IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO.

3, MARCH 1997 221

Nonstandard Electrical Networks

and the Resurrection of
Kirchhoff’s Laws
Armen H. Zemanian, Life Fellow, IEEE

Abstract— Kirchhoff’s laws fail to hold in general for infinite

electrical networks. Standard calculus is simply incapable of
resolving this paradox because it cannot provide the infinitesimals
and more generally the hyperreal currents and voltages that such
networks often require. However, nonstandard analysis can do
precisely this. The idea of a nonstandard electrical network is in-
troduced in this paper and is used to reestablish Kirchhoff’s laws
for a fairly broad class of infinite electrical networks. The second
section herein presents a fairly brief tutorial on infinitesimals,
hyperreal numbers, and the key ideas of nonstandard analysis
needed for a comprehension of this paper.
Index Terms—Hyperreal voltages and currents, infinitesimals,
resistors fed by a source branch
resistor. The
infinite networks, nonstandard analysis, transfinite networks.
I. INTRODUCTION
I T IS FAIRLY WELL-KNOWN now that Kirchhoff’s laws
need not always hold in infinite networks (see [15, Sections
1.6 and 3.4]). Perhaps the simplest example of this is provided
by Fig. 1, wherein an infinite parallel circuit of 1 resistors is
fed by a 1 V voltage source in series with another 1 resistor.
The infinite parallel circuit must be a short ( ), and thus
the voltage across it is V. Hence, the current in each
resistor ( 1, 2, 3, ) is 0 A, and therefore
A, too. On the other hand, since , we have that
A. We conclude that 1 A flows toward node but 0 A flows
away from it—in violation of Kirchhoff’s current law.
How to explain this discrepancy would have been no prob-
lem two hundred years ago, for a conventional mathematical
argument at that time (had electrical circuits been extant then)
would assert that the current in each parallel resistor is
not exactly 0 A but is instead an infinitesimal and that the
sum of those infinitesimals is 1 A. Though intuitive
and useful, infinitesimals were not legitimate—at least by later
standards of mathematical rigor; in fact, they simply served
as a “fudge factor,” allowing one to achieve apparently valid
results. During the nineteenth century, infinitesimals were
expunged from rigorous mathematics and replaced by limit
processes based upon cumbersome epsilon-delta arguments.
But, the facility of infinitesimals insured their survival at least
for informal arguments. Not surprisingly, they were eventually
Manuscript received March 13, 1995; revised March 27, 1996. This work
was supported by the National Science Foundation under Grant DMS-9200738
and MIP-9423732. This paper was recommended by Associate Editor S. Mori.
The author is with the Department of Electrical Engineering, The State
University of New York, Stony Brook, NY 11794 USA.
Publisher Item Identifier S 1057-7122(97)01506-7.
1057–7122/97$10.00  1997 IEEE
Fig. 1. An infinite parallel circuit of 1
consisting of a 1 V voltage source in series with another 1
input resistance of the infinite parallel circuit is r .
reinvented, this time rigorously by Robinson [10], [11] about
35 years ago. Not only did this justify at last techniques
that were so useful in the early development of calculus and
differential equations during the 17th and 18th centuries, it also
established an alternative method, called nonstandard analysis,
for generating new mathematics, both pure an applied. As
the literature survey of Lindstrom [9, pp. 90–98] indicates,
nonstandard analysis has been effectively used in a variety of
mathematical subjects. Furthermore, it has appeared in at least
one place within mathematical circuit theory in providing an
explanation of certain jump behavior in the chaotic dynamics
of nonlinear circuits [5].
The objective of this paper is to introduce into mathematical
circuit theory the idea of a “nonstandard electrical network.”
Not only conventionally infinite but also many transfinite
electrical networks will have nonstandard representations. (A
reason for considering transfinite networks is explained in
[16].) One consequence of our introduction of nonstandard
networks will be the reestablishment of Kirchhoff’s laws
as inviolate rules for certain infinite networks. The kind of
nonstandard network we shall consider is a finite intercon-
nection of one-ports having nonstandard parameters. We shall
show that certain one-ports, which are internally infinite, have
nonstandard descriptions. The network of Fig. 1 is of this sort,
the infinite parallel circuit being a one-port with a nonstandard
input resistance.
In Section II, we provide some information about nonstan-
dard analysis; this may serve as a brief tutorial on that subject.
In Section III, we lift Kirchhoff’s laws and Ohm’s law into the
nonstandard realm and thereby obtain nonstandard versions
of nodal analysis. This discussion also indicates that much

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
222 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO. 3, MARCH 1997

more of conventional network analysis, such as fundamental- This idea can be reworked to define infinitesimals as certain
loop analysis and Thevenin’s and Norton’s theorems, can equivalence classes of sequences of real numbers that approx-
be extended in a nonstandard way. Certain one-ports having imate 0 in a certain way. More generally, all real sequences
infinite internal circuits, such as infinite parallel circuits and can be partitioned into equivalence classes, called hyperreal
infinite series circuits, are introduced in Section IV and are numbers, and they include not only infinitesimals but also
shown to have nonstandard descriptions. These examples are numbers infinitesimally close to any real number, as well as
generalized in Section V to obtain three broad classes of infinitely large numbers.
internally infinite one-ports that not only have nonstandard To be more specific, let 1, 2, 3, be the set of all
port-terminal descriptions but also satisfy Kirchhoff’s laws at positive natural numbers , and let denote
their internal infinite nodes and infinite loops in a nonstandard a sequence of real numbers. Two (not necessarily convergent)
way. Upon combining the results of Sections III–V, we finally sequences and will be called “equivalent” if they
obtain in Section VI a nonstandard theory for a broad class are the same on a “large enough” subset of . In order
of infinite networks in which Kirchhoff’s laws are always to specify which subsets are “large enough,” we choose a
satisfied and determine voltage–current regimes. This final measure for all the subsets of as follows:
conclusion is stated succinctly in the third paragraph of that Conditions 2.1:
section. i) For each , either or .
ii) .
II. SOME ELEMENTS OF NONSTANDARD ANALYSIS iii) If is a finite subset of , then .
iv) is finitely additive; that is, if and are disjoint
Let us explain some of the elements of nonstandard analysis, subsets, then .
enough to provide an understanding of the nonstandard circuit
The “large enough” subsets are those having measure
theory we introduce below. This section presents but a bare
1. These four conditions imply the following results: If
minimum of ideas. More detailed introductions to nonstandard
[or ], then, for the complement
analysis can be found in a number of sources. Davis and
of , [respectively, ].
Hersh [2] provide a historical perspective along with a short
If is a finite partition of , (that is, if
description of some key ideas. Henle and Kleinberg [3] and
these finitely many subsets are pairwise disjoint and if their
Keisler [6] give elementary expositions. Introductions at a
union is ), then exactly one of the has measure 1 and all
somewhat higher level are those of Keisler [7] and Lindstrom
the others have measure 0. If and ,
[9]. Far more is presented in the books by Davis [1] and Hurd
and Loeb [4], but it may be advisable for the novice to read the then . If and if , then
aforementioned introductions before attempting these books. . A proof that such a measure can be assigned
Robinson’s original book [11] is at an advanced level and to the subsets of is given in [9, pp. 84–85]. In fact, there
requires a knowledge of symbolic logic. are many such measures.
There are two approaches to nonstandard analysis, the Having chosen a particular measure of this sort, we
axiomatic and the constructive. Actually, each subsumes the can partition the set of all sequences of real numbers into
other, the only difference being the order in which ideas are equivalence classes by taking two such sequences as being
introduced. The axiomatic approach is based upon symbolic equivalent if the subset of on which they agree has measure
logic and is the method used by Robinson in his seminal 1; in symbols, and are considered equivalent if
works [10], [11]. As a result, the earlier expositions were of . An equivalence class will be denoted
this kind but did not provide much intuition as to the nature by or by , where is any one
of infinitesimals. The constructive approach is probably more of the sequences in that class. Any such sequence is
accessible to those accustomed to mathematical analysis rather called a representative of that class. These equivalence classes
than mathematical logic, in particular, to circuit theorists. represent new entities, called hyperreal numbers or simply
So how can one construct an infinitesimal? Well, how are hyperreals. (Similarly, we often say just “real” in place of “real
the real numbers constructed? Answering the latter question number.”) We let denote the set of reals and the set of
may give us a clue about the former one. The constructions of hyperreals. can be viewed as an extension of as follows.
the rational numbers as ratios of integers was accomplished by If , then the equivalence class
the ancients, but it was disconcerting when the Pythagoreans is the image of in ; is also called the hyperreal image
discovered that the diagonal of the unit square is not such of . On the other hand, the hyperreal number 1, 1/2, 1/3,
a ratio [8, pp. 104–105]. This situation remained unresolved 1/4, is not the image of any real number but is in fact an
from ancient times to the late 19th century, when at last Cantor “infinitesimal” (see below). Also, 1, 2, 3, is an “infinite
and Dedekind [8, p. 179] proposed different but equivalent hyperreal” (in particular, an “infinite integer” because all its
definitions for the real numbers. The definition most pertinent entries are integers). Furthermore, 1, 1/2, 1/3, 1/4, 1/5, 1/6,
for us is that whereby a real number is defined as an equivalent and 1, 0, 1/3, 0, 1/5, 0, denote the same hyperreal
class of Cauchy sequences of rational numbers, two Cauchy if the measure we have chosen assigns 1 to the set of odd
sequences being considered equivalent if their terms approach positive natural numbers.
each other. In this way, the real numbers expand and fill out We shall use the following symbolism. will always be an
the set of rational numbers. index varying through . Thus, is a

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
ZEMANIAN: NONSTANDARD ELECTRICAL NETWORKS 223

sequence, and is the hyperreal having the resurrection of Kirchhoff’s laws will be accomplished
as one of its representatives. A constant will have whatever be the choice of the measure .
as its hyperreal image (the symbol Another peculiarity arises from the fact (pointed out before)
is also used for ). Moreover, a hyperreal symbol, such as that, for any finite partition of , any chosen measure
, involving a subscript other than will always denote assigns 1 to exactly one of the subsets in the partition and 0
the hyperreal image of a constant ; thus, means to all the others. Consider for example the nested subsets of
, not . Also, real numbers of the form 1, 2, 3, , where
will be denoted by lower-case Roman letters, and hyperreals by is an even positive natural number; thus, if , then
upper-case Roman letters. We will usually use corresponding . We can choose a measure such that
lower-case and upper-case letters, as for example for all . Also, for any fixed , let for , and
or . If the let for . Because of our choice of ,
hyperreal carries an index, that index will never be ; thus,
we may write “ ” but never “ .”
The hyperreals are ordered as follows. If and
and if , then . Thus,
the real numbers and their images in have the same order,
and that order extends to all of . In this way, we have
positive hyperreals and negative hyperreals .
This indicates that a sequence of progressively sparser sam-
Moreover, we define the arithmetic operations componentwise
plings of a given hyperreal identifies that hyperreal; in
using any representatives for the hyperreals. In particular,
fact, the can be changed outside any particular subset of
, , ,
measure 1 without altering that identification. Actually, for
and if in the last case. The usual
a fixed measure , it is the asymptotic behavior of
arithmetic laws hold in ; in fact, is an ordered field with
that determines , and that asymptotic behavior can be
the zero element and the unit element . Therefore, we
determined by a sampling within an arbitrarily sparse subset
can manipulate the hyperreals as we do the real numbers. The
of of measure 1.
absolute value of a hyperreal is if , is
It was noted above that, with the arithmetic operations
if , and is if .
transferred into by means of componentwise definitions,
A hyperreal is called infinitesimal (respectively, infinite)
is a field in quite the same way as is a field. Conse-
if (respectively, ) for every positive
quently, rational functions of hyperreals and equations between
real number . A hyperreal that is not infinite is called finite.
such rational functions can be set up in the hyperreal realm.
Two hyperreals and are said to be infinitesimally close
More specifically, consider two rational functions that in-
if is infinitesimal, in which case we write .
volve some real coefficients and some real variables ,
Altogether, the set of all hyperreals is called the hyperreal line,
which we may display together by means of a vector
and that line consists of the infinitesimals (those hyperreals
. Then, the equation between the
infinitesimally close to ), the finite hyperreals (each of
rational functions will be true in the real realm when and
which is infinitesimally close to some image of a real number),
only when lies in some subset of the Cartesian product
and the infinite hyperreals. In fact, for each hyperreal (finite
of taken times. (Of course, that subset may
or infinite), there is a set of infinitely close hyperreals called
be void.) It is an important property of nonstandard analysis
the monad for , and there is a larger set of finitely close
that equation will again be true in the subset of
hyperreals (i.e., is finite) called the galaxy for
consisting of all such that . is
. Furthermore, each finite hyperreal has a standard part
called an “internal set” in , but there are more general
st such that st . There is only one such real
kinds of “internal sets” in [7, p. 43], [9, pp. 10 and
st having an image in to which the finite hyperreal
24]. All this is a result of the so-called “transfer principle” of
is infinitesimally close.
nonstandard analysis [7, pp. 34–35], [9, p. 77].
Let us at this point discuss the arbitrariness arising from
As a simple example, consider in the real realm.
the choice among many possibilities of the measure . For
This equation is true whenever , , and
example, consider the hyperreal
. Thus, the set mentioned above is given by
(1)

One of the measures we may choose will assign 1 to the Consequently, any of the form
set of even positive integers, in which case (1) will be equal , where , ,
to the hyperreal . On the other hand, another permissible and for all in a subset of of measure 1,
measure will assign 1 to the set of odd positive integers, in will be a member of . We can succinctly restate this by
which case (1) will be equal to the hyperreal , and setting , , and and asserting
will now be equal to 0, 1, 0, 1, . We can view this that has the solution when
arbitrariness as simply different ways of arriving at the same and . Here, and need not
hyperreal line. Moreover, the final result of this paper, namely, be the images of real numbers.

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
224 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO. 3, MARCH 1997

This is a critically important result, for it allows us to find This result may be either infinitesimal, finite, or infinite. For
solutions in the hyperreal realm that do not have preimages in instance
the real realm. For instance, Kirchhoff’s current law does not
hold for the network of Fig. 1 because the total resistance of
the infinite parallel circuit cannot be obtained through finitely
many arithmetic operations. Actually, the result can be
obtained in by passing to a limit as the number of parallel
resistances increases toward infinity. But, that passage to a
limit through an infinite sequence of parallel combinations is
an operation falling outside of the theory of finite networks. On
the other hand, we shall show in the next section that the total
resistance of the infinite parallel circuit can be represented by
an infinitesimal obtained by means of an infinite sequence of
finite parallel combinations. Moreover, the other parameters
of Fig. 1 can be taken as the hyperreal images of constants.
So, by finite circuit theory lifted into the hyperreal realm, we where the last series is an infinite hyperreal. Note however
shall find a hyperreal solution for the current in the source. that an alteration in the order in which the terms of a hyperreal
Furthermore, the current in each branch of the infinite parallel series is taken will in general change the hyperreal value of
circuit will be found to be an infinitesimal, and the sum of all the series.
those infinitesimals will be equal to the hyperreal solution for We will also encounter hyperreal infinite series whose terms
. This latter result will be a consequence of how we define are not images of reals. These too can be summed in if
the hyperreal image of an infinite series. Thus, Kirchhoff’s each has the form , where when
current law will hold in at the nodes and . . Any hyperreal can be put into this form just
But how, one may persist in asking, does nonstandard by changing finitely many of its representative components.
analysis of some other infinite network provide a hyperreal In this case, we have
solution when standard analysis provides none at all? It does so
as follows. The kind of infinite network we shall consider will
be a finite network of one-ports, some or all of which consist
internally of infinite circuits. Each infinite circuit may in turn
be viewed as the end result of a sequence of finite circuits. So,
standard analysis, applied to finite networks of internally finite
one-ports, will provide solutions for the currents entering and (3)
within the one-ports at each stage of the sequence. Thus, upon
considering the sequence all at once, those currents become (The series (2) has just this form with replaced by
sequences of real numbers. The current sequences need not whenever .) Summing the componentwise is the
converge, but they will always be representatives of hyperreal same as summing the rectangular array in (3) columnwise. This
currents. Moreover, sufficiently many components of those works because each column has only finitely many nonzero
representatives will satisfy the standard Kirchhoff current law entries. The result is
to warrant the satisfaction of the nonstandard Kirchhoff current
law. The nonstandard Kirchhoff voltage law will be satisfied in
the same way. The hyperreals that we obtain may depend upon
the ways in which the internally finite one-ports are expanded
into internally infinite ones. Nonetheless, whatever ways are (4)
chosen, the resulting hyperreals will satisfy the nonstandard
Kirchhoff laws. This then is our hyperreal definition of when the
One last matter: Subsequently we will meet infinite series have the stated form. Moreover, this is just the form the
of hyperreals, as for instance when we sum the conductances will have when we build our infinite circuits by inserting
in the infinite parallel circuit of Fig. 1 or the currents in those one branch at a time. Here too, the final result will depend
conductances. We can sum an infinite series of hyperreals just in general on the order in which we insert the branches. Note
as we do a series of reals whenever the hyperreals are images that we are not free to alter arbitrarily the representatives of
of reals. In particular, let for infinitely many of the for this may change the hyperreal
each 1, 2, 3, . Then, define as the hyperreal value of the series. This difficulty will be avoided by choosing
having as a representative the sequence of partial sums of the and fixing a particular order for the insertion of branches.
reals taken in their natural order: Here ends our brief tutorial. Our explanation of the hyper-
reals has been cursory but hopefully informative. We have left
out many essential concepts that one should know if one is to
be knowledgeable about nonstandard analysis. We recommend
(2) [7] or [9] for an introduction to this subject.

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
ZEMANIAN: NONSTANDARD ELECTRICAL NETWORKS 225

one of these nodes may be ground. Let

(8)

be the vector of voltages at all the nodes other than ground.

Then, the relationship between node voltages and branch
voltages can be written succinctly as

(9)
(a) It is a fact that Kirchhoff’s voltage law holds around every
loop of the network if and only if is given by (9) for some
vector of node voltages.
Furthermore, let be the diagonal matrix whose
main-diagonal entries are and whose other entries are
0; possibly, some of the main diagonal entries may be 0, too
(b) (i.e., for some ). By Ohm’s law, the vector of currents
Fig. 2. (a) A standard Norton branch containing both a conductance gj and in the conductances is . With being
an independent current source hj . The orientation of the branch is from left to the vector of branch current sources, the branch current vector
right, and vj ; ij , and hj are measured accordingly with the polarities shown.
uk and ul are node voltages. (b) A standard Thevenin branch containing a
is then
resistance rj and an independent voltage source ej . If 0 < rj <
01
1 , if
rj = g
j , and if ej = 0 rj hj , then this Thevenin branch is equivalent to (10)
the Norton branch so far as terminal conditions are concerned.
So, given the finite network, that is, its graph and all the
values and , a combination of (7), (9), and (10) comprises
III. FINITE NONSTANDARD ELECTRICAL NETWORKS a nodal analysis yielding the following expression for :
Our objective in this section is to show how nonstandard
analysis can be applied to finite electrical networks. In other (11)
words, we wish to lift the theory of such networks from the so long as is nonsingular. A sufficient condi-
real realm into the hyperreal realm. tion for that nonsingularity is that every branch have a positive
For the sake of specificity, let us do so for a nodal analysis conductance: for all . These are all conventional
of a particular kind of network. We start in the real realm with results in the real realm. In summary, if Ohm’s law (10)
a finite connected network of nodes and branches, with is satisfied and if is nonsingular—or, more
each branch assigned an orientation and consisting of either particularly, if every branch has a positive conductance, then
a positive conductance alone, or an independent current Kirchhoff’s laws are satisfied throughout the network if and
source alone, or a parallel combination of and as only if (11) holds.
shown in Fig. 2(a). This is the Norton form of a branch. The Note that the right-hand side of (11) involves no more than
branch index varies from 1 to . We will call this the standard a finite number of arithmetic operations. Since these have
network. With the superscript denoting matrix transpose, we been lifted into the hyperreal realm through componentwise
let the column vectors definitions, we can transfer (11) into that realm under the
caveat of nonsingularity. Indeed, we can now analyze in
(5)
the very same way a finite connected network whose every
and branch is either a hyperreal positive conductance , or
(6) a hyperreal independent current source , or a parallel
combination of these two elements. Such a network will be
represent, respectively, the vector of branch voltages and the called a nonstandard network or synonymously a network
vector of branch currents. Next, choose a ground node and in the hyperreal realm. All voltages and currents will now
index the remaining nodes by . Let be the be hyperreals. So, upon replacing all lower-case symbols for
incidence matrix, wherein (or 1) if branch the source currents, conductances, and voltages by upper-case
is incident away from (respectively, toward) node and symbols, we have the following results. Kirchhoff’s current
if is not incident to . It follows that Kirchhoff’s law becomes
current law holds at every node if and only if
(12)
(7)
where
Next, assign V as the voltage at the ground node
, and choose the other node voltages (13)
arbitrarily—at least for the moment. If branch is incident Kirchhoff’s voltage law is assured whenever
away from node and toward node , then its branch voltage
is related to the corresponding node voltages by ; (14)

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
226 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO. 3, MARCH 1997

for some node voltage vector . Finally, (11) becomes is the hyperreal voltage in the th
branch, and the plus (minus) sign is used if the orientation of
(15) branch agrees (respectively, disagrees) with the orientation
of loop . Finally, the equality in (17) holds componentwise
(We could also replace the 0’s, 1’s, and 1’s in by
for almost all , as indicated above.
’s, ’s, and ’s to get in place of , but
Finally, let us note that not all finite networks of hyperreal
the operations with the merely represent additions and
one-ports are amenable to a nonstandard analysis—as, for
subtractions, which we have already defined on the hyperreals
example, when there is a loop of pure voltage sources that
componentwise; thus, there is no need for this latter change
violate (17), or a node with only pure current sources as
in symbols.)
its incident branches which violate (16), or simply a three-
In order for (15) to be meaningful in the hyperreal realm,
branch network consisting of a parallel circuit of a
we need to confine the determinant of to
resistor, a resistor, and a V voltage source. So,
the internal set . This will certainly be so if is
we have to restrict the kinds of nonstandard networks we
a positive hyperreal for every . Other than this, there is no
consider. With regard to nodal analysis, our assumption about
restriction on the hyperreal elements of and . We can
the nonsingularity of suffices for this purpose.
conclude as follows: If Ohm’s law (13) is satisfied and if
the determinant of is not equal to or,
more particularly, if every branch of the nonstandard network IV. NONSTANDARD ONE-PORTS AND
has a positive hyperreal conductance , then, for every THEIR HYPERREAL REPRESENTATIONS
choice of the hyperreal independent current-source vector , We have seen in the last section that a nonstandard finite
(15) holds, and the hyperreal branch-voltage vector and network of one-ports with hyperreal parameters has a hyperreal
the hyperreal branch-current vector will satisfy Kirchhoff’s voltage-current regime, but we have not looked inside the one-
voltage and current laws throughout the network. Note that, the ports to see if their hyperreal parameters represent any standard
and need not be hyperreal images of real numbers, and electrical circuits, finite or infinite. If the hyperreal parameters
therefore the nonstandard network need not be the hyperreal of a given nonstandard one-port are images of real parameters,
image of a standard network. In particular, any conductance, then the one-port is internally the hyperreal image of a real
voltage, or current in the nonstandard network may be either circuit. For example, if the nonstandard one-port is described
an infinitesimal, finite, or infinite hyperreal. by , where and are hyperreal
We have been considering a nonstandard nodal analysis. images of the reals and , then the nonstandard one-port is
Nevertheless, much more of standard network theory can be just the hyperreal image of the real Norton branch such as that
transferred into the hyperreal realm to get, for example, a of Fig. 2(a). If in addition , then the nonstandard
nonstandard mesh analysis or more generally a nonstandard one-port is also the hyperreal image of the equivalent Thevenin
version of a fundamental-loop analysis [13]. Also, many of branch of Fig. 2(b).
the usual network theorems, such as Thevenin’s and Norton’s We will now examine several cases wherein the parameters
theorems, can be transferred into the hyperreal realm as well. of a nonstandard one-port are hyperreals that need not be
We will not explicitly do so here because the needed arguments images of reals.
are much the same as those used for a nodal analysis. All the
nonstandard finite-network theory we employ in this paper is A. An Infinite Parallel Circuit of Conductances
truly available. Let us assume that the nonstandard one-port is described
Let us now restate Kirchhoff’s laws in terms of hyperreals. by , where , , and are the one-port’s hyperreal
His current law (12) can be rewritten as current, voltage, and input conductance. In particular, let
be any hyperreal. By virtue of the definition
(16)
(2) of an infinite series of hyperreals, is synthesized as
the infinite parallel circuit of hyperreal images of real
where is any node in the nonstandard finite network, the conductances , where and for
summation is over the indexes for the branches incident to . Thus,
, is the hyperreal current in the th (18)
branch, and the plus (minus) sign is used if branch is incident
away from (respectively, toward) node . The equality in This is illustrated in Fig. 3(a). There are no restrictions on
(16) holds componentwise for almost all ; that is, there is the and thus none on the either; they can be positive or
a subset of of measure 1 such that negative and have any order of growth or decay. Consequently,
for all . can be either an infinitesimal, finite, or infinite hyperreal.
As for Kirchhoff’s hyperreal voltage law, we have Since each is the hyperreal image of the real conductance
, we may take the preimage of the hyperreal circuit branch
(17) by branch to obtain the real parallel circuit shown in Fig. 3(b),
but now that real circuit may not have a real input conductance.
If the sequence
where is any oriented loop in the nonstandard finite network,
the summation is over the indexes for the branches in , (19)

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
ZEMANIAN: NONSTANDARD ELECTRICAL NETWORKS 227

and , then , and so forth. That is, we continue choosing

sequentially two odd-indexed conductances and then one even-
indexed conductance, repeating this cycle indefinitely. This
corresponds to reindexing the conductances according to
, , , , , , , . With
this reindexing, becomes

(21)
(a)
No two corresponding subsequences of (20) and (21) will be
identical, and therefore and must represent different
hyperreals. This shows that one cannot rearrange the con-
ductances of an nonstandard infinite parallel circuit without
in general changing its input conductance. Because of this,
(b) it is not enough to specify the set of conductances in a
Fig. 3. (a) The synthesis of an arbitrary G 2 3 R by means of an infinite nonstandard infinite parallel circuit if the hyperreal input
parallel circuit of hyperreal images hak i of real conductances ak . The order conductance is to be uniquely determined; the order in which
of insertion of the conductances is taken to be the same as that of the
indexes k = 1; 2; 3; 1 1 1 . (b) The corresponding infinite parallel circuit of the conductances are introduced into the parallel circuit must
real conductances ak . Now, there may be no real input conductance. also be specified—except in certain special cases, such as the
case where all the conductances have the same value.
converges, then its limit will be the standard part st of [7, Another discrepancy can arise when one tries to simplify
Theorem 8(i), p. 56], and the difference between and the an infinite parallel circuit by combining conductances. For
hyperreal image of st will be the infinitesimal instance, consider again the infinite parallel circuit of 1-
conductances. Its input conductance is 1, 2, 3, .
st However, if we combine pairs of conductances into 2- con-
ductances, the resulting parallel circuit of 2- conductances
will have the input conductance 2, 4, 6, . Clearly,
This will be so whatever be the choice of the measure . and are different hyperreals.
However, if the sequence (19) strictly increases in absolute It is noteworthy however that, if
value unboundedly, then will be infinite whatever be and
therefore will not have a standard part [7, Theorem 9(i), p. 57].
Finally, if the sequence (19) oscillates boundedly, then will
be finite as long as the measure is chosen appropriately [7, if converges absolutely, and if is obtained by
Theorem 9(ii), p. 57], in which case it will have a standard part. combining some of the conductances before inserting them,
A particular case of interest is when all the are positive; then will be an infinitesimal. (To see this, compare the
now (19) either increases to a limit or increases unboundedly, terms of with .)
and is either a positive finite hyperreal or respectively, a All this indicates that there is more subtlety to infinite
positive infinite hyperreal. parallel circuits in the hyperreal realm than there is in the real
Now, let . The hyperreal input resistance of the realm. Actually, it is this subtlety that allows us to resurrect
circuit of Fig. 3(a) is Kirchhoff’s current law. In the real realm that law simply
collapses at times.
So, let us consider the currents in the nonstandard infinite
where . has a nonzero parallel circuit of Fig. 3(a), where the order of the conduc-
standard part when and only when has a nonzero standard tances is taken to be that of its indexes 1, 2, 3, .
part; otherwise, st when is infinite, and is infinite Its input conductance is given by (18), and its input current
when st . is related to its input voltage by Ohm’s law: .
There is an important feature of these infinite parallel By virtue of (3), this can be expanded into
circuits that must be taken into account: The hyperreal
depends in general upon the order in which the conductances
are connected into the parallel circuit. We may alter that order
for finitely many of the without changing , but infinitely
many such changes may affect . For example, let for (22)
all and insert these in accordance with the natural
order of their indexes. Then, Each row in this array is the hyperreal current in the th
becomes hyperreal conductance . Thus, we now have Kirchhoff’s
current law in the form of a nonstandard infinite series
(20)
.
On the other hand, let us instead reorder those conductances We can use this technique to resolve the paradox concerning
by first connecting , then and , then , then Fig. 1. Consider the more general case where the positive

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
228 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO. 3, MARCH 1997

conductances and voltage source are arbitrary reals. Thus, in

the hyperreal realm we have the infinite parallel circuit of
Fig. 3(a) being fed from a source branch with the voltage
source in series with the conductance .
Upon making a Thevenin-to-Norton conversion, that source Fig. 4. The synthesis of any hyperreal Norton branch, for which
branch becomes a current source in parallel
I = VG 0 H , as an infinite parallel circuit of images of reals. Again,
the order of insertion of the elements is taken to be the same as that of the
with . As before, we take it that the conductances k=
indexes 111
1, 2, 3, .
of the parallel circuit are built up in the order of their
indexes. By the current division law for a finite nonstandard
network

Fig. 5. The synthesis of an arbitraryR2 R 3 by means of an infinite series

hx i x
circuit of hyperreal images k of real resistors k . The order of insertion
of resistors is that of the indexesk= 111
1, 2, 3, .

Norton branch with the representation

where

can be synthesized by the infinite parallel circuit of Fig. 4,

whose elements are images of reals; it is understood that the
conductances and the current sources in Fig. 4 are summed
separately. As before, the order in which the elements are
summed cannot be altered in general if those alterations
(In the last expression, there are initial 0’s.) So again, extend over infinitely many elements. In particular, it is also
the nonstandard Kirchhoff’s current law is satisfied. understood that the infinite parallel circuit is built up through
In the special case of Fig. 1 where a sequence of finite parallel circuits by introducing each pair
, we have of elements and a pair at a time in the order of their
subscripts.

C. An Infinite Series Circuit of Resistances

The kind of circuit we now wish to consider is a one-port
consisting of a one-way infinite series circuit of resistances
and
whose infinite extremity is one of the port’s terminals. This is
illustrated in Fig. 5. The theory of transfinite graphs justifies
such a circuit [15, Sections 3.2 and 3.3], [17, Sections 2.1 and
5.2], but under standard analysis Kirchhoff’s voltage law may
fail if the resistances are not appropriately chosen [15, Section
3.4]. On the other hand, any hyperreal resistance
We can interpret all of this by saying that, at every stage of
can be synthesized by an infinite series circuit of images of
the construction of the parallel circuit, Kirchhoff’s current law
real resistances as follows. Let . Set
is satisfied in the real realm, and therefore, for the infinite
and for . Then, by
network, Kirchhoff’s current law is satisfied in the hyperreal
(2), .
realm.
Actually, all of the ideas we have discussed for infinite
parallel circuits of conductances carry over to the present
B. An Infinite Parallel Circuit of Norton Branches case in a dual fashion. For example, in order to specify the
Let us reverse our approach now and start with an arbitrary input resistance uniquely, the infinite series circuit should
nonstandard Norton branch consisting of a current source be viewed as the result of inserting the resistances into a
in parallel with a conductance finite series circuit one at a time, and the order in which this is
. Set and done may affect ; therefore, that order should be specified.
for 2, 3, 4, . Then, . Similarly, combining some of the resistances before insertion
Also, set and for 2, 3, 4, may affect . However, once these specifications are made,
. Then, . It follows that any nonstandard the nonstandard Kirchhoff voltage law will

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
ZEMANIAN: NONSTANDARD ELECTRICAL NETWORKS 229

The reason we have presented this example is that it does

not fit into the class of “finitely constructible” one-ports, which
we will be introducing in the next section. By definition, the
finitely constructible one-ports have internally no more than
countably many branches. Even though the network of Fig. 7
Fig. 6. The synthesis of any hyperreal Thevenin branch, for which is constructed out of a sequence of finite circuits, the final
= 0
V IR E , as an infinite series circuit of images of reals. Elements
are inserted in the order of their indexes k=1, 2, 3, .111 result has uncountably many branches. The difficulty here is
that no single terminating resistor in the final result can be
identified with a particular terminating resistor of an interme-
hold around the infinite series circuit, where the summation is diate finite circuit. Such an identification is essential for the
defined by (3) and an equation dual to that of (22) occurs. finitely constructible one-ports. Note also that the symmetry
in resistance values was essential to our analysis; without such
D. An Infinite Series Circuit of Thevenin Branches symmetry, the infinitesimal currents in the terminating resistors
More generally, given any nonstandard Thevenin branch are difficult (impossible?) to determine.
with the representation , where ,
, , and , we can set F. Other One-Ports
as before and in addition , where and Nonstandard one-ports can be constructed out of other
for . Thus, infinite circuits by using a sequences of finite subcircuits that
“fill out” the infinite circuits. This is done in the next section
in three different ways.

This expression is realized by an infinite series circuit whose V. FINITELY CONSTRUCTIBLE ONE-PORTS
elements are images of reals; this is illustrated in Fig. 6. Here Henceforth, we will always assume that every standard
too, it is understood that the infinite series circuit is built up branch has a positive resistance. Thus, every branch can be
through a sequence of finite series circuits by introducing each converted either into the Norton form of Fig. 2(a) with
pair of elements and a pair at a time in the order or into the Thevenin form of Fig. 2(b) with . This
of their subscripts. prevents the occurrence of a node incident only to current
sources that do not satisfy Kirchhoff’s current law; it also
E. An Infinite Binary Tree with Terminating Resistances prevents the occurrence of a loop containing only voltage
Consider the infinite resistive binary tree connected at sources that do not satisfy Kirchhoff’s voltage law.
infinity through 1-nodes to many resistors of value . We shall now show how some more general kinds of
Fig. 7 illustrates this network. The resistors at the th level internally infinite one-ports can be taken as the final result
of the tree all have the same value . What is the input of a sequence of internally finite one-ports. We have three
resistance between nodes and ? versions of this.
Since there are no restrictions on and the , this question
can only be answered through a nonstandard analysis. To do A. Internally Infinite One-Ports Finitely
so, we must first specify the order in which the network Constructible From Opens
was built up. Let us assume that this was done through a Let be a one-port, which is internally a countably
sequence of finite binary trees of levels ending infinite, 0-connected 0-network (in the terminology of [15]
in many resistances. It follows that the currents in each and [17]); and will be the nodes of that serve
level of every are all the same, and therefore the same as the port terminals. 0-connectedness is the same as the
is true for the infinite binary tree . At each step of this ordinary connectedness of customary graph theory. That
construction, let us drive the network with a constant current is a 0-network means that there are no connections at infinity,
source ; thus, the infinite network will be driven by the that is, there are no nodes of ranks larger than 0 (again in
hyperreal current source . By the symmetry of the infinite the terminology of [15] and [17]). Thus, has no infinite
network, the current in each resistor is and loops. In general, will have infinite nodes (i.e., nodes with
the current in each resistor is . (Remember that infinitely many incident branches). We may say that is
, whereas 2, 4, 8, .) “finitely constructible from opens” because of the following
Now, the sum of the voltages along any path from node construction.
to node is . Thus, To open a branch will mean that, with respect to a Norton
representation [Fig. 2(a)], the branch’s conductance , current
source , and current are all set equal to 0: .
Since , it follows that the branch’s voltage is
indeterminate. We shall arbitrarily set when is opened.
This will not affect the validity of the following arguments.
With regard to the conditions , the same result
can be obtained simply by deleting the branch.

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
230 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO. 3, MARCH 1997

Fig. 7. An infinite binary tree connected at its uncountably many infinite extremities to resistors. At each horizontal level, the resistance values are all
the same. The entire circuit is fed by a current source h.

Let us number the branches of with the indexes For each branch of , we have hereby assigned a se-
. Also, for each , let be the quence of currents and a sequence
one-port obtained from by opening every branch with of voltages and thereby a hyperreal current
but leaving unchanged when . We shall and a hyperreal voltage . Moreover, and
think of as having the same infinite graph as that of whenever . Therefore, we can invoke (3)
but with branch values as stated. Thus, may have one to obtain the hyperreal version of Kirchhoff’s current law at
or more isolated nodes whereby all the branches incident every (finite or infinite) node of :
to have been opened. In fact, the terminal nodes and (23)
may be isolated for some sufficiently small . Moreover,
the finite subnetwork of induced by those branches
that have not been opened at some step may have [The summation symbol and the choices of signs are the same
many components. as that for (16).]
Now let the port terminals and be excited by an In the same way, we also have established the hyperreal
external branch in the Thevenin or Norton form (Fig. 2); version of Kirchhoff’s voltage law around any loop of
:
in this case, we may allow to be a pure voltage source
(with ) but not a pure current source. Whatever be the (24)
source value or and whatever be ,a
standard nodal or fundamental-loop analysis can be applied to [The summation symbol and the choices of signs are the same
to obtain a unique voltage and a unique current as that for (17).] This is because there are only finite loops
for branch for each . For each fixed in since has no nodes of ranks larger than 0 (i.e.,
and for any (finite or infinite) internal node in has no connections at infinity). Thus, for each fixed loop ,
, Kirchhoff’s current law in the form Kirchhoff’s voltage law will be violated only for finitely many
will be satisfied—even at the isolated nodes of . The indexes . The set of those has measure 0, whence (24).
same will be true at the port nodes when the current in Finally, we can apply either the standard Thevenin theorem
is taken into account. On the other hand, with still or the standard Norton theorem to to get the components
fixed, Kirchhoff’s voltage law will be satisfied around those of a hyperreal Thevenin representation [Fig. 2(b)],
loops that remain within (i.e., those loops that pass or a hyperreal Norton representation
only through branches with indexes no larger than )—and at the port terminals and of ; here, and
trivially around those loops that remain outside of . Note that a different numbering of the branches
(i.e., all branch indexes for the loop being larger than ). In of by the indexes will lead in general to
general, Kirchhoff’s voltage law will not be satisfied for loops different hyperreal representations. To be sure, there are special
in having branch indexes both larger and no larger cases, such as the infinite parallel circuit of identical resistors,
than . for which the representation is independent of the numbering.

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
ZEMANIAN: NONSTANDARD ELECTRICAL NETWORKS 231

B. Internally Infinite One-Ports Finitely established the hyperreal form of Kirchhoff’s current law (23)
Constructible From Shorts at every ordinary 0-node of . This is because we
This time let be a one-port, which internally is a are now assuming that all ordinary 0-nodes of are of finite
countably infinite, -connected -network for some natural degree. Thus, for each node , Kirchhoff’s current law can
number and whose ordinary 0-nodes are all of finite degree. be violated at for only finitely many .
Again, and will be the port terminals of . The precise As before, by the standard Thevenin or Norton theorem
applied to , we obtain the components of the hyperreal
definition of a -connected -network is given in [15, ch. 5]
Thevenin representation or equivalently the
and [17, chs. 2 and 5]. Let us merely say here that a -network
hyperreal Norton representation for at
is obtained by first connecting infinite 0-networks together at
its port terminals and . Here again, these representations
their infinite extremities to get a 1-network, then by connecting
may depend upon the order in which the branches of are
infinite 1-networks together at their infinite extremities to get
numbered.
a 2-network, and so forth through the countable ordinals up
(We should also note that each “end” of [12] will be
to . The connections at the infinite extremities are made
shorted in this nonstandard representation. Thus, for example,
through -nodes (i.e., generalized nodes) of ranks up to .
if the one-port is a one-way-infinite ladder with port terminals
That is -connected means that every two branches in
at its two disconnectable 0-tips, that one-port will appear as a
are connected through possibly transfinite paths passing
short—that is, and .)
through the -nodes. In general, will have infinite loops.
That a 0-node is ordinary means that it is not part of a node of
higher rank. We will apply Kirchhoff’s current law only to the C. Internally Infinite One-Ports Finitely
ordinary 0-nodes, and these we assume are all of finite degree. Constructible From Trees
In this case, we may say that is “finitely constructible from
The procedure of Section V-A does not allow the one-port
shorts” again because of the following construction.
to have transfinite loops and the procedure of Section V-B
To short a branch will mean that, with respect to a
prohibits infinite ordinary nodes. But, there is a circumstance
Thevenin representation the branch’s resistance , voltage
when both transfinite loops and infinite ordinary nodes can
source , and voltage are all set equal to 0: .
be allowed in . This is when is countable and has
Since , it follows that the branch’s current is
an “essentially spanning tree.” In particular, we define an
indeterminate. This time we arbitrarily set when is
essentially spanning tree of a transfinite one-port to be a
shorted. This too will not lead to any difficulties.
subnetwork within the one-port that is transfinitely connected,
After numbering the branches of by has no (finite or transfinite) loops, meets the two port-terminal
and fixing , let now be the one-port obtained from nodes, and meets every nonsingleton (finite or transfinite)
by shorting every branch with but leaving node. Presently, it has not been established whether every
unchanged when . Here too, we think of transfinite network has an essentially spanning tree. But, many
as having the same infinite graph as that of but with the of them do. In fact, this is certainly the case for ordinary,
stated branch values. This time will denote the connected locally countable, infinite networks [15, Section 2.4] and for
finite network obtained from by coalescing into a single ordinary infinite networks of any cardinality [14, Theorem
node every maximal set of nodes that are connected by paths 9.1]. Another class of transfinite networks having essentially
of short circuits in , a different coalesced node for each spanning trees are the “finitely structured” ones [17, Section
maximal set. 4.6]; these do not have infinite ordinary nodes, but they can
Let the port terminals and be excited by an external be easily altered to have them. When a one-port has internally
branch in the Thevenin or Norton form (Fig. 2); here, we countably many branches and an essentially spanning tree, we
may allow to be a pure current source but not a pure shall say that it is “finitely constructible from a tree.” Let this
voltage source. Whatever be or and whatever be so for the internally countably infinite one-port , and
be , a standard nodal or fundamental-loop analysis can let and be the port terminals as before. Number the
be applied to to obtain a unique voltage and a branches of by 1, 2, 3, . Then choose an essentially
unique current for the th branch , where . spanning tree for . We call the branches of tree
For each finite or infinite loop in , Kirchhoff’s branches and the branches of chords, as is customary.
voltage law will be satisfied around . However, Kirchhoff’s For each 1, 2, 3, short every tree branch of index
current law may not be satisfied at any node of having and open every chord of index to get the infinite
incident branches with indexes both no larger than and larger network as before. Then, let be the corresponding
than . finite network. We can now append a source branch to
Once again, we have a sequence of currents at and and analyze to get a unique voltage
and a sequence of voltages for each branch and current for each branch with index . Because the
of and thereby a hyperreal current and voltages on the shorted tree branches are all 0, the voltages
a hyperreal voltage . Here too, and on all the opened chords are determined through Kirchhoff’s
whenever . So, we can invoke (3) to voltage law applied to (finite or transfinite) loops in .
get the hyperreal form of Kirchhoff’s voltage law (24) around Similarly, because the currents in the opened chords are all
every (finite or infinite) loop of . Similarly, we have 0, the currents in all the shorted tree branches are determined

Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
232 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 44, NO. 3, MARCH 1997
through Kirchhoff’s current law applied to (finite or infinite)
nodes in . Upon doing this for every , we obtain
hyperreal currents and hyperreal voltages throughout , and
these satisfy nonstandard versions of Kirchhoff’s laws. The
rest of the analysis proceeds as before.
VI. CONCLUSIONS
By combining the results of Sections III–V, we can draw the
following conclusions. Given any standard infinite (perhaps
transfinite) network , whose every branch has a positive
resistance, and having chosen a measure according to
Conditions 2.1, we may be able to partition into finitely
many one-ports and then analyze the interior circuitry of each
one-port by applying Kirchhoff’s laws and Ohm’s law to a
sequence of finite networks arising from opens or shorts in
order to find hyperreal solutions to that circuitry. In particular,
we may obtain a hyperreal Thevenin or Norton representation
for each one-port at its port terminals. Then, a nonstandard
nodal or fundamental-loop analysis applied to the finite circuit
of hyperreal one-ports will yield the hyperreal port voltages
and port currents at all the one-ports of our partition of .
These in turn determine all the hyperreal currents and voltages
within the one-ports, and the latter will satisfy nonstandard
versions of Kirchhoff’s laws at all (finite and infinite) ordinary
nodes and around all (finite and transfinite) loops.
Actually, the separate analyzes of the individual one-ports
followed by the analysis of Section III can be combined into a
single analysis as follows. After having partitioned into one-
ports each of which is internally finite or of a type discussed
in Section V, we can for each make a standard
analysis of the finite network obtained from by opening
and shorting branches in accordance with the types of one-
ports the branches are in. This will yield the th component
of a representative for each hyperreal voltage or current in
, and thereby the hyperreal voltages and currents throughout
. The latter will satisfy nonstandard versions of Kirchhoff’s
laws and Ohm’s law. Furthermore, those laws determine in
this way the hyperreal voltage–current regime for . This is
our conclusion. Let us restate it as follows:
Let be a standard countably infinite network whose
every branch has a positive resistance. Assume that can be
partitioned into a finite number of one-ports such that each
one-port is either internally finite or internally infinite and
finitely constructible either from opens, or from shorts, or from
a tree. (The one-ports need not be all of the same kind.) Then,
for each choice of such a partitioning, for each choice of a
branch numbering within each one-port, for each choice of
the measure , and for each choice of an essentially spanning
tree in the last case, will have a nonstandard image in
which a unique set of hyperreal branch voltages and currents
satisfy nonstandard versions of Kirchhoff’s laws and Ohm’s
law. The nonstandard image of each internally infinite, finitely
constructible one-port is obtained from a sequence of internally
finite one-ports as stated in Section V. Finally, the standard
Kirchhoff’s laws and Ohm’s law determine a unique set of
currents and voltages throughout each finite truncation of
obtained at each stage of the replacements of internally infinite
Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
one-ports by internally finite ones and in this way determine
the hyperreal voltages and currents throughout the nonstandard
image of .
It is perhaps disconcerting that the hyperreal voltage-current
regime depends upon our choices of the partitioning of into
one-ports, of the numbering of the branches within each one-
port, of the measure , and of an essentially spanning tree in
the last case. One might say that an infinite electrical network
is an abstraction that resists a unique comprehension. On the
one hand, standard analysis is simply not delicate enough to
force the satisfaction of Kirchhoff’s laws at all infinite nodes
and around all transfinite loops. On the other hand, it appears
that nonstandard analysis is overly delicate, for it requires
more specifications than is provided by just the graph and
the element values of an infinite network. The constructive
approach to nonstandard analysis does resolve the paradoxes
about Kirchhoff’s laws, at least for certain infinite networks,
but it may also produce many different nonstandard images
of a given infinite network. The circuit theorist is free to
choose among them. (An analog to this is the infinitely many
versions of the delta function in nonstandard analysis [11, pp.
137–138].)
Finally, let us comment on what has not been achieved. The
sentences of symbolic logic are of finite length and therefore
cannot encompass an infinity of terms. As a result, assertions
concerning an infinity of real terms cannot in general be trans-
ferred into the hyperreal realm. However, there are exceptions.
The infinite series (2) and (3) do deal with infinitely many
hyperreals, but they do so through sequences of finite partial
sums [add columnwise in (3)]. This is why we partitioned
into a finite network of internally infinite one-ports and then
dealt with each one-port through a sequence of finite circuits
obtained by opening or shorting branches. However, not all
transfinite networks have been encompassed. How to proceed
when every finite partitioning of the network results in at
least one one-port having both infinite nodes and transfinite
loops but no assurance of an essentially spanning tree remains
an open problem.
Another open problem is the restoration of Kirchhoff’s laws
for uncountable networks like that of Section IV-E but without
symmetry.
REFERENCES
[1] M. Davis, Applied Nonstandard Analysis. New York: Wiley, 1977.
[2] M. Davis and R. Hersh, “Nonstandard analysis,” Scientific Amer., vol.
226, pp. 78–86, June 1972.
[3] J. M. Henle and E. M. Kleinberg, Infinitesimal Calculus. Cambridge,
MA: The MIT Press, 1979.
[4] A. E. Hurd and P. A. Loeb, An Introduction to Nonstandard Real
Analysis. New York: Academic, 1985.
[5] M. Itoh and L. O. Chua, “Canards and chaos in nonlinear systems,”
in Proc. 1992 IEEE Int. Symp. Circuits Sys., San Diego, CA, 1992, pp.
2789–2792.
[6] H. J. Keisler, Elementary Calculus. Boston, MA: Prindle, Weber &
Schmidt, 1976.
[7] , Foundations of Infinitesimal Calculus. Boston, MA: Prindle,
Weber & Schmidt, 1976.
[8] M. Kline, Mathematics. New York: Oxford Univ. Press, 1980.
[9] T. Lindstrom, “An invitation to nonstandard analysis,” in Nonstandard
Analysis and Its Application, N. Cutland, Ed. New York: Cambridge
Univ. Press, 1988.
ZEMANIAN: NONSTANDARD ELECTRICAL NETWORKS
[10] A. Robinson, “Non-standard analysis,” in Proc. Royal Academy of
Sciences. Amsterdam, The Netherlands, 1961, ser. A, vol. 64, pp.
432–440.
[11] , Non-Standard Analysis. Amsterdam, The Netherlands: North-
Holland, 1966.
[12] P. M. Soardi, Potential Theory on Infinite Networks. New York:
Springer-Verlag, 1994, Lecture Notes in Mathematics 1590.
[13] J. L. Synge, “The fundamental theorem of electrical networks,” Quar-
terly of Applied Mathematics, vol. IX, pp. 113–127, 1951.
[14] C. Thomassen, “Infinite graphs,” Graph Theory, vol. 2, pp. 129–160,
1983.
Authorized licensed use limited to: Corporacion Universitaria de la Costa. Downloaded on March 15,2020 at 17:45:02 UTC from IEEE Xplore. Restrictions apply.
233
[15] A. H. Zemanian, Infinite Electrical Networks. New York: Cambridge
Univ. Press, 1991.
[16] , “Why transfiniteness?” Graph Theory Notes of New York. New
York Acad. Sciences, vol. XXVIII, pp. 30–32, 1995.
[17] , Transfiniteness—for Graphs, Electrical Networks, and Random
Walks. Boston, MA: Birkhauser, 1996.
Armen H. Zemanian (M’48–SM’55–F’70–LF’91), for a photograph and
biography, see p. 54 of the January 1997 issue of this TRANSACTIONS.