Emalm 2005 05 02 Ffts PDF

The Discrete Fourier Transform
Group-Theoretic DFTs
Fast Fourier Transforms for Sn
Computational Results
Decimation-in-frequency Fast Fourier Transforms

for the Symmetric Group
Eric Malm
Presentation Days ’05

Thesis Advisor: Michael Orrison
Harvey Mudd College
02 May 2005
a·r·t
har vey·mudd·college
Eric Malm FFTs for the Symmetric Group

Overview
1 The Discrete Fourier Transform
Signal Analysis
Example
Significance of DFT
2 Group-Theoretic DFTs
Symmetries and Groups
Wedderburn’s Theorem
Generalized DFTs
3 Fast Fourier Transforms for Sn
Decimation-In-Frequency FFTs
Representation Theory of Sn
Construction of Factorization
4 Computational Results a·r·t

Signal Analysis
Example
Significance of DFT
Signal Analysis
The Setup
Suppose we want to analyze some periodic signal f
f HtL
t
1 2
a·r·t

Signal Analysis
Example
Significance of DFT
Signal Analysis
The Setup
• Pick some full time period of f
f HtL
t
1 2
a·r·t

Signal Analysis
Example
Significance of DFT
Signal Analysis
The Setup
• Pick some full time period of f
• Take N samples f0 , f1 , . . . , fN-1 of f in this time period
f HtL
t
1
a·r·t

Signal Analysis
Example
Significance of DFT
Discrete Fourier Transforms
æç f0 ö÷
çç ÷
çç f1 ÷÷÷
çç ÷
çç ¶ ÷÷÷
çç ÷÷
èfN-1 ø
• Process f0 , . . . , fN-1 with the Discrete Fourier Transform
a·r·t

Signal Analysis
Example
Significance of DFT
æç f0 ö÷
çç ÷ N-1
çç f1 ÷÷÷ input 1
çç
çç ¶ ÷÷÷
÷ ----® f̂k = â fj e-2Πijk / N
çç ÷÷ N
j= 0
èfN-1 ø
a·r·t

Signal Analysis
Example
Significance of DFT
æç f0 ö÷ æç f̂0 ö÷
çç ÷ N-1 çç ÷
çç f1 ÷÷÷ input 1 output çç f̂1 ÷÷÷
çç
çç ¶ ÷÷÷
÷ ----® f̂k = â fj e-2Πijk / N -----® çç
çç ¶ ÷÷÷
÷
çç ÷÷ N çç ÷÷
j= 0 ç ÷
èfN-1 ø f̂
è N-1 ø
• Get N complex numbers f̂0 , . . . f̂N-1 such that
N-1
2Πk 2Πk
f (t ) » â f̂k Kcos t + i sin tO
N N
k =0
a·r·t

Signal Analysis
Example
Significance of DFT
çç ÷ N-1 çç ÷
çç
çç ¶ ÷÷÷
÷ ----® f̂k = â fj e-2Πijk / N -----® çç
çç ¶ ÷÷÷
÷
çç ÷÷ N çç ÷÷
j= 0 ç ÷
èfN-1 ø f̂
è N-1 ø
N-1
2Πk 2Πk
N N
k =0
“pure” frequency
a·r·t

Signal Analysis
Example
Significance of DFT
çç ÷ N-1 çç ÷
çç
çç ¶ ÷÷÷
÷ ----® f̂k = â fj e-2Πijk / N -----® çç
çç ¶ ÷÷÷
÷
çç ÷÷ N çç ÷÷
j= 0 ç ÷
èfN-1 ø f̂
è N-1 ø
N-1
2Πk 2Πk
N N
k =0
amplitude
a·r·t

Signal Analysis
Example
Significance of DFT
Example
Example
Our original signal is secretly the sum of three “pure” frequencies:
1
t
1 2
f HtL
2 1
= + t
1
1 2
t
1 2
1
+ t
1 2
a·r·t

Signal Analysis
Example
Significance of DFT
Significance of the DFT

Time-Shift Invariance
Suppose we sampled our signal f over a different time period
• The samples f0 , . . . , fN-1 could be much different
• But the Fourier coefficients f̂0 , . . . , f̂N-1 will not be
The DFT is therefore invariant under translational symmetry
f HtL
t
12 32 a·r·t

Generalized DFTs

• Different spaces have different symmetries
Space Symmetry
time domain time translations
sphere rotations
lists permutations
A (1 3) C (1 3 2) B (2 3) B
B ---® B ----® A ---® C
C A C A a·r·t

Generalized DFTs

• Different spaces have different symmetries
• Write symmetries abstractly as groups
Space Symmetry Group

time domain time translations Z / NZ
sphere rotations SO(3)
lists permutations Sn
A (1 3) C (1 3 2) B (2 3) B
B ---® B ----® A ---® C
C A C A a·r·t

Generalized DFTs
Group Algebras
Reformulation as Group Algebra
• Treat functions on spaces as functions on groups
• Rewrite functions on group as group algebra elements:
f : X ® C --® f : G ® C --® â f (g) g

g ÎG
f HtL
t
0 1 2 3 4 5 6 7
a·r·t

Generalized DFTs
Theorem (Wedderburn)
The group algebra CG of a finite group G is isomorphic to an algebra
of block diagonal matrices:
h
CG @ Å Cdj ´dj
j= 1
Example (CS3 )
æçê ö÷
çç ÷
çç ê ê ÷÷÷
CS3 @ C1´1 Å C2´2 Å C1´1 = ççç ÷÷
çç ê ê ÷÷÷
ç ÷
è êø
a·r·t

Generalized DFTs
Generalized Discrete Fourier Transforms (DFTs)
Definition (Generalized DFT)

Any such isomorphism D on CG is a generalized DFT for G
• Coefficients in matrix D(f ): generalized Fourier coefficients
• Blocks along diagonal: smallest CG-invariant spaces in CG
Change of Basis
DFT a change of basis into a symmetry-invariant basis
• Picking standard bases on CG, matrix algebra gives DFT matrix
• Naïve bound of O(|G|2 ) on complexity of DFT evaluation
a·r·t

Decimation-In-Frequency Fast Fourier Transforms (FFTs)
Decimation-In-Frequency FFT
• Fix chain of subgroups of G:
1 = G0 < G1 < × × × < Gn-1 < Gn = G.
• Project into successively smaller subspaces in stages

corresponding to subgroups
• Goal: Obtain sparse factorization of change-of-basis matrix D
Sn an Ideal Proof-of-Concept Group
Nonabelian, representation theory well understood, natural chain of
subgroups
1 < S2 < S3 < × × × < Sn a·r·t har vey·mudd·college

Each block in matrix algebra for CSn corresponds to a different

non-increasing (proper) partition of n
Example (CS3 )
ê ê
CS3 @ IêM Å Kê êO Å IêM
3#
a·r·t

Character Graph for 1 < S2 < S3
Graded diagram of proper partitions for 1 < S2 < × × × < Sn
æçê ö÷
çç ÷
çç ê ê ÷÷÷
çç ÷
çç ê ê ÷÷÷
çç ÷÷
è êø
CS3
a·r·t

Each pathway through the diagram corresponds to a filled-in partition

and a row/column in matrix
1 2 3
æçê ö÷
çç ÷
çç ê ê ÷÷÷
çç ÷
çç ê ê ÷÷÷
çç ÷÷
è êø
1 2
CS3
1
a·r·t

Each pathway through the diagram corresponds to a filled-in partition

and a row/column in matrix
1 3
æçê ö÷ 2
çç ÷
çç ê ê ÷÷÷
çç ÷
çç ê ê ÷÷÷
çç ÷÷
è êø 1
2
CS3
1
a·r·t

A pair of paths ending at same diagram specifies a 1-D Fourier space
¯
æçê ö÷
ç ÷
® ççç ê ê ÷÷÷
çç ÷
çç ê ê ÷÷÷
çç ÷÷
è êø
CS3
a·r·t

Stages of Subspace Projections

• Partial paths give CSn subspaces
• At stage for Sk , project onto
these subspaces
• Build sparse factor from
projections
• Full paths by stage for Sn
ê ê
CS3 @ I ê M Å K O Å IêM
ê ê
a·r·t

Stages of Subspace Projections

• Partial paths give CSn subspaces
• At stage for Sk , project onto
these subspaces
• Build sparse factor from
projections
• Full paths by stage for Sn
ê ê
CS3 @ I ê M Å K O Å IêM
ê ê
a·r·t

Factorization of CS3 DFT Matrix
Full DFT matrix for CS3 :
æç1 1 1 1 1 1 ö÷
çç ÷÷
1 1 1
çç1 -1
çç 2
-2 2
- 12 ÷÷÷÷
çç ÷÷
çç0 0 - 3 3 3
- 32 ÷÷÷÷
çç 2 2 2
çç 1÷
÷
çç0 0 - 1 - 1 1 ÷
çç 2 2 2 2÷ ÷÷
çç 1 1 1 1 ÷÷
çç1 1 - 2
- 2
- 2
- 2÷
÷
çç ÷÷
ç ÷
è1 -1 -1 1 -1 1ø
a·r·t

Factorization of CS3 DFT Matrix
Three factors (CS2 on rows, CS2 on columns, CS3 on rows):
æç11 1
- 21
ö÷ æç1 1 1
ö÷ æç1 1
1 1
ö÷
çç ÷÷ çç ÷÷ çç ÷÷
çç 1 ÷÷ çç 1 -1 ÷÷ çç 1 1÷ ÷÷
çç ÷÷÷ ççç ÷ çç
çç 1 ÷ çç 1 -1÷ ÷÷ çç1 -1 ÷÷
÷÷
çç 1 1÷ ÷ç 1
÷÷ çç 1 -1 ÷÷
ç 2÷ ç ÷ç
è 1 -1 ø è 1 1ø è 1 -1ø
plus a permutation matrix and a row-scaling diagonal matrix
a·r·t

Factorization of CSn DFT Matrices
a·r·t

Factorization of CSn DFT Matrices
a·r·t

Results from Prototype Implementation
Prototype þÿ Implementation
Can compute exact FFT up to n = 6
Operation Counts For Evaluation

1
n tnfull tnDIF tnMaslen [1] 2
n(n - 1)
3 4.7 2.7 2.7 3
4 18.8 5.3 5.4 6
5 87.9 8.8 9.1 10
6 486.4 13.8 13.6 15
a·r·t

Future Directions
Theory
• Prove O(n2 |Sn |) bounds on operation counts
• Deduce better bases for blocks in factors
• Relate FFT on Sn to FFTs on Sn-1
Implementation
• Improve efficiency of þÿ implementation
• Port to MATLAB or GAP
• Parallelize decimation-in-frequency FFT algorithm
a·r·t

References
[1] David K. Maslen.
The efficient computation of fourier transforms on the symmetric group.
Mathematics of Computation, 67(223):1121–1147, July 1998.
[2] David K. Maslen and Daniel N. Rockmore.

Generalized FFTs - a survey of some recent results.
DIMACS Series in Discrete Mathematics and Computer Science, 28:183–237,
1997.
[3] Bruce E. Sagan.
The Symmetric Group: Representations, Combinatorial Algorithms, and
Symmetric Functions.
Wadsworth & Brooks/Cole, Pacific Grove, CA, 1991.
On The Web
Senior thesis website: Xhttp://www.math.hmc.edu/~emalm/thesis/\
a·r·t

Emalm 2005 05 02 Ffts PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Emalm 2005 05 02 Ffts PDF

Uploaded by

Copyright:

Available Formats

The Discrete Fourier Transform

Decimation-in-frequency Fast Fourier Transforms

Presentation Days ’05

Eric Malm FFTs for the Symmetric Group

Eric Malm FFTs for the Symmetric Group

Eric Malm FFTs for the Symmetric Group

Eric Malm FFTs for the Symmetric Group

Eric Malm FFTs for the Symmetric Group

Discrete Fourier Transforms

• Process f0 , . . . , fN-1 with the Discrete Fourier Transform

Eric Malm FFTs for the Symmetric Group

Discrete Fourier Transforms

• Process f0 , . . . , fN-1 with the Discrete Fourier Transform

Eric Malm FFTs for the Symmetric Group

Discrete Fourier Transforms

Eric Malm FFTs for the Symmetric Group

Discrete Fourier Transforms

Eric Malm FFTs for the Symmetric Group

Discrete Fourier Transforms

Eric Malm FFTs for the Symmetric Group

Eric Malm FFTs for the Symmetric Group

Significance of the DFT

Eric Malm FFTs for the Symmetric Group

Symmetries and Groups

Eric Malm FFTs for the Symmetric Group

Symmetries and Groups

Space Symmetry Group

Eric Malm FFTs for the Symmetric Group

f : X ® C --® f : G ® C --® â f (g) g

Eric Malm FFTs for the Symmetric Group

Eric Malm FFTs for the Symmetric Group

Generalized Discrete Fourier Transforms (DFTs)

Definition (Generalized DFT)

Eric Malm FFTs for the Symmetric Group

Decimation-In-Frequency Fast Fourier Transforms (FFTs)

1 = G0 < G1 < × × × < Gn-1 < Gn = G.

• Project into successively smaller subspaces in stages

Eric Malm FFTs for the Symmetric Group

Each block in matrix algebra for CSn corresponds to a different

Eric Malm FFTs for the Symmetric Group

Character Graph for 1 < S2 < S3

Graded diagram of proper partitions for 1 < S2 < × × × < Sn

Eric Malm FFTs for the Symmetric Group

Character Graph for 1 < S2 < S3

Each pathway through the diagram corresponds to a filled-in partition

Eric Malm FFTs for the Symmetric Group

Character Graph for 1 < S2 < S3

Each pathway through the diagram corresponds to a filled-in partition

Eric Malm FFTs for the Symmetric Group

Character Graph for 1 < S2 < S3

A pair of paths ending at same diagram specifies a 1-D Fourier space

Eric Malm FFTs for the Symmetric Group

Stages of Subspace Projections

Eric Malm FFTs for the Symmetric Group

Stages of Subspace Projections

Eric Malm FFTs for the Symmetric Group

Factorization of CS3 DFT Matrix

Full DFT matrix for CS3 :

Eric Malm FFTs for the Symmetric Group

Factorization of CS3 DFT Matrix

Three factors (CS2 on rows, CS2 on columns, CS3 on rows):

plus a permutation matrix and a row-scaling diagonal matrix

Eric Malm FFTs for the Symmetric Group