Numerical continuation scheme for tracing the

double bounded homotopy for analysing nonlinear

circuits
Hkctor V&zquez-Leal,Luis Hernhdez-Martinez, Arturo Samiento-Reyes and Roberto Castaiieda-Sheissa
National Institute for Astrophysics, Optics and Electronics
Electronics Department, CAD Group
P.O. Box 51, 72000, Puebla, Pue., Mexico
E-mail [email protected]

Abstrmt-A numerical continuation For tracing the double

bounded homotopy (DBH) for obtaining DC solutions of nonlinear circuits is proposed. The double bounded homotopy is used
to find multiple DC solutions with the advantage of having a
stop criterion which is based on the property of having a double
bounded trajectory. The key aspects of the implementation of
the numerical continuation are presented in this paper. Besides,
in order to trace and apply the stop criterion some blocks of the
numerical Continuation are modified and explained.

I.

INTRODUCTION

The numerical continuation methods (also called path folLowing and path tracking) are numerical tools used to trace the
homotopy trajectories. They are a combination of a variety of
numerical methods focussed on drawing a path in order to accomplish specific needs of a particular homotopy formulation.
Due to the specific features of the DBH, it becomes necessary to devise a well-suited numerical continuation method
in order to trace the homotopy trajectory having n robust stop
criterion. The next section is devoted to explain the traditional
numerical continuation methods.

Homotopy methods, [ 1 J, [2], [3], have an important role in

the analysis of circuits exhibiting multiple operating points.
Although, homotopy methods are able to find more than one
solution to the equilibrium equation of the circuit, they still
show several problems. Among them, it is worthy to mention
the lack of a reliable stop criterion.
It is well-known that there are two types of paths of
solutions, open and closed paths, the main problem is when

Solutions

"1

I
I

to stop searching f o r more solutions. For closed paths. this

can be solved by testing whether a new solution is not indeed

a previously found solution. For open paths, this is a serious
drawback, because there is no reasonable and reliable stop
criterion to decide when to stop seeking for more solutions.
The Double Bounded Homotopy, [4], [SI,has been proposed
as an alternative to circumvent the problem of the stop
criterion. The DBH formulation can be recast as follows:

H ( f ( s ) ,A) = CQ

+ eQ ln(Df2(z) + 1)

(1)

Homotopy Trijectory

where f(z)is the original set of nonlinear algebraic equations,

X is the homotopy parameter, C y D are positive constants of
the DBH, and Q is given by:

Q = (A

11. NUMERICAL
CONTINUATION METHODS

- a)(A - b )

where a and b are values of the double solution lines.

This homotopy possesses symmetrical branches that are
bounded by the solution lines. The symmetry and bounding
properties [6] of the trajectory of the DBH are depicted in
Figure 1 . These properties are useful in order to implement a
reliable stop criterion.

0-7803-901 5-6/05/$20.0002005 IEEE.

Fig. 1. Symmetry and bounding of DBH

Some homotopies have global convergence when applied to

solve the equilibrium equation of certain type of circuits. However, without an appropriated numerical continuation method
is not possible to ensure that all of the solutions may be found.
There are some reasons for this problem, one of them is related
to the predictor-corrector steps, if the coefficients of these
steps are not properly selected, then the numerical continuation

1122

fails and Ioses track on the homotopy trajectory. The other of the homotopy trajectory begins at X = 0 and ends' at X = 1.
reason is that once the numerical continuation crosses the When the tracing is close to X = 1,.the find Zem Srrategy
solution line, the algorithm fails to determine the solution takes over. The simplest example of strategy is to use [zf,A,]
because it diverges. It clearly results that it is important to (the last iteration) as the initial point to solve the equilibrium
study the characteristics of the numerical continuation in order equation f(s)with a Newton-like method.
to use them appropriately. The numerical continuation scheme
Because the Newton method possesses local convergence,
consists of a predictor, a corrector, a step control, a find zero ,it still could fail to find the solution. In [9] some techniques
strategy and a stop criterion.
are reported that implement the find Zem Strategy accurately
J ) Predictor: The predictor point for fd,
X j ) is given by: and reliably. The basic idea is to use two points (X < 1 and
X 2 1) in the vicinity of X = 1, and interpolate the point at
A = 1 in order to obtain a point close to the real solution and
(rCJ+',
= (ZJ,
A,)
h*t
use a Newton-like method to find the solution to the original
where ti is an appropriate step length and t is a normalised system f(z).
tangent vector to the homotopy trajectory (see Figure 2). This
5 ) Srop criterion: In fact, there are not stop criteria in the
predictor can be considered as a step of the Euler method traditional numerical continuation methods when applied to
(or any other integration method) for solving the diflerential homotopy trajectory tracing. The most common way to stop
equation that describes the homotopy trajectory (continuation tracing the trajectory is to set a maximum allowed number
path). Predictor steps are usually based on tangent predictions ( I T M A X ) of predictor-corrector steps without finding any
but there are several alternatives like the secant predictor 171, solution. This technique is inefficient because it usually fails
interpolation predictor [8], Taylor polynomial predictor [8].
to find some solutions on the homotopy trajectory.

xJ+l)

111. MODIFIED
NUMERICAL
CONTINUATION

Fig. 2.

This section explains the modifications accomplished on

the scheme above with the idea o f providing a reliable stop
criterion to the numerical continuation. Modifications are
introduced on both the find zero strategy and the stop criterion.
1) Find zero strategy. The DBH has the characteristic
of never crossing X = 1 [4], hence the findig zero
strategy should start after the trajectory bounces on the
bounding line. An efficient way to achieve this process
is by monitoring the change of sign of AA produced in
the predictor step. This can be done by inultiplying AX
of two consecutive predictor steps.

Predictor-comcior steps

2 ) Corrector: When the predictor step finishes, it is necessary to rectify the homotopy trajectory by using a corrector
step. This step solves the homotopy formulaiion by starting
from (Zj+r.X,+l) (see Figure 2) , A common way io solve
this equation is by using the Newton-Gauss method [SI,which
can be solved for systems of the type RN+l3 RN.
3) Step Conrml: The Predictor-Corrector scheme can be
optimized for tracing acceleration by using a step length
control. A small constant step length can trace the curve
successfully but not efficiently, because this process involves
too many steps along flat branches. Therefore, it is necessary
to adapt the step length to the convergence behaviour at each
predictor-correclor step. The basic criterion is to control the
step by observing the convergence quality of the corrector
step. A change on the number of iterations in the corrector
step produces n compensation factor 1, which affects the step
kength as follows:
hj+l =

Chj

4 ) Find Zem Stmtegy: Without an efticient finding zero

strategy, the numerical continuation is incomplete and the
hotnotopy could fail to converge to some soiutions. The tracing

This procedure is depicted in Figure 3, where ihe sign

ol' AA changes after bouncing from point A to point B .
Besides, in order to apply a quadratic interpolation the
algorithm needs three points ( A ,B j C).
2) Stop criterion. The stop criterion for this homotopy is
depicted in Figure 4. The homotopy trajectory starts at
the symmetry axis of rhe homotopy trajectory at the
point S. Then it traces the half of the trayectory (the
symmetrical branch) and stops when it returns to the
symmetry axis at the point E.
The modified numerical continuation is depicted in Figure
5 where the dashed blocks are the specific characteristics
added to the procedure. The scheme is explained a s follows: It
starts when the predictor calculates the tangent at (zj!Xj) and
&+I) over the
using n step length calculates the point (Zj+l!
tangent. Then, the corrector uses the solution of the predictor
i n order to obtain a new point on the homotopy trajectory,
given by ( d + ' , X j + l ) . At this point, the step control is
' I n lkct if it is wanted to find muhiplc solutions the numericnl continuation
should follow beyond X = 1.

1123

I af symmetry

MS
I

Predictor

I
Interpolation points

e Solirfions
Fig. 3. -Find zero strategy

Sofufions

Finding Zero Strategy

Interpolation

Fig. 4. Stop criterion

Fig. 5 . Block diagram of the modified numerical continuation

applied in order to accelerate the tracing. Next, the find zero

straregy is applied which is triggered at each solution when the
trajectory bounces on the solution line. Finally, the numerical
continuation stops tracing when the trajectory returns to the
symmetry axis.

The DBH formulation yields:

Hl(fi.:A) = lOOQ e@In(O.OOlf~ -I- 1) = 0

A) = lO0Q -k eQ ln(0.001f~ 1) = 0

H*(f2,

IV. EXAMPLES

In order to illustratethe use of the DSH with the modifications, a first example is used JO solve the system of equations
given as:
fl(51122) =
Z.(

- 1 ) ( m- 4 ) ( q - 6 )

f 2 ( ~ 1 ,ZZ) = (XI - 3)(21 - G)(zI

- 9)

fs1

=o

+2 2 = 0

The graphic solution of the sysiem is shown in Figure 6.

where &.= X(X - 1); i.e. a = 0 and b = 1.

The homotopy trajectories are depicted in Figure 7. The
starting points lie on the plane defined by X = 0.5, while, the
solutions are obtained when X reaches the value of 1 .
A second example is given by the latch circuit of Figure
8, which contains two NMOS transistors ( M I and M z ) , two
linear resistors ( R I and Rz) and a voltage source ( E ) .The
model of the transistors is the unified MOS model reported

1124

E3V

Fig. 8. Example circuit

Fig. 6. System of five solutions

Fig. 9. Graphic of the equilibrium equation of the circuit and homotopy

Lrajectory

X = 0.5
End oftrc4iectot-y X = 0.5

v. CONCLUSIONS

0 Startingpoint

.*

Fig. 7. Hoinotopy trajectory

A numerical continuation for tracing a DBH hns been

presented. The numerical continuation scheme exhibits an
improved performance regarding the stop criterion. Several
examples illustrating the aplication of the scheme 10 nonlinear
resistive circuits were also presenred.

in [IO] which is a modified version of the well-known BSIM

ACKNOWLEDGEMENTS
model.
Hector Vazquez Leal i; holder of a scholarship from CONAFigure 9 shows the graph of the equilibrium equation and
the homotopy trajectory of the circuit in the space (q;
712,I E ) . CyT MCxico under contract 143907. This work has been
partially supported by a CONACyT Mexico research project
The homotopy finds all three solutions of the circuit.
A last example is the well-known benchmark circuit re- under grant 42588-Y.
ported in [ 1 I]. This circuit has 4 bipolar transistors modeled
REFERENCES
by the half-sided Ebers-Moll model.
R.
C.
Melville
and
L.
Tnjkovif.
Artificial parameter homotopy methods
The formulation is the same of [11] which is based on
for the de operating point problem, lEEE ~rnnrocrionson cornptircrthe junction voltages vlr v2, v3, v4. Figure 10 depicts the
oirled design ofinregmed circirirs and systems, vol. 12. no. 6. pp. $61homotopy trajectory and the six found solutions versus V I .
877, June 1997.

I I25

YII
1

-0.2

Fig. 10. Solution of the Chua's circuit

[21 Denise M. Wolf and Seth R. Sanders, "Multiparameter homotopy

methods for finding dc operating points of nonlinear circuits," IEEE
frarisncrions on circuits ond sysreiiwl: fundaiirentcil rheuiy ciiid nplicnlions. vol. 43. no. 10, pp. 824-837, October 1996.
131 Lieven Vandenberghe and Joos Vaitdewalle. "Variable dimension algorithms for solviiig resistive circuits:' liiiernnrionol Joiinml of Circtiir
T h o q and Applicclrions, vol. 18. no. 5, pp, 443474, September 1990.
141 Hector V'Szquez-teal. Luis Hem6ndez-Martinez. and Anuro SnnnientoReyes. "Double-bounded homotopy for analysing nonlinear resistive
circuits:' fiuernariorinl Synpusirrrir on Circtrirr and Sys~eins.hqay 2005
151 Hector VSzquez-Leal, Luis Hemfinder-Martinez, and Anuro SannienloReyes, "Numerical path following for double boundcd homotopy scheme
for analysing nonlinear resistive circuits," IBERCHZP. 2003.
[6] Hector Vhzquez-Leal, Luis HemSndez Martinez,and Anuro Sanniento
Reyes, "Double bounded hombtopy applied to nontinear circuits
simulation,"
Springer Verlag. Applicabfe AI,qeb,a ifr Engineering.
Con~municationand Coolpaliirg. 2004.
171 R. Seydel. 'Tncing branches,'' World of B$mnlion. June 1999.
[ 8 ] Eugene L. Allgowcr and Kuit Gcorg. Nurriericnl Parb fullurvii~g. 1994.
191 Maria Sosonkina. Layne T. Watson, and David E. Stewart. "Notc on
rhe end game in homotopy zero curve tracking:' ACM l?flnscinrions U/i
Malhcrrtorical ~0fhvol-e.vol. 22, no. 3. pp. 281-287, September 1996.
[IO] S.H Jen, B.J. Sheu, and Y. Oshima. "A unified approach lo submicroii dc
mas transis%ormodeling for low-voltage ics," Analog /nrepnted Circrrirs
nnd Signal Piocessins, vol. 12, pp. 107-1 18, 1997.
[ I I] A. Ushida and L. 0. Chua. 'Tracing solution curves of non-linear
equations with sharp turning poinu,'' Circuir 7 % c u r ~cmd Applicotiorts,
vol. 12, pp. 1-21, 1984.

R I , =30 k R

R y l kR

CL

Fig. I I .

1126

vcc

Clma's circuit