Lecture 21 Continuous Problems Fr é chet Derivatives
description
Transcript of Lecture 21 Continuous Problems Fr é chet Derivatives
![Page 1: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/1.jpg)
Lecture 21
Continuous Problems
Fréchet Derivatives
![Page 2: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/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 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/3.jpg)
Purpose of the Lecture
use adjoint methods to compute
data kernels
![Page 4: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/4.jpg)
Part 1
Review of Last Lecture
![Page 5: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/5.jpg)
a function
m(x)
is the continuous analog of a vector
m
![Page 6: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/6.jpg)
a linear operator
ℒis the continuous analog of a matrix
L
![Page 7: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/7.jpg)
a inverse of a linear operator
ℒ-1
is the continuous analog of the inverse of a matrix
L-1
![Page 8: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/8.jpg)
a inverse of a linear operatorcan be used to solve
a differential equation
if ℒm=f then m=ℒ-1fjust as the inverse of a matrix
can be used to solvea matrix equation
if Lm=f then m=L-1f
![Page 9: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/9.jpg)
the inner product
is the continuous analog of dot product
s= aTb
![Page 10: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/10.jpg)
the adjoint of a linear operator
is the continuous analog of the transpose of a matrix
LT
ℒ †
![Page 11: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/11.jpg)
the adjoint can be used tomanipulate an inner product
just as the transpose can be used to manipulate the dot product
(La) Tb= a T(LTb)
(ℒa, b) =(a, ℒ†b)
![Page 12: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/12.jpg)
table of adjoints
c(x)-d/dxd2/dx2
c(x)d/dxd2/dx2
![Page 13: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/13.jpg)
Part 2
definition of the Fréchet derivatives
![Page 14: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/14.jpg)
first rewrite the standard inverse theory equation in terms of perturbations
a small change in the modelcauses a small change in the data
![Page 15: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/15.jpg)
second compare with the standard formula for a
derivative
![Page 16: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/16.jpg)
third identify the data kernel as
a kind of derivative
this kind of derivative is called aFréchet derivative
![Page 17: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/17.jpg)
definition of a Fréchet derivative
this is mostly lingothough perhaps it adds a little insight about
what the data kernel is
,
![Page 18: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/18.jpg)
Part 2
Fréchet derivative of Error
![Page 19: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/19.jpg)
treat the data as a continuous function d(x)then the standard L2 norm error is
![Page 20: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/20.jpg)
let the data d(x) be related to the model m(x)by
could be the data kernel integral
=
because integrals are linear operators
![Page 21: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/21.jpg)
to
a perturbation in the modelcauses
a perturbation in the error
now do a little algebra to relate
![Page 22: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/22.jpg)
ifm(0) implies d(0) with error E(0)
then ...
![Page 23: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/23.jpg)
ifm(0) implies d(0) with error E(0)
then ...
all thisis just
algebra
![Page 24: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/24.jpg)
ifm(0) implies d(0) with error E(0)
then ...
use δd = ℒδm
![Page 25: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/25.jpg)
ifm(0) implies d(0) with error E(0)
then ...
use adjoint
![Page 26: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/26.jpg)
ifm(0) implies d(0) with error E(0)
then ...
Fréchet derivative of Error
![Page 27: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/27.jpg)
you can use this derivative to solve and inverse problem using the
gradient method
![Page 28: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/28.jpg)
example
![Page 29: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/29.jpg)
example
this is the relationship between model and data
d(x) =
![Page 30: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/30.jpg)
example
construct adjoint
![Page 31: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/31.jpg)
example
Fréchet derivative of Error
![Page 32: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/32.jpg)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-2
0
2
x
m(x
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
5
x
d(x)
10 20 30 40 50 60 70 80 90 100-4
-2
0
iteration
log1
0 er
ror,
Em(x)
d(x) x(A)
(B)
(C)
log 10
E/E 1 xiteration
![Page 33: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/33.jpg)
Part 3
Backprojection
![Page 34: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/34.jpg)
continuous analog of least squares
![Page 35: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/35.jpg)
now define the identity operator ℐm(x) = ℐ m(x)
![Page 36: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/36.jpg)
view as a recursion
![Page 37: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/37.jpg)
view as a recursion
using the adjoint as if it
were the inverse
![Page 38: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/38.jpg)
example
backprojection
exactm(x) = ℒ-1 dobs = d dobs / dx
![Page 39: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/39.jpg)
example
backprojection
exactm(x) = ℒ-1 dobs = d dobs / dx
crazy!
![Page 40: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/40.jpg)
0 20 40 60 80 100 120 140 160 180 200
-0.50
0.5
x
m
0 20 40 60 80 100 120 140 160 180 200
-0.50
0.5
x
m
0 20 40 60 80 100 120 140 160 180 200-1
0
1
x
m
(A)
(B)
(C)
m(x)true
mest (x )m(1) (x
)
xxx
![Page 41: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/41.jpg)
interpretation as tomography
backprojection
m is slownessd is travel time of a ray from –∞ to x
integrate (=add together) the travel times of all rays that pass through the point x
![Page 42: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/42.jpg)
discrete analysis
Gm=dG= UΛVT G-g= VΛ-1UT GT= VΛUT
if Λ-1≈ Λ then G-g≈ GT backprojection works when the
singular values are all roughly the same size
![Page 43: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/43.jpg)
suggests scalingGm=d → WGm=Wdwhere W is a diagonal matrix chosen to make the singular values more equal in overall size
Traveltime tomography:Wii = (length of ith ray)-1
so [Wd]i has interpretation of the average slowness
along the ray i.Backprojection now adds together the average slowness of all rays that
interact with the point x.
![Page 44: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/44.jpg)
0 20 40 60
0
10
20
30
40
50
60
y
x
true model
0 20 40 60
0
10
20
30
40
50
60
y
x
LS estimate
0 20 40 60
0
10
20
30
40
50
60
y
x
BP estimate(A) (B) (C)
y
x
yx
yx
![Page 45: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/45.jpg)
Part 4
Fréchet Derivative
involving a differential equation
![Page 46: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/46.jpg)
Part 4
Fréchet Derivative
involving a differential equation
seismic wave equationNavier-Stokes equation of fluid flow
etc
![Page 47: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/47.jpg)
data d is related to field u via an inner product
field u is related to model parameters m via a differential equation
![Page 48: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/48.jpg)
pertrubation δd is related to perturbation δu via an inner product
write in terms of perturbations
perturbation δu is related to perturbation δm via a differential equation
![Page 49: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/49.jpg)
what’s the data kernel ?
![Page 50: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/50.jpg)
easy using adjoints
data inner product with field
![Page 51: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/51.jpg)
easy using adjoints
data is inner product with field
field satisfies ℒδu= δm
![Page 52: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/52.jpg)
easy using adjoints
data is inner product with field
field satisfies ℒδu= δm
employ adjoint
![Page 53: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/53.jpg)
easy using adjoints
data is inner product with field
field satisfies ℒδu= δm
employ adjoint
inverse of adjoint is adjoint ofinverse
![Page 54: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/54.jpg)
easy using adjoints
data is inner product with field
field satisfies ℒδu= δm
employ adjoint
inverse of adjoint is adjoint ofinverse
data kernel
![Page 55: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/55.jpg)
easy using adjoints
data is inner product with field
field satisfies ℒδu= δm
employ adjoint
inverse of adjoint is adjoint ofinverse
data kernel
data kernel satisfies “adjoint differential equation
![Page 56: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/56.jpg)
most problem involving differential equations are solved numerically
so instead of just solving
you must solve
and
![Page 57: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/57.jpg)
so there’s more work
but the same sort of work
![Page 58: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/58.jpg)
exampletime t instead of position x
field solves a Newtonian-type heat flow equationwhere u is temperature and m is heating
data is concentration of chemical whose production rate is proportional to temperature
![Page 59: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/59.jpg)
exampletime t instead of position x
field solves a Newtonian-type heat flow equationwhere u is temperature
data is concentration of chemical whose production rate is proportional to temperature
= (bH(ti-t), u)so hi= bH(ti-t)
![Page 60: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/60.jpg)
we will solve this problemanalytically
using Green functions
in more complicated casesthe differential equation
must be solved numerically
![Page 61: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/61.jpg)
Newtonian equation
its Green function
![Page 62: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/62.jpg)
adjoint equation
its Green function
![Page 63: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/63.jpg)
note that the adjoint Green function
is the original Green function
backward in time
that’s a fairly common patternwhose significance will be pursued in a homework problem
![Page 64: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/64.jpg)
we must perform a Green function integral to compute the data kernel
![Page 65: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/65.jpg)
![Page 66: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/66.jpg)
time, trow,
i
![Page 67: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/67.jpg)
0 50 100 150 200 250 300 350 400 450 5000
0.5
1
t
H(t
,10)
0 50 100 150 200 250 300 350 400 450 5000
1
2
t
m(t
)
0 50 100 150 200 250 300 350 400 450 5000
20
t
u(t)
0 50 100 150 200 250 300 350 400 450 5000
1000
2000
t
d(t)
(A)
(B)
(C)
(D)
time, ttime, t
time, ttime, t
F(t,τ=30
)m(t)
u(t )d(t )
![Page 68: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/68.jpg)
Part 4
Fréchet Derivative
involving a parameter indifferential equation
![Page 69: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/69.jpg)
Part 4
Fréchet Derivative
involving a parameter indifferential equation
![Page 70: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/70.jpg)
previous example
unknown functi on is “forcing”
another possibility
forcing is knownparameter
is unknown
![Page 71: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/71.jpg)
linearize around a simpler equation
and assume you can solve this equation
![Page 72: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/72.jpg)
the perturbed equation is
subtracting out the unperturbed equation,ignoring second order terms, and rearranging gives ...
![Page 73: Lecture 21 Continuous Problems Fr é chet Derivatives](https://reader035.fdocuments.in/reader035/viewer/2022081503/56815d6e550346895dcb760b/html5/thumbnails/73.jpg)
then approximately
pertubation to parameter acts as
an unknown forcing
so it is back to the form of a forcingand the previous methodology can be applied