Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Numerical Fourier transform based on hyperfunction theory

25 views

Published on

This is the PC slide of a contributed talk in the conference "ECMI2018 (The 20th European Conference on Mathematics for Industry)", 18-20 June 2018, Budapest, Hungary. In this talk, we propose a numerical method of Fourier transforms based on hyperfunction theory.

Published in: Engineering
  • Be the first to comment

  • Be the first to like this

Numerical Fourier transform based on hyperfunction theory

  1. 1. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Numerical Fourier Transform Based on Hyperfunction Theory Hidenori Ogata Dept. Computer and Network Engineering, The Graduate School of Informatics and Engineering, The Univerisity of Electro-Communications, Tokyo, Japan 21 June, 2018 Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  2. 2. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Aim of the study Aim of the study Numerical Fourier transform based on hyperfunction theory Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  3. 3. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Aim of the study Aim of the study Numerical Fourier transform based on hyperfunction theory Fourier transform F[f ](ξ) = ∞ −∞ f (x) exp(−2πiξx) dx. It is very familiar in science and engineering. But, it is difficult to compute it by conventional methods, especially, if f (x) decays slowly as x → ±∞. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  4. 4. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Fourier transforms are difficult to compute! Numerical integration by the DE rule. (1) ∞ 0 dx 1 + x2 , (2) ∞ 0 cos x 1 + x2 dx. -16 -14 -12 -10 -8 -6 -4 -2 0 0 10 20 30 40 50 60 70 log10(error) N (1) (2) vertical axis: log10(error) horizontal axis: number of sampling points N Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  5. 5. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Aim of the study Aim of the study Numerical Fourier transform based on hyperfunction theory Fourier transform F[f ](ξ) = ∞ −∞ f (x) exp(−2πiξx) dx. It is very familiar in science and engineering. But, it is difficult to compute it by conventional methods, especially, if f (x) decays slowly as x → ±∞. It is easy to compute it using hyperfunction theory. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  6. 6. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Contents Contents 1 Basis of hyperfunction theory 2 Fourier transform in hyperfunction theory 3 Numerical Fourier transform 4 Numerical examples 5 Summary Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  7. 7. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 1. Hyperfunction theory M. Sato, 1958 a theory of generalized functions based on complex function theory Hyperfunction the difference between the boundary values of an analytic function F(z) f (x) = [F(z)] ≡ F(x + i0) − F(x − i0). F(z) : defining function of the hyperfunction f (x) Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  8. 8. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 1. Hyperfunction: f (x) = F(x + i0) − F(x − i0) Examples δ(x) = − 1 2πi 1 x + i0 − 1 x − i0 = lim ǫ↓0 1 π ǫ x2 + ǫ2 , χ(−1,1)(x) = 1 x ∈ (−1, 1) 0 x ∈ [−1, 1] = − 1 2πi log (x + i0) + 1 (x + i0) − 1 − log (x − i0) + 1 (x − i0) − 1 . log z: the principal value, i.e., log x ∈ R for x > 0. -1 -0.5 0 0.5 1Re z -1 -0.5 0 0.5 1 Im z -6 -4 -2 0 2 4 6 Re F(z) -2 -1.5 -1 -0.5 0 0.5 1 1.5 2Re z -2-1.5-1-0.500.511.52 Im z -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Re F(z) defining function F(z) of δ(x) defining function F(z) of χ(−1,1)(x) Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  9. 9. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 2. Fourier transform in hyperfunction theory How to treat Fourier transforms in hyperfunction theory. F[f ](ξ) = ∞ −∞ f (x)e−2πiξx dx. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  10. 10. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 2. Fourier transform in hyperfunction theory How to treat Fourier transforms in hyperfunction theory. F[f ](ξ) = 0 −∞ f (x)e−2πi(ξ+i0)x dx + +∞ 0 f (x)e−2πi(ξ−i0)x dx . We can define F[f ](ξ) thanks to e−2πǫ|x| even if the integral is not convergent in the conventional sense. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  11. 11. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 2. Fourier transform in hyperfunction theory How to treat Fourier transforms in hyperfunction theory. F[f ](ξ) = 0 −∞ f (x)e−2πi(ξ+i0)x dx + +∞ 0 f (x)e−2πi(ξ−i0)x dx . We can define F[f ](ξ) thanks to e−2πǫ|x| even if the integral is not convergent in the conventional sense. (Example) F[1](ξ) = 0 −∞ e−2πi(ξ+i0)x dx + ∞ 0 e−2πi(ξ−i0)x dx = − 1 2πi 1 ξ + i0 − 1 ξ − i0 = δ(ξ). Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  12. 12. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 2. Fourier transform in hyperfunction theory How to treat Fourier transforms in hyperfunction theory. F[f ](ξ) = 0 −∞ f (x)e−2πi(ξ+i0)x dx + +∞ 0 f (x)e−2πi(ξ−i0)x dx . We can define F[f ](ξ) thanks to e−2πǫ|x| even if the integral is not convergent in the conventional sense. Fourier transform F[f ](ξ) Hyperfunction with the defining functions F±(ζ) F[f ](ξ) = F+(ξ + i0) − F−(ξ − i0), F+(ζ) = 0 −∞ f (x)e−2πiζx dx ( Im ζ > 0 ), F−(ζ) = − ∞ 0 f (x)e−2πiζx dx ( Im ζ < 0 ). Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  13. 13. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy We use the definition of Fourier transforms in hyperfunction theory for numerical Fourier transforms. F[f ](ξ) = ∞ −∞ f (x)e−2πiξx dx = F+(ξ + i0) − F−(ξ − i0), where F+(ζ) = 0 −∞ f (x)e−2πiζx dx ( Im ζ > 0 ), F−(ζ) = − ∞ 0 f (x)e−2πiζx dx ( Im ζ < 0 ). We can compute F±(ζ) in C because the integrands include the exponentially decaying factor exp(−2π| Im ζ||x|). Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  14. 14. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy We use the definition of Fourier transforms in hyperfunction theory for numerical Fourier transforms. F[f ](ξ) = ∞ −∞ f (x)e−2πiξx dx = F+(ξ + i0) − F−(ξ − i0), where F+(ζ) = 0 −∞ f (x)e−2πiζx dx ( Im ζ > 0 ), F−(ζ) = − ∞ 0 f (x)e−2πiζx dx ( Im ζ < 0 ). Strategy 1 Get F±(ζ) in { ζ ∈ C | ± Im ζ > 0 }. 2 Get F±(ξ ± i0) on R by analytic continuation. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  15. 15. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy 1/2 1. Get F±(ζ) in { ζ ∈ C | ± Im ζ > 0 }. F+(ζ) = 0 −∞ f (x)e−2πiζx dx ( Im ζ > 0 ), F−(ζ) = − ∞ 0 f (x)e−2πiζx dx ( Im ζ < 0 ). Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  16. 16. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy 1/2 1. Get F±(ζ) in { ζ ∈ C | ± Im ζ > 0 }. F±(ζ) = ± ∞ 0 f (∓x)e±2πiζx dx ( ± Im ζ > 0 ). Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  17. 17. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy 1/2 1. Get F±(ζ) in { ζ ∈ C | ± Im ζ > 0 }. F±(ζ) = ± ∞ 0 f (∓x)e±2πiζx dx ( ± Im ζ > 0 ). We get F±(ζ) in Taylor series. F±(ζ) = ∞ n=0 c (±) n (ζ − ζ (±) 0 )n ( ± Im ζ (±) 0 > 0 ), c (±) n = 1 n! F (n) ± (ζ (±) 0 ) = ± 1 n! ∞ 0 (±2πix)n f (∓x)e±2πiζ (±) 0 x exponential decay dx. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  18. 18. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy 2/2 2. Get F±(ξ ± i0) on R by analytic continuation. Taylor series → continued fraction F±(ζ) = ∞ n=0 cn(ζ − ζ (±) 0 )n = a (±) 0 1 + a (±) 1 (ζ − ζ (±) 0 ) 1 + a (±) 2 (ζ − ζ (±) 0 ) 1 + ... R R ζ (±) 0 convergence region c (±) n → a (±) n by quotient difference (QD) algorithm The QD algorithm is unstable. → multiple precision arithmetic Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  19. 19. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 3. Numerical Fourier transform: strategy 2/2 2. Get F±(ξ ± i0) on R by analytic continuation. Taylor series → continued fraction F±(ζ) = ∞ n=0 cn(ζ − ζ (±) 0 )n = a (±) 0 1 + a (±) 1 (ζ − ζ (±) 0 ) 1 + a (±) 2 (ζ − ζ (±) 0 ) 1 + ... R R ζ (±) 0 convergence region F[f ](ξ) = F+(ξ + i0) − F−(ξ − i0). Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  20. 20. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples (1) F[tanh(πx)](ξ) = −i cosech(πξ), (2) F[(1 + x2 )−ν−1/2 ] = 2πν+1/2 Γ(ν + 1/2) |ξ|ν Kν(2π|ξ|) ( ν = 1.5 ), (3) F[log |x|](ξ) = −γδ(ξ) − 1 2|ξ| . The center of the Taylor series of F±(ζ): ζ (±) 0 = ±i. C++ programs, multiple precision arithmetic (100 decimal digits). exflib: multiple precision arithmetic library. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  21. 21. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: errors of our method -60 -50 -40 -30 -20 -10 0 -4 -2 0 2 4 log10(error) xi (1) (2) (3) vertical axis: log10(error) horizontal axis: ξ (1) F[tanh(πx)](ξ), (2) F[(1+x2 )−ν−1/2 ](ξ), (3) F[log |x|]. The number of sampling points for numerical integration (the DE rule) (1) 1330 (2) 1416 (3) 1430. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  22. 22. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: an interesting fact We do not need to compute oscillatory integrals for Fourier transforms. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  23. 23. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: an interesting fact We do not need to compute oscillatory integrals for Fourier transforms. The defining function ( F[f ](ξ) = F+(ξ + i0) − F−(ξ − i0) ) F±(ζ) = ± ∞ 0 f (∓x)e±2πiζx dx = ∞ n=0 c (±) n (ζ ∓ i)n ( ± Im ζ > 0 ), c (±) n = ± 1 n! ∞ 0 (±2πix)n f (∓x) exp(−2πx)dx. including no oscillatory function. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  24. 24. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: comparisons Comparison with the previous methods 1 DE rule & Richardson extrapolation M. Sugihara, J. Comp. Appl. Math., 17 (1987) 47–68. F[f ](ξ) = lim n→∞ ∞ −∞ f (x) exp(−2πiξx)exp(−2−n x2 )dx. 2 DE-type rule for oscillatory integrals T. Ooura & M. Mori, J. Comp. Appl. Math., 38 (1991) 353–360. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  25. 25. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: comparisons F[tanh(πx)](ξ) = −i cosech(πξ), ξ = 1. multiple precision arithmetic (100 decimal digits, exflib) the center of the Taylor series ζ (±) 0 = 1 ± i. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  26. 26. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: comparisons F[tanh(πx)](ξ) = −i cosech(πξ), ξ = 1. multiple precision arithmetic (100 decimal digits, exflib) the center of the Taylor series ζ (±) 0 = 1 ± i. number of sampling points for integration error our method 2610 2.8 × 10−28 DE & Richardson 10060 9.1 × 10−20 DE for oscillatory integrals 948 5.0 × 10−25 Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  27. 27. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: comparisons number of sampling points for integration error our method 2610 2.8 × 10−28 DE & Richardson 10060 9.1 × 10−20 DE for oscillatory integrals 948 5.0 × 10−25 Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  28. 28. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary 4. Numerical examples: comparisons number of sampling points for integration error our method 2610 2.8 × 10−28 DE & Richardson 10060 9.1 × 10−20 DE for oscillatory integrals 948 5.0 × 10−25 Our method > DE & Richardson Our method < DE for oscillatory integrals Our method computes F[f ](ξ) as a function, i.e., once we obtain the coefficients a (±) n of the continued fraction, we can compute F[f ](ξ)’s for many ξ’s. The conventional methods compute F[f ](ξ) as an integrals. Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  29. 29. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Summary 1 Numerical Fourier transform based on hyperfunction theory 2 Fourier transform as a hyperfunction F[f ](ξ) = 0 −∞ f (x)e−2πi(ξ+i0)x dx + ∞ 0 f (x)e−2πi(ξ−i0)x dx. defining functions F±(ζ) Get F±(ζ) in { ζ ∈ C | ± Im ζ > 0 }. Get F±(ξ ± i0) on R by analytic continuation (by the continued fraction). 3 Numerical examples shows the effectiveness of our method. Problems for future studies 1 Theoretical error estimate 2 Analytic continuation by continued fractions Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory
  30. 30. Hyperfunction theory Fourier transform in hyperfunction theory Numerical Fourier transform Numerical examples Summary Thank you very much! Hidenori Ogata Numerical Fourier Transform Based on Hyperfunction Theory

×