JOURNAL OF MATHEMATICAL ANALYSIS AND APPLlCATIONS 55, 394406 (1976)

The Voltage and Current Transfer Ratios

of RLC Operator Networks

A. H. ZEMANIAN*

Department of Applied Mathematics and Statistics,

State University of New York at Stony Brook,
Stony Brook, New York 11794

Submitted by Lo@ Zadeh

The subject of this work is the voltage and current transfer ratios of three-
terminal networks having no mutual coupling and whose impedances are
analytic functions taking their values in an abelian self-adjoint algebra of
bounded linear operators on a Hilbert space. Each such value is also assumed
to be invertible. It is shown that these ratios have the form [I $- .4(l)]-i,
where, for each 5 in a sufficiently small open cone in the right-half complex
plane with apex at the origin and the real axis as its bisector, the numerical
range of A([) is contained in a compact subset of the open right-half plane.
This implies that the ratios are strictly contractive for each (; in the cone. The
angle of the cone is n/(2k + 2), w h ere k is the number of internal nodes of
a certain “surrogate” network; this result is best possible. For two-terminal-
pair networks the ratios are shown to be strictly contractive for each 5 in a
similar cone with angle ?r/(2k + 4).

1. INTRODUCTION

The concept of an electrical network whose parameters are bounded linear

operators on a Hilbert space H arises in the analysis of distributed or infinite
electrical networks [ll, 141. Moreover, it provides a new and fairly unex-
plored area for theoretical research. There have been, to be sure, many papers
dealing with the interconnections of blocks whose input-output relations
are operators on H; the interconnections dictate various relationships between
these operators. Such, for example, is a feedback loop with forward and
reverse branches characterized by operators. The systems that are the subject
of this work are distinguished as follows. They are electrical networks whose
currents and voltages are H-valued analytic functions that satisfy Kirchhoff’s
node and loop laws respectively and a generalized Ohm’s law as well; the

* This work was supported by the National Science Foundation Grant

PO33568-X001.
394
Copyright tC 1976 by -academic Press. Inc.
All rights of reproduction in any form reserved.
 VOLTAGE AND CURRENT TRANSFER RATIOS 395

branch impedances are operator-valued analytic functions that imitate

resistive, inductive, and capacitive impedances.We shall refer to this
configuration as an operator network. Such a system arises as the Laplace
transform of an electrical network whosevoltages and currents are H-valued
functions of time and whose resistances,inductances, and capacitancesare
fixed operators [ll, Corollary la]. Prior works dealing with this concept are
[2, 3, 4, 11, 121.
Our objective in the present work is to investigate the voltage and current
transfer ratios of three-terminal finite operator networks having no mutual
coupling or internal sources.We shall show that these ratios have the form
[I + -4(<)]-l, where, for each 5 in a sufficiently small open cone in the right-
half complex plane with its apex at the origin and the real axis as its bisector,
the numerical range of S(c) is contained in a compact subset of the open
right-half plane. This implies that the voltage and current transfer ratios are
strictly contractive for the same restriction on 5. The angle of the cone is
equal to z-/(2k + 2), where k is the number of internal nodesoccurring in a
surrogate network defined below. Similar results are establishedfor two-
terminal-pair networks.
In order to establish these results we assumethat the branch impedance
values are invertible and commute with each other. This allows us to use
Kirchhoff’s third and fourth laws for our operator networks. Although
commutativity is a strong restriction, we shall show by example that it is a
necessaryone.
In this work, H denotesa complex Hilbert spacewith inner product (., .)
and norm I/ .I1 . By an “operator” we always mean a bounded linear operator
on H. f * andf-i denote respectively the adjoint and inverse of an operatorf,
and I is the identity operator on H. (Currents are always denoted with
subscripts; for example, Ik([).) C is the comples plane, 5 E C is a complex
variable, and [* is the complex conjugate of 5.

2. THE NUMERICAL RANGES OF CERTAIN OPERATORS

In subsequent sections we will analyze a finite network whose branch

impedanceshave the form of either r, 15,or d[-l, where the network param-
eters Y, 1, and d are positive invertible operators and each such parameter
commutes with all the other parametersof the network. Let J& be the set
of all the network parametersand & the collection of all abelian self-adjoint
algebrasof operatorsthat contain As. rd is nonempty and is partially ordered
by set inclusion. By Hausdorff’s maximality theorem [8, p. 3921,& contains a
maximal totally ordered subcollection g. Let & be the union of all the sets
in a. Then, 4 is a maximal abelian self-adjoint algebra of operators. Thus,
396 A. H. ZEMANIAN

given any operator network with parameters of the type indicated, there
exists such an ~5’ containing every branch impedance. In the following, we
take&7 to be fixed.
& contains every branch admittance (i.e., the inverse of every branch
impedance) as well. Indeed, each branch admittance is a positive invertible
operator f-t multiplied possibly by 5 or 5-l. Thus, its adjoint is the same
positive operator multiplied possibly by 5” or (5*)-l. Now, f-l commutes
with every operator that commutes withf. Our assertion now follows from
the fact that &’ is maximal.
It should also be noted that every operator in J? is normal.
& is closed as a subspace of the Banach space B(H) of all operators on H.
Indeed, iffn +fin B(H),fn EA, andg is any member of&?‘, thengftgf, =
fng -+ fg, Hence, f commutes with every g E,&‘. Similarly, so too does f‘“.
Thus, f EM by the maximality of&‘. This means that,.& is a Banach space in
itself and is in fact an abelian C*-algebra.
We now invoke the Gelfand-Neumark theorem [9, p. 3321, which states
that there exists an isometric *-isomorphism of 4 onto the abelian B*-
algebra ‘6(Y) of all continuous complex-valued functions on the compact
Hausdorff space 4 of all maximal ideals in A. Letpdenote the image off
under this isometric *-isomorphism, and let fl( f) be the spectrum off. It is a
fact [9; p. 3201 that A(f) equals the range R( f ) of & Consequently, if f,
g E.,z’Z, then 41( fg) = I?( &) C R( f ) R(f) = A(f) A(g). \Ve will use this
result in the proof of Lemma 2 below.
A few more definitions are needed. The numerical range W(f) of an!
operator f is the bounded set of complex numbers

W(f) =z [(fa, a): a E H, jj a jl = I)-.

Let uz be any positive integer. C(m) will denote the cone {A E C: 1 arg h [ <
42~2) with the origin excluded. Thus, C(m) is an open set. (Throughout
this work. the principal branch of the ‘(arg” function will always be under-
stood.) For any fixed 5 E C(m), we set

Q(m,2;)={hrC:IargXj <mIarg<i)

and take the origin to be a member of Q(m, 5). Thus, Q(m, 5) is a closed cone,
whose sides form an angle less than n, with the real positive axis as its bisector.
Q(m) will denote the set of all analytic operator-valued functions F on
C(m) such that, for each 5 E C(m), the following three conditions hold.

(i) F(i) Ed@

(ii) W[F([)] C Q(YFZ,1)
(iii) W[F(<)] is bounded away from the origin.
 VOLTAGE AND CURRENT TRANSFER RATIOS 397

Note that every branch impedance and admittance of the network con-
sidered at the beginning of this section is a member of Q(1). Also,
Q(1) c Q(2) c Q(3) c ... .

I~EiXlMA 1. If F and G are membersof Q(m), then F + G and F(.)-l are

also membersof Q(m).
Proof. It was shown in [13, Lemmas 1 and 21 that, if F and G are analytic
operator-valued functions on C(m) satisfying conditions (ii) and (iii) for each
5 E C(m), then so too are F + G and F(.)-l. SinceA is closed under addition,
it follows immediately that F + G EQ(~). On the other hand, for each
5 E C(m), F(c)-l commutes with every operator in&?, and so too does[F(<)-l]*.
By the maximality of &‘, F(c)-’ E&‘. Thus, F(e)-l EQ(m) too.
It follows from Lemma 1 that the impedance of a series or parallel con-
nection of impedances in Q(m) is also in Q(m).

LEMMA 2. If Fl cQ(m,) and F2 EQ(m,), then Fl(.) F2(.) EQ(m, i- m2).

Proof. Choose any 5 E C(m, $- mz). Since C(m, + m,) is contained in
both C(m,) and C(m,), we have that F,(c), Fa(& and -- F,(c) F2(<) are all
members of ,H and therefore normal. Now, the closure W(f) of the numerical
range of a normal operator f is the convex hull of its spectrum [6, p. 1121.
Consequently, A[F,([)] C W[F&)] C Q(mj , 5) for i = 1, 2. But,
4FI(OF2(5)1C 4Fd5)14F2(5)1CQh + mz,5). Since Q(m,+ ~2,, 5) is
an acute cone, the convex hull of A[F,({)F,(l)] is also contained in
Q(m, + m2, 5). Therefore, so to is W[Fl([)F2(<)]. Finally, since IV[F,(c)]
and W[F,(lJ] are both bounded away from the origin, this argument also
shows that TV[F,([)F,(~)] is b ounded away from the origin too. Altogether
then, Fd.1 FA.) E oh + +J.

3. THREE-TERMINAL NETWORKS

In the following we shall assume that all the branch impedances of the
network at hand are members of Q( 1). W e note again that any branch impe-
dance having one of the forms indicated at the beginning of the preceding
section is a member of Q(1).
An important consequence of our assumption and in particular of the
commutativity of the values of the members of Q( 1) for each fixed 5 E C( 1)
is that we can manipulate these operators in much the same way as complex
numbers are manipulated. For example, determinants of such operators exist
as operators and have their customary properties. Also, n simultaneous linear
equations in n unknown members of H with coefficients in Q(1) can be solved
398 A. H. ZEMANIAN

by Cramer’s rule so long as the determinant of coefficients has an inverse.

Ratios can be written in fraction form without indicating in what order the
numerator and the inverse of the denominator are to be composed. Finally,
and most important for our purposes, Kirchhoff’s third and fourth laws
[IO; p. 1581, [7, p. 2811 extend to our operator networks so long as 5 is
restricted to a C(m) with a sufficiently large m to insure that the sum of
the cotree impedanceproducts or the sumof the tree admittance products are
invertible. To verify this assertionwe need merely trace through the proofs
of Kirchhoff’s third and fourth laws [lo, pp. 160-163; 7, pp. 260-2841;
every step extends to our more general operator networks.
Here are most of the conditions and someof the terminology we shall use
for our operator networks in the subsequenttheorems.

Conditions A. The netword YI is connected and has a finite number of

nodes and branches. Each branch is incident at two distinct nodes. Either
three or four nodesare designatedasexternal nodes, and external connections
can be made to the network only at external nodes and only in conformity
with the three-terminal connections of Fig. 1 or the two-terminal-pair con-
nections of Fig. 2. All the other nodesof YI are called internal nodes and all the
branchesinsideYI are called internal branches. Branchesarising from external
connections to YI are called external branch. There is no mutual coupling
between any pair of branches. Also, there are no sourcesinside Yt. Every
internal branch hasan impedancein Q( 1), which implies that its admittance

ns
FIGURE 1

FIGURE 2
 VOLTAGE AND CURRENT TRANSFER RATIOS 399

is in Q(1) as well (Lemma 1); thus, no internal branch is a short circuit, and
there does not exist inside tll a path of short circuits between any two external
nodes. This ends Conditions A.
A maximal series connection in the network % is a path of internal branches
whose nodesare of degree two except for the first and last nodeswhich are
either external nodes or have degreesno lessthan three. A maximal parallel
connection in CJtis a parallel connection of internal branchescontaining every
internal branch that is incident to both nodesof the parallel connection. We
can replace a maximal seriesconnection (or a maximal parallel connection)
by a single branch whose impedance(or admittance) is the sum of the impe-
dances(or admittances) in the maximal series (or parallel) connection. Ry
Lemma 1, the resulting impedance(or admittanee) is in Q(1) whenever the
original impedances(or admittances)are in Q( 1).
Given an % satisfying Conditions A, we generate a certain equivalent
network, which we will call the “surrogate network”, as follows. Replace
all maximal series connections by equivalent single branches. In the
resultingnetwork replaceall maximal parallel connectionsby equivalent single
branches. Continue repeating these two steps. Since % is finite, the process
must stop, at which point we will have a three-terminal or two-terminal-pair
network%’ possessing the following three properties: (i) It satisfiesConditions
A; (ii) all its internal nodes have degreesno less than three; (iii) no two
internal branchesare connected in parallel. %’ is uniquely determined by ‘8;
we will call %I’ the surrogate network of %. %’ will have precisely the same
behavior at its external nodes as does‘%. Throughout the following, k will
denote the number of internal nodes in the surrogate network 91’.
The voltage transfer ratio T,,(c) of a three-terminal network is defined by
the equation
(2)
under the condition Is({) = 0. Here, V,(t) is the voltage rise of a voltage
source connected acrossthe input and V,(c) is the voltage drop acrossthe
output with polarities as shown in Fig. 3. The condition 1,(c) = 0 on the

II(C) k&)=0
- -
elm l n2

In

I +
WI

.---.I-
n3

FIGURE 3
400 A. H. ZEMANIAN

output current is achieved by maintaining an open circuit externally across

the output.
A cut-node of a connected network is a node whose deletion coupled with
the deletion of all the branchesincident to that node resultsin a disconnected
network. An external node of ‘% is a cut-node if and only if it is a cut-node
in W.

THEOREM 1. Let the three-terminal network !R of Fig. 1 satisfy Conditions A.

Assume that, with no external connections to %, n, is not a cut-node. Then, for all
5 E Wk + 3, II T2k311< 1 and T,,(5) has the form

T&J = [I + 45)1-‘,
where A ~Q(2k + 2) and I denotes the identity operator in H.
Proof. We replace% by its surrogate‘9’. Since 92’ hasthe sameterminal
behavior as does%, we may write VI(<) =.Z,,([) 1r([) and Va([) =Zar([) 1r(c),
where 4,(S) and Gl(5) are customary open-circuit driving-point and transfer
impedancesof %‘. In theseequations, it is understood that the input voltage
sourceof Figure 3 is replaced by an input current source delivering the same
current asdoesthe voltage source.This is permissiblesinceZ,,(l) is invertible
for every 5 E C(1) [ll, Th eorem21. The result is that we can take the admit-
tance of the external branch as being zero.
By Kirchhoff’s fourth law [7; p. 2811

Here, X(t) is the sum of all tree admittance products for %‘. W,,,(<) (re-
spectively, W,,,,([)) is th e sum of all 2-tree admittance products such that 7~~
is (respectively, n, and n2 are) in one componentof each2-tree and n3is in the
other component. Note that, since eachinternal branch admittance of ‘92’is in
Q(1) and sinceeachspanningtree in W hask + 2 branches,Lemmas 1 and 2
imply that [X(5)1-l exists for at least each Z;f C(k + 2). From (2) we get

In this equation 1r(<) can be any member of H. Indeed, by [II ; Theorem 21

Z,,({) is invertible for every 5 E C(l), and therefore we can always choose
V,(t) to obtain the desiredIr(b) = Z,,(J)-i V,(t). Thus, we can conclude that

T,,(5)= Z&J Z&)-’ = $+1.3 . (4

 VOLTAGE AND CURRENT TRANSFER RATIOS 401

Since each 2-tree has k + 1 branches, both IVi,,,(l;) and IV,,,(~) are sums of
products of K + 1 branch admittances each and are therefore members of
Q(k + 1) according to Lemmas 1 and 2. By these two lemmas again, T,,(c)
exists for every 5 E C(2K + 2).
We next observe that every product in IVia,, appears in IV,,,([). On the
other hand, IViSa has more products than does IV,,,,(<). Indeed, n, is not a
cut-node even when ‘W has no external connections. Therefore, there exists a
path from n2 to ns inside% not containing n, [ 1; p. 241. This implies that there
exists at least one 2-tree in ‘W such that n, is in one component and n, and 1za
are in the other component of the 2-tree. The corresponding 2-tree admittance
product is in Wi,,(<) but not in W,,,,(c).
We now divide the numerator and denominator of (4) by IV,,,,(<). This
yields (3) where A(c) = M(t) IV1,,,(&l and M(5) is the sum of those
admittance products appearing in W,*,(iJ but not in W,,,,(c). It follows from
Lemmas 1 and 2 that A l Q(2k + 2).
Finally, by the definition of Q(wz), for each 5 E C(2k + 2), FY[A([)] is
contained in the acute cone S2(2k + 2, 5) and is bounded away from the
origin. Therefore there exists a S > 0 such that inf Re W[A(5)] > 8. Con-
sequently, for any nonzero a E H,

(1 4- 6)II a II2< Re([l+ A(C)1@,a)< IV + 4511a IIII a II .

This implies that 11Tsi(5)11 = I~[1 + A(<)]-‘// < (1 + 8)-l < 1. The proof
is complete.
The assumption used in the definition of Q(1) and therefore in the hypo-
thesis of Theorem 1, that all branch impedance operators in W commute is a
severe one. The question arises whether it can be relaxed without violating
the condition that 11T,i(c)\\ < 1. Th e answer in general is no. Indeed, in the
network of Fig. 4 assume that the resistors y1 and r2 are positive invertible
operators that do not commute. But, T,, = (I + J-l, where A = ylypl.
Now, inf W(A) < 0 because the commutativity of two positive operators is
necessary for their composition to be positive. Upon multiplying by a suitable
number, we can choose rr such that inf W(A) > -1. This yields
402 A. H. ZEMANIAN

0 < inf ?#‘(I + A) < 1. But, the closure of the numerical range of a positive
invertible operator is the convex hull of its spectrum, and therefore the
spectral mapping theorem [6, p. 2251 yields /I Tar 11= sup W[(I + A)-l] =
[inf W(1+ S)]-’ > 1.
It is also worth noting that the integer 2K + 2 in the conclusion
;4 E Q(2K + 2) of Theorem 1 cannot be decreased. This can be seen by
computing Tsi([) for the three-terminal ladder network of Fig. 5.

----r-T------I-,.
“22n4 FIGURE 5

The current transfer ratio Jai([) of a three-terminal network is defined by

the equation
-uJ = Jz&) 4(5) (5)
under the condition IT&<) = 0. The polarities for the currents are indicated
in Figure 6, where the current It({) results from an external current source
across the input and the condition I’,({) = 0 is maintained by a short circuit
q

across the output.

THEOREM 2. Let the three terminal network % of Fig. 1 satisfy Conditions A.

Assume that, with no external connections to ‘3, n, is not a cut-node. Then, for all
5 E C(2k + 9, II Ja1(5)ll-=c1 and J&l hastheform

M,,(5)= v + wrl, (6)

where B ~Q(2k + 2).
Proof. With the connections indicated in Fig. 6, we again replace 91
by 91’ and write Ii(c) = Iin V,(c) and --1,(c) = Y,,(l) I’,([), where
 VOLTAGE AND CURRENT TRANSFER RATIOS 403

YI,({) and Yar(<) are short-circuit driving-point and transfer admittances of

9X’. This is permissible because by the dual of [ll, Theorem 21 Y,,({) is
invertible so that the input current source in Fig. 3 can be replaced by a
voltage source having the same voltage as that appearing across the current
source. With this replacement, both external branches have zero impedance.
Let b, be the external input branch, b, the external output branch, and b, ,
b 4 ,..., b, the internal branches of ‘S’. By Kirchhoff’s third law [lo, p. 1581,

Yll(5) = %(5)/W), 1’21(5)= ~21(5)/~(5)’

Here, D(c) is the sum of all cotree impedanceproducts. By Lemmas 1 and 2,
[0(.)1-l exists for at least every 5 E C(b - k - 2). IV,,(~) (respectively,
IV,,(~)) is the sum of all impedance products having the following three
properties. (i) Each product is the product of b - k - 3 distinct branch
impedances. (ii) The removal of the corresponding branches results in a
network having exactly one loop. (iii) That loop passesthrough b, (re-
spectively, b, and b,).
Now, according to (5)

1;1(0 V,(5) = Ml Y,,(5) ~1W

Becauseof the invertibility of Y,,(c), we can make VI([) be any memberof H
by appropriately selecting a value for II(c). Hence, we can solve for Jar(<) to
obtain
.Jm = Y,,(5) YllW = ~21(5w11(0* (7)
Clearly, every product in ~Var(~)also appearsin N,,(~). However, since n,
is not a cut-node even when branchesb, and b, are removed, there exists a
path inside !JI’ from n, to 1zanot containing n2 . Consequently, there is at least
one product in N,,(c) that does not appear in N.,,(c). We may therefore
divide the numerator and denominator of (7) by IV,,([) to obtain (6) where
B(5) = K(5) [NdW and W) is the sum of those impedance products
appearing in N,,(c) but not in N,r(c).
Next, we divide the numerator and denominator of B(l) by the product
of the impedances of the branches b, , b, ,..., b, . This yields B(c) =
P(5) [Q(5)]-‘, where both f’(5) and O(5) are sums of admittance products,
each product being the product of (b - 2) - (b - K - 3) = K + 1 distinct
branch admittances. Thus, B ~Q(2k + 2). Finally, a repetition of the last
paragraph in the proof of Theorem 1 shows that 11JzI({)j/ < 1 for
[ E C(2k + 2). This endsthe proof.
A ladder network like that of Fig. 5, where now the element nearest the
input is a shunting capacitor and the element nearestthe output is a series
inductor, shows that the integer 2K + 2 in the conclusion of Theorem 2
cannot be decreased.
404 A. H. ZEMANIAN

4. TWO-TERMINAL-PAIR NETWORKS

In this section we show that the values of the voltage and current transfer
ratios of two-terminal-pair networks are also strictly contractive operators
under certain conditions.
In the following I/ij([) will denote the voltage drop from node ni to node nj .
The voltage transfer ratio T,, of the two-terminal-pair network of Fig. 2
is defined by V,,(c) = T,,(c) Vr,(<) under the requirement that the output
current Ia entering the network through node na is zero.

THEOREM 3. Let the two-terminal-pair network % of Fig. 2 satisfy con-

ditions A. dssume that, with no external connections to 9l, either n, or n3 (but
not necessarily both) is not a cut-zode. Then, 11Tzl(<)jl < 1 for every
5 E C(2k + 4).

Proof. As usual, we first replace 91 by its surrogate a’. We may write

Vz,(Q - V&l) -- I;a(& where T7a,([) = T(t) I’,,(~) and b;a(<) =
S(5) 1Gi1&-). Thus, T&‘) = T(i) - S(i). I-I ere, T (or S) is the voltage transfer
ratio of 91’ acting as a three-terminal network with n4 (respectively, n,) taken
as an internal node. This three-terminal network has k + 1 internal nodes
and, if n, is not a cut-node, satisfied the hypothesis of Theorem 1. In this
case, T (or S) is a member of Q(2k + 4). Moreover, for each [ E C(2k + 4),
WIT(&l] (respectively, IVIS([)-l]) is contained in the acute cone

Co = {A E C: j arg(h - 1)1 ,< (2k + 4) 1 arg 5 11

and is bounded away from the point h = 1.
Similarly, we may write V;,,(t) = V,,(c) - V4r({), Va,(t) = -S(5) P’,,(t;),
V4r(c) = -T(c) Vr,([), and Tzl(<) = T(t) - S(t), where once again T and
S are the voltage transfer ratios of three-terminal networks having k + 1
internal nodes. If na is not a cut-node, we can conclude again that W[T(&l]
and W[5‘(&1] are bounded as stated above.
So, now assume that either n, or ns is not a cut-node. Let (1(f) denote the
spectrum of an operator f, as before. Given any 5 E C(2k + 4), A[T(iJ-‘1 is
closed and contained in the closure of WIT(lJ-l] [6, p, 1111. By the spectral
mapping Theorem [6, p. 2251, fl[T([)] .IS a compact subset of the interior D
of the circle centered at X = h and of radius 4. Since T E Q(2k + 4), T(c)
is normal, and therefore the closure of W[T({)] is the convex hull of A[T(<)]
[6, p. 1121. Hence, M’[T({)] is contained in a compact subset of D.
By the same argument, lV[S({)] is also contained in a compact subset of D.
But, WP”,&)l = WFWI - ~+‘[%-)I,and therefore W[T,,({)] is contained
in a compact subset of the open unit disc. But, T,,(1) is normal because both
T(c) and S(c) are normal and commute [5]. Consequently, II T,,(c)/1 =
sup W[T,,(QJ < 1, which is what we wished to prove.
 VOLTAGE AND CURRENT TRANSFER RATIOS 405

To define the current transfer ratio ]sr([) of the two-terminal-pair network

% of Figure 2, we connect an external current source from node nsto node n,
and an external short-circuit between nodes n2 and n4 . With II(<) and Is(c)
denoting respectively the current entering 111through the external nodes n,
and n2, we set 1,(c) = --J,,(iJ 1,(l) under the condition V,,(c) = 0 and
thereby defined Jzl(Q.
After replacingCJtby %’ but maintaining the external short circuit between
n2 and n4, we can imposean equivalent external excitation by removing the
input current source1r([) and then connecting two external current sources
equal to II(<) in either of two ways: (i) One current source feeds1r([) from
ns to n4and the other feedsII({) from n3to n, . (ii) One current source feeds
1r(c) from ns to na and the other feeds1r(<) from nz to n, . Under the connec-
tions (i), Jsr([) can be shown to be the difference of two current transfer
ratios of three-terminal networks having k + 1 internal nodesand satisfying
the hypothesis of Theorem 2, so long asna is not a cut-node when !R hasno
external connections. The sameresult is true under connections (ii) escept
that now tz4must not be a cut-node when there are no external connections
on %. The proof of Theorem 3 can now be applied to this difference of two
current transfer ratios to establishthe following.

THEOREM 4. Let the two-terminal-pair network ofFig. 2 satisfy conditions A.

Assume that, with no external connections to 9l, either n2 OYn3 (but not necessarily
both) is not a cut-node. Then, 11Je,(Qj < 1 for every 5 E C(2k + 4).
A Jinal note. If someof the external nodes are cut-nodes or if there exist
paths of short circuits inside ‘% connecting the external nodes, then it is
possibleT,,(c) and J,r(LJ to have norms equal to but not larger than one.

ACKNOWLEDGMENT

The author is indebted to Ron Douglas for pointing out a simpler proof of Lemma 2.

REFERENCES

1. RI. BEHZAD AND G. CHARTRAND, “Introduction to the Theory of Graphs,” .411yn

and Bacon, Boston, 1971.
2. V. DOLEZAL, Hilbert networks I, SIAM I. Control 12 (1974), 755-778.
3. V. DOLE~AL AND A. H. ZEMANIAN, Hilbert networks. II. “Some qualitative
properties,” SIAM 1. Control 13 (1975), 153-161.
4. V. DOLEZAL, “Generalized Hilbert networks,” to appear.
5. B. FUGLEDE, “A commutativity theorem for normal operators,” Proc. Nat. Acad.
Sri. U.S.A. 36 (1950), 35-40.
6. P. R. HALMOS, “A Hilbert Space Problem Book,” Van Nostrand, Princeton,
New Jersey, 1967.
406 A. H. ZEMANIAN

7. W. MAYEDA, “Graph Theory,” Wiley-Interscience, New York, 1972.

8. W. RUDIN, “Real and Complex Analysis,” McGraw-Hill, New York, 1966.
9. G. F. SIMMONS, “Topology and Modern Analysis,” McGraw-Hill, New York,
1963.
10. L. WEINBERG, “Network Analysis and Synthesis,” McGraw-Hill, New York, 1962.
11. A. H. ZEMANIAN, Passive operator networks, IEEE Trans. Circuits and Systems
CAS-21 (1974), 184-193.
12. -4. H. ZEMANIIZN, Infinite networks of positive operators, Circuit Theovy and
Applications 2 (1974), 69-78.
13. A. H. ZEMANIAN, “Continued fractions of operator-valued analytic functions,”
J. Approx. Theory 11 (1974), 3 19-326.
14. A. H. ZEMANIAN, Infinite electrical networks, PYOC IEEE 64 (1976), 6-l 7.