CIRCUIT THEORY AND APPLICATIONS, VOL.

12, 111-122 (1984)

INTERPOLATION WITH POSITIVE REAL FUNCTIONS OF

SEVERAL VARIABLES
J. GREGOR
Faculty of Electrical Engineering, Czech Technical University, Prague, Czechoslouakia

SUMMARY
The interpolation problem for n-dimensional positive real functions is formulated in two different settings: with
interpolation knots either contained on the distinguished boundary of the Cartesian product of open right half planes
or in this n-dimensional 'open right half plane'. A complete solution of the first problem is given. Necessary conditions
of existence of a solution of the second problem are proved and the well known Nevanlinna-Pick algorithm is
generalized so that construction of interpolating n-dimensional positive real functions becomes possible at least in
case when the interpolation knots are real vectors. Examples illustrate the suggested methods, show the complexity
and significance of the problem and methods under discussion.
Interpolation in special classes of functions has for some time been an important problem in mathematics. General
schemes of solution in linear functional spaces with orthogonality are well known. For classes of functions which do
not satisfy the axioms of a linear space special methods have been studied in detail, because they are closely connected
to other important parts of mathematics such as approximation theory, numerical analysis and the theory of moments.
Considerable efforts have been put into interpolation in the complex domain with functions analytic in the unit disc
and with positive real functions. The significance of interpolation procedures in the class of one-variable positive real
functions in network theory, e.g. in broadband matching and approximation problems has been recognized in
References 1-3. An excellent survey paper has recently been devoted to these problem^;^ in this paper some new
ideas for the use of interpolation in system theory were outlined. The growing interest in multivariable network
theory calls for generalizations of interpolation procedures to multivariable cases. Many of the one-variable applications
can easily be extended to the multivariable case and it is hoped that in passive network theory and in problems of
n-dimensional digital systems the solution of the interpolation problem will contribute to the difficult synthesis
problems in the multivariable theory.
In this paper conditions for the existence of multivariable positive real interpolating functions will be investigated
and some procedures for the construction of such functions will be described.

1. INTRODUCTION
In this paper the following notations and definitions will be used: z = ( zl, z 2 , .. . , z,) are complex row
vectors, z = x + j y E C", j 2 = -1, Z = x - jy. Here, x, y are real vectors, x = (xl,x2,. . . ,x"), x, E R . The
symbol ( , ) denotes the scalar product, ( z , w )=C:=l z,G,, ))-)Iis the norm defined by ) ) z / ) ~( z=,z ) . For
real vectors x = (xl, x2,. . . ,x"), x > 0 means x, > 0 for all i = 1, 2,. . . , n. Relations 3 , = , < are also
component-wise defined. The set r" = { z E C", R e z > 0) will be called the (open) right half plane,
K " = { z E C", Re z = 0) its distinguished boundary. r"
= { z E C", R e z 2 0) U {a}, where (00) denotes the
set? compactifying C". The class of functions analytic on the set r" will be denoted d (r).Similarly, the
set of functions continuous on r"
except perhaps at infinityt will be C(r). K" is isomorphic to a real
n-dimensional Euclidean space and, when appropriately completed, it forms the Bergman-Shilov boundary
of r". Considering intervals I in this space and denoting t, = Im z, we may set
dt1 dt2. . . dt,
mes (I) = 7
(l+t:)(l+t:). . . ( 1 + t 2 , )

t 'Infinity'here means the set {a}compactifying the space C",i.e. {a}= {(zl, z 2 , . . . , z , ) : l / z , = 0 for at least one value of i, 1 s i s n}.

0098-98 86/84/ 0201 11-1 2$0 1.20 Received 19 September 1983

@ 1984 by John Wiley & Sons, Ltd. Revised 30 November 1983
112 J . GREGOR

which can be extended to a measure p on K". We now have mes ( K " )= 1.The set of functionsf: K " + C ' ,
for which j K n ( f ( * d p < + will~ be denoted by L 2 ( K ) .When the scalar product (f, g ) for functions f,
g E L 2 ( K )is defined by ( f , g) =IK" fg d p , L 2 ( K )is a Hilbert space. In the following G = { f &(I')
~ l l C(T);
flK e L 2 ( K ) } A. function f:r"+ C will be called positive real ( f 24)~ if f~ d(I'),f(r")c r' and f ( x ) is
real for any real vector x > 0. A rational positive real function is a reactance function ( f %R) ~ iff f~ 24
and f(z)+ f ( - z ) = 0 for all z at which f is analytic.
In the class of positive real functions various interpolation problems can be considered; we shall
concentrate our attention on those which are significant in network theory; they have one-dimensional
counterparts, which have been already used in solving various system-theoretical problems.
1. Let a set E, E c K" and a function fE : E + C be given. The existence, uniqueness and construction
of its positive real extension is to be found, i.e. a positive real function f, continuous on the set I'" U a,
E c R c K " , assuming the given values f E on the set E, is sought. For n = 1 this is sometimes called
interpolation on the real frequency axis.
2 . Let a set E with E c I'" and a function f E :E + C be given; again we are looking for existence and
uniqueness conditions, and for a method of construction of a positive real extension f of the given function
f E . To avoid trivialities we will assume in both cases that no partial derivative of the function f extending
f E is identically zero, i.e. the function f for which we are looking actually depends on all of the variables
involved.
The problem of positive real extension is of rather a general nature; its solution seems to be rather
difficult for functions of several variables. It would be, for instance, highly desirable to have a theoretically
founded procedure, like the one by Bode and Gewertz for one-variable functions, which would construct
a network function from its values at the distinguished boundary K " of I"'. Some results concerning
analytic extensions of functions analytic on the unit disc from values on an infinite set of points on its
distinguished boundary can be found in Rudin's book;' these results show the complexity of these problems.
In this paper the sets E will be considered at most countable, in many cases finite. Even under this
restriction the two formulated problems have to be modified, e.g. mixed versions (with E c K " U 1-I') or
other generalizations of the known one-variable results (e.g. with E c 61'", where ST" means the boundary
of the set r" will not be considered here.
The 'symmetry' of positive real functions imposes some mild restrictions upon the function fL which
has to be extended. A set E will be called -- symmetric if z E E implies Y E E. A function fE defined on a
symmetric set will be called symmetric if ffi( z )= fE ( 2 ) . Any set E can be extended into a symmetric set;
this need not be true for a function f defined on a subset of E. The pair ( E , f E ) where , E is a symmetric
set and fE a symmetric function defined on E, will be called an admissible pair (compare with Reference
6.
Let us now recall some known results for one-variable functions.

Lemma
Let E c K ' = 61,E finite and let ( E ,f E ) be an admissible pair, R e fE 2 0. Then a positive real extension
f E 9? with flli = f E always exists. If Re f E = 0 for all z E E, this extension can be found in the class of
reactance functions ( f ~ 3If ~ E )consists
. of N points, the function f can be rational and of degree
s2*[(N+1)/2]-1.

Lemma 2.'
Let E c r', E finite, and let fE c r' so that ( E ,f E ) forms an admissible pair, card E = n. Then a positive
real extension f of fk exists if and only if the Hermitian matrix Q = [qik], i, k = 1, 2, . . . , n with elements

is positive semidefinite. If det Q = 0, then this extension is unique and rational of degree at most 2n.
 INTERPOLATION 113

Procedures of effective construction of the interpolating one-variable functions have been

details of these calculations will be omitted here and in forthcoming examples.

2 . INTERPOLATION WITH KNOTS IN K "

Let us first consider the 'reactance' case.

Proposition 3
If E c K " is a finite symmetric set, fE :E + C a symmetric function with R e fE = 0 , then there always
exists a reactance function f :I?" + C of the form

f ( ~ ) = f i ( z l ) + f 2 ( z J.+. +. f n ( Z n ) (3)
such that flE =fE for all z E E, and all the functions f i , i = 1, 2 , . . . ,n in (3) are one-variable reactance
functions.
Proof. Let E = ( Z ( ~ ) : Z ( ~ ) =z(i kZ), $. .~. ,)zkk'), , R e z j k ' = O , i = l , 2 , . . . , n, k = l , 2 , . . . , N } , where
N = card E. It is best viewed in matrix form: in the following we shall refer to the N X n matrix with
elements Z k r as to the matrix E of kn0ts.t
If each column of the matrix E consists only of numbers of mutually different absolute values ( 1 z k l ( # / z k , (
for all fixed numbers k ) , we can formulate n different one-variable interpolation problems with f i ( Z k 1 ) =
( l / n ) f E ( z ' k ' ) k, = 1 , 2 , . . . , n, where f E ( ~ ( k ) ) are given values. Evidently from (3) we have
" 1
f(z'k')= fl(zkl)= -fE(z'k')=fE(Z(k))
r=l r = ~n

and f is a sum of reactance functions.

If the above condition is not satisfied we will consider a system of N" linear equations with ( N " n - p )
unknowns ark' f i ( z k i ) , where p is the total number of equal pairs in the columns of matrix E R , N* is the
number of its rows. The system can be written as
n

1 aik=fE(z'k'), k = 1, 2,. . . , N* (4)

r=l

, number p does not exceed N * ( n - 1); otherwise at least two of the knots would
Because z ' ~#) z ' ~ )the
be equal. The matrix of the system is quasidiagonal and therefore the system (4) always has a solution.
Next, n one-dimensional interpolation problems for functions fi with fi(zik)= arkhave to be solved, which
is always possible according to Lemma 1 . This completes the proof.

Example 4
Let a reactance function f : r4+ C be found so that

f (i,2i, 4 j 7 i )= 2 i f(2j, -j,i, - j ) =i f(3i, 2J,i,i)= O

the matrix E R reads

ER=j[i -: 2 4 -I]
~

t In some cases it is sufficient to consider the matrix E reduced to E R containing only one of any two complex conjugate rows,
i.e. E R is an ( N + r ) / 2 X n matrix, where r s N is the number of 'real' knots; d4)= 2'4', q = 1, 2 , . . . , r N.
114 J . GREGOR

the system (4) after multiplying each equation by ( - j ) is

~l,+~I,+a,,+a,,=2
a21 + a,,+ a23- a14 = 1

a31 +a,,+ a2,+ a,, = 0

One of its solutions is
1 3
a,, = a,z= a13 = a,,=; u*1= u22= a,3=2 a31 = -5
We now have to solve four one-dimensional interpolation problems:
1. f l ( i ) = j / 2 , f1(2j)=j/2, f1(3j)= -3j/2
2. f 2 ( 2 j ) = j / 2 , f2(-j)= j / 2
3. ~ 4 j=)j / 2 , =i/2
4. f 4 ( j ) = j / 2
The final answer is

Here the degree of f, is smaller than the degree estimated in Lemma 1. The given solution is not unique
and a procedure resulting in functions f i of minimum degree would be desirable.
Let us now consider the more general case of the extension of a non-reactance positive real function
from its values on K " . The following example shows that the pattern of Proposition 3 cannot be used.

Example 5
Find a positive real function f :Tz+ C of the form
f(Zl,Z,) =fl(zl)+f2(22)

satisfying f ( j, 2j) = 2 + j, f( j , - j ) = 1 - j, f ( - Zj, - 2 j ) = 3j.

The values ark of f l and f,, with the notation similar to that of Example 4,have to satisfy the following
system of equations:
all + a 1 2 =2+j
all + a,, = 1- j
a31 +612=3j

where a , k are complex numbers with R e a r k 3 0. Denote R e ark= b , k and look for the non-negative solution
of the following system of equations.
bIl+b,Z =2
611 +b22 =1
b12 +b,1=0
We have b I 2= b,, = 0, which implies 6, = 2 and b,, < 0. This contradiction leads to the conclusion that
an interpolating function of the given form does not exist.
It is interesting to compare this result with example 6.3.1 in Kudin's book.' The apparent contradiction
of these two examples shows the significahce of our symmetry requirements. We shall see later in Section
4, that these requirements also have other unpleasant consequences.
The problem under consideration can be solved using another way of reasoning. Let g be any reactance
function of n variables. This function maps the set E c K " into a set G c K ' . According to Lemma 1 it
 INTERPOLATION 115

is always possible to find a one-variable positive real function F, which assumes the desired values fE at
the points of G so that f ( z ) = F ( g ( z ) )is the extension of fE. The degree of f depends on the choice of
the function g. Thus, we obtain

Proposition 6
Let ( E , f E ) be an admissible pair, E c K " ; to any reactance function g such that f E ( d k ) ) #
f i 3 ( ~ ( 4 ) ) 3 g ( ~#( g(z'4')
k)) for all z ( ~ )z(4'
, E E, there exists a one-variable positive real function F (with
f~ !iBR when Ref, = 0 for all z E E ) , such that the function f ( z ) = F ( g ( z ) )is a positive real extension of
fE from the set E onto r".

Example 7
Let us again consider Example 5. Taking g( z ) = al z1+ a2z2with a , > 0, u2> 0, the mappings E R+ G
and G + fE are described as follows:

F
(j,-j) 3 j(a,-u,) -+ I-j

According to Lemma 1, the resulting function f = F ( g ) has a degree not exceeding 5 . For some values
of a l , a2 this degree can be lowered.
This example yields the following

Corollary 8
Let ( E ,fE) be an admissible pair, E c K " , card E = N. Then there exists a positive real extension f of
fE which has a degree not exceeding 2 * [ ( N+ 1)/2] - 1.
The proof follows from Proposition 6 with g(z) aizi and Lemma 1.
We may conclude that interpolation with, a finite number of knots on the distinguished boundary K "
of r" is always possible by positive real functions for any admissible pair ( E ,f E ) ; by simple manipulation
this interpolation can be reduced to that of one-variable functions.

3. INTERPOLATION WITH KNOTS IN rn

When E c r",the problem of interpolation is much more difficult. To prove the basic result formulating
necessary conditions, we need an integral representation of positive real functions and some lemmas,
which have already been p u b l i ~ h e d We
. ~ shall briefly repeat them here for convenience.
Consider the Hilbert space L 2 ( K )introduced earlier:

Lemma 9
The set S of functions

is complete and orthonormal in L 2 ( K ) .

116 J. GREGOR

The proof of orthonormality is obvious, e.g. from

the completeness follows from more general results, e.g. in Reference 3, Chap. XVII.
The subset S + c S, consisting of functions with mi 3 0, is evidently orthonormal in the set G c L z ( K )
of functions which are analytic in the set r" and continuous on its closure. S' is countable; let its elements
be denoted by (Pk, k = 0,1, . . . , cpo = 1. Direct calculation gives

for R e u > 0, Re z 2 0.

Lemma 10
If f E G, then for all z E r" there is

'
Here, H , denotes the function assuming values H ( u , z ) for any z E r'', , ) is the scalar product and
j t = ( j t l , . . . , jt " ) .
The proof follows from the series expansion f ( z ) =I?="( Y k ( P k ( 2 ) and Lemma 9.

Lemma I 1
If f E G, then

for all z, U E ~ " .

Proof. For all z E r" there is H ( u, z ) f ( u ) = (H,f)(u ) E G; therefore Lemma 10 yields H ( u, z)f(u ) =
(Hzf,
Hu)= (f, H,,Gz)and similarly

The result is now obtained by adding these two equations.

Proposition 12
Let ( E , f E )be an admissible pair, E = r". If there exists a positive real extension f of fk, then the
Hermitian matrix Q = [ q i k ] ,

E i = 1, 2,. . . are the elements of E ) is positive semidefinite.

(where z ( ~ )E,
Proof. Let f~ 92,f l E = f E . For the real vector p = ( p , , p 2 , . . . , p" ), satisfying 0 < p, < 1 for all i = 1 ,
2 , . . . , n, form the function
 INTERPOLATION 117

Let the mapping K, : r"X r"-+ C be defined by

Let us now consider the sum

s = c K,( z ( i )Z, ( k ) ) ajdk
i. k

with arbitrary complex numbers ak and elements 2") of the set E. Since f 3,we have f, E 3 ;moreover
f,E G and therefore according to Lemma 11 we can write

i, k \ i, k I

Since f, E 3,it must be that S 2 0 for all complex values ak. The mapping K , is therefore positive
semidefinite for any considered vector p. Taking p + 0 we obtain the positive semidefinitness of K

(10)
Hence

and K is positive semidefinite iff the statement under discussion holds true, which has to be proven.
The above proposition is a generalization of the 'only- if statement' of Lemma 2 to positive real functions
of n variables. Unfortunately, this proposition only gives the necessary conditions. The following example
shows that these conditions are not sufficient.

Example 13
For n = 2 let the set E be E = ((1, l), ( 3 , 3 ) }and let fE be f E ( l1)
r = 1, f E ( 3 ,3) =4. Evidently, ( E , f E )is
an admissible pair, the matrix
Q=[li2 6/16 ]
5/16 219
is positive semidefinite, but no positive real extension f of f E can exist. Assuming the existence of the
function f(z,, zz),the one-variable function rp ( z )= f ( z , z ) is also positive real, with rp(1) = 1, rp(3) =4.
The corresponding matrix

is not positive semidefinite, which according to Lemma 2, contradicts the statement cp E 3.Therefore no
positive real extension of fE can exist and the conditions of Proposition 12 are not sufficient.

Proposition 14
Let ( E ,fE) be an admissible pair, E c r". A positive real extension f~ 3 of f E exists if and only if
there exists a positive real function g E 3 such that the matrix Q, = [ q l k ] ,

is positive semidefinite.
118 I. GREGOR

If this condition is satisfied, the function f can be given the form f (z ) = F ( g ( z)), where F is a one-variable
positive real function.
Proof. The 'if statement' follows from Lemma 2, the 'only if' part follows, when substituting g = f .
The resulting matrix Q, in this case has all its elements equal to 1 and is positive semidefinite.
Unfortunately, a non-trivial choice of the function g is not always possible, and therefore the 'only if'
part of the last statement is of little practical value.
Before turning t o construction methods for interpolating functions, let us demonstrate some more
examples which will reveal the complexity of the problem.

Example 15
Let (E,f,) be an admissible pair and let all the columns of the matrix of knots E=[zkr] consist of
mutually different values. Further let the matrices Q,, j = 1, 2,. . . ,n

be positive semidefinite. Then there exists a positive real extension F of the form

where fi E 3 are one variable positive real functions, i.e. FIE = fE

Example 16
In some cases a 'three-point interpolation problem' can be solved supposing the interpolating function
f to be in the form

a+p+y+6>0, f f s Z p y , a>0.
Taking, e.g. f ( l , l )= 7, f(3,3) = 3, f(4, I ) = 8, direct substitution yields a rather complicated system of
three equations with 6 unknowns. One of its solutions gives the positive real function

whereas the given values of fE were obtained by evaluating the function f(z,, zz) = 28(z1+ z2)/
(21+ 4-7,z2+ + 2). The following partial result is of more general interest.

Lemma 17
Let zo = (ZOI,202, . . . , z O nE)r" and fo E rl. Then for any positive real function g, which satisfies the
condition larg g( zo)l a larg fol, the function f, defined by the equation

where
 INTERPOLATION 119

is positive real, f E 9,and f( 20) = fo, provided that

(15)

If (15) is not satisfied, then f~ 9 with f(zo)=fo does not exist.

Proof From (14), f can be expressed as

By denoting g ( z o )= r eip, f o = p el', for zo # Zo we obtain 0 s S IcpI < ~ / 2 therefore

;
sin ( c p - +),
A=
sin (cp + CL)
and f~ 3. From (14) we immediately obtain f(zo) =fo.
Note that in this lemma we can always choose g ( z ) = ( a , z ) for a suitable vector a > 0.

4. THE NEVANLINNA-PICK PROCEDURE

We shall start with a simple proposition.

Proposition 18
Let g, fi be positive real functions of IZ variables, zo a real vector, zo> 0. Then the function f, defined
by

is positive real and f(z0) = a for any positive value of b,.

Proof. Denoting [f(z )- a ] / [ f ( z +) a ] = cp(z),we obtain Ip(z)l< 1 for all z E r",because g, fi E %. cp
is analytic in I'" and therefore f(z)= a[l+cp(z)]/[l -cp(z)] is analytic in r",f(r")c .'?I Evidently for
any vector x > 0, f ( x ) is real, which completes the proof.
In principle, this proposition enables us to carry out a procedure aimed at solving the interpolation
problem with the set E of knots consisting of a finite number of real and positive vectors z ( ~ )E, = { z ' ~ ) ,
i = l , 2 ,..., N } .
Select m e of the points, say zN,and choose a suitable function g. Then the interpolation problem will
be solved by the function f in (17) provided the function ( f i ( t ) - b , ) / ( f , ( z )+ b,) assumes the values
f(zi)-f(zhi) g(zi)+g(zN)
f(Zi)+f(ZN) g(z;)-g(z,)
at the remaining points ti,i = 1, 2, . . . , N - 1. This is again an interpolation problem with the number of
knots reduced by one. Repeating the procedure ( N - 1) times, we obtain the result.
This is essentially the classical scheme of the Nevanlinna-Pick procedure. Nevertheless, some important
differences (which we shall discuss later) make the use of this procedure rather difficult in the multivariable
case.
Let first an example be shown.

Example 19
Find a positive real function f E 3 in two variables, satisfying the following conditions
f(l,1)=7 f(l,2)=? f(3,3)=3 f(&1)=8
120 J. GREGOR

Let here the point (1,2) be chosen as the first interpolation point and let the function g be of the 'simplest'
form
g(z) = a121 + a222
A simple straightforward calculation results in a formula for the values of f , at the remaining interpolation
knots

Here, z") are the points of the set E, w, = f ( z ( ' ) ) (a,

, z ) is the scalar product. Choosing a = (1.7, l ) , we get

Detailed reasons for the given choice of the vector a as well as detailed calculations are omitted here.
For the function f , E 3 we get from (17) the values
f l ( l ,1)=0~16883116886,
f,(3,3) =0*1044950379bl
fl(i,1)=0.2614840989bl
Proceeding in a similar way with the point (3,3) and g ( z )= 22, + z2, we get
f l ( ~ ) - 0 . 1 0 4 4 9 . . . 22,+2,-9f2(Z)-b2
fl(Z)+O.IO499 . . . 2 ~ 1 + Z 2 + 9 f Z ( ~ ) + b 2
and
f Z ( 1 , 1)=0-3598370549b2 f2(:, 1)=0.1946901718b2.
These conditions can be satisfied by
f z ( Z 1 , 22) = (O.33O2937661~1+0.0295432888)b2
with f 2 E 9:Hence, the interpolation problem has been solved; the condition f E % is also satisfied in
virtue of Proposition 18.
It has to be pointed out that this proposition, besides some of its disadvantages, has a positive feature:
it gives such an interpolating function f which can easily be synthesized by series-parallel connections
because, in fact,

The main difficulty in applying the procedure lies in the choice of the function g in each of the steps.
These functions g k E 9
2 must satisfy a rather severe inequality

for all i = 1, 2 , . . . , k - 1. This inequality does not depend explicitly on the knots z ~ + Z~k +,2 , . . . , z, and
therefore the ordering of knots may influence the choice of g k . A good choice of the 'reference knot' z k , '
hence a suitable ordeIing of knots will be that which results in the smallest possible value of

The function g k E %! need not depend on all the variables; simpler choices, if possible, are preferable.
Let these suggestions be illustrated by one more example.
 INTERPOLATION 121

Example 20
Lettheretobefoundafunctionf 9withf(1, 1, l ) = l O , f ( 3 , 2 , 1 ) = 1 4 , f ( 8 , 2 , 6 ) = 3 9 , f ( 6 , 9 , 3 ) = 4 0 .
Calculating first the values

we find that mk attains its minimum for the point = (8,2,6). It equals 29/49. Trying at the first step
the simplest choice g ( z ) = z1 we find that the corresponding inequality for fl(z) cannot be satisfied. With
z ( ~=)(1, 1, 1) we obtain mk = 3/5, which is close to the previous value; in this case, however, g ( z ) = z ,
yields
f(z)-lO --21-1 f , ( z ) - l
f ( z ) + l o - z,+ 1 f l ( Z ) + 1
with

' (8,2,6) and g,(z)

Proceeding in a similar manner, we choose z ' ~ = = 3z2+z3 and we obtain
2 ) - 302 - 322+ 23 - 12 f i ( Z ) - 1
41f,( 2 ) + 302 - 322 f z3+ 12 f2( z ) + 1
with f2(3, 2 , l ) =w,
f,(6, 9,3) =#. To find f2E 9,we take g 3 ( z )= z1+ z3, so that

346fi ( 2) - 2 189 - Z 1 + 2 3 - 4 f 3 ( Z ) - 1
346f2( z ) + 2 189 - z + z3+ 4 f3 ( z ) + 1
with f3(6,9,3) = 0.0566981216, and the function f3 can be identically equal to this constant. Therefore
the formula

together with (21) and (20) yields the final answer.

The consideration above shows that on the basis of Proposition 18 the Nevanlinna-Pick interpolation
algorithm for positive real functions can be generalized. The procedure does not depend on the number
of variables. The volume of necessary calculations does not depend essentially on the number of variables,
although at present trial and error methods must be involved to find suitable auxiliary functions, which
are necessary to carry out the procedure. When the interpolation problem has a solution in the class 92
of functions, it is principally always possible to find one by the described method, at least when the set
of interpolation points consists only of real vectors. With complex interpolation points the procedure can
be applied in a broader class of functions; for positive real functions it remains unclear how to satisfy the
'symmetry' requirements. It is hoped that future research may result in a systematic search procedure for
the involved auxiliary functions even in these more general cases.

5. CONCLUDING REMARKS
The interpolation problem for multivariable network functions, even in its specialized formulation given
in the introductory section of this paper, is far from being completely solved, although some basic results
have been summarized. Some of the results presented here invoke the idea of using the well-known
'bilinear' transformation of variables and use results in interpolation with functions analytic in the unit
disc. These result would be closer to n-dimensional digital system theory. Especially, such extension of
the Nevanlinna-Pick procedure could be of more general interest, but some aspects, important in the
above explanations would be lost. Here perhaps the close relationship between the Nevanlinna-Pick
122 J. GREGOR

algorithm and problems of multivariabie impedance synthesis has to be mentioned. We hope to devote
another paper to these problems in the near future.

REFERENCES
1. V. Relevitch, Interpolatinn matrices, Philip Rex Repts., 25, 337-369 ( 1 970).
2. L. 1. Smilen, Interpolation on the real frequency axis, IEEE Int. Cotiv. Kec., 13(7), 42-SO (1965).
3. A. Zygmund, Trigonometric Series, Cambridge University Press, 1959.
4. Ph. Delsarte, Y. Genin and Y. Kamp, On the role of the Nevanlinna-Pick problem in circuit and system theory, Int. J. Cir.
Zbeor.Appl., 9, 177-187 (1981).
5. W. Rudin, Funcfion Theory in Polydiscs, W. A. Benjamin, Inc., New York, 1969.
6 . [I). C. Youla and M. Saito, Interpolation with positive real functions, J. Frankl. Inst., 284, 77-108 (1967).
7. J. Gregor, Integral representation of n-variable positive real functions, Comm. Math. Univ. Carolinae, Prague, 21(4), 707-717
(1980).
8. J. Gregor, Interpolatiun with rnultivariable positive real functions, Proc. ECCTD83, Stuggart, 1983.
9. A. Koriinyi and L. Pukrinszky, Holomorphic functions with positive real part a n polycylinders, Trans. Amer. Math. Soc., 108,
449-456 (1963).