INTERNATIONAL JOURNAL OF CIRCUIT THEORY AND APPLICATIONS, VOL.

25, 141—145 (1997)

SOME APPLICATION NOTES ON REDUCTION OPERATORS

T. TUMA
Faculty of Electrical Engineering, University of Ljubljana, Trz\ as\ ka 25, 1000 Ljubljana, Slovenia

(No. of Figures: 1; No. of Tables: 0; No. of Refs: 4)

1. INTRODUCTION
In earlier publications1,2 the theoretical background of reduction operators has been thoroughly discussed.
This letter presents an interesting case study of the reduction operator approach to circuit equations. Based
on an extremely simple example circuit, the relation between the nodal equations and the state variable
equations is examined.

2. REVIEW OF REDUCTION OPERATOR

Let R be the reduction operator of rank n. Its formal definitions including all properties and proofs can be
n
found in References 1 and 2. For convenience though, we will repeat its most essential definition:

C D C D C D
I] 0 A ]n A A ] A
R (A)¢ n n 11n 12 " 11n n 12 ¢A3 (1)
n !A A~1 I ] A A 0 A3 U
21 11 r r 21 22 r]c ]
22 r c
where

A3 ¢A !A A~1A (2)
22 22 21 11 12
As can be seen from (1), the reduction operator of rank n reduces all elements in the (r!n)]n
southwestern corner of A to zero, transforming it to an upper block triangular form A3 .3 A may be any
U
matrix with more than n rows and columns provided that there is a non-singular square submatrix A of
11
order n in its northwestern corner. The application of the reduction operator R actually implies the inversion
n
of the northwestern portion in A. Complementary to the reduction operator, we further define the transposed
reduction operator

AC DC DB
I 0 AT AT
11 21 T
RT (A)¢(R (AT))T"
n n !AT AT~1 I AT AT
12 11 12 22

C DC D C D
A A I !A~1A A 0
" 11
12 11 12 " 11 ¢A3 (3)
A A 0 I A A3 L
21 22 21 22
where A3 is identically equal to (2). As expected, the transposed reduction operator transforms the matrix
22
A into a lower block triangular form A3 .3
L

CCC 0098—9886/97/020141—05$17.50 Received 12 July 1995

( 1997 by John Wiley & Sons, Ltd. Revised 27 February 1996
142 LETTERS TO THE EDITOR

R and its close relative RT are then used together with three general permutation matrices P, Q and R to
m n
constitute a general operator equation mapping any matrix equation
AB"0 (4)
into
R (PAQT)RT (QBRT )¢R (A3 )RT (Bª )¢A3ª B3ª "0 (5)
m n m n
It has been established1,2 that the repeated application of this operator equation suffices to describe entire
circuit analysis approaches. For instance, a series of reduction operators can be used to execute a Gaussian
forward elimination, a back substitution, a triangular decomposition or a Gauss—Jordan elimination. Also,
a circuit’s nodal, cutset or loop equation set may formulated solley by using reduction operators and
permutation matrices.

3. APPLICATION EXAMPLE
Consider the extremely simple two-node (n"2) circuit in Figure 1. It is composed of four branches (b"4),
namely one capacitor branch (c"1), three resistive branches (r"3) and no inductive branch (l"0).
This circuit requires a basic tableau (6) of 10th order.4 Since we intend to apply reduction operators to the
tableau, we use a rather unusual notation (6). The equation right-hand side has been appended to the
coefficient matrix on the left-hand side, leaving a zero matrix behind. The notation also necessitates an
additional !1 constant element in the variable vector. The blank areas in (6) correspond to zero matrices
and should be considered full of zero coefficients.
The tableau in (6) is now suitable for reduction operators since it has the required form (4). The matrix in
(6) containing derivative operators actually makes (6) a linear differential equation set rather than a linear
one. In order to produce the correct nodal equations, we must replace all reactances in the circuit by
adequate linear time-invariant models. In our case the capacitor branch equation (d/dt)v !i "0
b3 b3
becomes g v !i "i . The current excitation i (t) should also be considered time-independent i .
3 b3 b3 3 1 1
1 0 !1 0 0 0 v
n1
1 0 0 0 0 v
!1 n2
v
1 !1 0 0 !1 0 b1
v
0 1 0 0 0 !1 b2
v
1 1 1 0 b3
v "0 (6)
0 0 !1 1 b4
ib
1
0 0 0 0 1 0 0 0 i (t) ib
1 2
0 !1 0 0 0 r 0 0 0 ib
2 3
0 0 d/dt 0 0 0 0 0 ib
!1/c
3 4
0 0 0 !1 0 0 0 r 0 !1
4

Figure 1. Sample circuit

( 1997 by John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl., Vol. 25, 141—145 (1997)
 LETTERS TO THE EDITOR 143

At this stage the entire nodal analysis approach including Gaussian elimination and extended back
substitution can be performed by a simple series of reduction operators in combination with some
permutation matrices:

AA AAA BBB BB A B
v
AT !I 0 0 n
v
R R P R R R P 0 0 A 0 QT QT Q Q b "0 (7)
2 1 2 9 8 4 1 1 2 2 1 i
0 Y Z e b
!1
hgggiggggj
tableau (6)
hgggigggj
nodal equations (9)
hgggggigggggj
gaussian elimination (10)
hgggggggigggggggj
extended back substitution

Let us discuss (7) gradually. The first step consists of transforming the basic circuit tableau to a nodal
equation set. This is done by the permutation matrices

0 I 0 0
I 0 0 4]4
4]4 0 0 I ] 0
P " 0 0 I ] , Q " 4 4 (8)
1 4 4 1 I 0 0 0
0 I ] 0 2]2
2 2 0 0 0 1

and the reduction operators R and R . The result can be seen in

4 8

!1 0 0 0 1 0 v
b1
0 0 0 1 0 v
!1 b2
v
0 0 !1 0 1 !1 b3
v
0 0 0 !1 0 1 b4
i
1 0 0 0 0 0 i b1
1 i "0 (9)
0 r 0 0 !1 0 0 b2
2 i
b3
0 0 !1 0 g !g i i
3 3 3 b4
0 0 0 r 0 !1 0 v
4 n1
1/r #g !g i !i v
2 3 3 3 1 n2
!g 1/r #g !i !1
3 4 3 3
The second-order nodal equation set is only a small part of (9), namely the last two rows. The triangular
equation set above it relates the independent nodal voltages v and v to all the other circuit variables.
n1 n2
In our second-order equation set the Gaussian elimination’s forward course consists of a single reduction
operator, i.e. R in (7). In this way only the last row of (9) is affected:
9
F F F
F
2 1/r #g i !i v
2 3
!g
3 3 1 n1 "0 (10)
v
1#r g r g2 r g (i !i ) n2
2 0 4 3! 2 3 2 3 3 1 !i !1
r 1#r g 1#r g 3
4 2 3 2 3

A second-order back substitution should produce the circuit nodal voltages from the last two rows in (10).
Considering the upper triangular matrix portion above the nodal equations, however, we can run an
extended 10th-order back substitution assembling the values of all circuit variables.

( 1997 by John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl., Vol. 25, 141—145 (1997)
144 LETTERS TO THE EDITOR

Using reduction operators,2 any back substitution is formally executed as a forward substitution of the
reversed matrix order. The reversing matrices in (7) are antidiagaonl unit matrices P and Q . The
2 2
substitution itself needs only the two reduction operators R and R in order to arrive at the final diagonal
1 2
form from which the solution vector is obtained by division.
Let us now go back to the circuit tableau (6) and use the state variable approach. This time the derivation
operators are kept in the equation set rather than being replaced with time-invariant models. The reduction
operator expression for state variable analysis yields a remarkable resemblance between the nodal equations
(7) and the state equations:

AA AAA BBB BBA B

v
AT !I 0 0 n
v
R R P R R R P 0 0 A 0 QT QT Q Q b "0 (11)
4 1 2 9 6 3 1 1 2 2 1 i
0 Y(t) Z(t) e(t) b
!1
hggggigggggj
Tableau (6)
hgggggggggiggggggggggj
state equations (13)
hggggggggggggiggggggggggggggj
output equations

In order to clarify (11), we will explicitly discuss its three stages. First, a permutation is needed to determine
the circuit’s order of complexity and to select the state variables. This is done by heuristic methods and leads
to the permutation matrices
0 0 1 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0
P " 0 0 0 0 0 0 0 1 0 0 , Q " 0
1 0 0 0 0 0 0 0 0 1 0 (12)
1
0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1

Secondly, three reduction operators R , R and R are applied in (11). The permuted equation set is
3 6 9
transformed to an upper block triangular form with the state equations in the last block:

!1 0 0 1 0 0 0 v
b1
0 0 1 0 0 0 v
!1 b2
v
0 0 !1 0 1 0 0 b4
i
1 0 0 0 0 0 i (t) b1
1 i
0 r 0 !1 0 0 0 b2
2 i
b4
"0
0 0 r 0 !1 0 0 v
4 1n
1/r 0 1 0 !i (t) v
2 1 n2
0 1/r !1 0 0 i
4 b3
1 !1 0 0 v
!1 b3
d/dt!( t(t) !1

(13)

( 1997 by John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl., Vol. 25, 141—145 (1997)
 LETTERS TO THE EDITOR 145

where
1/r 0 1 ~1 0 !i (t)

C D C D
1 2 1 1 !i (t)r
[!( t(t)]"! 0 0 ! 0 1/r !1 0 0 " 1 2
c 4 c (r #r ) c (r #r )
3 1 !1 0 !1 0 3 2 4 3 2 4
(14)
As expected, we find in this simple case only one state equation involving the branch voltage across our
only capacitor v . The modified nodal equations of the circuit’s resistive portion can be easily identified
b3
directly above the state equation. In order to compute the state equation coefficients ( and t, the inversion
of this nodal equation set is compulsory (equation (14)). The final first-order differential equation in the last
row of (13) hence reads
d !1 !i (t)r
v " v # 1 2 (15)
dt b3 c (r #r ) b3 c (r #r )
3 2 4 3 2 4
It is not until now that the circuit’s time dependence is resolved by an integration algorithm. Once the
circuit transient response of the state variables is determined, the output equations above the state equation
in (13) are employed to get the values of all the other circuit variables. This is done by a simple blockwise
back substitution similar to the one in (7). Again an antidiagonal permutation P , Q is used to reverse the
2 2
equation order. Then the last two reduction operators R and R from (11) take effect and yield the values of
1 4
the output variables.
At this point, perhaps some application notes are necessary in order to explain the inclusion of time
derivative operators (6) in a computer data structure. Today most programming tools are equipped with
complex arithmetics, which is indispensable for frequency domain analysis anyway. By using complex
arithmetics, the time derivative operators in (6) can be accommodated as unit imaginary parts of coefficients.
All subsequent operations are simply performed in a complex domain.

REFERENCES
1. T. Tuma and F. Bratkovic8 , ‘A mathematical model for network analysis methods’’, Proc. Eur. Conf. on Circuit ¹heory and
Design—93, Davos, 1993, pp. 397—402.
2. T. Tuma and F. Bratkovic8 , ‘A general approach to circuit equations’, Int. j. cir. theor. appl. 22, 431—445 (1994).
3. R. P. Tewarson, Sparse Matrices, Academic, New York, 1973.
4. L. O. Chua and P. M. Lin, Computer-Aided Analysis of Electronic Circuits, Prentice-Hall, Englewood Cliffs, NJ, 1975.

( 1997 by John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl., Vol. 25, 141—145 (1997)