FMM CMSC 878R/AMSC 698R - UMD
Transcript of FMM CMSC 878R/AMSC 698R - UMD
![Page 1: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/1.jpg)
1
CMSC 858M/AMSC 698R Fast Multipole Methods
Nail A. Gumerov & Ramani Duraiswami
Lecture 7
Outline
• Representation of functions in the space of coefficients
• Matrix representation of operators
• Truncation and truncated operators
• Translation operator
• Reexpansion coefficients
• R|R and S|S translation operators
• Examples
• S|R and R|S translation operators
• Properties of translation operators
![Page 2: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/2.jpg)
2
Why do we need represent
functions in different spaces?
• Functions should be efficiently summed up;
• Sums of functions should be compressed;
• Error bounds should be established;
• Functions should be translated and expanded over different bases;
• For computations we need discrete and finite function representations.
• Some functions measured experimentally or approximated by splines, and there is no explicit analytical representation in the whole space.
Linear Spaces
![Page 3: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/3.jpg)
3
Linear Operators
Linear Spaces
Operator
Linear Operator
An example of linear operator: Differential Operator.
Representation of Functions and
Operators Bases
Reexpansion Coefficients
Matrix Representation
of operator A
![Page 4: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/4.jpg)
4
Function Representation in the Space of
Coefficients
Function Representation in the Space of
Coefficients (2)
C(W)
F(W)
R
A(W)
![Page 5: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/5.jpg)
5
p-Truncated Vectors
F(W)
Fp(W)
A(W)
Rp
Dense in F(W)
Matrix Representation of Linear Operators
F(W) A(W)
A(W’) F(W’)
F
Representation of a Linear Operator
![Page 6: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/6.jpg)
6
p-Truncation (Projection) Operator
F(W) A(W)
Rp(W) Fp(W) Pr(p)
Norm of p-Truncation Operator
(important for error bounds)
![Page 7: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/7.jpg)
7
p-Truncated Operator
Norm of p-Truncated Operator
(important for error bounds)
![Page 8: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/8.jpg)
8
Translation Operator
y t
y+t
Example of Translation Operator
y
F(y) F(y+t)
t
T(t)
![Page 9: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/9.jpg)
9
R|R-reexpansion
Example of R|R-reexpansion
![Page 10: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/10.jpg)
10
R|R-translation operator
Why the same operator named
differently?
The first letter shows
the basis for F(y)
The second letter
shows the basis
for F(y +t)
Needed only to show the expansion basis
(for operator representation)
![Page 11: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/11.jpg)
11
Matrix representation of
R|R-translation operator
Consider
Coefficients of
shifted function
Coefficients of
original function
Reexpansion of the same
function over shifted basis
We have:
![Page 12: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/12.jpg)
12
R|R-reexpansion of the same
function over shifted basis (2)
|y - x* - t| < r 1 = r - |t|
W 1
W 2
x *1
x *2
R (R|R)
x R
Wr(x*)
x *
(R|R)
y x*+t
t
r
Wr1(x*+t)
Since Wr1(x*+t) Wr(t) !
Original expansion
Is valid only here! r1
Example of power series
reexpansion
![Page 13: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/13.jpg)
13
S|S-reexpansion
S|S-translation operator
![Page 14: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/14.jpg)
14
S|S and R|R-translation operators
are very similar,
(actually, this is just two representations of
the same translation operator in different domains and bases)
But picture is different…
x i
W 1
W 2
x *1
x *2
S
(S|S)
S
x i
x *
x*+t
(S|S)
y
t
Wr1(x*+t)
Wr(x*)
r r1
|y - x* - t| > r 1 = r + |t|
Since
Wr1(x*+t) Wr(t) !
Original expansion
Is valid only here!
Also
|xi - x* | < r
singular point !
![Page 15: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/15.jpg)
15
S|R-reexpansion
Does R|S reexpansion exist?
• Theoretically yes (in some cases, e.g. analytical continuation);
• In practice, since the domain of S-expansion is larger
then the domain of R-expansion, this either
not useful (due to error bounds), or can be avoided in algorithms;
• We will not use R|S-reexpansions in the FMM algorithms.
![Page 16: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/16.jpg)
16
S|R-translation operator
S|R-operator has almost the same
properties as S|S and R|R
(t cannot be zero)
![Page 17: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/17.jpg)
17
Picture is different…
x i
W 1
x *1
x *2
S
(S|S)
S
x i
x *
x*+t
(S|R)
y
t
Wr1(x*+t)
Wr(x*)
r
r1
|y - x* - t| < r 1 = |t| - r
Since
Wr1(x*+t) Wr(t) !
Original expansion
Is valid only here!
Also
|xi - x* | < r
singular point !
Properties of the translation
operator
![Page 18: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/18.jpg)
18
Spectrum of the translation
operator eigen function eigen value
derivative in direction s
R-expansion
S-expansion
x*
xi R S
Example from previous lectures
![Page 19: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/19.jpg)
19
In this case we have
Norm of the Translation Operator
![Page 20: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/20.jpg)
20
Active and Passive points of view
on translation operator
y
F(y) F(y+t)
t
T(t)
y
F(y)
t
T(t)
``Active” point of view:
Operator transforms
function.
The reference frame
Does not change.
``Passive” point of view:
Function does not change.
Operator transforms
the reference frame.
Norms of R|R, S|S, and S|R-
operators (1)
y
F(y)
t
T(t)
xi
singular point of F(y)
W
W’
![Page 21: FMM CMSC 878R/AMSC 698R - UMD](https://reader034.fdocuments.us/reader034/viewer/2022050420/62711722072cfd5be73b6944/html5/thumbnails/21.jpg)
21
Norms of R|R, S|S, and S|R-
operators (2)
From the passive point of view, the translation operator does nothing,
but just changes the reference frame. So if we consider that R|R, S|S,
and S|R do just change of the reference frame PLUS they shrink
the domain, where the function is bounded, then their norms
do not exceed 1.
This is the difference
between general
translation operator
and R|R, S|S, and S|R
operators.
Error of exact R|R, S|S, and S|R-
translation