chebyshev expansions for solutions of linear differential
TRANSCRIPT
![Page 1: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/1.jpg)
1 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Chebyshev Expansions for Solutions ofLinear Differential Equations
Alexandre Benoit,Joint work with Bruno Salvy
INRIA
October 28, 2009
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 2: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/2.jpg)
2 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
I Introduction
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 3: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/3.jpg)
3 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
How to evaluate a function f in [−1, 1]?
Two representations of f :
in Taylor series
f =+∞∑n=0
cnxn, cn =f (n)(0)
n!,
or in Chebyshev series
f =+∞∑n=0
tnTn(x),
tn =2
π
∫ 1
−1
Tn(t)f (t)√1− t2
dt.
Projects using Chebyshev series to represent functions in Matlab :Chebfun, Miscfun.
How to compute tn?
General case: numerical computation of the integral. Slow.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 4: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/4.jpg)
3 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
How to evaluate a function f in [−1, 1]?
Two representations of f :
in Taylor series
f =+∞∑n=0
cnxn, cn =f (n)(0)
n!,
or in Chebyshev series
f =+∞∑n=0
tnTn(x),
tn =2
π
∫ 1
−1
Tn(t)f (t)√1− t2
dt.
Basic properties of Chebyshev polynomials
Tn(cos(θ)) = cos(nθ)
∫ 1
−1
Tn(x)Tm(x)√1− x2
dx =
0 if m 6= nπ if m = 0π2 otherwise
Tn+1 = 2xTn − Tn−1
T0(x) = 1
T1(x) = x
T2(x) = 2x2 − 1
T3(x) = 4x3 − 3x
Projects using Chebyshev series to represent functions in Matlab :Chebfun, Miscfun.
How to compute tn?
General case: numerical computation of the integral. Slow.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 5: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/5.jpg)
3 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
How to evaluate a function f in [−1, 1]?
Two representations of f :
in Taylor series
f =+∞∑n=0
cnxn, cn =f (n)(0)
n!,
or in Chebyshev series
f =+∞∑n=0
tnTn(x),
tn =2
π
∫ 1
−1
Tn(t)f (t)√1− t2
dt.
Projects using Chebyshev series to represent functions in Matlab :Chebfun, Miscfun.
How to compute tn?
General case: numerical computation of the integral. Slow.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 6: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/6.jpg)
3 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
How to evaluate a function f in [−1, 1]?
Two representations of f :
in Taylor series
f =+∞∑n=0
cnxn, cn =f (n)(0)
n!,
or in Chebyshev series
f =+∞∑n=0
tnTn(x),
tn =2
π
∫ 1
−1
Tn(t)f (t)√1− t2
dt.
Projects using Chebyshev series to represent functions in Matlab :Chebfun, Miscfun.
How to compute tn?
General case: numerical computation of the integral. Slow.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 7: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/7.jpg)
4 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Computation of Coefficients with Recurrences
Theorem (60’s)
If f is solution of a linear differential equation with polynomialcoefficients, then the Chebyshev coefficients are cancelled by alinear recurrence with polynomial coefficients.
Applications:
Numerical computation of the coefficients.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 8: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/8.jpg)
4 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Computation of Coefficients with Recurrences
Theorem (60’s)
If f is solution of a linear differential equation with polynomialcoefficients, then the Chebyshev coefficients are cancelled by alinear recurrence with polynomial coefficients.
Applications:
Numerical computation of the coefficients.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 9: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/9.jpg)
4 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Computation of Coefficients with Recurrences
Theorem (60’s)
If f is solution of a linear differential equation with polynomialcoefficients, then the Chebyshev coefficients are cancelled by alinear recurrence with polynomial coefficients.
Applications:
Numerical computation of the coefficients.
Computation of closed-form for coefficients.
Example (f (x) = arctan(x/2))
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 10: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/10.jpg)
5 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
State of the Art
Clenshaw (1957): numerical scheme to compute the Chebyshevcoefficients without computing all these integrals.
Fox and Parker (1968): method for the computation of theChebyshev recurrence relations for differential equations of smallorders.
Paszkowski (1975): algorithm for computing the Chebyshevrecurrence relation.
Lewanowicz (1976): algorithm for computing a smaller orderChebyshev recurrence relation in some cases.
Rebillard (1998): new algorithm for computing the Chebyshevrecurrence relation.
Rebillard and Zakrajsek (2006): algorithm for computing a smallerorder Chebyshev recurrence relation compared with Lewanowiczalgorithm.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 11: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/11.jpg)
6 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
New Results (2009)
Simple unified presentation of these algorithms using fractions ofrecurrence operators.
Complexity analysis of the existing algorithms (order k, degree k)
Paszkowski’s and Lewanowicz’s algorithms: O(k4) arithmeticoperations in Q.Rebillard’s algorithm: O(k5) arithmetic operations in Q.
New fast algorithm: O(kω) arithmetic operations. Here, ω is afeasible exponent for matrix multiplication with coefficients in Q(ω ≤ 3).
Implementation of algorithm in Maple.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 12: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/12.jpg)
7 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
II Fractions of Recurrence Operators
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 13: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/13.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Taylor series (f :=∑
cnxn)
xf =X
cnxn+1 =X
cn−1xn,
f ′ =X
ncnxn−1 =X
(n + 1)cn+1xn
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 14: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/14.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Taylor series (f :=∑
cnxn)
xf =X
cnxn+1 =X
cn−1xn,
f ′ =X
ncnxn−1 =X
(n + 1)cn+1xn
x 7→X := S−1,ddx 7→D := (n + 1)S .
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 15: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/15.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Taylor series (f :=∑
cnxn)
xf =X
cnxn+1 =X
cn−1xn,
f ′ =X
ncnxn−1 =X
(n + 1)cn+1xn
x 7→X := S−1,ddx 7→D := (n + 1)S .
(4 + x2)
„d
dx
«2
+ 2x ddx
7→(4+S−2)(n+1)(n+2)S2+2S−1(n + 1)S
= (n + 1)`4(n + 2)S2 + n
´4(n + 2)cn+2 + ncn = 0
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 16: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/16.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Monomial Basis xn = Mn(x)
xMn(x) = Mn+1(x),
(Mn(x))′ = nMn−1(x).
Chebyshev series
xTn(x) =1/2 (Tn+1(x) + Tn−1(x))
T ′n(x) =n (Tn−1(x)− Tn+1(x))
2(1− x2).
x 7→X := S−1,
d
dx7→D := (n + 1)S .
x 7→X :=S + S−1
2,
d
dx7→D :=
(n + 1)S − (n − 1)S−1
2(1− X 2)=
2n
S−1 − S.
(4 + x2)
„d
dx
«2
+ 2xd
dx
7→(4+S−2)(n+1)(n+2)S2+2S−1(n+1)S
= (n + 1)`4(n + 2)S2 + n
´4(n + 2)cn+2 + ncn = 0
(n − 1)(n + 1)`(n + 2)S2 + 18n + (n − 2)S−2
´((n − 1)S2 − 2n + (n + 1)S−2)
,
(n + 2)tn+2 + 18ntn + (n − 2)tn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 17: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/17.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Monomial Basis xn = Mn(x)
xMn(x) = Mn+1(x),
(Mn(x))′ = nMn−1(x).
Chebyshev series
xTn(x) =1/2 (Tn+1(x) + Tn−1(x))
T ′n(x) =n (Tn−1(x)− Tn+1(x))
2(1− x2).
x 7→X := S−1,
d
dx7→D := (n + 1)S .
x 7→X := S+S−1
2 ,
d
dx7→D :=
(n + 1)S − (n − 1)S−1
2(1− X 2)=
2n
S−1 − S.
(4 + x2)
„d
dx
«2
+ 2xd
dx
7→(4+S−2)(n+1)(n+2)S2+2S−1(n+1)S
= (n + 1)`4(n + 2)S2 + n
´4(n + 2)cn+2 + ncn = 0
(n − 1)(n + 1)`(n + 2)S2 + 18n + (n − 2)S−2
´((n − 1)S2 − 2n + (n + 1)S−2)
,
(n + 2)tn+2 + 18ntn + (n − 2)tn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 18: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/18.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Monomial Basis xn = Mn(x)
xMn(x) = Mn+1(x),
(Mn(x))′ = nMn−1(x).
Chebyshev series
xTn(x) =1/2 (Tn+1(x) + Tn−1(x))
T ′n(x) = n(Tn−1(x)−Tn+1(x))2(1−x2) .
x 7→X := S−1,ddx 7→D := (n + 1)S .
x 7→X :=S + S−1
2,
ddx 7→D := (n+1)S−(n−1)S−1
2(1−X 2) =2n
S−1 − S.
(4 + x2)
„d
dx
«2
+ 2xd
dx
7→(4+S−2)(n+1)(n+2)S2+2S−1(n+1)S
= (n + 1)`4(n + 2)S2 + n
´4(n + 2)cn+2 + ncn = 0
(n − 1)(n + 1)`(n + 2)S2 + 18n + (n − 2)S−2
´((n − 1)S2 − 2n + (n + 1)S−2)
,
(n + 2)tn+2 + 18ntn + (n − 2)tn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 19: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/19.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Monomial Basis xn = Mn(x)
xMn(x) = Mn+1(x),
(Mn(x))′ = nMn−1(x).
Chebyshev series
xTn(x) =1/2 (Tn+1(x) + Tn−1(x))
T ′n(x) =n (Tn−1(x)− Tn+1(x))
2(1− x2).
x 7→X := S−1,
d
dx7→D := (n + 1)S .
x 7→X :=S + S−1
2,
d
dx7→D :=
(n + 1)S − (n − 1)S−1
2(1− X 2)=
2n
S−1 − S.
(4 + x2)“
ddx
”2+ 2x d
dx
7→(4+S−2)(n+1)(n+2)S2+2S−1(n+1)S
= (n + 1)`4(n + 2)S2 + n
´4(n + 2)cn+2 + ncn = 0
(n−1)(n+1)((n+2)S2+18n+(n−2)S−2)((n−1)S2−2n+(n+1)S−2)
,
(n + 2)tn+2 + 18ntn + (n − 2)tn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 20: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/20.jpg)
8 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Morphisms of Rings of Operators (S · un = un+1)
Monomial Basis xn = Mn(x)
xMn(x) = Mn+1(x),
(Mn(x))′ = nMn−1(x).
Chebyshev series
xTn(x) =1/2 (Tn+1(x) + Tn−1(x))
T ′n(x) =n (Tn−1(x)− Tn+1(x))
2(1− x2).
x 7→X := S−1,
d
dx7→D := (n + 1)S .
x 7→X :=S + S−1
2,
d
dx7→D :=
(n + 1)S − (n − 1)S−1
2(1− X 2)=
2n
S−1 − S.
(4 + x2)
„d
dx
«2
+ 2xd
dx
7→(4+S−2)(n+1)(n+2)S2+2S−1(n+1)S
= (n + 1)`4(n + 2)S2 + n
´4(n + 2)cn+2 + ncn = 0
(n − 1)(n + 1)`(n + 2)S2 + 18n + (n − 2)S−2
´((n − 1)S2 − 2n + (n + 1)S−2)
,
(n + 2)tn+2 + 18ntn + (n − 2)tn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 21: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/21.jpg)
9 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Application to Chebyshev recurrences relations
Definition
Let ϕ be “the Chebyshev morphism”:
ϕ(x) =1
2
(S + S−1
)et ϕ
(d
dx
)=
2n
−S + S−1.
Theorem (BenoitSalvy2009)
f ∈ Ck , L is a differential operator of order k such that L · f = 0.Suppose that either of the following holds:
(i).
Z 1
−1
f (k)(x)√1− x2
dx is convergent;
(ii).
Z 1
−1
(1− x2)k f (k)(x)√1− x2
dx is convergent and (1− x2)i |pi , i = 0, . . . , k.
Then, the Chebyshev coefficients of f are cancelled by a numerator of ϕ(L).
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 22: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/22.jpg)
10 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Ore Polynomials: Framework for Recurrence Operators
∑ai (n)un+i is represented by
∑ai (n)S i .
These polynomials are non-commutative.
Multiplication defined by: Sn = (n + 1)S .
Ring denoted Q(n)〈S〉.
Main property: the degree in S of a product is the sum of thedegrees of its factors.
Algorithm for (left or right) euclidian division.
gcld algorithm (Ore 1933)
INPUT recurrence operators A and BOUTPUT The “greatest”G such thatA = G A and B = G B
lclm algorithm (Ore 1933)
INPUT recurrence operators A and BOUTPUT The “smallest ”U and Vsuch that UA = VB
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 23: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/23.jpg)
10 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Ore Polynomials: Framework for Recurrence Operators
∑ai (n)un+i is represented by
∑ai (n)S i .
These polynomials are non-commutative.
Multiplication defined by: Sn = (n + 1)S .
Ring denoted Q(n)〈S〉.Main property: the degree in S of a product is the sum of thedegrees of its factors.
Algorithm for (left or right) euclidian division.
gcld algorithm (Ore 1933)
INPUT recurrence operators A and BOUTPUT The “greatest”G such thatA = G A and B = G B
lclm algorithm (Ore 1933)
INPUT recurrence operators A and BOUTPUT The “smallest ”U and Vsuch that UA = VB
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 24: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/24.jpg)
11 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Fractions of Recurrence Operators (Ore 1933)
Field of fractions of Q(n)〈S〉 defined by:
A
B=
C
D⇔ ∃(U,V ) such that UA = VC and UB = VD.
Addition:A
B+
C
D=
UA
UB+
VC
VD=
UA + VC
UB,
Multiplication:D
C· A
B=
VD
VC· UA
UB=
UA
VC.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 25: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/25.jpg)
11 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Fractions of Recurrence Operators (Ore 1933)
Field of fractions of Q(n)〈S〉 defined by:
A
B=
C
D⇔ ∃(U,V ) such that UA = VC and UB = VD.
Addition:A
B+
C
D=
UA
UB+
VC
VD=
UA + VC
UB,
Multiplication:D
C· A
B=
VD
VC· UA
UB=
UA
VC.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 26: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/26.jpg)
11 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Fractions of Recurrence Operators (Ore 1933)
Field of fractions of Q(n)〈S〉 defined by:
A
B=
C
D⇔ ∃(U,V ) such that UA = VC and UB = VD.
Addition:A
B+
C
D=
UA
UB+
VC
VD=
UA + VC
UB,
Multiplication:D
C· A
B=
VD
VC· UA
UB=
UA
VC.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 27: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/27.jpg)
12 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Example:√
1− x2
f =√
1− x2 is cancelled by the differential operator : x + (1− x2) ddx .
ϕ
(x + (1− x2)
d
dx
)=
S + S−1
2+
(1− S2 + 2 + S−2
4
)2n
−S + S−1
=
(−S + S−1
) (S + S−1
)2 (−S + S−1)
− (n + 2)S2 + 2n − (n − 2)S−2
2 (−S + S−1)
=−(n + 3)S2 + 2n − (n − 3)S−2
2 (−S + S−1)
The Chebyshev coefficients cn satisfy :
(n + 3)cn+2 − 2ncn + (n − 3)cn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 28: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/28.jpg)
12 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Example:√
1− x2
f =√
1− x2 is cancelled by the differential operator : x + (1− x2) ddx .
ϕ
(x + (1− x2)
d
dx
)=
S + S−1
2+
(1− S2 + 2 + S−2
4
)2n
−S + S−1
=
(−S + S−1
) (S + S−1
)2(−S + S−1)
− (n + 2)S2 + 2n − (n − 2)S−2
2(−S + S−1)
=−(n + 3)S2 + 2n − (n − 3)S−2
2 (−S + S−1)
The Chebyshev coefficients cn satisfy :
(n + 3)cn+2 − 2ncn + (n − 3)cn−2 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 29: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/29.jpg)
13 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Normalization
Definition
A fraction AB is called normalized when the gcld of A and B is 1.
Example: Normalized fraction for√
1− x2
we have:
ϕ
(−x + (−1 + x2)
d
dx
)=−(n + 3)S2 + 2n − (n − 3)S−2
2 (−S + S−1)
=
(−S + S−1
)((n + 2)S − (n − 2)S−1
)2(−S + S−1)
.
Smaller order⇒ (n + 2)cn+1 − (n − 2)cn−1 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 30: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/30.jpg)
13 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Normalization
Definition
A fraction AB is called normalized when the gcld of A and B is 1.
Example: Normalized fraction for√
1− x2
we have:
ϕ
(−x + (−1 + x2)
d
dx
)=−(n + 3)S2 + 2n − (n − 3)S−2
2 (−S + S−1)
=
(−S + S−1
)((n + 2)S − (n − 2)S−1
)2(−S + S−1)
.
Smaller order⇒ (n + 2)cn+1 − (n − 2)cn−1 = 0.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 31: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/31.jpg)
14 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
III Algorithms
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 32: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/32.jpg)
15 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Lewanowicz’s algorithm (1976)
Horner+Normalize at each step.Example with f =
√1− x2
(1− x2)d
dx+ x .
ϕ(1− x2) =−S2 + 2− S−2
4=
(S + S−1)(−S + S−1)
4
ϕ(1− x2)ϕ
(d
dx
)=
((n + 1)S − (n − 1)S−1
)2
ϕ(1− x2)ϕ
(d
dx
)+ ϕ(x) =
(n + 2)S − (n − 2)S−1
2
A recurrence verified by the Chebyshev coefficients of f is:
(n + 2) cn+1 − (n − 2) cn−1 = 0
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 33: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/33.jpg)
15 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Lewanowicz’s algorithm (1976)
Horner+Normalize at each step.Example with f =
√1− x2
(1− x2)d
dx+ x .
ϕ(1− x2) =−S2 + 2− S−2
4=
(S + S−1)(−S + S−1)
4
ϕ(1− x2)ϕ
(d
dx
)=
((n + 1)S − (n − 1)S−1
)2
ϕ(1− x2)ϕ
(d
dx
)+ ϕ(x) =
(n + 2)S − (n − 2)S−1
2
A recurrence verified by the Chebyshev coefficients of f is:
(n + 2) cn+1 − (n − 2) cn−1 = 0
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 34: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/34.jpg)
15 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Lewanowicz’s algorithm (1976)
Horner+Normalize at each step.Example with f =
√1− x2
(1− x2)d
dx+ x .
ϕ(1− x2) =−S2 + 2− S−2
4=
(S + S−1)(−S + S−1)
4
ϕ(1− x2)ϕ
(d
dx
)=
(S + S−1)(−S + S−1)
4
2n
−S + S−1
=
((n + 1)S − (n − 1)S−1
)2
ϕ(1− x2)ϕ
(d
dx
)+ ϕ(x) =
(n + 2)S − (n − 2)S−1
2
A recurrence verified by the Chebyshev coefficients of f is:
(n + 2) cn+1 − (n − 2) cn−1 = 0
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 35: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/35.jpg)
15 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Lewanowicz’s algorithm (1976)
Horner+Normalize at each step.Example with f =
√1− x2
(1− x2)d
dx+ x .
ϕ(1− x2) =−S2 + 2− S−2
4=
(S + S−1)(−S + S−1)
4
ϕ(1− x2)ϕ
(d
dx
)=
((n + 1)S − (n − 1)S−1
)2
ϕ(1− x2)ϕ
(d
dx
)+ ϕ(x) =
((n + 1)S − (n − 1)S−1
)2
+S + S−1
2
=(n + 2)S − (n − 2)S−1
2
A recurrence verified by the Chebyshev coefficients of f is:
(n + 2) cn+1 − (n − 2) cn−1 = 0
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 36: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/36.jpg)
15 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Lewanowicz’s algorithm (1976)
Horner+Normalize at each step.Example with f =
√1− x2
(1− x2)d
dx+ x .
ϕ(1− x2) =−S2 + 2− S−2
4=
(S + S−1)(−S + S−1)
4
ϕ(1− x2)ϕ
(d
dx
)=
((n + 1)S − (n − 1)S−1
)2
ϕ(1− x2)ϕ
(d
dx
)+ ϕ(x) =
(n + 2)S − (n − 2)S−1
2
A recurrence verified by the Chebyshev coefficients of f is:
(n + 2) cn+1 − (n − 2) cn−1 = 0
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 37: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/37.jpg)
16 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Algorithms of Paszkowski (1975) and Rebillard (1998)
Observation: if D = ϕ( ddx ) = 2n
−S+S−1 then D−1 is a polynomial.
INPUT :
L =k∑
i=0
pi (x)
(d
dx
)i
OUTPUT : A numerator of ϕ(L)
Computation with polynomials only.
Paszkowski
Compute qi (x) such that
k∑i=0
pi (x)
(d
dx
)i
=k∑
i=0
(d
dx
)i
qi (x).
k∑i=0
pi (X )D i =
k∑i=0
D−k+iqi (X )
D−k.
Rebillard
Xk := D−kXDk .
k∑i=0
pi (X )D i =
k∑i=0
pi (Xk)D−k+i
D−k.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 38: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/38.jpg)
16 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Algorithms of Paszkowski (1975) and Rebillard (1998)
Observation: if D = ϕ( ddx ) = 2n
−S+S−1 then D−1 is a polynomial.
INPUT :
L =k∑
i=0
pi (x)
(d
dx
)i
OUTPUT : A numerator of ϕ(L)
Computation with polynomials only.
Paszkowski
Compute qi (x) such that
k∑i=0
pi (x)
(d
dx
)i
=k∑
i=0
(d
dx
)i
qi (x).
k∑i=0
pi (X )D i =
k∑i=0
D−k+iqi (X )
D−k.
Rebillard
Xk := D−kXDk .
k∑i=0
pi (X )D i =
k∑i=0
pi (Xk)D−k+i
D−k.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 39: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/39.jpg)
16 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Algorithms of Paszkowski (1975) and Rebillard (1998)
Observation: if D = ϕ( ddx ) = 2n
−S+S−1 then D−1 is a polynomial.
INPUT :
L =k∑
i=0
pi (x)
(d
dx
)i
OUTPUT : A numerator of ϕ(L)
Computation with polynomials only.
Paszkowski
Compute qi (x) such that
k∑i=0
pi (x)
(d
dx
)i
=k∑
i=0
(d
dx
)i
qi (x).
k∑i=0
pi (X )D i =
k∑i=0
D−k+iqi (X )
D−k.
Rebillard
Xk := D−kXDk .
k∑i=0
pi (X )D i =
k∑i=0
pi (Xk)D−k+i
D−k.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 40: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/40.jpg)
17 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Our algorithm: Divide and conquer
D−i is of bidegree (2i , 2i).
New, fast algorithm
Step 1: Compute qi (x) such that
k∑i=0
pi (x)
(d
dx
)i
=k∑
i=0
(d
dx
)i
qi (x).
Step 2 : Divide and conquer
k∑i=0
D−k+iqi (X ) =
D−k2
k/2∑i=0
D−k2 +iqi (X )+
k∑i=k/2+1
D−k+iqi (X ).
Balanced products.
Theorem
If the degrees of pi are at most k,
New: O(kω) arithmeticoperations.
Paszkowski and Lewanowiczalgorithms : O(k4) arithmeticoperations.
Rebillard : O(k5) arithmeticoperations.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 41: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/41.jpg)
17 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Our algorithm: Divide and conquer
D−i is of bidegree (2i , 2i).
New, fast algorithm
Step 1: Compute qi (x) such that
k∑i=0
pi (x)
(d
dx
)i
=k∑
i=0
(d
dx
)i
qi (x).
Step 2 : Divide and conquer
k∑i=0
D−k+iqi (X ) =
D−k2
k/2∑i=0
D−k2 +iqi (X )+
k∑i=k/2+1
D−k+iqi (X ).
Balanced products.
Theorem
If the degrees of pi are at most k,
New: O(kω) arithmeticoperations.
Paszkowski and Lewanowiczalgorithms : O(k4) arithmeticoperations.
Rebillard : O(k5) arithmeticoperations.
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 42: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/42.jpg)
18 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
IV Conclusion and Future Works
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations
![Page 43: Chebyshev Expansions for Solutions of Linear Differential](https://reader033.vdocuments.net/reader033/viewer/2022050214/626e424d45c2fe643864233e/html5/thumbnails/43.jpg)
19 / 19
Introduction Fractions of Recurrence Operators Algorithms Conclusion and Future Works
Conclusion and Future works
Contributions:
Use of fractions of recurrence operators.
New algorithm.
Maple code.
Available in http://ddmf.msr-inria.inria.fr/
Perspectives:
Numerical computation of the coefficients.
Compare our algorithm with the algorithm of Rebillard andZakrajsek.
Recurrence in other bases (Jacobi, Hermite and Laguerrepolynomials, Bessel functions)
Alexandre Benoit Chebyshev Expansions for Solutions of Linear Differential Equations