Professional Documents
Culture Documents
Silvia Bertoluzza
IMATI Enrico Magenes - CNR, Pavia
Joint work with Val erie Perrier, Lab. Jean Kuntzmann, INP Grenoble
1 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996]
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy Applied to a wide class of problems
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy Applied to a wide class of problems
Fluid dynamics
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy Applied to a wide class of problems
Fluid dynamics Elasticity
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy Applied to a wide class of problems
Fluid dynamics Elasticity Fluid structure interaction
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy Applied to a wide class of problems
Fluid dynamics Elasticity Fluid structure interaction Semiconductors
2 / 24
Wavelet Collocation
Introduced O. Vasilyev & S. Paolucci [1995] and by S.B. & G. Naldi [1996] Two ingredients
Use of Deslaurier-Dubuc interpolating functions Use of a collocation approach
Pointwise nonlinerities are easily treated (no integral computation) Very simple yet eective rening/derening strategy Applied to a wide class of problems
Fluid dynamics Elasticity Fluid structure interaction Semiconductors
Almost no theory
Silvia Bertoluzza (IMATI-CNR) Wavelet collocation for fourth order problems 2 / 24
Wavelet Collocation
Basis: translated of
f f f f
3 / 24
Wavelet Collocation
Basis: translated of
f f f f
coarser space Vj 1
T t t t t t t t t t t t t t t t t t t t t
Silvia Bertoluzza (IMATI-CNR)
Basis: translated of
t t t t
3 / 24
Wavelet Collocation
4 / 24
Wavelet Collocation
4 / 24
Wavelet Collocation
Dierence
T f f f f f ff f f f f f ff f f f f f ff f f f f f ff
4 / 24
Wavelet Collocation
5 / 24
Wavelet Collocation
6 / 24
Wavelet Collocation
6 / 24
Wavelet Collocation
6 / 24
Wavelet Collocation
is the autocorrelation function of the Daubechies scaling function with L 1 zero moments
6 / 24
Wavelet Collocation
is the autocorrelation function of the Daubechies scaling function with L 1 zero moments is interpolatory ((k ) = k 0 )
Silvia Bertoluzza (IMATI-CNR) Wavelet collocation for fourth order problems 6 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N}
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1 detail space: Wj = (Ij +1 Ij )Vj +1
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1 detail space: Wj = (Ij +1 Ij )Vj +1 Wj : span of {j ,k = (2j x k ) = (2(2j x k ) 1), k N}
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1 detail space: Wj = (Ij +1 Ij )Vj +1 Wj : span of {j ,k = (2j x k ) = (2(2j x k ) 1), k N} Well dened correspondence between dyadic points and basis functions:
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1 detail space: Wj = (Ij +1 Ij )Vj +1 Wj : span of {j ,k = (2j x k ) = (2(2j x k ) 1), k N} Well dened correspondence between dyadic points and basis functions:
at xed coarse level j0 x = k 2j0 j0 ,k
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1 detail space: Wj = (Ij +1 Ij )Vj +1 Wj : span of {j ,k = (2j x k ) = (2(2j x k ) 1), k N} Well dened correspondence between dyadic points and basis functions:
at xed coarse level j0 x = k 2j0 j0 ,k for j > j0 and odd k = 2n + 1 x = k 2j j 1,n
7 / 24
Wavelet Collocation
DD interpolating wavelets
Vj : span of {jk , k N} Ij : C 0 (R) Vj Langrangian interpolation operator Vj Vj +1 detail space: Wj = (Ij +1 Ij )Vj +1 Wj : span of {j ,k = (2j x k ) = (2(2j x k ) 1), k N} Well dened correspondence between dyadic points and basis functions:
at xed coarse level j0 x = k 2j0 j0 ,k for j > j0 and odd k = 2n + 1 x = k 2j j 1,n
Wavelet Collocation
8 / 24
Wavelet Collocation
8 / 24
Wavelet Collocation
given values at xj ,k = k 2j , k = 0, . . . , 2j you want to uniquely select one function in Vj obtain values at xj ,k outside [0, 1] by extrapolating values in [0, 1]
8 / 24
Wavelet Collocation
given values at xj ,k = k 2j , k = 0, . . . , 2j you want to uniquely select one function in Vj obtain values at xj ,k outside [0, 1] by extrapolating values in [0, 1] this yields subspace Ej Vj
8 / 24
Wavelet Collocation
given values at xj ,k = k 2j , k = 0, . . . , 2j you want to uniquely select one function in Vj obtain values at xj ,k outside [0, 1] by extrapolating values in [0, 1] this yields subspace Ej Vj Vj [0, 1] = Ej |[0,1]
8 / 24
Wavelet Collocation
given values at xj ,k = k 2j , k = 0, . . . , 2j you want to uniquely select one function in Vj obtain values at xj ,k outside [0, 1] by extrapolating values in [0, 1] this yields subspace Ej Vj Vj [0, 1] = Ej |[0,1]
8 / 24
Wavelet Collocation
given values at xj ,k = k 2j , k = 0, . . . , 2j you want to uniquely select one function in Vj obtain values at xj ,k outside [0, 1] by extrapolating values in [0, 1] this yields subspace Ej Vj Vj [0, 1] = Ej |[0,1]
Vj [0, 1] Vj +1 [0, 1]
8 / 24
Wavelet Collocation
given values at xj ,k = k 2j , k = 0, . . . , 2j you want to uniquely select one function in Vj obtain values at xj ,k outside [0, 1] by extrapolating values in [0, 1] this yields subspace Ej Vj Vj [0, 1] = Ej |[0,1]
Wavelet Collocation
9 / 24
Wavelet Collocation
9 / 24
Wavelet Collocation
Problem Look for uh Vj [0, 1]d such that Auh () = f () Gj ]0, 1[,
Silvia Bertoluzza (IMATI-CNR)
B u () = g () Gj ]0, 1[d
9 / 24
Wavelet Collocation
10 / 24
Wavelet Collocation
10 / 24
Wavelet Collocation
10 / 24
Wavelet Collocation
10 / 24
Wavelet Collocation
Problem Look for uh Vh such that Auh () = f () h ]0, 1[, B u () = f () h ]0, 1[d
10 / 24
Wavelet Collocation
Adaptivity
Simple adaptive strategy: look at coecients of tentative solution to design the next grid
11 / 24
Wavelet Collocation
Adaptivity
Simple adaptive strategy: look at coecients of tentative solution to design the next grid
if u small (|u | < r ): delete it from the grid
11 / 24
Wavelet Collocation
Adaptivity
Simple adaptive strategy: look at coecients of tentative solution to design the next grid
if u small (|u | < r ): delete it from the grid if u big (|u | > a > r ): add neighbouring points at higher level
11 / 24
Wavelet Collocation
Adaptivity
Simple adaptive strategy: look at coecients of tentative solution to design the next grid
if u small (|u | < r ): delete it from the grid if u big (|u | > a > r ): add neighbouring points at higher level
solve a sequence of problems with greed designed by looking at the solution at the previous step
11 / 24
Wavelet Collocation
Adaptivity
Simple adaptive strategy: look at coecients of tentative solution to design the next grid
if u small (|u | < r ): delete it from the grid if u big (|u | > a > r ): add neighbouring points at higher level
solve a sequence of problems with greed designed by looking at the solution at the previous step other strategies might be applied but no theory available
11 / 24
Wavelet Collocation
Adaptivity
Simple adaptive strategy: look at coecients of tentative solution to design the next grid
if u small (|u | < r ): delete it from the grid if u big (|u | > a > r ): add neighbouring points at higher level
solve a sequence of problems with greed designed by looking at the solution at the previous step other strategies might be applied but no theory available strategy eectively applied to dierent equations (Burgers equation, Convection-Diusion problems, Euler-Poisson system for semiconductors,. . . )
11 / 24
Wavelet Collocation
B u = g on ]0, 1[d
Wavelet Collocation
13 / 24
Wavelet Collocation
13 / 24
Wavelet Collocation
13 / 24
Wavelet Collocation
13 / 24
Wavelet Collocation
13 / 24
Wavelet Collocation
13 / 24
Wavelet Collocation
conditioning of the matrices is so bad that round-o is killer solution: increase the precision of (selected) computations
13 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
14 / 24
Wavelet Collocation
Multiresolution analysis
Vj [0, 1] dened as the span of such functions
15 / 24
Wavelet Collocation
Multiresolution analysis
Vj [0, 1] dened as the span of such functions Interpolation Ij : C 1 (0, 1) Vj [0, 1]
2j
Ij f = f (0)j ,1 +
k =0
15 / 24
Wavelet Collocation
Multiresolution analysis
Vj [0, 1] dened as the span of such functions Interpolation Ij : C 1 (0, 1) Vj [0, 1]
2j
Ij f = f (0)j ,1 +
k =0
Vj [0, 1] Vj +1 [0, 1]
15 / 24
Wavelet Collocation
Multiresolution analysis
Vj [0, 1] dened as the span of such functions Interpolation Ij : C 1 (0, 1) Vj [0, 1]
2j
Ij f = f (0)j ,1 +
k =0
Vj [0, 1] Vj +1 [0, 1]
15 / 24
Wavelet Collocation
Multiresolution analysis
Vj [0, 1] dened as the span of such functions Interpolation Ij : C 1 (0, 1) Vj [0, 1]
2j
Ij f = f (0)j ,1 +
k =0
Vj [0, 1] Vj +1 [0, 1]
15 / 24
Wavelet Collocation
Smoothness charachterization
Theorem Let f =
2j0 +1
2j
fk j0 ,k +
k =1 j k =1 ,p Bq
2j
(
k =1
|fk | )
p q /p
+
j
jsq /p
16 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
17 / 24
Wavelet Collocation
try with the multiscale basis basis reconstructs exactely polynomials of order 11, results not satisfactory at all!!
Silvia Bertoluzza (IMATI-CNR) Wavelet collocation for fourth order problems 17 / 24
Wavelet Collocation
18 / 24
Wavelet Collocation
18 / 24
Wavelet Collocation
18 / 24
Wavelet Collocation
18 / 24
Wavelet Collocation
18 / 24
Wavelet Collocation
good! but very expensive solve linear system in double precision much more satisfactory!!
Silvia Bertoluzza (IMATI-CNR) Wavelet collocation for fourth order problems 18 / 24
Wavelet Collocation
19 / 24
Wavelet Collocation
19 / 24
Wavelet Collocation
19 / 24
Wavelet Collocation
19 / 24
Wavelet Collocation
19 / 24
Wavelet Collocation
19 / 24
Wavelet Collocation
matrix and right hand side assembled in double precision linear system solved in double precision
19 / 24
Wavelet Collocation
Equation: w (iv ) + 4w = 1, Boundary conditions w (1) = w (1) = 0, w (1) = w (1) = Exact solution w= 1 2(sin(1)sinh(1)sin(x )sinh(x ) + cos (1)cosh(1)cos (x )cosh(x ) 4(cosh(2) + cos (2)) sinh(2) sin(2) 4(cosh(2) + cos (2)) in (1, 1)
20 / 24
Wavelet Collocation
Errors h .0625 .0313 .0156 .0078 .0039 L2 error 4.8859e-10 2.087e-12 1.2167e-14 3.2334e-17 4.5559e-17 L error 8.071e-10 4.6301e-12 1.9113e-14 9.2613e-17 9.395e-17
21 / 24
Wavelet Collocation
Errors h .0625 .0313 .0156 .0078 .0039 L2 error 4.8859e-10 2.087e-12 1.2167e-14 3.2334e-17 4.5559e-17 L error 8.071e-10 4.6301e-12 1.9113e-14 9.2613e-17 9.395e-17 L2 error 4.8859e-10 2.8972e-12 5.7798e-09 4.4425e-09 3.1474e-08 L error 8.0717e-10 4.6153e-12 2.0393e-08 2.0393e-08 2.0393e-08
21 / 24
Wavelet Collocation
Equation: EIw (iv ) = q Boundary conditions f , w (0) = w (2L) = w (0) = w (2L) = 0 Exact solution is known
Silvia Bertoluzza (IMATI-CNR) Wavelet collocation for fourth order problems
Wavelet Collocation
Equation: EIw (iv ) = 0 Boundary conditions w (0) = w (0) = w (L) = 0, EIw (L) = P
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions First tests show that the method behaves as expected
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions First tests show that the method behaves as expected Plenty of open problems!
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions First tests show that the method behaves as expected Plenty of open problems! Adaptivity needs to be studied
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions First tests show that the method behaves as expected Plenty of open problems! Adaptivity needs to be studied Preconditioning is of paramount importance
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions First tests show that the method behaves as expected Plenty of open problems! Adaptivity needs to be studied Preconditioning is of paramount importance No theory!
24 / 24
Wavelet Collocation
Conclusions
Modied interpolating wavelets on [0, 1] well suited for 4th order problems Computations must be (partially) carried out in multiple precisions First tests show that the method behaves as expected Plenty of open problems! Adaptivity needs to be studied Preconditioning is of paramount importance No theory! For whom might be interested: the Advanpix Multiple Precision Toolbox for Matlab works beautifully!! http://www.advanpix.com
Silvia Bertoluzza (IMATI-CNR) Wavelet collocation for fourth order problems 24 / 24