative:
f (x+dx)−f (x)dx = f ′(x) + 1
2! f ′′(x) dx + O(dx2)
Heiner Igel Computational Seismology 11 / 32
![Page 17: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/17.jpg)
Finite Differences and Taylor Series Finite Difference Definition
Finite Differences and Taylor Series
Using the same approach - adding the Taylor Series for f (x + dx)and f (x − dx) and dividing by 2dx leads to:
f (x+dx)−f (x−dx)2dx = f ′(x) + O(dx2)
This implies a centered finite-difference scheme more rapidlyconverges to the correct derivative on a regular grid=⇒ It matters which of the approximate formula one chooses=⇒ It does not imply that one or the other finite-differenceapproximation is always the better one
Heiner Igel Computational Seismology 12 / 32
![Page 18: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/18.jpg)
Finite Differences and Taylor Series Higher Derivatives
Higher Derivatives
The partial differential equations have often 2nd (seldom higher)derivativesDeveloping from first derivatives by mixing a forward and abackward definition yields to
∂2x f ≈
f (x+dx)−f (x)dx − f (x)−f (x−dx)
dxdx
=f (x + dx)− 2f (x) + f (x − dx)
dx2
Heiner Igel Computational Seismology 13 / 32
![Page 19: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/19.jpg)
Finite Differences and Taylor Series Higher Derivatives
Higher Derivatives
Determining the weights with which the function values have to bemultiplied to obtain derivative approximationsThe Taylor series for two grid points at x ± dx , include the functionat the central point as well and multiply each by a real number
af (x + dx) = a[f (x) + f ′(x) dx + 1
2! f ′′(x) dx2 + . . .]
bf (x) = b [f (x)]cf (x − dx) = c
[f (x)− f ′(x) dx + 1
2! f ′′(x) dx2 − . . .]
Heiner Igel Computational Seismology 14 / 32
![Page 20: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/20.jpg)
Finite Differences and Taylor Series Higher Derivatives
Higher Derivatives
Summing up the right-hand sides of the previous equations andcomparing coefficients leads to the following system of equations:
a + b + c = 0
a− c = 0
a + c =2!
dx2
⇒a =
1dx2
b = − 2dx2
c =1
dx2
Heiner Igel Computational Seismology 15 / 32
![Page 21: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/21.jpg)
Finite Differences and Taylor Series High-Order Operators
High-Order Operators
What happens if we extend the domain of influence for thederivative(s) of our function f(x)?Let us search for a 5-point operator for the second derivative
f ′′ ≈ af (x + 2dx) + bf (x + dx) + cf (x) + df (x − dx) + ef (x − 2dx)
a + b + c + d + e = 02a + b − d − 2e = 0
4a + b + d + 4e =1
2dx2
8a + b − d − 8e = 016a + b + d + 16e = 0
Heiner Igel Computational Seismology 16 / 32
![Page 22: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/22.jpg)
Finite Differences and Taylor Series High-Order Operators
High-Order Operators
Using matrix inversion we obtain a unique solution
a = − 112dx2
b =4
3dx2
c = − 52dx2
d =4
3dx2
e = − 112dx2 .
with a leading error term for the 2nd derivative is O(dx4)
=⇒ Accuracy improvement
Heiner Igel Computational Seismology 17 / 32
![Page 23: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/23.jpg)
Finite Differences and Taylor Series High-Order Operators
High-Order Operators
Graphical illustration of theTaylor Operators for thefirst derivative for higherordersThe weights rapidlybecome small forincreasing distance tocentral point of evaluation
Heiner Igel Computational Seismology 18 / 32
![Page 24: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/24.jpg)
Finite-Difference Approximation of Wave Equations
Finite-Difference Approximation ofWave Equations
Heiner Igel Computational Seismology 19 / 32
![Page 25: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/25.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
To solve the wave equation, we start with thesimplemost wave equation:The constant density acoustic wave equationin 1D
p = c2 ∂2x p + s
impossing pressure-free conditions at the twoboundaries as
p(x) |x=0,L= 0
Heiner Igel Computational Seismology 20 / 32
![Page 26: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/26.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
The following dependencies apply:
p → p(x, t) pressurec → c(x) P-velocitys → s(x, t) source term
As a first step we need to discretize space and time and we do thatwith a constant increment that we denote dx and dt.
xj = jdx , j = 0, jmax
tn = ndt , n = 0, nmax
Heiner Igel Computational Seismology 21 / 32
![Page 27: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/27.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
Starting from the continuous description of the partial differentialequation to a discrete description. The upper index will correspond tothe time discretization, the lower index will correspond to the spatialdiscretization
pn+1j → p(xj , tn + dt)
pnj → p(xj , tn)
pn−1j → p(xj , tn − dt)
pnj+1 → p(xj + dx , tn)
pnj → p(xj , tn)
pnj−1 → p(xj − dx , tn)
.
Heiner Igel Computational Seismology 22 / 32
![Page 28: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/28.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
pn+1j − 2pn
j + pn−1j
dt2 = c2j
[pn
j+1 − 2pnj + pn
j−1
dx2
]+ sn
j .
the r.h.s. is defined at sametime level nthe l.h.s. requires informationfrom three different time levels
n − 1i
n + 1
j − 1 j j + 1
Heiner Igel Computational Seismology 23 / 32
![Page 29: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/29.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
Assuming that information at time level n (the presence) and n − 1 (thepast) is known, we can solve for the unknown field pn+1
j :
pn+1j = c2
jdt2
dx2
[pn
j+1 − 2pnj + pn
j−1
]+ 2pn
j − pn−1j + dt2sn
j
The initial condition of our wave simulation problem is such thateverything is at rest at time t = 0:
p(x , t)|t=0 = 0 , p(x , t)|t=0 = 0.
Heiner Igel Computational Seismology 24 / 32
![Page 30: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/30.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
Waves begin to radiate as soon as the source term s(x , t) starts toactFor simplicity: the source acts directly at a grid point with index jsTemporal behaviour of the source can be calculated by Green’sfunction
s(x , t) = δ(x − xs) δ(t − ts)
where xs and ts are source location and source time and δ()corresponds to the delta function
Heiner Igel Computational Seismology 25 / 32
![Page 31: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/31.jpg)
Finite-Difference Approximation of Wave Equations
Acoustic waves in 1D
A delta function contains all frequencies and we cannot expect that ournumerical algorithm is capable of providing accurate solutionsOperating with a band-limited source-time function:
s(x , t) = δ(x − xs) f (t)
where the temporal behaviour f(t) is chosen according to our specificphysical problem
Heiner Igel Computational Seismology 26 / 32
![Page 32: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/32.jpg)
Finite-Difference Approximation of Wave Equations
Example
Simulating acoustic wave propagation in a 10km column (e.g. theatmosphere) and assume an air sound speed of c = 0.343m/s. Wewould like to hear the sound wave so it would need a dominantfrequency of at least 20 Hz. For the purpose of this exercise weinitialize the source time function f (t) using the first derivative of aGauss function.
f (t) = −8 f0 (t − t0) e− 1
(4f0)2 (t−t0)2
where t0 corresponds to the time of the zero-crossing, f0 is thedominant frequency
Heiner Igel Computational Seismology 27 / 32
![Page 33: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/33.jpg)
Finite-Difference Approximation of Wave Equations
Example
What is the minimum spatialwavelength that propagates inside themedium?What is the maximum velocity insidethe medium?What is the propagation distance ofthe wavefield (e.g., in dominantwavelengths)?
Heiner Igel Computational Seismology 28 / 32
![Page 34: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/34.jpg)
Finite-Difference Approximation of Wave Equations
Example
Sufficient to look at the relation between frequency and wavenumber:
c =ω
k=
λ
T= λ f
where c is velocity, T is period, λ is wavelength, f is frequency, andω = 2πf is angular frequency
dominant wavelength of f0 = 20Hzsubstantial amount of energy in the wavelet is at frequenciesabove 20 Hz=⇒ λ = 17m and λ = 7m for frequencies 20Hz and 50Hz,respectively
Heiner Igel Computational Seismology 29 / 32
![Page 35: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/35.jpg)
Finite-Difference Approximation of Wave Equations
Matlab code fragment
% Time extrapolationfor it = 1 : nt,(...)% Space derivativesfor j=2:nx-1d2p(j)=(p(j+1)-2*p(j)+p(j-1))/dxˆ 2;end% Extrapolationpnew=2*p-pold+cˆ 2*d2p*dtˆ 2;% Source inputpnew(isrc)=pnew(isrc)+src(it)*dtˆ 2;% Time levelspold=p;p=pnew;(...)
Heiner Igel Computational Seismology 30 / 32
![Page 36: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/36.jpg)
Finite-Difference Approximation of Wave Equations
Result
Choosing a grid increment ofdx = 0.5m −→ about 24 points perspatial wavelength for the dominantfrequencySetting time increment dt = 0.0012 −→around 40 points per dominant period
Heiner Igel Computational Seismology 31 / 32
![Page 37: The Finite Difference Method - uni-muenchen.de · The Finite Difference Method ... Finite Differences and Taylor Series Finite Difference Definition ... backward definition yields](https://reader031.vdocuments.us/reader031/viewer/2022021612/5b889f0b7f8b9a435b8e2ac0/html5/thumbnails/37.jpg)
Finite-Difference Approximation of Wave Equations
Summary
Replacing the partial derivatives by finite differences allows partialdifferential equations such as the wave equation to be solved directly for(in principle) arbitrarily heterogeneous media
The accuracy of finite-difference operators can be improved by usinginformation from more grid points (i.e., longer operators). The weightsfor the grid points can be obtained using Taylor series
Heiner Igel Computational Seismology 32 / 32