Lecture 1 Describing Inverse Problems
description
Transcript of Lecture 1 Describing Inverse Problems
![Page 1: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/1.jpg)
Lecture 1
Describing Inverse Problems
![Page 2: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/2.jpg)
SyllabusLecture 01 Describing Inverse ProblemsLecture 02 Probability and Measurement Error, Part 1Lecture 03 Probability and Measurement Error, Part 2 Lecture 04 The L2 Norm and Simple Least SquaresLecture 05 A Priori Information and Weighted Least SquaredLecture 06 Resolution and Generalized InversesLecture 07 Backus-Gilbert Inverse and the Trade Off of Resolution and VarianceLecture 08 The Principle of Maximum LikelihoodLecture 09 Inexact TheoriesLecture 10 Nonuniqueness and Localized AveragesLecture 11 Vector Spaces and Singular Value DecompositionLecture 12 Equality and Inequality ConstraintsLecture 13 L1 , L∞ Norm Problems and Linear ProgrammingLecture 14 Nonlinear Problems: Grid and Monte Carlo Searches Lecture 15 Nonlinear Problems: Newton’s Method Lecture 16 Nonlinear Problems: Simulated Annealing and Bootstrap Confidence Intervals Lecture 17 Factor AnalysisLecture 18 Varimax Factors, Empircal Orthogonal FunctionsLecture 19 Backus-Gilbert Theory for Continuous Problems; Radon’s ProblemLecture 20 Linear Operators and Their AdjointsLecture 21 Fréchet DerivativesLecture 22 Exemplary Inverse Problems, incl. Filter DesignLecture 23 Exemplary Inverse Problems, incl. Earthquake LocationLecture 24 Exemplary Inverse Problems, incl. Vibrational Problems
![Page 3: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/3.jpg)
Purpose of the Lecture
distinguish forward and inverse problems
categorize inverse problems
examine a few examples
enumerate different kinds of solutions to inverse problems
![Page 4: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/4.jpg)
Part 1
Lingo for discussing the relationship between observations and the things
that we want to learn from them
![Page 5: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/5.jpg)
three important definitions
![Page 6: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/6.jpg)
things that are measured in an experiment or observed in nature…
data, d = [d1, d2, … dN]T
things you want to know about the world …
model parameters, m = [m1, m2, … mM]T relationship between data and model parameters
quantitative model (or theory)
![Page 7: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/7.jpg)
gravitational accelerationstravel time of seismic waves
data, d = [d1, d2, … dN]T model parameters, m = [m1, m2, … mM]T quantitative model (or theory)
densityseismic velocity
Newton’s law of gravityseismic wave equation
![Page 8: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/8.jpg)
Quantitative Model
Quantitative Model
mest dpre
mest dobs
Forward Theory
Inverse Theory
estimates predictions
observationsestimates
![Page 9: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/9.jpg)
Quantitative Model
Quantitative Model
mtrue dpre
dobs≠
mest
due to observational
error
![Page 10: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/10.jpg)
Quantitative Model
Quantitative Model
mtrue dpre
dobs≠
mest
due to observational
error≠ due to error
propagation
![Page 11: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/11.jpg)
Understanding the effects ofobservational error
is central to Inverse Theory
![Page 12: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/12.jpg)
Part 2
types of quantitative models(or theories)
![Page 13: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/13.jpg)
A. Implicit Theory
L relationships between the data and the model are known
![Page 14: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/14.jpg)
Example
mass = density ⨉ length ⨉ width ⨉ height
MH
M = ρ ⨉ L ⨉ W ⨉ HLρ
![Page 15: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/15.jpg)
weight = density ⨉ volumemeasure
mass, d1size, d2, d3, d4,
want to knowdensity, m1 d1 = m1 d2 d3 d4 or d1 - m1 d2 d3 d4 = 0
d=[d1, d2, d3, d4]T and N=4m=[m1]T and M=1f1(d,m)=0 and L=1
![Page 16: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/16.jpg)
note
No guarantee thatf(d,m)=0contains enough information
for unique estimate m
determining whether or not there is enoughis part of the inverse problem
![Page 17: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/17.jpg)
B. Explicit Theory
the equation can be arranged so that d is a function of m
L = N one equation per datum
d = g(m) or d - g(m) = 0
![Page 18: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/18.jpg)
Example
Circumference = 2 ⨉ length + 2 ⨉ heightL
rectangle H
Area = length ⨉ height
![Page 19: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/19.jpg)
C = 2L+2Hmeasure
C=d1A=d2
want to knowL=m1 H=m2
d=[d1, d2]T and N=2m=[m1, m2]T and M=2
Circumference = 2 ⨉ length + 2 ⨉ heightArea = length ⨉ heightA=LH
d1 = 2m1 + 2m2d2 m1m2 d=g(m)
![Page 20: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/20.jpg)
C. Linear Explicit Theory
the function g(m) is a matrix G times m
G has N rows and M columns
d = Gm
![Page 21: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/21.jpg)
C. Linear Explicit Theory
the function g(m) is a matrix G times m
G has N rows and M columns
d = Gm“data kernel”
![Page 22: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/22.jpg)
Example
M = ρg ⨉ V g+ ρq ⨉ V q
gold quartz
total mass = density of gold ⨉ volume of gold+ density of quartz ⨉ volume of quartzV = V g+ V q
total volume = volume of gold + volume of quartz
![Page 23: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/23.jpg)
M = ρg ⨉ Vg+ ρq ⨉ V qV = V g+ V q
measure V = d1M = d2
want to know Vg =m1 Vq =m2assume ρg ρg
d=[d1, d2]T and N=2m=[m1, m2]T and M=2
d = 1 1ρg ρq
mknown
![Page 24: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/24.jpg)
D. Linear Implicit Theory
The L relationships between the data are linear
L rowsN+M columns
![Page 25: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/25.jpg)
in all these examples m is discrete
one could have a continuous m(x) instead
discrete inverse theory
continuous inverse theory
![Page 26: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/26.jpg)
as a discrete vector m
in this course we will usually approximate a continuous m(x)
m = [m(Δx), m(2Δx), m(3Δx) … m(MΔx)]T
but we will spend some time later in the course dealing with the continuous
problem directly
![Page 27: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/27.jpg)
Part 3
Some Examples
![Page 28: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/28.jpg)
1965 1970 1975 1980 1985 1990 1995 2000 2005 2010-0.5
0
0.5
1
calendar year
tem
pera
ture
ano
mal
y, d
eg C
time, t (calendar years)
temperatu
re anomal
y, T i (deg
C)A. Fitting a straight line to data
T = a + bt
![Page 29: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/29.jpg)
each data pointis predicted by a
straight line
![Page 30: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/30.jpg)
matrix formulation
d = G m M=2
![Page 31: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/31.jpg)
B. Fitting a parabola
T = a + bt+ ct2
![Page 32: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/32.jpg)
each data pointis predicted by a
strquadratic curve
![Page 33: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/33.jpg)
matrix formulation
d = G m M=3
![Page 34: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/34.jpg)
straight line
note similarity
parabola
![Page 35: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/35.jpg)
in MatLab
G=[ones(N,1), t, t.^2];
![Page 36: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/36.jpg)
C. Acoustic Tomography
1 2 3 4
5 6 7 8
13 14 15 16
hh
source, S receiver, R
travel time = length ⨉ slowness
![Page 37: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/37.jpg)
collect data along rows and columns
![Page 38: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/38.jpg)
matrix formulation
d = G m M=16N=8
![Page 39: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/39.jpg)
In MatLabG=zeros(N,M);for i = [1:4]for j = [1:4] % measurements over rows k = (i-1)*4 + j; G(i,k)=1; % measurements over columns k = (j-1)*4 + i; G(i+4,k)=1;endend
![Page 40: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/40.jpg)
D. X-ray Imaging
S
R1R2R3 R4 R5enlarged
lymph node
(A) (B)
![Page 41: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/41.jpg)
theory
I = Intensity of x-rays (data)s = distancec = absorption coefficient (model parameters)
![Page 42: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/42.jpg)
Taylor Seriesapproximation
![Page 43: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/43.jpg)
Taylor Seriesapproximationdiscrete pixelapproximation
![Page 44: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/44.jpg)
Taylor Seriesapproximationdiscrete pixelapproximation length of beam i in pixel jd = G m
![Page 45: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/45.jpg)
d = G m
matrix formulation
M≈106N≈106
![Page 46: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/46.jpg)
note that G is huge106⨉106
but it is sparse(mostly zero)
since a beam passes through only a tiny fraction of the total number of
pixels
![Page 47: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/47.jpg)
in MatLab
G = spalloc( N, M, MAXNONZEROELEMENTS);
![Page 48: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/48.jpg)
E. Spectral Curve Fitting
0 2 4 6 8 10 120.65
0.7
0.75
0.8
0.85
0.9
0.95
1
velocity, mm/s
coun
ts
![Page 49: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/49.jpg)
single spectral peak
0 5 100
0.1
0.2
0.3
0.4
0.5
d
p(d)
0 5 100
10
20
30
d
coun
ts
0 5 100
0.1
0.2
0.3
0.4
0.5
d
p(d)
area, Aposition, f
width, czp(z)
![Page 50: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/50.jpg)
q spectral peaks“Lorentzian”
d = g(m)
![Page 51: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/51.jpg)
e1e2e3e4e5
e1e2e3e4e5
s1 s2
ocean
sediment
F. Factor Analysis
![Page 52: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/52.jpg)
d = g(m)
![Page 53: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/53.jpg)
Part 4
What kind of solution are we looking for ?
![Page 54: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/54.jpg)
A: Estimate of model parameters
meaning numerical values
m1 = 10.5m2 = 7.2
![Page 55: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/55.jpg)
But we really need confidence limits, too
m1 = 10.5 ± 0.2m2 = 7.2 ± 0.1 m1 = 10.5 ± 22.3m2 = 7.2 ± 9.1orcompletely different implications!
![Page 56: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/56.jpg)
B: probability density functions
if p(m1) simplenot so different than confidence intervals
0 5 100
0.1
0.2
0.3
0.4
m
p(m
)
0 5 100
0.1
0.2
0.3
0.4
m
p(m
)
0 5 100
0.1
0.2
0.3
0.4
m
p(m
)
![Page 57: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/57.jpg)
0 5 100
0.1
0.2
0.3
0.4
m
p(m
)
0 5 100
0.1
0.2
0.3
0.4
m
p(m
)0 5 10
0
0.1
0.2
0.3
0.4
m
p(m
)
m is about 5plus or minus 1.5m is eitherabout 3plus of minus 1or about 8plus or minus 1but that’s less likely
we don’t really know anything useful about m
![Page 58: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/58.jpg)
C: localized averages
A = 0.2m9 + 0.6m10 + 0.2m11might be better determined than eitherm9 or m10 or m11 individually
![Page 59: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/59.jpg)
Is this useful?
Do we care aboutA = 0.2m9 + 0.6m10 + 0.2m11?
Maybe …
![Page 60: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/60.jpg)
Suppose if m is a discrete approximation of m(x)
m(x)x
m10 m11m9
![Page 61: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/61.jpg)
m(x)x
m10 m11m9
A= 0.2m9 + 0.6m10 + 0.2m11weighted average of m(x)
in the vicinity of x10
x10
![Page 62: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/62.jpg)
m(x)x
m10 m11m9
average “localized’in the vicinity of x10
x10 weightsof weighted
average
![Page 63: Lecture 1 Describing Inverse Problems](https://reader035.fdocuments.us/reader035/viewer/2022081502/568161a5550346895dd162c2/html5/thumbnails/63.jpg)
Localized average meancan’t determine m(x) at x=10
but can determineaverage value of m(x) near x=10 Such a localized average might very
well be useful