Multiwavelets based on hermite cubic splines
Abstract
the convection-diffusion equation. We use an implicit scheme for the time discretization and
Full text
MULTIWAVELETS BASED ON HERMITE CUBIC SPLINES
Dana ˇ
Ce n´
a
*V´
acla Finˇ
ek
**Ge a Plaˇ
cko ´
a
Technical Uni e si y o Libe ec
Facul y o Science, Humani ies and Educa ion
Depa men o Ma hema ics and Didac ics o Ma hema ics
S uden sk´
a 2, 461 17, Libe ec, Czech Republic
[email p o ec ed]
*Technical Uni e si y o Libe ec
Facul y o Science, Humani ies and Educa ion
Depa men o Ma hema ics and Didac ics o Ma hema ics
S uden sk´
a 2, 461 17, Libe ec, Czech Republic
acla [email p o ec ed]
**Technical Uni e si y o Libe ec
Facul y o Science, Humani ies and Educa ion
Depa men o Ma hema ics and Didac ics o Ma hema ics
S uden sk´
a 2, 461 17, Libe ec, Czech Republic
ge a.placko [email p o ec ed]
Abs ac
Fi s mul iwa ele s ha e appea ed a ound he ea ly 1990s. The basic idea behind mul iwa ele s
is simple: o eplace he single scaling unc ion φby he mul iscaling unc ion Φ
Φ
Φ o ha e some
addi ional desi ed p ope ies. I seems o be an in e es ing ade o because mul iwa ele s
p o ide highe o de app oxima ion wi h sho e suppo han single scaling unc ion. Mo eo e ,
i is possible o ha e bo h symme ic and o hogonal mul iwa ele s while his is no possible o
single wa ele s. In ecen yea s, se e al simple cons uc ions o wa ele bases based on He mi e
cubic splines we e p oposed. In his con ibu ion, we sho ly e iew hese cons uc ions, use
hese wa ele s o sol e nume ically di e en ial equa ions, and compa e hei pe o mance.
Keywo ds: Wa ele ; He mi e cubic splines; ellip ic di e en ial equa ions.
In oduc ion
A ec o - alued unc ion
Φ
Φ
Φ(x)=(φ1(x),φ2(x),...,φ (x))T,φ1(x),φ2(x),...,φ (x)∈L2(R)
will be called a mul iscaling unc ion i i sa is ies he ollowing e inemen equa ion
Φ
Φ
Φ(x) =
∑
k=s
HkΦ
Φ
Φ(ax −k),
whe e s, ∈Z,s< ,Hka e some × ma ices, a∈Nand a≥2. The mul iscaling unc ion
gene a es a mul i esolu ion analysis o L2(R)in a simila way as o scala wa ele s. Subspaces
Vja e de ined
Vj=closL2(R){φl,j,k=aj/2φl(ajx−k): 1 ≤l≤ ,k∈Z},j∈Z.
46
Then Wj,j∈Zdeno es he o hogonal complemen a y subspaces o Vjin Vj+1and he ec o -
alued unc ion
Ψ
Ψ
Ψ(x)=(ψ1(x),ψ2(x),...,ψ(a−1) (x))T,ψ1(x),ψ2(x),...,ψ(a−1) (x)∈L2(R).
The e o e he e exis ma ices Qksuch ha
Ψ
Ψ
Ψ(x) =
∑
k=s
QkΦ
Φ
Φ(ax −k),
and as decomposi ion and econs uc ion algo i hms can be cons uc ed like o scala wa ele s.
Mo eo e , mul iwa ele s usually ha e sho e suppo han co esponding single wa ele s, i is
possible o cons uc bo h symme ic and o hogonal mul iwa ele s while his is no possible
o single wa ele s, and inally, i is possible o ha e a basis o med by unc ions wi h di e en
o de s (see Example 1). Howe e , mul iwa ele s ha e also some disad an ages: he disc e e
mul iwa ele ans o m usually equi es p ep ocessing and pos p ocessing, and hei cons uc-
ion is also mo e complica ed. The eason, why p ep ocessing is necessa y, is in he ac ha
o scala wa ele s, we ha e
2−n/2 2−nk≈Z (x)φn,k(x)dx
and his does no hold in gene al o mul iwa ele s. Mul iwa ele s which do no equi e p ep o-
cessing a e o ins ance he so called balanced mul iwa ele s. They we e cons uc ed in [7, 8].
Fo mo e de ails on mul iwa ele s, we e e o [6].
Example 1 Simple example o piecewise linea mul iwa ele s is aken om [1]:
φ1(x) = 1 0 ≤x≤1
0o he wise ,φ2(x) = 2√3x−1
20≤x≤1
0o he wise ,
ψ1(x) =
6x−1 0 ≤x<1
2
6x−51
2≤x≤1
0o he wise
,ψ2(x) =
2√32x−1
20≤x<1
2
−2√32x−3
21
2≤x≤1
0o he wise
.
The emainde o his pape is o ganized as ollows. In he nex sec ion, we e iew se -
e al simple cons uc ions o wa ele s based on He mi e cubic splines. Speci ically, wa ele s
wi h wo anishing wa ele momen s p oposed in [5], a hie a chical basis based on He mi e cu-
bic splines, wa ele s wi h ou anishing wa ele momen s p oposed in [2], and wa ele s wi h
espec o which bo h he mass and s i ness ma ices co esponding o he one-dimensional
Laplacian a e spa se [3]. In he las sec ion, we use hese wa ele s o sol e nume ically di e -
en ial equa ions and compa e hei pe o mance.
1 He mi e Cubic Spline Wa ele s
In he yea 2000, W. Dahmen e al. [4] p oposed a cons uc ion o bio hogonal mul iwa ele s
adap ed o he in e al [0,1]on he basis o He mi e cubic splines. They s a ed wi h He mi e
cubic splines as he p imal scaling bases on R. Then, hey cons uc ed dual scaling bases on
Rconsis ing o con inuous unc ions wi h small suppo s and wi h polynomial exac ness o
o de 2. Consequen ly, hey de i ed p imal and dual bounda y scaling unc ions e aining he
polynomial exac ness. This ensu es anishing momen s o he co esponding wa ele s. Finally,
47
Sou ce: Own
Fig. 1.Piecewise linea mul iwa ele s
hey applied he me hod o s able comple ions o cons uc he co esponding p imal and dual
mul iwa ele s on he in e al.
These He mi e cubic splines a e de ined by
φ1(x) =
(x+1)2(1−2x)−1≤x≤0
(1−x)2(2x+1)0≤x≤1
0 o he wise
,φ2(x) =
(x+1)2x−1≤x≤0
(1−x)2x0≤x≤1
0 o he wise
.
They possess he ollowing in e pola ion p ope y:
φ1(0) = 1,φ0
1(0) = 0,φ2(0) = 0,φ0
2(0) = 1
and hen o any unc ion ∈C1(R)
u:=∑
j∈Z
(j)φ1(x−j)+ ∑
j∈Z
0(j)φ2(x−j)
is a He mi e in e polan o on Z.
1.1 Wa ele s p oposed by R. Q. Jia and S. T. Liu
An in e es ing cons uc ion was p oposed in [5]. The au ho s cons uc ed mul iwa ele s based
on He mi e cubic splines wi h con inuous de i a i es and wi h a suppo on he in e al [−1,1].
One o he cons uc ed wa ele s is symme ic, and he second one is an isymme ic. They also
adap ed hem o he in e al [0,1]and hei cons uc ion o bounda y wa ele s is e y simple
unlike he cons uc ion om [4]. In compa ison wi h he semi-o hogonal wa ele s, he wa ele s
a di e en le els a e o hogonal wi h espec o hu0, 0iins ead o hu, i.They a e gi en by
ψ1(x) = −2φ1(2x+1)+4φ1(2x)−2φ1(2x−1)−21φ2(2x+1)+21φ2(2x−1),
48
φ1φ2
Sou ce: Own
Fig. 2.The He mi e cubic spline
ψ1ψ2
Sou ce: Own
Fig. 3.Wa ele s p oposed by R. Q. Jia and S. T. Liu
ψ2(x) = φ1(2x+1)−φ1(2x−1)+9φ2(2x+1)+12φ2(2x)+9φ2(2x−1).
Bo h wa ele s a e suppo ed on [−1,1]and i s adap a ion o he in e al [0,1]is pe o med
by he es ic ion o he an isymme ic wa ele o he in e al [0,1]. Fo n≥1,le Vnbe he space
o piecewise cubic splines ∈C1(0,1)∩C[0,1] o which (0) = (1) = 0.The dimension o
Vnis 2n+1and he se
Φn:={φ1(2nx−j):j=1,...,2n−1}∪φ2(2nx−j)|[0,1]:j=0,...,2n
is he basis o Vn.Le Wnbe he complemen o Vnin Vn+1wi h he basis de ined by
Ψn:={ψ1(2nx−j):j=1,...,2n−1}∪ψ2(2nx−j)|[0,1]:j=0,...,2n.
Then, we ha e he ollowing decomposi ion o H1
0(0,1)
H1
0(0,1) = V1+W1+W2+W3....
1.2 Hie a chical Basis
The second possibili y o de ine complemen wa ele spaces, we use he ollowing hie a chical
app oach:
ψ1(x) = φ1(2x+1)and ψ2(x) = φ2(2x+1).
49
Bo h unc ions a e suppo ed on [−1,0]and he e o e no bounda y unc ions a e necessa y.
The complemen space Wnis hen gi en by
Ψn:={ψ1(2nx−j):j=1,...,2n}∪{ψ2(2nx−j):j=1,...,2n}.
ψ1ψ2
Sou ce: Own
Fig. 4.Hie a chical wa ele s
1.3 Wa ele s wi h Momen s
The hi d possibili y is o de ine wa ele s o ha e maximal numbe o anishing momen s o
he gi en suppo [−1,1]. Wa ele s a e gi en hen by
ψ1(x) = φ1(2x+1)−φ1(2x−1)+ 39
7φ2(2x+1)+ 132
7φ2(2x)+ 39
7φ2(2x−1),
and
ψ2(x) = −1
2φ1(2x+1)+φ1(2x)−1
2φ1(2x−1)−15
4φ2(2x+1)+ 15
4φ2(2x−1).
ψ1ψ2
Sou ce: Own
Fig. 5.Wa ele s wi h he maximal numbe o anishing momen s
50
Bounda y wa ele s a e cons uc ed o ha e he same numbe o anishing momen s as inne
wa ele s. Fo mo e de ails, see [2]. Again le Wnbe he complemen o Vnin Vn+1wi h he basis
de ined by
Ψn:={ψ1(2nx−j),ψ2(2nx−j):j=1,...,2n−1}
∪ψ3(2nx−1)|[0,1]∪ψ4(2n(x−1)+1)|[0,1].
1.4 Wa ele s P oposed by T. J. Dijkema and R. S e enson
Fu he cons uc ion was p oposed in [3]. They i s p o ed ha i is no possible o cons uc
con inuous piecewise smoo h wa ele s which ha e a compac suppo , o m Riesz basis o
L2(I), p ope ly scaled wa ele s o m he Riesz basis o H1(I),and wi h
Dψ0
λ,ψ0
µE=0,and ψλ,ψµ=0∀λ,µ.
Consequen ly, hey cons uc ed cubic He mi e wa ele s wi h he con inuous i s de i a i e such
ha among he p imal mul i esolu ion spaces Vjand he dual mul i esolu ion spaces ˜
Vj he
ollowing ela ion holds
Vj+V00
j⊂˜
Vj+1.
As a consequence
Dψ0
λ,ψ0
µE=0,ψλ,ψµ=0,∀λ,µ:|λ|>|µ|+1.
Thei wa ele s a e hen gi en by
ψ1(x) = φ1(2x+1)−φ1(2x−1)+ 39
7φ2(2x+1)+ 132
7φ2(2x)+ 39
7φ2(2x−1),
ψ2(x) = −1
2φ1(2x+1)+φ1(2x)−1
2φ1(2x−1)−15
4φ2(2x+1)+ 15
4φ2(2x−1),
ψ3(x) =
3
∑
i=−3
ciφ1(2x−i)+diφ2(2x−i),ψ4(x) =
3
∑
i=−3
eiφ1(2x−i)+ iφ2(2x−i)
wi h
c=−4595
13728,7
65,−18737
68640,1,−18737
68640,7
65,−4595
13728,
d=−68741
22880,−69
40,−204701
22880 ,0,204701
22880 ,69
40,68741
22880,
e=417
22880,−7
2340,5443
205920,0,−5443
205920,7
2340,−417
22880,
=723
4576,1
8,8153
13728,1
2,8153
13728,1
8,723
4576.
Again Wnis he complemen o Vnin Vn+1wi h he basis de ined by
Ψn:=ψ1(2nx−2j−1),ψ2(2nx−2j−1):j=0,...,2n−1−1
∪ψ3(2nx−2j):j=1,...,2n−1−1∪ψ4(2nx−2j)|[0,1]:j=0,...,2n−1.
51
ψ3ψ4
Sou ce: Own
Fig. 6.Wa ele s p oposed by T. J. Dijkema and R. S e enson
2 Nume ical Expe imen s
In his sec ion, he wa ele s in oduced in he p e ious sec ion a e used o sol e nume i-cally di -
e en ial equa ions. We will es ic ou sel es o he equa ion −pu00+qu = wi h he Di ichle
bounda y condi ions u(0) = u(1) = 0 and wi h posi i e cons an coe icien s. The co esponding
Gale kin app oxima ion p oblem is he ollowing: Find un=
2n+2
∑
i=1
ci isuch ha
Z1
0
pu0
n 0+qun dx =Z1
0
dx ∀ ∈Vn.
By he Lax-Milg am lemma, his app oxima ion p oblem has he unique solu ion. We also use
he s anda d wa ele p econdi ioning consis ing in no malizing each basis unc ion wi h espec
o he abo e bilinea o m. To sol e he a ising sys em o linea equa ions, we use he conjuga e
g adien me hod. The i e a ions a e e mina ed i he di e ence o wo consecu i e i e a ions is
less han 10−n−2/cond(An),whe e cond(An)deno es he condi ion numbe o he co esponding
s i ness ma ix.
Tab. 1.Ob ained esul s o he p oblem wi h p =q=1.
JL H M DS
n||un−u||L2NZ IT NZ IT NZ IT NZ IT
1 6.0e-02 48 4 40 7 54 7 54 5
2 1.7e-02 172 5 128 13 162 11 154 7
3 2.6e-03 552 6 368 20 418 16 410 9
4 2.7e-04 1580 7 976 31 1010 21 986 10
5 2.3e-05 4168 8 2448 41 2306 24 2202 13
6 1.8e-06 10396 10 5904 52 5042 25 4698 14
7 1.2e-07 24936 10 13840 74 10690 28 9754 16
8 8.3e-09 58283 11 31760 84 22194 31 19967 17
9 5.4e-10 135302 12 71696 110 46963 33 41627 19
Sou ce: Own
52
Tab. 2.Ob ained esul s o he p oblem wi h p =1and q =0.
JL H M DS
n||un−u||L2NZ IT NZ IT NZ IT NZ IT
1 6.0e-02 24 3 36 7 50 7 54 5
2 1.7e-02 56 5 116 12 154 11 154 7
3 2.6e-03 128 6 340 20 406 16 410 8
4 2.7e-04 280 6 916 31 994 20 986 10
5 2.3e-05 592 8 2324 41 2286 24 2202 12
6 1.8e-06 1224 10 5652 52 5018 25 4698 14
7 1.2e-07 2847 10 13332 69 10677 28 9762 15
8 8.3e-09 8818 10 30740 84 22347 30 20019 16
9 5.4e-10 35863 12 69652 110 46613 33 41421 19
Sou ce: Own
Fi s we sol e he abo e p oblem wi h p=q=1 and he exac solu ion u=x(1−e50x−50)
which exhibi s a s eep g adien nea he poin 1.Resul s a e summa ized in Table 1. Then we
sol e he abo e p oblem wi h p=1,q=0 and wi h he exac solu ion u=x(1−e50x−50)which
exhibi s a s eep g adien nea he poin 1.Resul s a e summa ized in Table 2. In bo h ables, NZ
is he numbe o nonze o elemen s in s i ness ma ices, IT ep esen s he numbe o i e a ions,
JL deno es wa ele s p oposed in [5], H deno es hie a chical basis, M deno es wa ele s p oposed
in [2], and inally DS deno es wa ele s p oposed in [3]. Achie ed app oxima ion e o was he
same o all bases.
Conclusion
P esen ed esul s a i m ha wa ele s p oposed in [5] ha e excellen condi ion numbe , and
especially o he Poisson equa ion, he a ising s i ness ma ices a e e y spa se. Wa ele s
p oposed in [3] a e bes sui ed o gene al di e en ial equa ions wi h cons an coe icien s.
Resul s o es ed hie a chical basis con i m he well known ac ha hese basis do no o m
he Riesz basis and he e o e hey need some addi ional p econdi ioning. Conce ning he basis
p oposed in [2], we suppose ha i will be be e ( han es ed bases) sui ed o non-cons an
di e en ial equa ions sol ed adap i ely. P esen ed esul s also sugges ha he e is p obably
some space o imp o e he condi ion numbe o he basis p oposed in [2].
Acknowledgmen s
This wo k has been suppo ed by he p ojec ESF No. CZ.1.07/2.3.00/09.0155 “Cons i u ion
and imp o emen o a eam o demanding echnical compu a ions on pa allel compu e s a TU
Libe ec”.
Li e a u e
[1] ALPER, B. K.; ROKHLIN, V.: A as algo i hm o he e alua ion o Legend e expan-
sions. Siam J. Sci. S a is . Compu ., 12 (1991), pp. 158-179.
[2] ˇ
CERN ´
A, D.; FIN ˇ
EK, V.: He mi e cubic spline wa ele s wi h momen s on he in e al. In
p epa a ion, 2012.
53
[3] DIJKEMA, T. J.; STEVENSON, R.: A spa se Laplacian in enso p oduc wa ele coo -
dina es. Nume . Ma h., ol. 115, no. 3 (2010), pp. 433-449.
[4] DAHMEN, W.; HAN, B.; JIA, R.-Q.; KUNOTH, A.: Bio hogonal Mul iwa ele s on he
In e al: Cubic He mi e Splines. CONSTR. APPROX., ol. 16, (1998), pp. 221-259.
[5] JIA, R.-Q.; LIU, S.-T.: Wa ele bases o He mi e cubic splines on he in e al. Ad . in
Comp. Ma h., 25 (2006), pp. 23-39.
[6] KEINERT, F.: Wa ele s and Mul iwa ele s. CHAPMAN & HALL/CRC, 2004.
[7] LEBRUN, J.; VETTERLI, M.: Balanced mul iwa ele s heo y and design. IEEE T ans.
Signal P ocess., 46 (1998), pp. 1119-1125.
[8] LEBRUN, J.; VETTERLI, M.: High-o de balanced mul iwa ele s: heo y, ac o iza ion,
and design. IEEE T ans. Signal P ocess., 49 (2001), pp. 1918-1930.
RND . Dana ˇ
Ce n´
a, Ph.D., RND . V´
acla Finˇ
ek, Ph.D., Ge a Plaˇ
cko ´
a, p om. ma .
54