0% found this document useful (0 votes)

48 views

Proiecte Functii Spline

The bspligui command launches a graphical user interface (GUI) that allows the user to interactively experiment with and observe the behavior of B-spline curves as their knot sequences are modified. Key observations that can be made include: the B-spline is positive on the interval defined by the first and last knots; it is zero outside this interval; its maximum value is never more than 1 and occurs at knots with high multiplicity; it is a piecewise polynomial of degree k-1 between knots; and adding or removing knots changes the degree of the polynomial pieces by 1. The GUI provides an interactive way to explore these fundamental properties of B-splines.

Uploaded by

Cristina Nico

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

48 views

Proiecte Functii Spline

Uploaded by

Cristina Nico

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 75

aptknt

Purpose 3aptknt
Acceptable knot sequence

Syntax knots = aptknt(tau,k)

[knots,k] = aptknt(tau,k)

Description Assume that tau has at least k entries, is nondecreasing, and satisfies
tau(i)<tau(i+k-1) for all i. Then the knot sequence knots returned by the
command aptknt(tau,k) is acceptable for interpolation to data at the sites tau
in the sense that there is exactly one spline of order k with knot sequence knots
that matches given values at those sites. This is so because the sequence knots
returned satisfies the Schoenberg-Whitney conditions
knots(i) < tau(i) < knots(i+k), i=1:length(tau)

with equality only at the extreme knots, each of which occurs with exact
multiplicity k.
If tau has fewer than k entries, then k is reduced to the value length(tau); the
k used is optionally returned. An error results if tau fails to be nondecreasing
and/or tau(i) equals tau(i+k-1) for some i.

Examples If tau is equally spaced, e.g., equal to linspace(a,b,n) for some n>=4, and y is
a sequence of the same size as tau, then sp = spapi(aptknt(tau,4),tau,y)
gives the cubic spline interpolant with the not-a-knot end condition. This is the
same cubic spline as produced by the command spline(tau,y), but in B-form
rather than ppform.

Algorithm The (k-1)-point averages sum(tau(i+1:i+k-1))/(k-1) of the sequence tau, as

supplied by aveknt(tau,k), are augmented by a k-fold tau(1) and a k-fold
tau(end). In other words, the command gives the same result as
augknt([tau(1),aveknt(tau,k),tau(end)],k), provided tau has at least k
entries and k is greater than 1.

Syntax [augknot,addl] = augknt(knots,k)

[augknot,addl] = augknt(knots,k,mults)

Description augknt(knots,k) returns a nondecreasing and augmented knot sequence that

has the first and last knot with exact multiplicity k. (This may actually shorten
the knot sequence.) Also optionally returns the number addl of knots added on
the left. (This number may be negative.)

augknt(knots,k,mults) makes sure that the augmented knot sequence

returned will, in addition, contain each interior knot mults times. If mults has
exactly as many entries as there are interior knots, then the j th one will
appear mults(j) times. Otherwise, the uniform multiplicity mults(1) is used.
If knots is strictly increasing, this ensures that the splines of order k with knot
sequence augknot satisfy k-mults(j) smoothness conditions across
knots(j+1), j=1:length(knots)-2.

Examples If you want to construct a cubic spline on the interval [a..b], with two
continuous derivatives, and with the interior break sequence xi, then
augknt([a,b,xi],4) is the knot sequence you should use.

If you want to use Hermite cubics instead, i.e., a cubic spline with only one
continuous derivative, then the appropriate knot sequence is
augknt([a,xi,b],4,2).

augknt([1 2 3 3 3],2) returns the vector [1 1 2 3 3], as does

augknt([3 2 3 1 3],2). In either case, addl would be 1.

3-8
aveknt

Purpose 3aveknt
Provide knot averages

Syntax tstar = aveknt(t,k)

Description aveknt(t,k) returns the averages of successive k-1 knots, i.e., the sites

ti : = ( t i + 1 + + t i + k 1 ) ( k 1 ) , i = 1 :n

which are recommended as good interpolation site choices when interpolating

n+k
from splines of order k with knot sequence t = ( t i ) i = 1 .

Examples aveknt([1 2 3 3 3],3) returns the vector [2.5000 3.0000], while

aveknt([1 2 3],3) returns the empty vector.

With k and the strictly increasing sequence breaks given, the statements
t = augknt(breaks,k); x = aveknt(t);
sp = spapi(t,x,sin(x));

provide a spline interpolant to the sine function on the interval

[breaks(1)..breaks(end)].

For sp the B-form of a scalar-valued univariate spline function, and with tstar
and a computed as
tstar = aveknt(fnbrk(sp,'knots'),fnbrk(sp,'order'));
a = fnbrk(sp,'coefs');

the points ( tstar(i), a(i) ) constitute the control points of the spline, i.e., the
vertices of the splines control polygon.

Syntax [nb,rows,ncols,last,blocks] = bkbrk(blokmat)

bkbrk(blokmat)

Description [nb,rows,ncols,last,blocks]=bkbrk(blokmat) returns the details of the

almost block-diagonal matrix contained in blokmat, with rows and last
nb-vectors, and blocks a matrix of size [sum(rows),ncols].

This utility program is not likely to be of interest to the casual user. It is used
in slvblk to decode the information, provided by spcol, about a spline
collocation matrix in an almost block diagonal form especially suited for
splines. But bkbrk can also decode the almost block-diagonal form used in [1].

bkbrk(blokmat) returns nothing, but the details are printed out. This is of use
when trying to understand what went wrong with such a matrix.

Syntax [knots,index] = brk2knt(breaks,mults)

Description The sequence knots is the sequence breaks but with breaks(i) occurring
mults(i) times, all i. In particular, breaks(i) will not appear unless
mults(i)>0. If, as one would expect, breaks is a strictly increasing sequence,
then knots contains each breaks(i) exactly mults(i) times.
If mults does not have exactly as many entries as does breaks, then all
mults(i) are set equal to mults(1).

If, as one would expect, breaks is strictly increasing and all multiplicities are
positive, then, for each i, index(i) is the first place in knots at which
breaks(i) appears.

Examples The statements

t = [1 1 2 2 2 3 4 5 5];
[xi,m] = knt2brk(t);
tt = brk2knt(xi,m)

give [1 2 3 4 5] for xi, [2 3 1 1 2] for m, and, finally, t for tt.

bspline(t,window) does the plotting in the subplot window specified by

window; see the MATLAB command subplot for details.

pp = bspline(t) plots nothing but returns the ppform of the B-spline.

Examples The statement pp=fn2fm(spmak(t,1),'pp') has the same effect as the

statement pp=bspline(t).

Syntax tau = chbpnt(t,k)

[tau,sp] = chbpnt(t,k,tol)

Description tau = chbpnt(t,k) are the extreme sites of the Chebyshev spline of order k
with knot sequence t. These are particularly good sites at which to interpolate
data by splines of order k with knot sequence t because the resulting
interpolant is often quite close to the best uniform approximation from that
spline space to the function whose values at tau are being interpolated.

[tau,sp] = chbpnt(t,k) also returns, in sp, the Chebyshev spline.

chbpnt(t,k,tol) also specifies the tolerance tol to be used in the iterative

process that constructs the Chebyshev spline. This process is terminated when
the relative difference between the absolutely largest and the absolutely
smallest local extremum of the spline is smaller than tol. The default value for
tol is .001.

Examples chbpnt([-ones(1,k),ones(1,k)],k) provides (approximately) the extreme

sites on the interval [-1 .. 1] of the Chebyshev polynomial of degree k-1.
If you have decided to approximate the square-root function on the interval
[0 .. 1] by cubic splines, with knot sequence t as given by
k = 4; n = 10; t = augknt(((0:n)/n).^8,k);

then a good approximation to the square-root function from that specific spline
space is given by
x = chbpnt(t,k); sp = spapi(t,x,sqrt(x));

as is evidenced by the near equi-oscillation of the error.

Algorithm The Chebyshev spline for the given knot sequence and order is constructed
iteratively, using the Remez algorithm, using as initial guess the spline that
takes alternately the values 1 and -1 at the sequence aveknt(t,k). The demo
chebdem gives a detailed discussion of one version of the process as applied to
a particular example.

pp = csape(x,y,conds) lets you choose the end conditions to be used, from a

rather large and varied catalog, by proper choice of conds. If needed, you supply
the corresponding end condition values as additional data values, with the first
(last) data value taken as the end condition value at the left (right) end. In
other words, in that case, s(x(j)) matches y(:,j+1) for all j, and the variable
endcondvals used in the detailed description below is set to y(:,[1 end]). For
some choices of conds, these end condition values need not be present and/or
are ignored when present.
conds may be a string whose first character matches one of the following:
'complete' or 'clamped', 'not-a-knot', 'periodic', 'second',
'variational', with the following meanings.

'complete' or Match endslopes (as given, with default as under

'clamped' default).
'not-a-knot' Make second and second-last sites inactive knots
(ignoring end condition values if given).
'periodic' Match first and second derivatives at left end with those
at right end.
'second' Match end second derivatives (as given, with default [0 0],
i.e., as in 'variational').
'variational' Set end second derivatives equal to zero (ignoring end
condition values if given).
default Match endslopes to the slope of the cubic that matches the
first four data at the respective end (i.e., Lagrange).

3-16
csape

By giving conds as a 1-by-2 matrix instead, it is possible to specify different

i
conditions at the two ends. Explicitly, the i th derivative, D s , is given the
value endcondvals(:, j ) at the left ( j is 1) respectively right ( j is 2) end in case
conds(j) is i, i = 1:2 . There are default values for conds and/or
endcondvals.
Available conditions are:

clamped Ds(e) = endcondvals(:,j) if conds(j) == 1

2
curved D s(e) = endcondvals(:,j) if conds(j) == 2

Lagrange Ds(e) = Dp(e) default

r r
periodic D s(a) = D s(b), r = 1, 2 if conds == [0 0]
2
variational D s( e) = 0 if conds(j) == 2 &
endcondvals(:,j) == 0

Here, e is a ( e is b ), i.e., the left (right) end, in case j is 1 (j is 2), and (in the
Lagrange condition) p is the cubic polynomial that interpolates to the given
data at e and the three sites nearest e .
If conds(j) is not specified or is different from 0, 1, or 2, then it is taken to be
1 and the corresponding endcondvals(:,j) is taken to be the corresponding
default value.
The default value for endcondvals(:,j) is the derivative of the cubic interpolant
at the nearest four sites in case conds(j) is 1, and is 0 otherwise.
It is also possible to handle gridded data, by having x be a cell array containing
m univariate meshes and, correspondingly, having y be an m -dimensional
array (or an m + r -dimensional array if the function is to be r -valued).
Correspondingly, conds is a cell array with m entries, and end condition
values may be correspondingly supplied in each of the m variables. This, as the
last example below, of bicubic spline interpolation, makes clear, may require
you to supply end conditions for end conditions.
This command calls on a much expanded version of the Fortran routine CUBSPL
in PGS.

3-17
csape

Examples csape(x,y) provides the cubic spline interpolant with the Lagrange end
conditions, while csape(x,y,[2 2]) provides the variational, or natural cubic
spline interpolant, as does csape(x,y,'v').
csape([-1 1],[3 -1 1 6],[1 2]) provides the cubic polynomial p for which
2 3
Dp( 1) = 3 , p( 1) = 1 , p(1) = 1 , D p(1) = 6 , i.e., p(x) = x .
Finally, csape([-1 1],[-1 1]) provides the straight line p for which
p(1) = 1, i.e., p ( x ) = x .
End conditions other than the ones listed earlier can be handled along the
following lines. Suppose that we want to enforce the condition
2
( s ) := aDs ( e ) + bD s ( e ) = c
for given scalars a , b , and c , and with e equal to x(1). Then one could compute
the cubic spline interpolant s 1 to the given data using the default end condition
as well as the cubic spline interpolant s 0 to zero data and some (nontrivial) end
condition at e , and then obtain the desired interpolant in the form

s = s1 + ( ( c ) ( s1 ) ) ( s0 ) s0

Here are the (not inconsiderable) details (in which the first polynomial piece of
s 1 and s 0 is pulled out to avoid differentiating all of s 1 and s 0 ):
pp1 = csape(x,y);
dp1 = fnder(fnbrk(pp1,1));
pp0 = csape(x,[1,zeros(1,length(y)),0],[1,0]);
dp0 = fnder(fnbrk(pp0,1));
e = x(1);
lam1 = a*fnval(dp1,e) + b*fnval(fnder(dp1),e);
lam0 = a*fnval(dp0,e) + b*fnval(fnder(dp0),e);
pp = fncmb(pp0,(c-lam1)/lam0,pp1);

As a multivariate vector-valued example, here is a sphere, done as a

parametric bicubic spline, 3D-valued, using prescribed slopes in one direction
and periodic end conditions in the other:
x = 0:4; y=-2:2; s2 = 1/sqrt(2);
clear v
v(3,:,:) = [0 1 s2 0 -s2 -1 0].'*[1 1 1 1 1];
v(2,:,:) = [1 0 s2 1 s2 0 -1].'*[0 1 0 -1 0];
v(1,:,:) = [1 0 s2 1 s2 0 -1].'*[1 0 -1 0 1];

3-18
csape

sph = csape({x,y},v,{'clamped','periodic'});
values = fnval(sph,{0:.1:4,-2:.1:2});
surf(squeeze(values(1,:,:)),squeeze(values(2,:,:)),...
squeeze(values(3,:,:))); axis equal, axis off

The lines involving fnval and surf could have been replaced by the simple
command: fnplt(sph). Note that v is a 3-dimensional array, with v(:,i+1,j)
the 3-vector to be matched at (x(i),y(j)), i=1:5, j=1:5. Note further that,
in accordance with conds{1} being 'clamped', size(v,2) is 7 (and not 5), with
the first and last entry of v(r,:,j) specifying the end slopes to be matched.
Here is a bivariate example that shows the need for supplying end conditions
of end conditions when supplying end conditions in both variables. We
reproduce the bicubic polynomial g(x,y) = x^3y^3 by complete bicubic
interpolation. We derive the needed data, including end condition values,
directly from g in order to make it easier for you to see just how the end
condition values must be placed. We also check the result.
sites = {[0 1],[0 2]}; coefs = zeros(4,4); coefs(1,1) = 1;
g = ppmak({bx,by},coefs);
Dxg = fnval(fnder(g,[1 0]),sites);
Dyg = fnval(fnder(g,[0 1]),sites);
Dxyg = fnval(fnder(g,[1 1]),sites);
f = csape(sites,[Dxyg(1,1), Dxg(1,:), Dxyg(1,2); ...
Dyg(:,1), fnval(g,sites), Dyg(:,2) ; ...
Dxyg(2,1), Dxg(2,:), Dxyg(2,2)], ...
{'complete','complete'});
if any(squeeze(fnbrk(f,'c'))-coefs), 'this is wrong', end

Algorithm The relevant tridiagonal linear system is constructed and solved using the
sparse matrix capabilities of MATLAB.

csapi(x,y,xx) is the same as fnval(csapi(x,y),xx), i.e., the values of the

interpolating cubic spline at the sites specified by xx are returned.
This command is essentially the MATLAB function spline, which, in turn, is
a stripped-down version of the Fortran routine CUBSPL in PGS, except that
csapi (and now also spline) accepts vector-valued data and can handle
gridded data.

Examples See the demo csapidem for various examples.

Up to rounding errors, and assuming that x is a vector with at least four
entries, the statement pp = csapi(x,y) should put the same spline into pp as
does the statement
pp = fn2fm(spapi(augknt(x([1 3:(end-2) end]),4),x,y),'pp');

except that the description of the spline obtained this second way will use no
break at x(2) and x(n-1).

3-20
csapi

Here is a simple bivariate example, a bicubic spline interpolant to the Mexican

Hat function being plotted:
x =.0001+[-4:.2:4]; y = -3:.2:3;
[yy,xx] = meshgrid(y,x); r = pi*sqrt(xx.^2+yy.^2); z = sin(r)./r;
bcs = csapi( {x,y}, z ); fnplt( bcs ), axis([-5 5 -5 5 -.5 1])

Note the reversal of x and y in the call to meshgrid, needed since MATLAB
likes to think of the entry z(i,j) as the value at (x(j),y(i)) while this toolbox
follows the Approximation Theory standard of thinking of z(i,j) as the value
at (x(i),y(j)). Similar caution has to be exerted when values of such a
bivariate spline are to be plotted with the aid of the MATLAB mesh function, as
is shown here (note the use of the transpose of the matrix of values obtained
from fnval).
xf = linspace(x(1),x(end),41); yf = linspace(y(1),y(end),41);
mesh(xf, yf, fnval( bcs, {xf, yf}).')

Algorithm The relevant tridiagonal linear system is constructed and solved, using the
MATLAB sparse matrix capability.
The not-a-knot end condition is used, thus forcing the first and second
polynomial piece of the interpolant to coincide, as well as the second-to-last and
the last polynomial piece.

Syntax [pp,p] = csaps(x,y)

[pp,p] = csaps(x,y,p)
[pp,p] = csaps(x,y,p,[],w)
[values,p] = csaps(x,y,p,xx)
[values,p] = csaps(x,y,p,xx,w)
[...] = csaps({x1,...,xm},y,...)

Description pp=csaps(x,y) returns the ppform of a cubic smoothing spline f to the given
data x,y, with the value of f at the data site x(j) approximating the data
value y(:,j), for j=1:length(x). The values may be scalars, vectors, matrices,
even ND-arrays. Data points with the same site are replaced by their
(weighted) average, with its weight the sum of the corresponding weights.
This smoothing spline f minimizes
n
2
w ( j ) y ( :, j ) f ( x ( j ) )
2 2
p + (1 p) (t) D f( t) dt
j=1
2
Here, z stands for the sum of the squares of all the entries of z , n is the
number of entries of x, and the integral is over the smallest interval containing
all the entries of x. The default value for the weight vector w in the error
measure is ones(size(x)). The default value for the piecewise constant weight
2
function in the roughness measure is the constant function 1. Further, D f
denotes the second derivative of the function f . The default value for the
smoothing parameter, p, is chosen in dependence on the given data sites x.

pp=csaps(x,y,p) lets you supply the smoothing parameter. The smoothing

parameter determines the relative weight you would like to place on the
contradictory demands of having f be smooth vs having f be close to the data.
For p = 0, f is the least-squares straight line fit to the data, while, at the other
extreme, i.e., for p = 1, f is the variational, or natural cubic spline interpolant.
As p moves from 0 to 1, the smoothing spline changes from one extreme to the
3
other. The interesting range for p is often near 1 ( 1 + h 6 ) , with h the
average spacing of the data sites, and it is in this range that the default value
for p is chosen. For uniformly spaced data, one would expect a close following
3
of the data for p = 1 ( 1 + h 60 ) and some satisfactory smoothing for

3-22
csaps

3
p = 1 ( 1 + h 0.6 ) . You can input a p > 1, but this leads to a smoothing
spline even rougher than the variational cubic spline interpolant.
If the input p is negative or empty, then the default value for p is used. Whether
or not you specify the smoothing parameter, you can always obtain the value of
p actually used, as the optional second output argument. This is important for
experimentation which you might start with [pp,p]=csaps(x,y) in order to
obtain a reasonable first guess for p.
If you have difficulty choosing p but have some feeling for the size of the noise
in y, consider using instead spaps(x,y,tol) which, in effect, chooses p in such
a way that the roughness measure
2
(t) D
2
s ( t ) dt

is as small as possible subject to the condition that the error measure

w j y ( :, j ) s ( x ( j ) )
2

does not exceed the specified tol . This usually means that the error measure
equals the specified tol .
The weight function in the roughness measure can, optionally, be specified
as a (non-negative) piecewise constant function, with breaks at the data sites
x , by inputing for p a vector whose ith entry provides the value of on the
interval (x(i-1) .. x(i)) for i=2:length(x). The first entry of the input vector
p continues to be used as the desired value of the smoothness parameter p. In
this way, it is possible to insist that the resulting smoothing spline be smoother
(by making the weight function larger) or closer to the data (by making the
weight functions smaller) in some parts of the interval than in others.

pp = csaps(x,y,p,[],w) lets you specify the weights w in the error measure,

as a vector of nonnegative entries of the same size as x.

[...] = csaps({x1,...,xm},y,...) provides the ppform of an m-variate

tensor-product smoothing spline to data on a rectangular grid. Here, the first
argument is a cell-array, containing the vectors x1, ..., xm, of lengths n1, ..., nm,
respectively. Correspondingly, y is an array of size [n1,...,nm] (or of size
[d,n1,...,nm] in case the data are d-valued), with y(:, i 1, , i m) the given
(perhaps noisy) value at the grid site x1(i 1), , xm(i m) ).

3-23
csaps

In this case, p if input must be a cell-array with m entries or else an m-vector,

except that it may also be a scalar or empty, in which case it is taken to be the
cell-array whose m entries all equal the p input. The optional second output
argument will always be a cell-array with m entries.
Further, w if input must be a cell-array with m entries, with w{i} either empty,
to indicate the default choice, or else a nonnegative vector of the same size as
xi.

csaps(x,y,p,xx) is the same as fnval(csaps(x,y,p),xx).

csaps(x,y,p,xx,w) is the same as fnval(csaps(x,y,p,[],w),xx).

Examples Example 1.
x = linspace(0,2*pi,21); y = sin(x)+(rand(1,21)-.5)*.1;
pp = csaps(x,y, .4, [], [ones(1,10), repmat(5,1,10), 0] );

returns a smooth fit to the (noisy) data that is much closer to the data in the
right half, because of the much larger error weight there, except for the last
data point, for which the weight is zero.
pp1 = csaps(x,y, [.4,ones(1,10),repmat(.2,1,10)], [], ...
[ones(1,10), repmat(5,1,10), 0]);

uses the same data, smoothing parameter, and error weight but chooses the
roughness weight to be only .2 in the right half of the interval and gives,
correspondingly, a rougher but better fit there, -- except for the last data point
which is ignored.
A plot showing both examples for comparison can now be obtained by
fnplt(pp); hold on, fnplt(pp1,'r'), plot(x,y,'ok'), hold off
title(['cubic smoothing spline, with right half treated ',...
'differently:'])
xlabel(['blue: larger error weights; ', ...
'red: also smaller roughness weights'])

The resulting plot is shown below.

3-24
csaps

cubic smoothing spline, with right half treated differently:

1.5

0.5

1.5
0 1 2 3 4 5 6 7
blue: larger error weights; red: also smaller roughness weights

Example 2. As a bivariate example, we add some uniform noise, from the

interval [-1/2 .. 1/2], to values of the MATLAB peaks function on a 51-by-61
uniform grid, obtain smoothed values for these data from csaps, along with the
smoothing parameters chosen by csaps, and then plot these smoothed values.
x = {linspace(-2,3,51),linspace(-3,3,61)};
[xx,yy] = ndgrid(x{1},x{2}); y = peaks(xx,yy);
rand('state',0), noisy = y+(rand(size(y))-.5);
[smooth,p] = csaps(x,noisy,[],x);
surf(x{1},x{2},smooth.'), axis off

Note the need to transpose the array smooth. For a somewhat smoother
approximation, use a slightly smaller value of p than the one, .9998889, used
above by csaps. The final plot is obtained by the following:
smoother = csaps(x,noisy,.996,x);
figure, surf(x{1},x{2},smoother.'), axis off

3-25
csaps

Algorithm csaps is an implementation of the Fortran routine SMOOTH from PGS.

The default value for p is determined as follows. The calculation of the

smoothing spline requires the solution of a linear system whose coefficient
matrix has the form p*A + (1-p)*B, with the matrices A and B depending on
the data sites x. The default value of p makes p*trace(A) equal
(1-p)*trace(B).

Syntax curve = cscvn(points)

Description cscvn(points) returns a parametric variational, or natural, cubic spline

curve (in ppform) passing through the given sequence points(:, j), j = 1 :end .
The parameter value t ( j ) for the jth point is chosen by Eugene Lees [1]
centripetal scheme, i.e., as accumulated square root of chord length:

points ( : ,i + 1 ) points ( : ,i ) 2
i<j

If the first and last point coincide (and there are no other repeated points), then
a periodic cubic spline curve is constructed. However, double points result in
corners.

Examples The following provides the plot of a questionable curve through some points
(marked as circles):
points=[0 1 1 0 -1 -1 0 0; 0 0 1 2 1 0 -1 -2];
fnplt(cscvn(points)); hold on,
plot(points(1,:),points(2,:),'o'), hold off

Here is a closed curve, good for 14 February, with one double point:
c=fnplt(cscvn([0 .82 .92 0 0 -.92 -.82 0; .66 .9 0 ...
-.83 -.83 0 .9 .66])); fill(c(1,:),c(2,:),'r'), axis equal

Algorithm The break sequence t is determined as

t = cumsum([0;((diff(points.').^2)*ones(d,1)).^(1/4)]).';

and csape (with either periodic or variational end conditions) is used to

construct the smooth pieces between double points (if any).

References [1] E.T.Y. Lee, Choosing nodes in parametric curve interpolation,

Computer-Aided Design 21 (1989), 363370.

3-27
fn2fm

Purpose 3fn2fm
Convert to specified form

Syntax g = fn2fm(f,form)
sp = fn2fm(pp,'B-',sconds)
g = fn2fm(f)

Description g = fn2fm(f,form) describes the same function as is described by f, but in the

form specified by the string form. Choices for form are 'B-', 'pp', 'BB', 'rB',
'rp', for the B-form, the ppform, the BBform, and the two rational spline
forms, respectively.
The B-form describes a function as a weighted sum of the B-splines of a given
order k for a given knot sequence, and the BBform (or, Bernstein-Bzier form)
is the special case when each knot in that sequence appears with maximal
multiplicity, k. The ppform describes a function in terms of its local polynomial
coefficients. The B-form is good for constructing and/or shaping a function,
while the ppform is cheaper to evaluate.
Conversion from a polynomial form to the corresponding rational form is
possible only if the function in the polynomial form is vector-valued, in which
case its last component is designated as the denominator. Converting from a
rational form to the corresponding polynomial form simply reverses this
process by reinterpreting the denominator of the function in the rational form
as an additional component of the piecewise polynomial function.
Conversion to or from the stform is not possible at present.
If form is 'B-' (and f is in ppform), then the actual smoothness of the function
in f across each of its interior breaks has to be guessed. This is done by looking,
for each interior break, for the first derivative whose jump across that break is
not small compared to the size of that derivative nearby. The default tolerance
used in this is 1.e-12.

g = fn2fm(f, B- ,sconds) permits you to supply, as the input argument

sconds, a tolerance (strictly between 0 and 1) to be used in the conversion from
ppform to B-form.
Alternatively, you can input sconds as a vector with integer entries, with at
least as many entries as the ppform in f has interior breaks. In that case,
sconds(i) specifies the number of smoothness conditions to be used across the

3-28
fn2fm

ith interior break. If the function in f is a tensor product, then sconds, if given,
must be a cell array.

fn2fm(f) converts a possibly old version of a form into its present version.

Examples sp = fn2fm(spline(x,y),'B-') gives the interpolating cubic spline provided

by the MATLAB command spline, but in B-form rather than in ppform.
p0 = ppmak([0 1],[3 0 0]);
p1 = fn2fm(fn2fm(fnrfn(p0,[.4 .6]),'B-'),'pp');

gives p1 identical to p0 (up to round-off in the coefficients) since the spline has
no discontinuity in any derivative across the additional breaks introduced by
fnrfn, hence conversion to B-form ignores these additional breaks, and
conversion to ppform does not retain any knot multiplicities (like the knot
multiplicities introduced, by conversion to B-form, at the endpoints of the
spline's basic interval).

Algorithm For a multivariate (tensor-product) function, univariate algorithms are applied

in each variable.
For the conversion from B-form (or BBform) to ppform, the utility command
sprpp is used to convert the B-form of all polynomial pieces to their local power
form, using repeated knot insertion at the left endpoint.
The conversion from B-form to BBform is accomplished by inserting each knot
enough times to increase its multiplicity to the order of the spline.
The conversion from ppform to B-form makes use of the dual functionals
discussed in Splines: An Overview on page 2-15. Without further
information, such a conversion has to ascertain the actual smoothness across
each interior break of the function in f.

Syntax [out1,...,outn] = fnbrk(f,part1,...,partm)

g = fnbrk(f,interval)
pp = fnbrk(pp,j)
fnbrk(f)

Description [out1,...,outn] = fnbrk(f,part1,...,partm) returns the part(s) of the

form in f specified by part1,...,partn (assuming that n<=m). These are the
parts used when the form was put together, in spmak or ppmak or rpmak or
rsmak or stmak, but also other parts derived from these.

You only need to specify the beginning character(s) of the revelant string.
Regardless of what particular form f is in, parti can be one of the following.

'form' The particular form used

'variables' The dimension of the functions domain
'dimension' The dimension of the functions target
'coefficients' The coefficients in that particular form
'interval' The basic interval of that form

Depending on the form in f, additional parts may be asked for.

If f is in B-form (or BBform or rBform), then additional choices for parti are.

'knots' The knot sequence

'coefficients' The B-spline coefficients
'number' The number of coefficients
'order' The polynomial order of the spline

3-30
fnbrk

If f is in ppform (or rpform), then additional choices for parti are

'breaks' The break sequence

'coefficients' The local polynomial coefficients
'pieces' The number of polynomial pieces
'order' The polynomial order of the spline
'guide' The local polynomial coefficients, but in the form
needed for PPVALU in PGS

If the function in f is multivariate, then the corresponding multivariate parts

are returned. This means, e.g., that knots, breaks, and the basic interval, are
cell arrays, the coefficient array is, in general, higher than two-dimensional,
and order, number and pieces are vectors.
If f is in stform, then additional choices for parti are

'centers' The centers

'coefficients' The coefficients
'number' Number of coefficients or terms
'type' The particular type

g = fnbrk(f,interval) with interval a 1-by-2 matrix [a b] with a<b does

not return a particular part. Rather, it returns a description of the univariate
function described by f and in the same form but with the basic interval
changed, to the interval given. If, instead, interval is [ ], f is returned
unchanged. This is of particular help when the function in f is m -variate, in
which case interval must be a cell array with m entries, with the i th entry
specifying the desired interval in the i th dimension. If that i th entry is [ ],
the basic interval in the i th dimension is unchanged.

pp = fnbrk(pp,j) , with pp the ppform of a univariate function and j a

positive integer, does not return a particular part, but returns the ppform of
the jth polynomial piece of the function in pp. If pp is the ppform of an

3-31
fnbrk

m -variate function, then j must be a cell array of length m . In that case, each
entry of j must be a positive integer or else an interval, to single out a
particular polynomial piece or else to specify the basic interval in that
dimension.

fnbrk(f) returns nothing, but a description of the various parts of the form is
printed at the command line instead.

Examples If p1 and p2 contain the B-form of two splines of the same order, with the same
knot sequence, and the same target dimension, then
p1plusp2 = spmak(fnbrk(p1,'k'),fnbrk(p1,'c')+fnbrk(p2,'c'));

provides the (pointwise) sum of those two functions.

If pp contains the ppform of a bivariate spline with at least four polynomial
pieces in the first variable, then ppp=fnbrk(pp,{4,[-1 1]}) gives the spline
that agrees with the spline in pp on the rectangle [b4 .. b5] x [-1 .. 1] , where
b4, b5 are the fourth and fifth entry in the break sequence for the first variable.

'dimension' The dimension of the functions target

'interval' The basic interval of that form

The specified value for part is not checked for consistency with the rest of the
description in f in case the string part terminates with the letter z.

Examples fndir(f,directions) returns a vector-valued function even when the function

described by f is ND-valued.You can correct this by using fnchg as follows:
fdir = fnchg( fndir(f,directions),...
'dim',[fnbrk(f,'dim'),size(directions,2)] );

fncmb(function,operation) returns (a description of) the function obtained

by applying to the values of the function in function the operation specified
by operation. The nature of the operation depends on whether operation is a
scalar, a vector, a matrix, or a string, as follows.

Scalar Multiply the function by that scalar.

Vector Add that vector to the functions values; this requires the
function to be vector-valued.
Matrix Apply that matrix to the functions coefficients.
String Apply the function or M-file, specified by that string, to the
functions coefficients.

The remaining options only work for univariate functions. See Limitations for
more information.

fncmb(function,function) returns (a description of) the pointwise sum of the

two functions. The two functions must be of the same form. This particular case
of just two input arguments is not included in the above table since it only
works for univariate functions.

fncmb(function,matrix,function) is the same as

fncmb(fncmb(function,matrix),function).

3-34
fncmb

fncmb(function,matrix,function,matrix) is the same as

fncmb((fncmb(function,matrix),fncmb(function,matrix)).

fncmb(function,op,function) returns the ppform of the spline obtained by

the pointwise combining of the two functions, as specified by the string op. op
can be one of the strings '+', '-', '*'. If the second function is to be a
constant, it is sufficient simply to supply here that constant.

Examples fncmb(fn,3.5) multiplies (the coefficients of) the function in fn by 3.5.

fncmb(f,3,g,-4) returns the linear combination, with weights 3 and -4, of the
function in f and the function in g.
fncmb(f,3,g) adds 3 times the function in f to the function in g.

If the function f in f happens to be scalar-valued, then f3=fncmb(f,[1;2;3])

contains the description of the function whose value at x is the 3-vector
( f(x), 2f(x), 3f(x) ) . Note that, by the convention throughout this toolbox, the
subsequent statement fnval(f3, x) returns a 1-column-matrix.
3
If f describes a surface in R , i.e., the function in f is 3-vector-valued bivariate,
then f2 = fncmb(f,[1 0 0;0 0 1]) describes the projection of that surface to
the ( x, z ) -plane.
The following commands produce the picture of a ... spirochete?
c = rsmak('circle');
fnplt(fncmb(c,diag([1.5,1]))); axis equal, hold on
sc = fncmb(c,.4);
fnplt(fncmb(sc,-[.2;-.5]))
fnplt(fncmb(sc,-[.2,-.5]))
hold off, axis off

If t is a knot sequence of length n+k and a is a matrix with n columns, then

fncmb(spmak(t,eye(n,n)),a) is the same as spmak(t,a).

fncmb(spmak([0:4],1),'+',ppmak([-1 5],[1 -1])) is the

piecewise-polynomial with breaks -1:5 that, on the interval [0 .. 4], agrees with
the function x B(x|0, 1, 2, 3, 4) + x (but has no active break at 0 or 1,
hence differs from this function outside the interval [0 .. 4]).
fncmb(spmak([0:4],1),'-',0) has the same effect as
fn2fm(spmak([0:4],1),'pp').

3-35
fncmb

Assuming that sp describes the B-form of a spline of order <k, the output of
fn2fm(fncmb(sp,'+',ppmak(fnbrk(sp,'interv'),zeros(1,k))),'B-')

describes the B-form of the same spline, but with its order raised to k.

Algorithm The coefficients are extracted (via fnbrk) and operated on by the specified
matrix or operation (and, possibly, added), then recombined with the rest of the
function description (via ppmak, spmak,rpmak,rsmak,stmak). To be sure, when
the function is rational, the matrix is only applied to the coefficients of the
numerator. Again, if we are to translate the function values by a given vector
and the function is in ppform, then only the coefficients corresponding to
constant terms are so translated.
If there are two functions input, then they must be of the same type (see
Limitations, below) except for the following.
fncmb(f1,op,f2) returns the ppform of the function

x f1(x) op f2(x)
with op one of '+', '-', '*', and f1, f2 of arbitrary polynomial form. If, in
addition, f2 is a scalar or vector, it is taken to be the function that is constantly
equal to that scalar or vector.

Limitations fncmb only works for univariate functions, except for the case
fncmb(function,operation), i.e., when there is just one function in the input.

Further, if two functions are involved, then they must be of the same type. This
means that they must either both be in B-form or both be in ppform, and,
moreover, have the same knots or breaks, the same order, and the same target.
The only exception to this is the command of the form
fncmb(function,op,function).

3-36
fnder

Purpose 3fnder
Differentiate a function

Syntax fprime = fnder(f)

fprime = fnder(f,dorder)

Description fnder(f,dorder) is the description of the dorderth derivative of the function

whose description is contained in f. The default value of dorder is 1. For
negative dorder, the particular |dorder|th indefinite integral is returned that
vanishes |dorder|-fold at the left endpoint of the basic interval.
The output is of the same form as the input, i.e., they are both ppforms or both
B-forms or both stforms. fnder does not work for rational splines; for them, use
fntlr instead. fnder works for stforms only in a limited way: if the type is
tp00, then dorder can be [1,0] or [0,1].

If the function in f is multivariate, say m -variate, then dorder must be given,

and must be of length m .

Examples If f is in ppform, or in B-form with its last knot of sufficiently high multiplicity,
then, up to rounding errors, f and fnder(fnint(f)) are the same.
If f is in ppform and fa is the value of the function in f at the left end of its
basic interval, then, up to rounding errors, f and fnint(fnder(f),fa) are the
same, unless the function described by f has jump discontinuities.
If f contains the B-form of f , and t 1 is its left-most knot, then, up to rounding
errors, fnint(fnder(f)) contains the B-form of f-f ( t 1 ) . However, its
left-most knot will have lost one multiplicity (if it had multiplicity > 1 to begin
with). Also, its rightmost knot will have full multiplicity even if the rightmost
knot for the B-form of f in f doesnt.
Here is an illustration of this last fact. The spline in sp = spmak([0 0 1], 1)
is, on its basic interval [0..1], the straight line that is 1 at 0 and 0 at 1. Now
integrate its derivative: spdi = fnint(fnder(sp)). As you can check, the
spline in spdi has the same basic interval, but, on that interval, it agrees with
the straight line that is 0 at 0 and -1 at 1.
See the demos spalldem and ppalldem for examples.

Algorithm For differentiation of either polynomial form, the derivatives are found in the
piecewise-polynomial sense. This means that, in effect, each polynomial piece

3-37
fnder

is differentiated separately, and jump discontinuities between polynomial

pieces are ignored during differentiation.
For the B-form, the formulas [PGS; (X.10)] for differentiation are used.
For the stform, differentiation relies on knowing a formula for the relevant
derivative of the basis function of the particular type.

Description df = fndir(f,y) is the ppform of the directional derivative, of the function f

in f, in the direction of the (column-)vector y. This means that df describes the
function D y f ( x ) := lim t 0( f ( x + t y )-f ( x ) ) /t .
If y is a matrix, with n columns, and f is d-valued, then the function in df is
prod(d)*n-valued. Its value at x , reshaped to be of size [d,n], has in its j th
column the directional derivative of f at x in the direction of the j th column
of y. If you prefer df to reflect explicitly the actual size of f , use instead
df = fnchg( fndir(f,y), 'dim',[fnbrk(f,'dim'),size(y,2)] );

Since fndir relies on the ppform of the function in f, it does not work for
rational functions nor for functions in stform.

Examples For example, if f describes an m-variate d-vector-valued function and x is some

point in its domain, then, e.g., with this particular ppform f that describes a
scalar-valued bilinear polynomial,
f = ppmak({0:1,0:1},[1 0;0 1]); x = [0;0];
[d,m] = fnbrk(f,'dim','var');
jacobian = reshape(fnval(fndir(f,eye(m)),x),d,m)

is the Jacobian of that function at that point (which, for this particular
scalar-valued function, is its gradient, and it is zero at the origin).
As a related example, the next statements plot the gradients of (a good
approximation to) the Franke function at a regular mesh:
xx = linspace(-.1,1.1,13); yy = linspace(0,1,11);
[x,y] = ndgrid(xx,yy); z = franke(x,y);
pp2dir = fndir(csapi({xx,yy},z),eye(2));
grads = reshape(fnval(pp2dir,[x(:) y(:)].'),...
[2,length(xx),length(yy)]);
quiver(x,y,squeeze(grads(1,:,:)),squeeze(grads(2,:,:)))

Here is the resulting plot.

3-39
fndir

1.2

0.8

0.6

0.4

0.2

0.2
0.2 0 0.2 0.4 0.6 0.8 1 1.2

Algorithm The function in f is converted to ppform, and the directional derivative of its
polynomial pieces is computed formally and in one vector operation, and put
together again to form the ppform of the directional derivative of the function
in f.

Syntax intgrf = fnint(f)

intgrf = fnint(f,value)

Description fnint(f,value) is the description of an indefinite integral of the univariate

function whose description is contained in f. The integral is normalized to have
the specified value at the left endpoint of the functions basic interval, with the
default value being zero.
The output is of the same type as the input, i.e., they are both ppforms or both
B-forms. fnint does not work for rational splines nor for functions in stform.
Indefinite integration of a multivariate function, in coordinate directions only,
is available via fnder(f,dorder) with dorder having nonpositive entries.

Examples The statement diff(fnval(fnint(f),[a b])) provides the definite integral

over the interval [a .. b] of the function described by f.
If f is in ppform, or in B-form with its last knot of sufficiently high multiplicity,
then, up to rounding errors, f and fnder(fnint(f)) are the same.
If f is in ppform and fa is the value of the function in f at the left end of its
basic interval, then, up to rounding errors, f and fnint(fnder(f),fa) are the
same, unless the function described by f has jump discontinuities.
If f contains the B-form of f , and t 1 is its left-most knot, then, up to rounding
errors, fnint(fnder(f)) contains the B-form of f-f ( t 1 ) . However, its
left-most knot will have lost one multiplicity (if it had multiplicity > 1 to begin
with). Also, its rightmost knot will have full multiplicity even if the rightmost
knot for the B-form of f in f doesnt.
Here is an illustration of this last fact. The spline in sp = spmak([0 0 1], 1)
is, on its basic interval [0..1], the straight line that is 1 at 0 and 0 at 1. Now
integrate its derivative: spdi = fnint(fnder(sp)). As you can check, the
spline in spdi has the same basic interval, but, on that interval, it agrees with
the straight line that is 0 at 0 and -1 at 1.
See the demos spalldem and ppalldem for examples.

Algorithm For the B-form, the formula [PGS; (X.22)] for integration is used.

3-41
fnint

Purpose Jumps, i.e., f(x+) f(x)

3fnjmp

Syntax jumps = fnjmp(f,x)

Description fnjmp(f,x) is like fnval(f,x), -- except that it returns the jump

f(x +) f(x ) across x (rather than the value at x) of the function f described
by f and that it only works for univariate functions.
This is a function for spline specialists.

Examples fnjmp(ppmak(1:4,1:3),1:4) returns the vector [0,1,1,0] since the pp

function here is 1 on [1 .. 2], 2 on [2 .. 3], and 3 on [3 .. 4], hence has zero jump
at 1 and 4 and a jump of 1 across both 2 and 3.
If x is cos([4:-1:0]*pi/4), then fnjmp(fnder(spmak(x,1),3),x) returns the
vector [12 -24 24 -24 12] (up to round-off). This is consistent with the fact
that the spline in question is a so called perfect cubic B-spline, i.e., has an
absolutely constant third derivative (on its basic interval). The modified
command
fnjmp(fnder(fn2fm(spmak(x,1),'pp'),3),x)

returns instead the vector [0 -24 24 -24 0], consistent with the fact that, in
contrast to the B-form, a spline in ppform does not have a discontinuity in any
of its derivatives at the endpoints of its basic interval. Note that
fnjmp(fnder(spmak(x,1),3),-x) returns the vector [12,0,0,0,12] since -x,
though theoretically equal to x, differs from x by roundoff, hence the third
derivative of the B-spline provided by spmak(x,1) does not have a jump across
-x(2),-x(3), and -x(4).

fnmin(f,interv) returns the minimum value on the interval [a..b] specified by

interv.

[minval,minsite] = fnmin(f,...) also returns a location, minsite, at

which the function in f takes that minimum value, minval.

Examples Example 1. We construct and plot a spline f with many local extrema, then
compute its maximum as the negative of the minimum of f . We indicate this
maximum value by adding a horizontal line to the plot at the height of the
computed maximum.
rand('seed',21);
f = spmak(1:21,rand(1,15)-.5);
fnplt(f)
maxval = -fnmin(fncmb(f,-1));
hold on, plot(fnbrk(f,'interv'),maxval([1 1])), hold off

0.3

0.2

0.1

0.2

0.3

0.4
0 5 10 15 20 25

3-44
fnmin

Example 2. Since spmak(1:5,-1) provides the negative of the cubic B-spline

with knot sequence 1:5, we expect the command
[y,x] = fnmin(spmak(1:5,-1))

to return -2/3 for y and 3 for x.

Algorithm fnmin first changes the basic interval of the function to the given interval, if
any. On the interval, fnmin then finds all local extrema of the function as left
and right limits at a jump and as zeros of the functions first derivative. It then
evaluates the function at these extrema and at the endpoints of the interval,
and determines the minimum over all these values.

Description fnplt(f) plots the function, described by f, on its basic interval.

If f is univariate, the following is plotted:

If f is scalar-valued, the graph of f is plotted.

If f is 2-vector-valued, the planar curve is plotted.
If f is d -vector-valued with d > 2 , the space curve given by the first three
components of f is plotted.

If f is bivariate, the following is plotted:

If f is scalar-valued, the graph of f is plotted (via surf).

If f is 2-vector-valued, the image in the plane of a regular grid in its domain
is plotted.
If f is d -vector-valued with d > 2 , then the parametric surface given by the
first three components of its values is plotted (via surf).

If f is a function of more than two variables, then the bivariate function,

obtained by choosing the midpoint of the basic interval in each of the variables
other than the first two, is plotted.

fnplt(f,arg1,arg2,arg3,arg4) permits you to modify the plotting by the

specification of additional input arguments. You can place these arguments in
whatever order you like, chosen from the following list:

A string that specifies a plotting symbol, such as '-.' or '*'; the default
is '-'.
A scalar to specify the linewidth; the default value is 1.
A string that starts with the letter 'j' to indicate that any jump in the
univariate function being plotted should actually appear as a jump. The
default is to fill in any jump by a (near-)vertical line.
A vector of the form [a,b], to indicate the interval over which to plot the
univariate function in f. If the function in f is m -variate, then this optional

3-46
fnplt

argument must be a cell array whose ith entry specifies the interval over
which the ith argument is to vary. In effect, for this arg, the command
fnplt(f,arg,...) has the same effect as the command
fnplt(fnbrk(f,arg),...). The default is the basic interval of f.
An empty matrix or string, to indicate use of default(s). You will find this
option handy when your particular choice depends on some other variables.

points = fnplt(f,...) plots nothing, but the two-dimensional points or

three-dimensional points it would have plotted are returned instead.

Algorithm The univariate function f described by f is evaluated at 101 equally spaced

sites x filling out the plotting interval. If f is real-valued, the points ( x, f(x) )
are plotted. If f is vector-valued, then the first two or three components of
f(x) are plotted.
The bivariate function f described by f is evaluated on a 51-by-51 uniform grid
if f is scalar-valued or d -vector-valued with d > 2 and the result plotted by
surf. In the contrary case, f is evaluated along the meshlines of a 11-by-11
grid, and the resulting planar curves are plotted.

If the function in f is m-variate, then addpts must be a cell array,

{addpts1,..., addptsm}, and the refinement is carried out in each of the
variables. If the ith entry in this cell array is empty, then the knot or break
sequence in the ith variable is unchanged.

Examples See fncmb for the use of fnrfn to refine the knot or break sequences of two
splines to a common refinement before forming their sum.

Algorithm The standard knot insertion algorithm is used for the calculation of the B-form
coefficients for the refined knot sequence, while Horners method is used for the
calculation of the local polynomial coefficients at the additional breaks in the
refined break sequence.

Syntax taylor = fntlr(f,dorder,x)

p = fntlr(f,dorder,x,interv)

Description fntlr(f,dorder,x) returns the unnormalized Taylor coefficients, up to the

given order dorder and at the given x, of the function described in f .
For a univariate function and a scalar x, this is the vector

dorder-1
T ( f, dorder, x ) := [f ( x ) ;Df ( x ) ; ;D f(x) ]

If, more generally, the function in f is d-valued with d>1 or even prod(d)>1
and/or is m-variate for some m>1, then dorder is expected to be an m-vector of
positive integers, x is expected to be a matrix with m rows, and, in that case,
the output is of size [prod(d)*prod(dorder),size(x,2)], with its j-th column
containing
i1-1 im-1
T ( f, dorder, x(:,j) ) ( i1, , im ) = D 1 D m f ( x(:,j) )
for i1=1:dorder(1), ..., im=1:dorder(m). Here, D i f is the partial derivative
of f with respect to its i th argument.

fntlr(f,dorder,x,interv) returns instead a ppform of the Taylor

polynomial at x of order dorder for the function described by f. The basic
interval for this ppform is as specified by interv. In this case and assuming
that the function described by f is m-variate, x is expected to be of size [m,1],
and interv is either of size [m,2] or else a cell array of length m containing m
vectors of size [1,2].

Examples If f contains a univariate function and x is a scalar or a 1-row matrix, then

fntlr(f,3,x) produces the same output as the statements

df = fnder(f); [fnval(f,x); fnval(df,x); fnval(fnder(df),x)];

As a more complicated example, look at the Taylor vectors of order 3 at 21

equally spaced points for the rational spline whose graph is the unit circle:
ci = rsmak('circle'); in = fnbrk(ci,'interv');
t = linspace(in(1),in(2),21); t(end)=[];
v = fntlr(ci,3,t);

3-49
fntlr

We plot ci along with the points v(1:2,:), to verify that these are, indeed,
points on the unit circle.
fnplt(ci), hold on, plot(v(1,:),v(2,:),'o')

Next, to verify that v(3:4,j) is a vector tangent to the circle at the point
v(1:2,j), we use the MATLAB quiver command to add the corresponding
arrows to our plot:
quiver(v(1,:),v(2,:),v(3,:),v(4,:))

Finally, what about v(5:6,:)? These are second derivatives, and we add the
corresponding arrows by the following quiver command, thus finishing the
figure.
quiver(v(1,:),v(2,:),v(5,:),v(6,:)), axis equal, hold off

0.8

0.6

0.4

0.2

0.4

0.6

0.8

1 0.5 0 0.5 1

Figure 3-1: First and Second Derivative of a Rational Spline Giving a Circle

Now, our curve being a circle, you might have expected the 2nd derivative
arrows to point straight to the center of that circle, and that would have been
indeed the case if the function in ci had been using arclength as its
independent variable. Since the parameter used is not arclength, we use the
formula, given in Example: A Spline Curve on page 2-31, to compute the
curvature of the curve given by ci at these selected points. For ease of
comparison, we switch over to the variables used there and then simply use the
commands from there.

3-50
fntlr

dspt = v(3:4,:); ddspt = v(5:6,:);

kappa = abs(dspt(1,:).*ddspt(2,:)-dspt(2,:).*ddspt(1,:))./...
(sum(dspt.^2)).^(3/2);
max(abs(kappa-1))
ans = 2.2204e-016

The numerical answer is reassuring: at all the points tested, the curvature is 1
to within roundoff.

Figure 3-2: The Function 1/(1+x^2+y^2) and Its Taylor Polynomial of

Order [3,3] at the Origin

As a final example, we start with a bivariate version of the Runge function,

2 2
obtaining, for variety, a ppform for its denominator, 1 + x + y , by bicubic
spline interpolation:
w = csapi({-1:1,-1:1},[3 2 3;2 1 2;3 2 3]);

Next, we make up the coefficient array for the numerator, 1, using exactly the
same size, and put the two together into a rational spline:
wcoefs = fnbrk(w,'coef');
scoefs = zeros(size(wcoefs)); scoefs(end)=1;
runge2 = rpmak(fnbrk(w,'breaks'),[scoefs;wcoefs]);

Then we enlarge the basic interval for this rational spline, plot it and plot, on
top of it, its Taylor polynomial at (0,0) of order [3,3].

3-51
fntlr

fnplt(fnbrk(runge2,{[-2 2],[-2 2]})); shading interp, hold on

fnplt(fntlr(runge2,[3 3],[0;0],[-.7 .7; -.7 .7]))
axis off, hold off

Since we shaded the function but not the Taylor polynomial, we can easily
distinguish the two in the previous figure. We can also see that, in contrast to
the function, the Taylor polynomial fails to be rotationally symmetric. This is
due to the fact that it is a polynomial of order [3,3] rather than a polynomial of
total order 3.
To obtain the Taylor polynomial of order 3, we get the Taylor polynomial of
order [3,3], but with (0,0) the left point of its basic interval, set all its
coefficients of total order bigger than 3 equal to zero, and then reconstruct the
polynomial, and plot it, choosing a different view in order to show off the Taylor
polynomial better. Here are the commands and the resulting figure.
taylor = fntlr(runge2,[3 3],[0;0],[0 1;0 1]);
tcoef = fnbrk(taylor,'coe'); tcoef([1 2 4]) = 0;
taylor2 = fnbrk(ppmak(fnbrk(taylor,'br'),tcoef),{[-1 1],[-1 1]});
fnplt(fnbrk(runge2,{[-2 2],[-2 2]})); shading interp, hold on
fnplt(taylor2), view(-28,-26), axis off, hold off

Figure 3-3: The Function 1/(1+x^2+y^2) and Its Taylor Polynomial of Order 3
at the Origin

If f is scalar-valued, then v is of the same size as x.

If f is [d1,...,dr]-valued, and x has size [n1,...,ns], then v has size
[d1,...,dr, n1,...,ns], with v(:,...,:, j1,...,js) the value of f at
x(j1,...,js), except that
(1) n1 is ignored if it is 1 and s is 2, i.e., if x is a row vector; and
(2) MATLAB ignores any trailing singleton dimensions of x.

For an m-variate f with m>1, with f [d1,...,dr]-valued, x may be either an

array, or else a cell array {x1,...,xm}.

If x is an array, of size [n1,...,ns] say, then n1 must equal m, and v has size
[d1,...,dr, n2,...,ns], with v(:,...,:, j2,...,js) the value of f at
x(:,j2,...,js), except that
(1) d1, ..., dr is ignored in case f is scalar-valued, i.e., both r and n1 are 1;
(2) MATLAB ignores any trailing singleton dimensions of x.

If x is a cell array, then it must be of the form {x1,...,xm}, with xj a vector,

of length nj, and, in that case, v has size [d1,...,dr, n1,...,nm], with
v(:,...,:, j1,...,jm) the value of f at (x1(j1), ..., xm(jm)), except that
d1, ..., dr is ignored in case f is scalar-valued, i.e., both r and n1 are 1.

3-53
fnval

If f has a jump discontinuity at x, then the value f(x +) , i.e., the limit from the
right, is returned, except when x equals the right end of the basic interval of
the form; for such x, the value f(x ) , i.e., the limit from the left, is returned.

fnval(f,x,'l') and fnval(x,f,'l') also provide that matrix f ( x ) , but

treating f as continuous from the left. This means that if f has a jump
discontinuity at x, then the value f(x ) , i.e., the limit from the left, is
returned, except when x equals the left end of the basic interval; for such x, the
value f(x +) is returned.
If the function is multivariate, then the above statements concerning
continuity from the left and right apply coordinatewise.

Examples The statement fnval(csapi(x,y),xx) has the same effect as the statement
csapi(x,y,xx).

Algorithm For each entry of x, the relevant break- or knot-interval is determined and the
relevant information assembled. Depending on whether f is in ppform or in
B-form, nested multiplication or the B-spline recurrence (see, e.g.,
[PGS; X.(3)]) is then used vector-fashion for the simultaneous evaluation at all
entries of x. Evaluation of a multivariate polynomial spline function takes full
advantage of the tensor product structure.
Evaluation of a rational spline follows up evaluation of the corresponding
vector-valued spline by division of all but its last component by its last
component.
Evaluation of a function in stform makes essential use of stcol, and tries to
keep the matrices involved to reasonable size.

z = fnzeros(f,[a b]) looks for zeros only in the interval [a .. b] specified by

the input.
Each column z(:,j) contains the left and right endpoint of an interval. These
intervals are of three kinds:

If the endpoints agree, then the function in f is relatively small at that point.
If the endpoints agree to many significant digits, then the function changes
sign across the interval, and the interval contains a zero of the function
provided the function is continuous there.
If the endpoints are not close, then the function is zero on the entire interval.

Examples Example 1. We construct and plot a piecewise linear spline that has each of the
three kinds of zeros, use fnzeros to compute all its zeros, and then mark the
results on that graph.
sp = spmak(augknt(1:7,2),[1,0,1,-1,0,0,1]);
fnplt(sp)
z = fnzeros(sp)
nz = size(z,2);
hold on
plot(z(1,:),zeros(1,nz),'>',z(2,:),zeros(1,nz),'<'), hold off

This gives the following list of zeros:

z =
2.0000 3.5000 5.0000
2.0000 3.5000 6.0000

In this simple example, even for the second kind of zero, the two endpoints
agree to all places.

3-55
fnzeros

0.8

0.6

0.4

0.2

0.4

0.6

0.8

1
1 2 3 4 5 6 7

Example 2. We generate a spline function with many extrema and locate all
that are in a certain interval by computing the zeros of the spline functions
first derivative there.
rand('seed',23)
sp = spmak(1:21,rand(1,16)-.5);
fnplt(sp)
z = mean(fnzeros(fnder(sp),[7,14]));
zy = fnval(sp,z);
hold on, plot(z,zy,'o'), hold off

0.3

0.2

0.1

0.2

0.3

0.4
0 5 10 15 20 25

3-56
fnzeros

Example 3. We construct a spline with a zero at a jump discontinuity and in

B-form and find all the splines zeros in an interval that goes beyond its basic
interval.
sp = spmak([0 0 1 1 2],[1 0 -.2]);
fnplt(sp)
z = fnzeros(sp,[.5, 2.7])
zy = zeros(1,size(z,2));
hold on, plot(z(1,:),zy,'>',z(2,:),zy,'<'), hold off

This gives the following list of zeros:

z =
1.0000 2.0000
1.0000 2.7000

Notice the resulting zero interval [2..2.7], due to the fact that, by definition, a
spline in B-form is identically zero outside its basic interval.

Algorithm fnzeros first converts the function to B-form. It locates zero intervals by the
corresponding sequence of consecutive zero B-spline coefficients. It locates the
sign changes in the B-spline coefficients for the function, isolates them from
each other by suitable knot insertion, and then uses the Modified Regula falsi
to locate the corresponding sign changes in the function, if any.

Syntax [xy,spcv] = getcurve

Description getcurve displays a gridded window and asks you for input. As you click on
points in the gridded window, the broken line connecting these points is
displayed. To indicate that you are done, click outside the gridded window.
Then a cubic spline curve, spcv, through the point sequence, xy, is computed
(via cscvn) and drawn. The point sequence and, optionally, the spline curve are
output.
If you want a closed curve, place the last point close to the initial point.
If you would like the curve to have a corner at some point, click on that point
twice (or more times) in succession.

Purpose 3knt2brk, knt2mlt

From knots to breaks and/or their multiplicities

Syntax knt2brk(knots)
[breaks,mults] = knt2brk(knots)
m = knt2mlt(t)
[m,sortedt] = knt2mlt(t)

Description The commands extract the distinct elements from a sequence, as well as their
multiplicities in that sequence, with multiplicity taken in two slightly different
senses.

knt2brk(knots) returns the distinct elements in knots, and in increasing

order.

[breaks,mults]=knt2brk(knots) additionally provides, in mults, the

multiplicity with which each distinct element occurs in knots. Explicitly,
breaks and mults are of the same length, and knt2brk is complementary to
brk2knt in that, for any knot sequence knots, the two commands
[xi,mlts] = knt2brk(knots); knots1 = brk2knt(xi,mlts); give knots1
equal to knots.

m=knt2mlt(t) returns a vector of the same length as t, with m(i) counting, in

the vector sort(t), the number of entries before its ith entry that are equal
to that entry. This kind of multiplicity vector is needed in spapi or spcol where
such multiplicity is taken to specify which particular derivatives are to be
matched at the sites in t. Precisely, if t is nondecreasing and z is a vector of the
same length, then sp = spapi(knots, t, z) attempts to construct a spline s
m(i)
(with knot sequence knots) for which D s ( t ( i ) ) equals z(i) , all i .

[m,sortedt]=knt2mlt(t) also returns the output from sort(t).

Neither knt2brk nor knt2mlt is likely to be used by the casual user of this
toolbox.

Examples [xi,mlts]=knt2brk([1 2 3 3 1 3]) returns [1 2 3] for xi and [2 1 3] for

mlts.

[m,t]=knt2mlt([1 2 3 3 1 3]) returns [0 1 0 0 1 2] for m and

[1 1 2 3 3 3] for t.

3-59
knt2brk, knt2mlt

Syntax newknots = newknt(f)

newknots = newknt(f,newl)
[newknots,distfn] = newknt(f,...)

Description newknt(f,newl) returns the knot sequence whose interior knots cut the basic
interval of f into newl pieces, in such a way that a certain piecewise linear
monotone function related to the high derivative of f is equidistributed.

The intent is to choose a knot sequence suitable to the fine approximation of a

function g whose rough approximation in f is assumed to contain enough
information about g to make this feasible.

newknt(f) uses for newl its default value, namely the number of polynomial
pieces in f.

[newknots,distfn]=newknt(f,...) also returns the ppform of that piecewise

linear monotone function being equidistributed.

Examples If the error in the least-squares approximation sp to some data x,y by a spline
of order k seems uneven, you might try for a more equitable distribution of
knots by using
spap2(newknt(sp),k,x,y);

For another example, see the last part of the demo difeqdem.

Algorithm This is the Fortran routine NEWNOT in PGS. With k the order of the
k
piecewise-polynomial function f in pp, the function D f is approximated by
a piecewise constant function obtained by local, discrete, differentiation of the
k-1
variation of D f . The new break sequence is chosen to subdivide the basic
interval of the piecewise-polynomial f in such a way that

newknots ( i + 1 ) k 1k
newknots ( i ) D f = const , i = k:k + newl 1

Syntax knots = optknt(tau,k)

knots = optknt(tau,k,maxiter)

Description t= optknt(tau,k,maxiter) provides the knot sequence t that is best for

interpolation from S k, t at the site sequence tau, with 10 the default for the
optional input maxiter that bounds the number of iterations to be used in this
effort. Here, best or optimal is used in the sense of [3] and [2], and this means
the following: For any recovery scheme R that provides an interpolant Rg that
matches a given g at the sites tau(1), ..., tau(n), we may determine the
k
smallest constant const R for which g-Rg const R D g for all smooth
functions g .

Here, f : = sup tau ( 1 ) < x < tau ( n ) f ( x ) . Then we may look for the optimal
recovery scheme as the scheme R for which const R is as small as possible.
Micchelli/Rivlin/Winograd have shown this to be interpolation from S k, t ,
with t uniquely determined by the following conditions:

1 t(1) = ... = t(k) = tau(1);

2 t(n+1) = ... = t(n+k) = tau(n);
3 Any absolutely constant function h with sign changes at the sites
t(k+1), ..., t(n) and nowhere else satisfies
tau ( n )
tau ( 1 )f ( x )h ( x )dx = 0 for all f S k ,t

Gaffney/Powell called this interpolation scheme optimal since it provides the

center function in the band formed by all interpolants to the given data that, in
addition, have their k th derivative between M and M (for large M ).

Examples See the last part of the demo spapidem for an illustration. For the following
highly nonuniform knot sequence
t = [0, .0012+[0, 1, 2+[0,.1], 4]*1e-5, .002, 1];

the command optknt(t,3) will fail, while the command optknt(t,3,20),

using a high value for the optional parameter maxiter, will succeed.

3-62
optknt

Algorithm This is the Fortran routine SPLOPT in PGS. It is based on an algorithm

described in [1], for the construction of that sign function h mentioned in (3)
above. It is essentially Newtons method for the solution of the resulting
nonlinear system of equations, with aveknt(tau,k) providing the first guess
for t(k+1), ...,t(n), and some damping used to maintain the
Schoenberg-Whitney conditions .

References [1] C. de Boor, Computational aspects of optimal recovery, in Optimal

Estimation in Approximation Theory, C.A. Micchelli & T.J. Rivlin eds., Plenum
Publ., New York, 1977, 69-91.
[2] P.W. Gaffney & M.J.D. Powell, Optimal interpolation, in Numerical
Analysis, G.A. Watson ed., Lecture Notes in Mathematics, No. 506,
Springer-Verlag, 1976, 90-99.
[3] C.A. Micchelli, T.J. Rivlin & S. Winograd, The optimal recovery of smooth
functions, Numer. Math. 80, (1974), 903-906.

3-63
ppmak

Purpose 3ppmak
Put together a spline in ppform

Syntax ppmak(breaks,coefs)
ppmak
pp = ppmak(breaks,coefs,d)
pp = ppmak(breaks,coefs,sizec)

Description The command ppmak(...) puts together a spline in ppform from minimal
information, with the rest inferred from that information. fnbrk provides any
or all of the parts of the completed description. In this way, the actual data
structure used for the storage of the ppform is easily modified without any
effect on the various fn... commands that use this construct. However, the
casual user is not likely to use ppmak explicitly, relying instead on the various
spline construction commands in the toolbox to construct particular splines.

ppmak(breaks,coefs) returns the ppform of the spline specified by the break

information in breaks and the coefficient information in coefs. How that
information is interpreted depends on whether the function is univariate or
multivariate, as indicated by breaks being a sequence or a cell array.
If breaks is a sequence, it must be nondecreasing, with its first entry different
from its last. Then the function is assumed to be univariate, and the various
parts of its ppform are determined as follows:

1 The number l of polynomial pieces is computed as length(breaks)-1, and

the basic interval is, correspondingly, the interval
[breaks(1) .. breaks(l+1)].
2 The dimension d of the functions target is taken to be the number of rows
in coefs. In other words, each column of coefs is taken to be one coefficient.
More explicitly, coefs(:,i*k+j) is assumed to contain the jth coefficient of
the (i+1)st polynomial piece (with the first coefficient the highest and the
kth coefficient the lowest, or constant, coefficient). Thus, with kl the number
of columns of coefs, the order k of the piecewise-polynomial is computed as
fix(kl/l).
After that, the entries of coefs are reordered, by the command
coefs = reshape(permute(reshape(coefs,[d,k,l]),[1 3 2]),[d*l,k])

3-64
ppmak

in order to conform with the internal interpretation of the coefficient array in

the ppform for a univariate spline.
If breaks is a cell array, of length m, then the function is assumed to be
m-variate (tensor product), and the various parts of its ppform are determined
from the input as follows:

1 The m-vector l has length(breaks{i})-1 as its ith entry and,

correspondingly, the m-cell array of its basic intervals has the interval
[breaks{i}(1) .. breaks{i}(end)] as its ith entry.
2 The dimension d of the functions target and the m-vector k of
(coordinate-wise polynomial) orders of its pieces are obtained directly from
the size of coefs, as follows.
a If coefs is an m-dimensional array, then the function is taken to be
scalar-valued, i.e., d is 1, and the m-vector k is computed as
size(coefs)./l. After that, coefs is reshaped by the command
coefs = reshape(coefs,[1,size(coefs)]).
b If coefs is an (r+m)-dimensional array, with sizec = size(c) say, then
d is set to sizec(1:r), and the vector k is computed as
sizec(r+(1:m))./l. After that, coefs is reshaped by the command
coefs = reshape(coefs,[prod(d),sizec(r+(1:m))]).
Then, coefs is interpreted as an equivalent array of size
[d,l(1),k(1),l(2),k(2),...,l(m),k(m)], with its
(:,i(1),r(1),i(2),r(2),...,i(m),r(m))th entry the coefficient of

m
( k( ) r( ) )
( x( ) breaks{ } ( i( ) ) )
=1

in the local polynomial representation of the function on the (hyper)rectangle

with sides

[ breaks{ } ( i ( ) ) .. breaks{ } ( i ( ) + 1 ) ], = 1:m

This is, in fact, the internal interpretation of the coefficient array in the ppform
of a multivariate spline.

ppmak prompts you for breaks and coefs.

3-65
ppmak

ppmak(breaks,coefs,d) with d a positive integer, also puts together the

ppform of a spline from the information supplied, but expects the function to be
univariate. In that case, coefs is taken to be of size [d*l,k], with l obtained
as length(breaks)-1, and this determines the order, k, of the spline. With this,
coefs(i*d+j,:) is taken to be the jth components of the coefficient vector for
the (i+1)st polynomial piece.

ppmak(breaks,coefs,sizec) with sizec a row vector of positive integers,

also puts together the ppform of a spline from the information supplied, but
interprets coefs to be of size sizec (and returns an error when
prod(size(coefs)) differs from prod(sizec)). This option is important only
in the rare case that the input argument coefs is an array with one or more
trailing singleton dimensions. For, MATLAB suppresses trailing singleton
dimensions, hence, without this explicit specification of the intended size of
coefs, ppmak would interpret coefs incorrectly.

Examples The two splines

p1 = ppmak([1 3 4],[1 2 5 6;3 4 7 8]);
p2 = ppmak([1 3 4],[1 2;3 4;5 6;7 8],2);

have exactly the same ppform (2-vector-valued, of order 2). But the second
command provides the coefficients in the arrangement used internally.
ppmak([0:2],[1:6]) constructs a piecewise-polynomial function with basic
interval [0..2] and consisting of two pieces of order 3, with the sole interior
break 1. The resulting function is scalar, i.e., the dimension d of its target is 1.
The function happens to be continuous at that break since the first piece is
2 2
x x + 2x + 3 , while the second piece is x 4(x 1) + 5 ( x 1 ) + 6 .
When the function is univariate and the dimension d is not explicitly specified,
then it is taken to be the row number of coefs. The column number should be
an integer multiple of the number l of pieces specified by breaks. For example,
the statement ppmak([0:2],[1:3;4:6]) leads to an error, since the break
sequence [0:2] indicates two polynomial pieces, hence an even number of
columns are expected in the coefficient matrix. The modified statement
ppmak([0:1],[1:3;4:6]) specifies the parabolic curve
2
x ( 1, 4 )x + ( 2, 5 )x + ( 3, 6 ) . In particular, the dimension d of its target is 2.
The differently modified statement ppmak([0:2],[1:4;5:8]) also specifies a

3-66
ppmak

planar curve (i.e., d is 2), but this one is piecewise linear; its first polynomial
piece is x ( 1, 5 )x + ( 2, 6 ) .
Explicit specification of the dimension d leads, in the univariate case, to a
different interpretation of the entries of coefs. Now the column number
indicates the polynomial order of the pieces, and the row number should equal
d times the number of pieces. Thus, the statement ppmak([0:2],[1:4;5:8],2)
is in error, while the statement ppmak([0:2],[1:4;5:8],1) specifies a scalar
3 2
piecewise cubic whose first piece is x x + 2x + 3x + 4 .
If you wanted to make up the constant polynomial, with basic interval [0..1]
say, whose value is the matrix eye(2), then you would have to use the full
optional third argument, i.e., use the command
pp = ppmak(0:1,eye(2),[2,2,1,1]);

Finally, if you want to construct a 2-vector-valued bivariate polynomial on the

rectangle [-1 .. 1] x [0 .. 1], linear in the first variable and constant in the
second, say
coefs = zeros(2,2,1); coefs(:,:,1) = [1 0; 0 1];

then the straightforward

pp = ppmak({[-1 1],[0 1]},coefs);

will fail, producing a scalar-valued function of order 2 in each variable, as will

pp = ppmak({[-1 1],[0 1]},coefs,size(coefs));

while the following command will succeed:

pp = ppmak({[-1 1],[0 1]},coefs,[2 2 1]);

See ppalldem for other examples.

Purpose 3rpmak, rsmak

Put together a rational spline

Syntax rp = rpmak(breaks,coefs)
rp = rpmak(breaks,coefs,d)
rs = rsmak(knots,coefs)
rs = rsmak(shape,parameters)

Description Both rpmak and rsmak put together a rational spline from minimal information.
rsmak is also equipped to provide rational splines that describe standard
geometric shapes. A rational spline must be scalar- or vector-valued.

rpmak(breaks,coefs) has the same effect as the command ppmak(breaks,

coefs), -- except that the resulting ppform is tagged as a rational spline, i.e.,
as a rpform.
To describe what this means, let R be the piecewise-polynomial put together
by the command ppmak(breaks,coefs), and let r(x) = s(x) w(x) be the
rational spline put together by the command rpmak(breaks,coefs). If v is the
value of R at x , then v(1:end-1)/v(end) is the value of r at x . In other words,
R(x) = [ s(x);w(x) ] . Correspondingly, the dimension of the target of r is one
less than the dimension of the target of R . In particular, the dimension (of the
target) of R must be at least 2, i.e., the coefficients specified by coefs must be
d-vectors with d > 1. See ppmak for how the input arrays breaks and coefs are
being interpreted, hence how they are to be specified in order to produce a
particular piecewise-polynomial.

rpmak(breaks,coefs,d) has the same effect as ppmak(breaks,coefs,d+1),

except that the resulting ppform is tagged as being a rpform. Note that the
desire to have that optional third argument specify the dimension of the target
requires different values for it in rpmak and ppmak for the same coefficient array
coefs.

rpmak(breaks,coefs,sizec) has the same effect as

ppmak(breaks,coefs,sizec) except that the resulting ppform is tagged as
being a rpform, and the target dimension is taken to be sizec(1)-1.

rsmak(knots,coefs) is similarly related to spmak(knots,coefs), and

rsmak(knots,coefs,sizec) to spmak(knots,coefs,sizec). In particular,
rsmak(knots,coefs) puts together a rational spline in B-form, i.e., it provides

3-68
rpmak, rsmak

a rBform. See spmak for how the input arrays knots and coefs are being
interpreted, hence how they are to be specified in order to produce a particular
piecewise-polynomial.

rsmak(shape,parameters) provides a rational spline in rBform that describes

the shape being specified by the string shape and the optional additional
parameters. Specific choices are:

rsmak('circle',radius,center)
rsmak('cone',radius,halfheight)
rsmak('cylinder',radius,height)
rsmak('southcap',radius,center)

From these, one may generate related shapes by affine transformations, with
the help of fncmb.
All fn... commands except fnint, fnder, fndir can handle rational splines.

Examples The commands

runges = rsmak([-5 -5 -5 5 5 5],[1 1 1; 26 -24 26]);
rungep = rpmak([-5 5],[0 0 1; 1 -10 26],1);
2
both provide a description of the rational polynomial r(x) = 1 ( x + 1 ) on the
interval [-5 .. 5]. However, outside the interval [-5 .. 5], the function given by
runges is zero, while the rational spline given by rungep agrees with
2
1 ( x + 1 ) for every x .
The figure of a rotated cone is generated by the commands
fnplt(fncmb(rsmak('cone',1,2),[0 0 -1;0 1 0;1 0 0]))
axis equal, axis off, shading interp

3-69
rpmak, rsmak

Figure 3-4: A Rotated Cone Given by a Rational Quadratic Spline

For further, illustrated examples, see NURBS and Other Rational Splines on
page 2-38.

slvblk(blockmat,b,w) returns the vector x that minimizes the weighted sum

2
j w ( j ) ( ( Ax b ) ( j ) ) .

Examples sp=spmak(knots,slvblk(spcol(knots,k,x,1),y.')) provides in sp the

B-form of the spline s of order k with knot sequence knots that matches the
given data (x,y), i.e., for which s(x) equals y.

Algorithm The command bkbrk is used to obtain the essential parts of the coefficient
matrix described by blokmat (in one of two available forms).
A QR factorization is made of each diagonal block, after it was augmented by
the equations not dealt with when factoring the preceding block. The resulting
factorization is then used to solve the linear system by backsubstitution.

Syntax pointer = sorted(meshsites,sites)

Description Various commands in this toolbox need to determine the index j for which a
given x lies in the interval [ t j ..t j + 1 ] , with ( t i ) a given nondecreasing
sequence, e.g., a knot sequence. This job is done by sorted in the following
fashion.

pointer=sorted(meshsites,sites) is the integer row vector whose j-th

entry equals the number of entries in meshsites that are ssites(j), with
ssites the vector sort(sites). Thus, if both meshsites and sites are
nondecreasing, then
meshsites(pointer(j)) sites(j) < meshsites(pointer(j)+1)

with the obvious interpretations when

pointer(j) < 1 or length(meshsites) < pointer(j) + 1

Specifically, having pointer(j) < 1 then corresponds to having sites(j)

strictly to the left of meshsites(1), while having length(meshsites) <
pointer(j)+1 then corresponds to having sites(j) at, or to the right of,
meshsites(end).

Examples The statement

sorted([1 1 1 2 2 3 3 3],[0:4])

will generate the output 0 3 5 8 8, as will the statement

sorted([3 2 1 1 3 2 3 1],[2 3 0 4 1])

Algorithm The indexing output from sort([meshsites(:).',sites(:).']) is used.

3-72
spap2

Purpose 3spap2
Least-squares spline approximation

Syntax sp = spap2(knots,k,x,y)
sp = spap2(l,k,x,y)
sp = spap2(knots,k,x,y,w)

Description spap2(knots,k,x,y) returns the B-form of the spline f of order k with the
given knot sequence knots for which
(*) y(:,j) = f(x(j)), all j

in the weighted mean-square sense, meaning that the sum

w(j) y(:,j) f ( x(j) )

2
is minimized, with default weights equal to 1. The
j
2
data values y(:,j) may be scalars, vectors, matrices, even ND-arrays, and z
stands for the sum of the squares of all the entries of z . Data points with the
same site are replaced by their average.
If the sites x satisfy the (Schoenberg-Whitney) conditions

knots(j) < x(j) < knots(j + k)

(**)
j = 1, , length(x) = length(knots) k
then there is a unique spline (of the given order and knot sequence) satisfying
(*) exactly. No spline is returned unless (**) is satisfied for some subsequence
of x.
sp = spap2(1,k,x,y), with 1 a positive integer, returns the B-form of a
least-squares spline approximant, but with the knot sequence chosen for you.
The knot sequence is obtained by applying aptknt to an appropriate
subsequence of x. The resulting piecewise-polynomial consists of l polynomial
pieces and has k-2 continuous derivatives. If you feel that a different
distribution of the interior knots might do a better job, follow this up with
sp1 = spap2(newknt(sp),k,x,y));

spap2(...,x,y,w) lets you specify the weights w in the error measure (given
above). w must be a vector of the same size as x, with nonnegative entries. All
the weights corresponding to data points with the same site are summed when
those data points are replaced by their average.

3-73
spap2

spap2({knorl1,...,knorlm},k,{x1,...,xm},y) provides a least-squares

spline approximation to gridded data. Here, each knorli is either a knot
sequence or a positive integer. Further, k must be an m-vector, and y must be
an (r+m)-dimensional array, with y(:,i1,...,im) the datum to be fitted at the
site [x{1}(i1),...,x{m}(im)], all i1, ..., im. However, if the spline is to be
scalar-valued, then, in contrast to the univariate case, y is permitted to be an
m-dimensional array, in which case y(i1,...,im) is the datum to be fitted at
the site [x{1}(i1),...,x{m}(im)], all i1, ..., im.

spap2({knorl1,...,knorlm},k,{x1,...,xm},y,w) also lets you specify the

weights. In this m-variate case, w must be a cell array with m entries, with w{i}
a nonnegative vector of the same size as xi, or else w{i} must be empty, in
which case the default weights are used in the ith variable.

Examples sp = spap2(augknt([a,xi,b],4),4,x,y)

is the least-squares approximant to the data x, y, by cubic splines with two

continuous derivatives, basic interval [a..b], and interior breaks xi, provided xi
has all its entries in (a..b) and the conditions (**) are satisfied in some
fashion. In that case, the approximant consists of length(xi)+1 polynomial
pieces. If you do not want to worry about the conditions (**) but merely want to
get a cubic spline approximant consisting of l polynomial pieces, use instead
sp = spap2(l,4,x,y);

If the resulting approximation is not satisfactory, try using a larger l. Else use
sp = spap2(newknt(sp),4,x,y);

for a possibly better distribution of the knot sequence. In fact, if that helps,
repeating it may help even more.
As another example, spap2(1, 2, x, y); provides the least-squares
straight-line fit to data x,y, while
w = ones(size(x)); w([1 end]) = 100; spap2(1,2, x,y,w);

forces that fit to come very close to the first data point and to the last.

Algorithm spcol is called on to provide the almost block-diagonal collocation matrix

( B j ,k ( x i ) ) , and slvblk solves the linear system (*) in the (weighted)
least-squares sense, using a block QR factorization.

3-74
spap2

Gridded data are fitted, in tensor-product fashion, one variable at a time,

taking advantage of the fact that a univariate weighted least-squares fit
depends linearly on the values being fitted.

Syntax spline = spapi(knots,x,y)

spline = spapi(k,x,y)
spapi({knork1,...,knorkm},{x1,...,xm},y)

Description spapi(knots,x,y) returns the spline f (if any) of order

k = length(knots) - length(x)

with knot sequence knots for which

(*) f(x(j)) = y(:,j), all j.

If some of the entries of x are the same, then this is taken in the osculatory
m(j)
sense, i.e., in the sense that D f ( x(j) ) = y(:,j) , with m(j) := #{ i<j : x(i)
m
= x(j) }, and D f the m -th derivative of f . Thus r -fold repetition of a site z
in x corresponds to the prescribing of value and the first r 1 derivatives of f
at z .
The data values, y(:,j), may be scalars, vectors, matrices, or even ND-arrays.

spapi(k,x,y), with k a positive integer, merely specifies the desired spline

order, k, in which case aptknt is used to determine a workable (though not
necessarily optimal) knot sequence for the given sites x. In other words, the
command spapi(k,x,y) has the same effect as the more explicit command
spapi(aptknt(x,k),x,y).

spapi({knork1,...,knorkm},{x1,...,xm},y) returns the B-form of a

tensor-product spline interpolant to gridded data. Here, each knorki is either
a knot sequence, or else is a positive integer specifying the polynomial order to
be used in the ith variable, thus leaving it to spapi to provide a corresponding
knot sequence for the ith variable. Further, y must be an (r+m)-dimensional
array, with y(:,i1,...,im) the datum to be fitted at the site
[x{1}(i1),...,x{m}(im)], all i1, ..., im , unless the spline is to be
scalar-valued, in which case, in contrast to the univariate case, y is permitted
to be an m-dimensional array.

Examples spapi([0 0 0 0 1 2 2 2 2],[0 1 1 1 2],[2 0 1 2 -1])produces the unique

cubic spline f on the interval [0..2] with exactly one interior knot, at 1, that
satisfies the five conditions

3-76
spapi

2
f(0+) = 2 , f ( 1 ) = 0 , Df(1) = 1 , D f(1) = 2 , f(2) = 1
These include 3-fold matching at 1, i.e., matching there to prescribed values of
the function and its first two derivatives.
Here is an example of osculatory interpolation, to values y and slopes s at the
sites x by a quintic spline:
sp = spapi(augknt(x,6,2),[x,x,min(x),max(x))],[y,s,ddy0,ddy1]);

with ddy0 and ddy1 values for the second derivative at the endpoints.
As a related example, if the function sin(x) is to be interpolated at the distinct
data sites x by a cubic spline, and its slope is also to be matched at a
subsequence x(s), then this can be accomplished by the command
sp = spapi(4,[x x(s)], [sin(x) cos(x(s))]);

in which a suitable knot sequence is supplied with the aid of aptknt. In fact, if
you wanted to interpolate the same data by quintic splines, simply change the
4 to 6.

As a bivariate example, here is a bivariate interpolant.

x = -2:.5.2; y = -1:.25:1; [xx, yy] = ndgrid(x,y);
z = exp(-(xx.^2+yy.^2));
sp = spapi({3,4},{x,y},z); fnplt(sp)

As an illustration of osculatory interpolation to gridded data, here is complete

bicubic interpolation, with the data explicitly derived from the bicubic
3 3
polynomial g ( u, v ) = u v , to make it easy for you to see exactly where the
slopes and slopes of slopes (i.e., cross derivatives) must be placed in the data
values supplied. Since our g is a bicubic polynomial, its interpolant, f , must be
g itself. We test this.
sites = {[0,1],[0,2]}; coefs = zeros(4,4); coefs(1,1) = 1;
g = ppmak(sites,coefs);
Dxg = fnval(fnder(g,[1,0]),sites);
Dyg = fnval(fnder(g,[0,1]),sites);
Dxyg = fnval(fnder(g,[1,1]),sites);
f = spapi({4,4}, {sites{1}([1,2,1,2]),sites{2}([1,2,1,2])}, ...
[fnval(g,sites), Dyg ; ...
Dxg.' , Dxyg]);

3-77
spapi

if any( squeeze( fnbrk(fn2fm(f,'pp'), 'c') ) - coefs )

'something went wrong', end

Algorithm spcol is called on to provide the almost-block-diagonal collocation matrix

( B j, k(x) ) (with repeats in x denoting derivatives, as described above), and
slvblk solves the linear system (*), using a block QR factorization.

Gridded data are fitted, in tensor-product fashion, one variable at a time,

taking advantage of the fact that a univariate spline fit depends linearly on the
values being fitted.

knots(j) < x(j) < knots(j + k) , all j

(with equality possible at knots(1) and knots(end)). In the multivariate case,

these conditions must hold in each variable separately.

3-78
spaps

Purpose 3spaps
Smoothing spline

Syntax sp = spaps(x,y,tol)
[sp,values] = spaps(x,y,tol)
[sp,values,rho] = spaps(x,y,tol)
[...] = spaps(x,y,tol,arg1,arg2,...)
[...] = spaps({x1,...,xr},y,tol,...)

Description spaps(x,y,tol) returns the B-form of the smoothest function f that lies
within the given tolerance tol of the given data points (x(j), y(:,j)),
j=1:length(x). The data values y(:,j) may be scalars, vectors, matrices,
even ND-arrays. Data points with the same data site are replaced by their
weighted average, with its weight the sum of the corresponding weights, and
the tolerance tol is reduced accordingly.

[sp,values] = spaps(x,y,tol) also returns the smoothed values, i.e.,

values is the same as fnval(sp,x).

Here, the distance of the function f from the given data is measured by

n
w ( j ) ( y ( :, j ) f ( x ( j ) ) )
2
E(f) =
j=1

with the default choice for the weights w making E ( f ) the composite

max(x) 2 2
trapezoidal rule approximation to y f , and z denoting the sum
min(x)
of squares of the entries of z .
Further, smoothest means that the following roughness measure is minimized:
max(x)
m m 2
F(D f ) = ( t ) D f ( t ) dt
min(x)
m
where D f denotes the mth derivative of f . The default value for m is 2, the
default value for the roughness measure weight is the constant 1, and this
makes f a cubic smoothing spline.
When tol is nonnegative, then the spline f is determined as the unique
m
minimizer of the expression E ( f ) + F ( D f ) , with the smoothing parameter
(optionally returned) so chosen that E ( f ) equals tol. Hence, when m is 2,

3-79
spaps

then, after conversion to ppform, the result should be the same (up to roundoff)
as obtained by csaps(x, y, ( + 1 )) . Further, when tol is zero, then the
natural or variational spline interpolant of order 2m is returned. For large
enough tol, the least-squares approximation to the data by polynomials of
degree <m is returned.
When tol is negative, then is taken to be -tol.
The default value for the weight function in the roughness measure is the
constant function 1. But you may choose it to be, more generally, a piecewise
constant function, with breaks only at the data sites. Assuming the vector x to
be strictly increasing, you specify such a piecewise constant by inputting tol
as a vector of the same size as x. In that case, tol(i) is taken as the constant
value of on the interval (x(i-1) .. x(i)), i=2:length(x), while tol(1)
continues to be used as the specified tolerance.

[sp,values,rho] = spaps(x,y,tol) also returns the actual value of used

as the third output argument.

[...] = spaps(x,y,tol,arg1,arg2,...) lets you specify the weight vector w

and/or the integer m, by supplying them as an argi. For this, w must be a
nonnegative vector of the same size as x; m must be 1 (for a piecewise linear
smoothing spline), or 2 (for the default cubic smoothing spline), or 3 (for a
quintic smoothing spline).

[...] = spaps({x1,...,xr},y,tol,...) returns the B-form of an r-variate

tensor-product smoothing spline that is roughly within the specified tolerance
to the given gridded data. Now y is expected to supply the corresponding
gridded values, with size(y) equal to [length(x1),...,length(xr)] in case
the function is scalar-valued, and equal to [d,length(x1),...,length(xr)]
in case the function is d-valued. Further, tol must be a cell array with r
entries, with tol{i} the tolerance used during the i-th step when a univariate
(but vector-valued) smoothing spline in the i-th variable is being constructed.
The optional input for m must be an r-vector (with entries from the set
{1,2,3}), and the optional input for w must be a cell array of length r, with
w{i} either empty (to indicate that the default choice is wanted) or else a
positive vector of the same length as xi.

Examples The statements

3-80
spaps

w = ones(size(x)); w([1 end]) = 100;

sp = spaps(x,y, 1.e-2, w, 3);

give a quintic smoothing spline approximation to the given data that close to
interpolates the first and last datum, while being within about 1.e-2 of the
rest.
x = -2:.2:2; y=-1:.25:1; [xx,yy] = ndgrid(x,y); rand('seed',39);
z = exp(-(xx.^2+yy.^2)) + (rand(size(xx))-.5)/30;
sp = spaps({x,y},z,8/(60^2)); fnplt(sp), axis off

produces the figure below, showing a smooth approximant to noisy data from a
smooth bivariate function. Note the use of ndgrid here; use of meshgrid would
have led to an error.

Algorithm Reinschs approach [1] is used (including his clever way of choosing the
equation for the optimal smoothing parameter in such a way that a good initial
guess is available and Newtons method is guaranteed to converge and to
converge fast).

Proiecte Functii Spline

Uploaded by

Proiecte Functii Spline

Uploaded by

aptknt

Syntax knots = aptknt(tau,k)

Algorithm The (k-1)-point averages sum(tau(i+1:i+k-1))/(k-1) of the sequence tau, as

See Also augknt, aveknt, newknt, optknt, spline

Syntax [augknot,addl] = augknt(knots,k)

Description augknt(knots,k) returns a nondecreasing and augmented knot sequence that

augknt(knots,k,mults) makes sure that the augmented knot sequence

augknt([1 2 3 3 3],2) returns the vector [1 1 2 3 3], as does

Syntax tstar = aveknt(t,k)

which are recommended as good interpolation site choices when interpolating

Examples aveknt([1 2 3 3 3],3) returns the vector [2.5000 3.0000], while

provide a spline interpolant to the sine function on the interval

See Also aptknt, chbpnt, optknt

Syntax [nb,rows,ncols,last,blocks] = bkbrk(blokmat)

Description [nb,rows,ncols,last,blocks]=bkbrk(blokmat) returns the details of the

See Also slvblk, spcol

Syntax [knots,index] = brk2knt(breaks,mults)

Examples The statements

give [1 2 3 4 5] for xi, [2 3 1 1 2] for m, and, finally, t for tt.

See Also augknt, knt2brk, knt2mlt

See Also bspline, chbpnt, spcol

bspline(t,window) does the plotting in the subplot window specified by

pp = bspline(t) plots nothing but returns the ppform of the B-spline.

Examples The statement pp=fn2fm(spmak(t,1),'pp') has the same effect as the

See Also bspligui, spalldem

Syntax tau = chbpnt(t,k)

[tau,sp] = chbpnt(t,k) also returns, in sp, the Chebyshev spline.

chbpnt(t,k,tol) also specifies the tolerance tol to be used in the iterative

Examples chbpnt([-ones(1,k),ones(1,k)],k) provides (approximately) the extreme

as is evidenced by the near equi-oscillation of the error.

See Also aveknt, chebdem

pp = csape(x,y,conds) lets you choose the end conditions to be used, from a

'complete' or Match endslopes (as given, with default as under

By giving conds as a 1-by-2 matrix instead, it is possible to specify different

clamped Ds(e) = endcondvals(:,j) if conds(j) == 1

Lagrange Ds(e) = Dp(e) default

As a multivariate vector-valued example, here is a sphere, done as a

See Also csapi, spapi, spline

csapi(x,y,xx) is the same as fnval(csapi(x,y),xx), i.e., the values of the

Examples See the demo csapidem for various examples.

Here is a simple bivariate example, a bicubic spline interpolant to the Mexican

See Also csape, spapi, spline, tspdem

Syntax [pp,p] = csaps(x,y)

pp=csaps(x,y,p) lets you supply the smoothing parameter. The smoothing

is as small as possible subject to the condition that the error measure

pp = csaps(x,y,p,[],w) lets you specify the weights w in the error measure,

[...] = csaps({x1,...,xm},y,...) provides the ppform of an m-variate

In this case, p if input must be a cell-array with m entries or else an m-vector,

csaps(x,y,p,xx) is the same as fnval(csaps(x,y,p),xx).

csaps(x,y,p,xx,w) is the same as fnval(csaps(x,y,p,[],w),xx).

The resulting plot is shown below.

cubic smoothing spline, with right half treated differently:

Example 2. As a bivariate example, we add some uniform noise, from the

Algorithm csaps is an implementation of the Fortran routine SMOOTH from PGS.

The default value for p is determined as follows. The calculation of the

See Also csape, spap2, spaps, tpaps

Syntax curve = cscvn(points)

Description cscvn(points) returns a parametric variational, or natural, cubic spline

Algorithm The break sequence t is determined as

and csape (with either periodic or variational end conditions) is used to

See Also csape, fnplt, getcurve, getcurv2, spcrvdem

References [1] E.T.Y. Lee, Choosing nodes in parametric curve interpolation,

Description g = fn2fm(f,form) describes the same function as is described by f, but in the

g = fn2fm(f, B- ,sconds) permits you to supply, as the input argument

Examples sp = fn2fm(spline(x,y),'B-') gives the interpolating cubic spline provided

Algorithm For a multivariate (tensor-product) function, univariate algorithms are applied

See Also ppalldem, ppmak, spalldem, spmak, rsmak, stmak

Syntax [out1,...,outn] = fnbrk(f,part1,...,partm)

Description [out1,...,outn] = fnbrk(f,part1,...,partm) returns the part(s) of the

'form' The particular form used

Depending on the form in f, additional parts may be asked for.

'knots' The knot sequence

If f is in ppform (or rpform), then additional choices for parti are

'breaks' The break sequence

If the function in f is multivariate, then the corresponding multivariate parts

'centers' The centers