On Cardinal Spline Smoothing

On Cardinal Spline Smoothing

Approximation Theory and Functional Analysis J.B. Prolla (Ed.) @North-Holland Publishing Company, 1979 ON CARDINAL SPLINE SMOOTHING I. J. SCHOENBER...

749KB Sizes 0 Downloads 118 Views

Approximation Theory and Functional Analysis J.B. Prolla (Ed.)

@North-Holland Publishing Company, 1979

ON CARDINAL SPLINE SMOOTHING

I. J. SCHOENBERG Mathematics Research Center University of Wisconsin Madison, Wisconsin 53706, USA.

INTRODUCTION The present paper describes the methods, with some changes be mentioned below, whereby I solved the numerical problem

to

assigned

to me at the Ballistics Research Laboratories in Aberdeen, Maryland, during the second World War (see [ 4

]

1.

The problem was to smooth very

extended equidistant tables of drag functions (or drag coefficients) by approximating them by very smooth functions that were easily computable with their first and second derivatives. The first question to be answered was this: When may a " m o w h g awefiage" o p e f i a t i o n L e g i t i m a t e L y b e c a L l e d a n m o o t h i n g 6ofimuLa ?

An

answer is given in Section 1 of Part I. That it is a reasonable

one

is shown in Section 2 of Part I. These ideas were later greatly generalized by Fritz John in his important work on parabolic differential equations (see [ 3 1 ) . The connection is briefly

mentioned

in

Section 2 of Part I. The second essential ingredient of our results of Part

I1

is

the process of c a f i d i n a e h p t i n e intehpolation (see [ 9 1 ). The required results are described in Section 3 of Part I. The changes from my war-time approach are developed completely in Part 11. This is the new contribution of the paper, and it arises 383

384

SCHOENBERG

i f w e a d a p t t h e i d e a o f E. T. Whittaker (see [lo 1 and [ 7

1 ) to

the

problem of smoothing a b i - i n f i n i t e sequence of e q u i d i s t a n t d a t a . The r e s u l t i s t h e cahd.inaL 4 p L i n e s m o o t h i n g phOces4. L e t m e thank P r o f e s s o r I l i o G a l l i g a n i f o r a s k i n g m e

to

visit

Rome i n May of 1 9 7 7 , where I gave a s h o r t 10-hours c o u r s e o n C a r d i n a 1 Spline Interpolation a t h i s I s t i t u t o per l e applicazioni d e l Calcolo "Mauro Picone".

PART I .

PRELIMINARIES

1. WHAT IS A SMOOTHING FORMULA?

The s e v e r a l smoothing ( o r gradua

-

t i o n ) methods d i s c u s s e d i n [lo] are a l l o f t h e "moving a v e r a g e " t y p . By t h i s w e mean t h a t w e are g i v e n a sequence o f r e a l and

symmetric

weights

which w e a p p l y t o t h e d a t a (x,)

t o produce t h e i r smoothedversicln (y,)

,

by t h e formula

yn

--

m

C

v=-co

an-Vxy

, (n=O,

f 1,

...) .

I n o t h e r words, w e c o n v o l u t e t h e sequences (a,)

and (x,)

,

an o p e r a

-

t i o n s y m b o l i c a l l y d e s c r i b e d by

(3)

By d e r i v i n g a formula ( 2 ) a c c o r d i n g t o a d e f i n i t e i d e a , s u c h a s a l e a s t s q u a r e s formula, o r W h i t t a k e r ' s formulae (see [ 1 0 , XI

1

),

Chapter

we may r e a s o n a b l y e x p e c t t h a t it w i l l t r a n s f o r m a g i v e n

se-

quence ( x ) i n t o a smoother sequence (y,). T h i s p o i n t becomes doubtn f u l l however, i n c a s e t h e formula ( 2 ) was o t h e r w i s e o b t a i n e d , e . g . ,

385

ON CARDINAL SPLINE SMOOTHING

by some a p p r o x i m a t i n g p r o c e d u r e t h a t i s n o t s t r i c t l y

interpolatory.

An example of such a p r o c e d u r e w i l l be o u r c a r d i n a l s p l i n e smoothing o f P a r t 11. A c r i t e r i o n f o r ( 2 ) t o be r e g a r d e d a s a b m o o t h i n g ~0hmlLea w a s proposed by t h e a u t h o r i n [ 4 , pp. 50

- 54 1

and p r o c e e d s a s f o l l w s .

To b e g i n w i t h , w e assume t h e L a u r a n t series

(4)

I

t o converge i n some r i n g c o n t a i n i n g t h e u n i t c i r c l e

z j =1. S e t t i n g

i u , we c a l l t h e even f u n c t i o n z = e

d(u)

(5)

iu

= F(e

)

=

a

0

+

2al cos u

+

2a2 c o s 2u

+

...

t h e chahactehintic ( u n c t i o n of t h e formula ( 2 ) . The r e g u l a r i t y of F(z) on

I

IIm

UI

z

I

= 1, i m p l i e s t h a t

$(u)

is regular

in

a

certain

strip

< a , whence t h e v a l i d T a y l o r e x p a n s i o n m

T h i s e x p a n s i o n a l l o w s us t o e x p r e s s e a s i l y an i m p o r t a n t p r o p e r t y ( 2 ) : I t s hephoductive p o w e f ~ o , r deghee

06

exactnebb.

We s a y t h a t ( 2 ) h a s t h e d e g r e e o f e x a c t n e s s i s a n a t u r a l number, p r o v i d e d t h a t ( 2 ) r e p r o d u c e s

yn

- xn

values

f o r a l l n , a sequence (x,) P(n) of

,

of

where

2m-1,

exactly,

m

i. e.

,

i f t h i s sequence r e p r e s e n t s t h e

a polynomial of d e g r e e

2m-1,

b u t d o e s n o t have

t h i s p r o p e r t y f o r any h i g h e r d e g r e e . T h a t t h e d e g r e e of e x a c t n e s s i s always odd, f o l l o w s from t h e symmetry c o n d i t i o n ( 1 ) . I n t e r m s of

(6)

w e have t h e f o l l o w i n g e a s i l y e s t a b l i s h e d p r o p o s i t i o n : T h e doamuta ( 2 ) hub t h e degkee

id t h e e x p a n n i o n ( 6 ) i n

06

the 6otm

06

eXaCtflebb

2m-1,

i6 and onLy

SCHOENBERG

386

$(u) = 1

(7)

-x

.,.

+

U2m

(A

#

0).

An example. L e t (1) be t h e sequence a

0

= 5/8,

al = 1 / 4 ,

a2 =

-

1/16,

an = 0

if

n > 2,

when t h e c o n v o l u t i o n ( 2 ) assumes t h e form

From (5) w e o b t a i n

(9)

$ ( u ) = ( 5 + 4 cox u

-

and ( 7 ) i s s e e n t o h o l d w i t h mula ( 8 ) has t h e d e g k e e

06

cos 2u)/8 = 1

rn = 2 , X

=

1/16.

- 7

u4

+

...

I t follows that the f o r -

exactness 3 .

May ( 8 ) be r e g a r d e d as a smoothing formula ? The g e n e r a l

cri-

t e r i o n i s d e s c r i b e d by t h e f o l l o w i n g .

DEFINITION:

kJe 4ay t h a t ( 2 ) Lo a s m o o t h i n g botmuLa, p h o v i d e d t h a t La2

chahactchihtic dunction

d(u) 4atiAdieb

This condition evidently implies t h a t t h e coed6icien-t

peaking i n ( 7 ) ,

A ,

ap-

io p o s i t i v e . The c h a r a c t e r i s t i c f u n c t i o n ( 9 ) i s e a s -

i l y s e e n t o s a t i s f y (lo), and so ( 8 ) i s a smoothing f o r m u l a a c c o r d i n g t o o u r d e f i n i t i o n of t h e t e r m . The c r i t e r i o n (10) raises t h e f o l l o w i n g k i n d o f e q u a t i o n :

Do

t h e numerous smoothing formulae a s g i v e n i n 110 1 s a t i s f y our criterion

301

ON CARDINAL SPLINE SMOOTHING

(lo)? S e e G r e v i l l e ' s p a p e r [ l 1 where s o m e of t h e s e q u e s t i o n s a r e answered a f f i r m a t i v e l y . I n [ 4 , p p . 50

-

54 ] good a r g u m e n t s i n s u p p o r t of

( 1 0 ) are p r e s e n t e d . Even m o r e c o n v i n c i n g r e a s o n s

the criterion

w e r e given i n

[5,

P a r t I ] . I n t h e n e x t s e c t i o n we r e p r o d u c e t h e s e a r g u m e n t s a s p r e s e n t e d i n [ 6 , pp. 2 0 0 - 2 0 4 1 .

2 . THE BEHAVIOR OF THE ITERATES O F A MOVING AVERAGE FORMULA:

THE

PROBLEM OF DE FOREST. A c o n c l u s i v e argument i n s u p p o r t o f t h e n e c e s s i t y of o u r c o n d i t i o n ( 1 . 1 0 )

i s f u r n i s h e d by t h e s o l u t i o n o f t h e f o l -

l o w i n g problem f i r s t s t a t e d a n d a t t a c k e d

by

Erasmus

L. D e

(1834-1888). I f we s u b j e c t t h e g i v e n s e q u e n c e ( x u ) n times

Forest

i n suc

-

c e s s i o n t o t h e same t r a n f o r m a t i o n ( 1 . 2 ) , we o b t a i n a l i n e a r transformation

which i s t h e n - f o l d 06

the

i t e r a t e of ( 1 . 2 ) . W h a t i n R h e a o y m p t o t i c

c a e d 6 i c i e n t n 0 6 (1) an

n

+

m ?

T h i s q u e s t i o n was

behaviah

answered

by D e F o r e s t a n d by G. B. D a n t z i g ( f o r r e f e r e n c e s see151 ) f o r t h e c a s e when a l l c o e f f i c i e n t s of that

m=l

( 1 . 2 ) a r e non - n e g a t i v e , h e n c e n e c e s s a r i l y

i n ( 1 . 7 ) . A g e n e r a l s o l u t i o n i s as follows.

L e t ( 1 . 2 ) be s u c h t h a t ( 1 . 1 7 )

hence t h a t

,

( l . l O ) , and (1.71, are s a t i s f i e d ,

X > 0. L e t

-V

which i s t h e normal f r e q u e n c y f u n c t i o n

2m

cos vx d v ,

SCHOENBERG

388

if

m =1, o t h e r w i s e (m = 2,3,.

.

.)

Gm(x) i s an e n t i r e f u n c t i o n

having

i n f i n i t e l y many zeros, a l l r e a l . T h e caeddicients a d (1) satis6q t h e a s y m p t o t i c h e L a t i o n h

--

--

1 1 --1 a ( n ) = ( i n ) 2m Gm(v(hn) 2m ) + , o ( n 2m)

(4)

a4

V

where t h e " l i t t e e

n +

m ,

v.

or' dymbok? hoed4 unidaamly d o h n l L i n t e g e h d

For a proof see ( 5 , P a r t I ] , where i t i s a l s o shown byexamples (1.10),

t h a t ( 4 ) no l o n g e r h o l d s i f t h e e q u a l i t y s i g n i s a l l o w e d i n and t h a t t h e c o e f f i c i e n t n = 2k

aAn) d i v e r g e s e x p o n e n t i a l l y t o

t e n d s t o i n f i n i t y t h r o u g h even v a l u e s , i f

( 1 . 1 0 ) are r e v e r s e d anywhere i n t h e i n t e r v a l

the

0 < u < 2.rr

+

m

,

as

inequalities

.

The f o l l o w i n g d i s c u s s i o n , w h i l e n o t d i r e c t l y r e l a t e d

to

our

s u b j e c t of smoothing, w i l l show t h e c o n n e c t i o n of t h e a s y m p t o t i c rel a t i o n ( 4 ) w i t h t h e w i d e r f i e l d of p a r a b o l i c d i f f e r e n t i a l e q u a t i o n s . Observe t h a t ( 2 ) i m p l i e s t h a t

(5)

--1

--1

U(x,t) = t 2m G m ( x t 2m) =

-

-tv

2m

+ ixv

d v , ( t > 0).

The f u n c t i o n under t h e i n t e g r a l s i g n i s immediately s e e n t o s a t i s f y for all v

, the

d i f f e r e n t i a l equation

which r e d u c e s t o t h e f a m i l i a r h e a t e q u a t i o n i f

also -plane

m = l . I t follows t h a t

U ( x , t ) , d e f i n e d by ( 5 ) , is a s o l u t i o n of (6) i n t h e upper h a l f t > 0 . On t h e o t h e r hand, a p p l y i n g t o ( 2 ) F o u r i e r ' s inversion

formula and s e t t i n g

v = O r we f i n d that

ON CARDINALSPLINE SMOOTHING

These r e m a r k s imply t h e f o l l o w i n g : 'I 6

1x 1

say, a s

+

LA a b o t u , t i o n

a,

06

f (x)

389

cantinuow and a ( I X I - * )

,

then

t h e d i , j d e ~ e n . t i a l e q u a t i o n ( 6 ) Aattin6qLng t h e boundmy

condition

This p a r t i c u l a r s o l u t i o n

u ( x , t ) may now a l s o be

approximated

by t h e f o l l o w i n g n u m e r i c a l p r o c e d u r e : Draw i n t h e ( x , t ) - p l a n e

the

rectangular lattice of p o i n t s

(WAX,

n At)

(w

= 0, k 1

, ...

;

D e f i n e on it a l a t t i c e f u n c t i o n

n = 0,1,2

u

v

, ...) .

by s t a r t i n g w i t h

uw ,o = f ( v Ax) ,

and computing t h e v a l u e s a l o n g e a c h h o r i z o n t a l l i n e from those on the l i n e below i t , by means o f t h e t r a n s f o r m a t i o n (1.2). T h i s

evidently

amounts t o i t e r a t i n g (1.21, a n d a f t e r n s t e p s w e o b t a i n

(10)

For any g i v e n x a n d

t > 0,

( 1 0 ) w i l l go o v e r i n t o ( 8 ) i f w e

f o l l o w i n g : We 6 h A t c o n n e c t the. m e o h - n i z e b

Ax and

A t

do the

by ,the h d a t i a n

SCHOENBERG

380

A t = X (Ax) 2m.

(11)

Id t h e i n t e g e k b

n ahe buch t h a t

v and

VAX

+

x,

and

n A t

.+

t

an

Ax

0.

+

then

U

v,n

T h i s follows r e a d i l y from r e l . a t i o n (4): ( 1 0 ) d i f f e r s

U(X,t).

+

1 0 ) and ( 8 1 , i n view o f t h e

asymptotic

from a Cauchy-Riemann sum f o r tk integral

( 8 1 , by a q u a n t i t y t h a t t e n d s t o z e r o due t o t h e u n i f o r m i t y i n

v of

t h e error t e r m o f ( 4 ) . I t i s i n t e r e s t i n g t o n o t e t h a t it d o e s n o t matter which

for-

mula ( 1 . 2 ) w e u s e i n t h i s c o n s t r u c t i o n , as l o n g as it i s o f the degree of exactness

2m-1,

i.e.,

i t s a t i s f i e s (1.71, and above a l l t h a t i t

s a t i s f i e s t h e s t a b i l i t y c o n d i t i o n (1.10)

,

t h e t e r m "stabi1ity"meaning

h e r e s t a b i l i t y on i t e r a t i o n . F o r t h e g e n e r a l t h e o r y of F. J o h n , which t h e e q u a t i o n ( 6 ) i s a s p e c i a l example, see [ 3 1

.

of

I n t h i s s e c t i o n w e d e a l t e x c l u s i v e l y w i t h f o r m u l a e ( 1 . 2 ) which s a t i s f y t h e symmetry r e l a t i o n . I n [ 2 ]

T.

N.

E. G r e v i l l e d e a l t

with

t h e more d i f f i c u l t c a s e o f unsymmetric f o r m u l a e .

3 . CARDINAL SPLINE INTERPOLATION (see [ 9 , L e c t u r e s 1

l e m o f caadinal intehpolation i s t o f i n d s o l u t i o n s

-

4 1 ) . T h e prob-

f ( x ) of t h e i n -

t e r p o l a t i o n problem

(1)

f ( v ) = Y"

,

for all i n t e g e r s

v

,

where ( y v ) are t h e d a t a . A f o r m a l s o l u t i o n i s f u r n i s h e d b y t h e series

391

ON CARDINAL SPLINE SMOOTHING

i n v e s t i g a t e d i n 1 9 0 8 by de l a V a l l G e P o u s s i n , also l a t e r

by

E. T.

W h i t t a k e r , who c a l l e d i t t h e cahdinad b e h i e b . The d i f f i c u l t y w i t h ( 2 )

“i: :y

i s t h e s l o w decay o f t h e f u n c t i o n

as

x

-. A

+

s o l u t i o n of (1) i s t h e p i e c e w i b e l i n e a h i n t e h p o t h z t

much s i m p l e r g i v e n by

S1(x)

m

(3)

where

M2(x) i s t h e roof f u n c t i o n d e f i n e d by

in

M2(x) = x + l

,

[-1,01

M (x) = 1

2

-x

i n [ O , l l and%(x)

=o

The p u r p o s e o f cahdinad b p l i n e i n t e h p o & z t i o n i s t o b r i d g e between t h e p i e c e w i s e l i n e a r

if 1x1 ’1.

the

gap

S1(x) d e f i n e d by (31, a n d t h e c a r d i n a l

series ( 2 ) . I t a i m s a t r e t a i n i n g s o m e of t h e s t u r d i n e s s a n d s i n p l i c i t y of ( 3 ) , a t t h e same t i m e c a p t u r i n g some of t h e s m o o t h n e s s a n d s o p h i s t i c a t i o n of

(2).

Le-t m be a natuhad numbeh, and d e b

(4)

S2rn-l

b e t h e cLadb

06

= {S(X)3

cahdinad b p d i n e d

S(x)

0 6 deghee

2m-1

dedined

by

the two conditionb:

(5)

The h e s t h i c t i o n whete

v

i d

04

S ( x ) -to e u e h y u n i t i n t e n v a l

a n i n X e g e h , i b apolynomia!.

(v ,v

0 6 deghee 2

2m

+11,

-

1.

392

SCHOENBERG

For

m =1

we f i n d

S1

t o be i d e n t i c a l w i t h t h e c l a s s ( 3 )

c o n t i n u o u s p i e c e w i s e l i n e a r f u n c t i o n s . Observe t h a t t h e c l a s s o f p o l y n o m i a l s of d e g r e e s n o t e x c e e d i n g The r o l e o f t h e r o o f - f u n c t i o n t h e s o - c a l l e d centha.! B-npLine

M 2 m ( ~ ) : Waiting

SZmml c o n t a i n s 2m-1.

of (3)

M2(x)

of

x+

i s t a k e n o v e r by = max ( x , O ) ,

it

may be d e d i n e d b y

Clearly port

M2m(~)



S2m-l; w e also f i n d t h a t

M2m(~) > 0

i n its

sup-

- m < x < m. The B - s p l i n e s h o u l d be f a m i l i a r i n view of the fun-

damental i d e n t i t y

which a l s o shows t h a t

IM2,(x)dx

= 1 if

w e choose

f (x) = x

The r e p r e s e n t a t i o n ( 3 ) a l s o g e n e r a l i z e s , and eueAny S ( x )

2m E

. SZm-l

admitb a unique hepheoentation m

S(x) = c c

~

M*m(X

--m

whehe t h e

-

v)

I

c v ahe c o n n t a n t n . T h i s i s t h e s o - c a l l e d ntandahd heptebefl-

t a t i o n . The c o n v e r s e i s clear: Every series (8) f u r n i s h e s an e l e m e n t of

SZm-1

ments o f

.

W e now t r y t o s o l v e t h e i n t e r p o l a t i o n problem (1) b y e l e -

S2m-l.

I n t h i s d i r e c t i o n t h e r e are t w o d i f f e r e n t k i n d s o f

results.

A. T h e d a t a (y,) s e q u e n c e (y,)

ahe

06 poweh

g h o w t h (See [ 8

i s of p o w e h g n o w t h , and w r i t e

1). W e s a y t h a t t h e

ON CARDINAL SPLINE SMOOTHING

(y,)

(9)

393

E PG,

provided t h a t

y,

(10)

v

+

E

PG,

= ~ ( l v l y ) as

f

m,

f o r some

y

2

0.

y1

2

Similarly, we w r i t e

f(x)

(11)

provided t h a t

f ( x ) = O ( l x l y l ) as

x

+

f o r some

f m,

Below w e e x c l u d e t h e t r i v i a l c a s e when

m=l,

l e m i s s o l v e d by ( 3 ) w i t h o u t any r e s t r i c t i o n on t h e

THEOREM 1:

16 t h e heqUenCt

(y,)

i h

0.

s i n c e o u r prob

-

(y,,).

a d pawet g t u w t h , t h e n t h e i n t e h -

palation p t a b l e m

huh a u n i q u e h o l u t i o n

S(x)

huch t h a t

The a s s u m p t i o n ( 9 ) o f Theorem 1 i s a rough one; i t admits,e.g., a l l bounded s e q u e n c e s ( y V ) , w i t h

y = 0 i n ( 1 0 ) . The s e c o n d assump-

t i o n t o which w e now p a s s , i s much more s e l e c t i v e , and

takes

a c c o u n t t h e f i n e r s t r u c t u r e of t h e sequence: i n f a c t i t a d m i t s a narrow subclass of t h e s e q u e n c e s of

PG.

into only

As u s u a l , w i t h strongeras-

sumptions, s t r o n g e r c o n c l u s i o n s are p o s s i b l e : The i n t e r p o l a n t w i l l e x h i b i t a n i m p o r t a n t extremum p r o p e r t y .

S(x)

SCHOENBERG

394

m

B. T h e c a n e when

IAmyv12 <

C

-m

(See [ 9 , L e c t u r e

m

6] )

.

We

i n t r o d u c e t h e classes o f s e q u e n c e s and f u n c t i o n s a s follows:

(14)

(15)

L:={f(x);

Li

Of c o u r s e

f,

...,f (m-l)

and

W e may also d e s c r i b e

ments of L :

a r e a b s o l u t e l y continuous, f(")(x) ELZ@)

are t h e f a m i l i a r

L;

l;

L2 and

1.

Lzl respectively.

as t h e c l a s s o f s e q u e n c e s o b t a i n e d f r a n e l e -

L2 by n s u c c e s s i v e summation. S i m i l a r l y t h e e l e m e n t s o f

are o b t a i n e d from t h o s e o f

THEOREM 2.

L2

by

n

successive integrations.

76

t h e n t h e i n t e h p o l a t i o n phobLem

has a u n i q u e n o L u t i o n n u c h t h a t

Jhio dolution f(x)

(19)

and

i b

S(x)

han t h e 6oLLowing exthemum p h o p e h t y :

an a t b i t a a h y 6unction ouch t h a t

76

396

ON CARDINAL SPLINE SMOOTHING

f ( v ) = y,

v ,

hat all

then

(21)

1-U7

f (x) =

UnLebh

J-m

x.

d o t aLL keaL

S(X)

I n words: If (Y,)



R2

t h e n t h e s p l i n e i n t e r p o l a n t S ( x ) mini-

r

mizes t h e i n t e g r a l

(22)

among a l l s u f f i c i e n t l y smooth i n t e r p o l a n t s o f If

y, = P ( v ) f o r a l l

v , where P ( x ) E

(y,).

7

,~t h e -n

~

m P(x) ~ ES2nrlnL2,

and t h e r e f o r e

S ( x ) = P ( x ) by t h e u n i c i t y of t h e s o l u t i o n i n Theorem m 2 . However, h e r e I(S) = 0 . I n t h e g e n e r a l c a s e of (y,) E L 2 wemay

therefore say t h a t

S ( x ) i s among a l l i n t e r p o l a n t s o f

t h a t " i s most n e a r l y " a p o l y n o m i a l o f d e g r e e If P(X)

y,

E SZm-1

= P ( u ) , where

n PG, and so

P

(XI

E

IT^^-^ ,

-

1.

P(x)

9

2 m

but

~

~

the

-

~

f ( x ) such t h a t

I (f) <

m

How d o we a c t u a L L y c o n n t h u c t t h e

9 :l

.

S(x)

04

E

ll ,

AOLUtiOnb

hence a f o r t i o r i

(y,)

E

L2

.

T h i s i n s u r e s t h e c o n t i n u i t y of t h e p e r i o d i c f u n c t i o n m

T(u) =

C -m

y,e

ivu

is

.

assume t h a t

(y,)

r again t h a

There

these

t e h p o l a t i a n p k o b l e m n ? To answer t h i s q u e s t i o n l e t u s f o r t h e

(23)

one

S ( x ) = P ( x ) i s t h e unique s o l u t i o n o f The-

orem 1. Theorem 2 does n o t a p p l y h e r e b e c a u s e (y,) no i n t e r p o l a n t

(y,),

in-

moment

SCHOENBERG

366

which w e c a l l t h e g e n e h a t i n g

6uncLLon of t h e sequence ( y v ) . Here and

below w e d e n o t e t h e r e l a t i o n s h i p between a sequenceand its g e n e r a t i n g f u n c t i o n s y m b o l i c a l l y by w r i t i n g

We a l s o r e q u i r e t h e g e n e r a t i n g f u n c t i o n o f t h e sequence ( M 2 m ( ~ ) ) r which

is

Z2,(u)

m- 1

=

C

v=- (m-1)

ivu MZm(v)e

T h i s i s a c o s i n e polynomial o f o r d e r

I

x

I 2

m-1,

m . I t i s r e a d i l y e v a l u a t e d by ( 7 )

@,(u) = 1 , p14(u) = -1 ( ~ + c o s u), 3

Z,(U) =

I

.

because

MZm(x) = 0

if

and w e f i n d t h a t

1 ~ ( 3 3 + 2 6cos

... .

U + C O S ~ U ) ,

It a l s o has the property t h a t

(27)

0 < d 2 m ( ~5) d2,.,,(u)

5

Z2m(0) = 1

for all

u.

I t f o l l o w s t h a t i t s r e c i p r o c a l h a s an expansion

w i t h real c o e f f i c i e n t s

w

,W-v

= W v # t h a t decay e x p o n e n t i a l l y .

Let

us f i n d t h e s t a n d a r d r e p r e s e n t a t i o n

o f t h e s o l u t i o n o f t h e i n t e r p o l a t i o n problem ( 1 7 ) , which r e q u i r e s that

ON CARDINAL SPLINE SMOOTHING

-

C c . M2m(v j J

(30)

j) = y,

397

v.

for all

Furthermore l e t

b e t h e a s y e t unknown g e n e r a t i n g f u n c t i o n o f t h e ( c . ) . S i n c e t h e con-

I

v o l u t i o n o f two s e q u e n c e s h a s a g e n e r a t i n g f u n c t i o n t h a t i s t h e produ c t o f t h e g e n e r a t i n g f u n c t i o n s o f t h e two s e q u e n c e s , w e see by (241, (26) , and (31) , t h a t t h e r e l a t i o n s

( 3 0 ) are e q u i v a l e n t t o t h e rela

-

tion

* (wv)

Now ( 2 8 ) shows t h a t ( c v ) = (y,)

c

(33)

V

c y j wv-j

=

and t h e r e f o r e

v.

for all

j

T h e b e ake t h e c o e 6 d i c i e n A b a d t h e intekpaLating s p l i n e ( 2 9 ) .

EXAMPLES:

1. 16 m = l ,

we o b t a i n

c v = y,

Section

v

f o r all

.

0

= 1 , wv = O ( v # 0 )

,

and

16 m = 2 , w e f i n d ( S e e [ 9 , L e c t u r e 4 ,

51 ) t h a t

W

V

XIv1,

=

2. I f w e choose

shows t h a t

(34)

t h e n $,(u) = 1 , hence w

cv = w

V

.

y,

-

where

6" , w h e r e

X

= -2

+

47

=

-.26795.

6 o = 1, 6 v = O ( V

Therefore t h e s p l i n e

# O),then (33)

388

SCHOENBERG

i s t h e s o l u t i o n o f t h e i n t e r p o l a t i o n problem

L*m-l(4

(35)

=

6v

I

for a l l

v.

The f u n c t i o n ( 3 4 ) i s t h e dundamental & u n c t i o n o f t h e p r o c e s s , and t h e S(X)

A O h t i O M

o f t h e g e n e h U l p k o b l e m (17) LO g i v e n b y m

T h i s c a r d i n a l i n t e r p o l a t i o n f o r m u l a b r i d g e s t h e gap between the linear i n t e r p o l a n t ( 3 ) a n d t h e c a r d i n a l series ( 2 ) . I n f a c t , n o t i c e t h a t i f

m = 1 t h e n ( 3 6 ) r e d u c e s t o ( 3 ) , w h i l e w e have

l i m S2m-1(~)= m+m

(37)

Also every d e r i v a t i v e

sin

TIX

TIx

(k) (x) c o n v e r g e s t o t h e corresponding derivaS2m-l

t i v e of t h e r i g h t s i d e of

(37) I uniformly f o r a l l real

x

I n o u r d i s c u s s i o n w e have assumed t h a t ( 2 3 ) h o l d s .

the

tULatbJMd

. However,

( 3 3 ) , ( 2 9 ) , and ( 3 6 ) a k e v a l i d d o % b o t h T h e o t e m d 1 and

2, undeh t h e i t t e n p e c t i w e a d d u m p t i a n d .

PART 11.

THE CARDINAL SMOOTHING SPLINE

1. STATEMENT OF THE PROBLEM:

We assume now t h a t

(1)

and r e s t r i c t o u r s e l v e s t o r e a l - v a l u e d d a t a and f u n c t i o n s .

We

also

r e c a l l t h e d e f i n i t i o n s ( 3 . 1 4 ) and ( 3 . 1 5 ) o f P a r t I , o f t h e c l a s s e s ly and

L:

.

I n view o f t h e i n c l u s i o n r e l a t i o n s

ON CARDINAL SPLINE SMOOTHING

( S e e [ 9 , p. 1 0 4 1 )

,

399

w e o b s e r v e t h a t (1) i m p l i e s t h a t (y,)

satisfiesthe

a s s u m p t i o n s o f Theorem 2 f o r a l l m .

We a t e g i v e n m and a n m o o t h i n g patrameteh

THE PROBLEM:

E < 0.

Among

aLL 6unctionn

we w i b h t o d i n d t h e . b o l u t i o n

I

m

J(f) =

(4)

E

*

06

t h e phobLem

2

m

+

( f ( m ) (x)) d x

C

-m

-03

-

y V l 2 = minimum.

I n bOlVing t h e minimum p t o b l e m ( 4 ) we may tenLkiot t h e choice

LEMMA 1: 06

(f (v)

adminnible dunctianb

f ( x ) t o t h e eeementb

06

(5)

PROOF:

If

f ( x ) is such t h a t

J(f) <

m

,

then ( f ( v )

a p p l y Theorem 2 t o t h e s e q u e n c e ( f ( v ) ), and l e t

be such t h a t

s(v)

= f

(v) for a l l

v

.

But t h e n

-

yv) E L2

.

We

4w

SCHOENBERG

and so

in view of the extremum property of

S ( x )

as expressed by (3.21)

Theorem 2. Therefore, for any

f ( x ) , the spline

f (x), produces a value

J(f).

Let

U6

J(s)

of

s(x) that interpolates

thehedote d i n d t h e nolutian

o d t h e m i n i m u m ptobLem

I, m

J(S) =

(8)

E

m

(S(m))2dx+ C

-w

(S(v)

-

Y,)~ = minimum.

Here we need another

LEMMA 2:

7 6 ( 7 ) o a t i d d i e n S(x)

E

L2 (R),

(S(m)(x))2dx = -00

PROOF: From (7) we find that

hence aLno (c.) E L2 , t h e n

C yj-” c . c j, v

3

,

ON CARDINAL SPLINE SMOOTHING

401

i s t h e e v e n s e q u e n c e d e f i n e d by

where (y,)

where, t o s i m p l i f y n o t a t i o n s w e dropped t h e s u b s c r i p t 2m o f M 2 m ( ~ ) . I n t e g r a t i o n s by p a r t s show t h a t

(-ilm-' M(2m-1) x)

Observe t h a t

Jm

-m

M i ( X ) M ( ~ ~ -( x~ )- r ) d x

.

i s a s t e p f u n c t i o n assuming i n c o n s e c u t i v e

u n i t i n t e r v a l s t h e v a ue s

... 1 0 ,

(14)

011,

-

(2m-1 1

)

I

(2m-1 2 )

...

I

1 - 1 1

0, 0,

... .

T h i s sequence h a s t h e g e n e r a t i n g f u n c t i o n

except f o r a s h i f t f a c t o r

eiuk which w e d i s r e g a r d . N o w ( 1 3 ) indicates

t h a t ( y ) i s t h e c o n v o l u t i o n o f t h e sequence ( 1 4 ) w i t h t h e sequence

r

However, i n (13) t h e s e q u e n c e Cu avbv-r

.

If

(yr)

appears

as a sum o f

w e pass from ( a v ) t o t h e r e v e r s e d s e q u e n c e

o b t a i n a genuine c o n v o l u t i o n

Cva-vbv-r

.

L e t us t h e r e f o r e

the

form

(a_"),

we

reverse

t h e f i r s t s e q u e n c e (14). As w e o b t a i n t h e g e n e r a i n g f u n c t i o n o f t h e r e v e r s e d sequence by c h a n g i n g

u into

- u i n its o r i g i n a l generating

function, we f i n d the generating function of factor

eiuk) t h e p r o d u c t

(yr

t o b e ( u p t o a shift

402

-ium

= e

S i n c e (y,)

(2 sin

u 2m 7)

Z2,(u).

i s an e v e n s e q u e n c e , i t s g e n e r a t i n g f u n c t i o n m u s t b e e v e n ,

and t h e r e f o r e

e s t a b l i s h i n g (10).

2 . SOLUTION OF THE PROBLEM:

From ( a ) ,

( 9 1 , and ( 7 ) w e f i n d t h a t

L e t us minimize t h i s f u n c t i o n of t h e (c,).

tions, we differentiate

-a 2 ack

J(S) =

E

To o b t a i n t h e normalequa-

J(S) obtaining

Z yj-kcj+

C { Z c . M ( w - j ) - y w ) M ( w - k ) = O (kEZ).

j

v

j

7

I f w e sum w i t h i n t h e double-sum o n l y w i t h r e s p e c t t o

where

v , we obtain

ON CARDINAL SPLINE SMOOTHING

(3)

(au)

403

2

.

(d2m(u))

+

The normal e q u a t i o n s t h u s become

or

(4)

C j

+

(clj-k

E

Y,

yj-k)~j=

M2m(~

-

k)

(k

E

However, by ( 3 ) a n d ( 1 . 1 0 ) w e f i n d

and w r i t i n g

(6)

(c,,)

+

C(u),

(y,)

+

T(u)

,

w e f i n d t h e normal e q u a t i o n s (4) t o b e e q u i v a l e n t t o t h e r e l a t i o n

i ( p ~ , ~ ( u +) )E~( 2 s i n + ) 2 m

whence

This e s t a b l i s h e s

~ , , ( u ) ) c ( u ) = ~ ( u 4) 2 m ( u ) t

if).

SCHOENBERG

404

THEOREM 3:

I n tehmh

06

whehe t h e c a e d d i c i e n t n dicientn (c.) 7

04

t h e expannion

w"(E)

=

w-"(E)

d e c a y e x p o n e n t i a L L y , t h e coed-

t h e naLution

0 6 t h e minimum p h o b l e m ,

ahe

W e c a l l t h e s o l u t i o n ( 9 ) t h e cahdinad smoothing n p l i n e .

3.

A dew p h o p e h t i e b A.

06

t h e cahdinad nmoothing n p d i n e

W e have assumed above t h a t

E

S(X) =s(x;E).

> 0 . However, i f w e s e t

E

=O

i n (2.81, i t becomes

and a comparison w i t h t h e e x p a n s i o n ( 3 . 2 8 ) o f P a r t I , w v ( 0 ) = wv

for all

v : T h i n nhawn t h a t S (x)

inteapolafing caadinal npline B.

What

n t h e eddect

a n t h e ahiginai? dada sequence

(S( v )

,

(y,) ?

06

06

shows

that

S ( x ; O ) = S ( x ) &educed A0 t h e

Theohem

2.

t h e nmoathing n p l i n e

S(x) = S(x;

E)

T h i s w e answer by determining the "smoothed"

t o compare i t w i t h ( y , ) .

By ( 2 . 9 ) and (2.10)we find

406

ON CARDINAL SPLINE SMOOTHING

a n d t h e r e f o r e , by ( 2 . 7 )

I

I n terms o f t h e e x p a n s i o n

1 (2 sin

(3)

=

u 2m

c

eivu

uv(E)

V

Z2,(U)

+

( 2 ) shows t h a t t h e s e q u e n c e ( S ( V ; E ) ) a h i n e n d h o m t h e d a t a (y,)

by t h e

n m o o t h i n g dohmuea

Observe t h a t by ( 2 . 8 ) a n d ( 3 ) t h e c o e f f i c i e n t s i n terms of

W v ( ~ )

by

u"(E)

= C MZm (v

j

-

j)

U ~ ( E )

are e x p r e s s e d

W.(E).

3

Is (4) a s m o o t h i n g f o r m u l a a c c o r d i n g t o o u r d e f i n i t i o n o f P a r t I , S e c t i o n l ? T h a t it i s o n e w e see i f w e i n s p e c t i t s c h a r a c t e r i s t i c

function

K(u;E)

(5)

1 u 2m (2 sin T )

=

!d2m(u)

+

f o r it is evident t h a t

0 < K(u;E)

(6)

C. cheabing

< K(O;E)

T h e b m o o t h i n g poweh E

.

06

=

1

for

0 < u < 2r

.

t h e 6ohmuLa ( 4 ) i n c h e a b e n w i t h

I n [ 4 , D e f i n i t i o n 2 , p . 5 3 1 w e g a v e good r e a s o n s

in-

for

406

SCHOENBERG

t h e f o l l o w i n g d e f i n i t i o n : Of two d i f f e r e n t smoothing formulae h a v i n g the characteristic functions

d ( u ) and

$(u)

, we

s a y t h a t t h e second

h a s g r e a t e r smoothing power, p r o v i d e d t h a t

(7)

/J(U)

However, i f

I 5

0 <

E

Id(u)I

<

El

for a l l

u, excluding e q u a l i t y f o r a l l

u.

i t i s c l e a r by ( 5 ) t h a t

and t h e c r i t e r i o n ( 7 ) i s s a t i s f i e d .

D.

The deghee

06

eXUCtneAA

0 6 ,the nmootking 6omonuRa

(4) A = h - l .

T h i s f o l l o w s from ( 1 . 7 ) o f P a r t I , b e c a u s e ( 5 ) shows t h a t w e h a v e t h e e x p a n s i o n i n powers o f

E.

06

u

I f w e d r o p o u r a s s u m p t i o n (1.1) , and assume o n l y t h a t ( y v )

poweh ghowth, Rhen

b y t h e dohmutae(2.8)

,

#in conb&ucfi#n

06 t h e bmvotking

(2.10) , and (2.9) , hemaind appticabLe.0f murse, J ( S ) , of

i t s e a r l i e r connection with t h e funtional holds. I n f a c t we w i l l f i n d t h a t sumably, it i s s t i l l t r u e t h a t o u r t h a t (y,)

npfine S ( x ) = S ( x ; E )

J(S) =

m

( 1 . 8 ) , no l o n g e r

f o r a l l s p l i n e s S . Pre-

S ( X ; E ) minimizes

J ( S ) , provided

s a t i s f i e s the condition

o f Theorem 2 . However, t h i s I was n o t a b l e t o e s t a b l i s h . I n any c a s e I recommend t h e c a r d i n a l smoothing s p l i n e ( S ( X ; E ) ) , which r e p r e s e n t s t h e m o d i f i c a t i o n , found more t h a n 30 y e a r s l a t e r , o f may war-time approach t o t h e problem o f c a r d i n a l smoothing.

407

ON CARDINALSPLINE SMOOTHING

REFERENCES

[ 11

T. N.

E.

GREVILLE, On s t a b i l i t y o f l i n e a r s m o o t h i n g

[ 21

T. N.

E.

GREVILLE, On a p r o b l e m of E .

SIAM J. N u m . A n a l y s i s , 3 ( 1 9 6 6 ) , p p . 1 5 7 - 1 7 0 .

s m o o t h i n g , SIAM J . Math. A n a l . [ 31

,

L.

De Forest i n iterated

5(1974) , pp.

376

FRITZ J O H N , On i n t e g r a t i o n o f p a r a b o l i c e q u a t i o n s by m e t h o d s , Corn. on P u r e a n d Appl. Math.

[ 41

formulas,

I.

J.

,

- 398. difference

5 ( 1 9 5 2 ) ,pp.155-211.

SCHOENBERG, C o n t r i b u t i o n s t o t h e p r o b l e m o f approximation

o f e q u i d i s t a n t d a t a by a n a l y t i c f u n c t i o n s , Q u a r t . o f Appl. Math., [ 51

4 ( 1 9 4 6 ) , P a r t A, p p . 45 - 9 9 ,

P a r t B , pp. 1 1 2 - 1 4 1 .

I. J. SCHOENBERG, Some a n a l y t i c a l a s p e c t s o f t h e p r o b l e m s

of

s m o o t h i n g , C o u r a n t A n n i v e r s a r y volume ".Sfidh% and en nay^", New York, 1 9 4 8 , p p .

351 - 3 7 0 .

[ 61

I . J. SCHOENBERG, On s m o o t h i n g o p e r a t i o n s a n d t h e i r g e n e r a t i n g

[ 71

I . J . SCHOENBERG, S p l i n e f u n c t i o n s a n d t h e p r o b l e m o f g r a d u a -

f u n c t i o n s , B u l l . Amer. Math. SOC., 59 ( 1 9 5 3 ) , p p . 1 9 9

t i o n , Proc. N a t . [ 81

Acad. S c i . 5 2 ( 1 9 6 4 ) , pp.

- 230.

947 - 9 5 0 .

I . J . SCHOENBERG, C a r d i n a l i n t e r p o l a t i o n and s p l i n e

functions

11. I n t e r p o l a t i o n o f d a t a o f power g r o w t h , J . Approx. The-

o r y , 6 ( 1 9 7 2 ) , pp. 4 0 4 [ 9

1

- 420.

I . J . SCHOENBERG, C a k d i n a b h p l i n e . i n t e t p o e a t i o n ,

Reg.

Conf.

Monogr. NQ 1 2 , 1 2 5 p a g e s , SIAM, P h i l a d e l p h i a , 1 9 7 3 .

[lo]

E . T . WHITTAKER a n d G.

ROBINSON, T h e caecueun o d o b n e n v a t i a n n ,

B l a c k i e a n d Son, London, 1924.

D e p a r t m e n t of M a t h e m a t i c s U n i t e d S t a t e s M i l i t a r y Academy

West P o i n t , N e w York 1 0 9 9 6