Chapter I11 Stability Theory 1.
Basic Concepts
Under the composition of a differential equation, describing some real phenomenon, it is always neces...
Under the composition of a differential equation, describing some real phenomenon, it is always necessary to simplify and idealize this phenomenon, isolating only some basic factors and neglecting the others. The initial conditions in real problems are usually the result of measurements and, consequently, of necessity are determined with some error. In order that this conditional differential equation even approximately describe the studied phenomenon, it is necessary that a small change of the initial function and small, in some sense, changes of the differential equation result in only small changes in the solution determined by this initial function. Stability theory studies conditions under which small, in some sense, changes or, as is often said, small perturbations of the differ ential equations and initial conditions result in small changes in the solutions. Let us introduce the basic definitions with regard to a differential equation with retarded arguments: %(t) = f (t,X(t) ,X(t-Tl(t))
,...,X(t-Tm(t))).
(1)
I. Definition of Stability. The solution x (t) of Eq. (1) is called stable, if for any E > O , 4 there exists a 6 ( ~ ) > 0 ,such that from the inequality I $ (t)- $ (t)I < 6 ( € 1 on the initial set, there follows Ix4 (t)-x$ (t) I < E for all tLto, where JI (t) is any continuous initial €unction.
Solutions not possessing this property are called unstable.
119
DIFFERENTIAL EQUATIONS WITH DEVIATING ARGUMENTS
11. Definition of Asymptotic Stability. A stable solution x (t) is called asymptotically 4 stable, if & i g Ix (t1-x (t)I = 0 for any continuous 4 JI initial function $(t), satisfying for sufficiently small 6 1> O the condition 1 (I (t)-$(t) 1<61111. Definition of Uniform Asymptotic Stability. The solution x (t) of Eq. (1) is called uniformly 4 asymptotically stable if there exists 6 > 0 , such that for each E > O there exists a T(E) such that for for any continuous init>tl+T(E), Ix4 (t)-xJ,(t) tial function $(t) satisfying the inequality 14 (t)-J,(t)I < 6 on the initial set Et , where 6 does 1 >t0' not depend on the choice of t1IV. Definition of Exponential Asymptotic Stability. The solution x (t) of Eq. (1) is called 4 exponentially asymptotically stable if there exist constants 6>0, u > O , B > 1 such that from the inequality 14 (t)-J, (t)1 < 6 , there follows Ix (t)-x (t)I < 4 J, B max I + - ~ , l-a(t-tO) e for t>T. tEEt
I
0
V. Definition of Asymptotic Stability in the Large. The solution of Eq. (1) x (t) is called 4 asymptotically stable in the large, if it is stable and'
for all continuous initial functions $(t). All definitions remain unchanged if, in Eq. (11, x(t) is an n-vector or even an element of a Banach space. In this case, only the modulus sign I I need be changed to the norm sign 1 1 I I.
120
Ill. STABILITY THEORY
Sometimes in the definitions of stability and asymptotic stability it is expedient to use not the metric space C but some other (cf. 11.131 1. 0' For an equation of neutral type, 9 (t) = f (t,X(t),X (t'T1
.
(t)) I . . ,X (t-T (t)) m
all definitions given above remain unchanged, but in place of 'the requirement 1 4 (t)-J, (t)I < 6 it is usually necessary to require nearness in the space C1 : I+(t)-J,(t)1.6
and I@'(t)- ~l'(t)1.6.
For investigations into the stability of some solution x (t) of Eq. (1) or (21, it is possible by the change4 of variable y(t)=x(t)-x (t) to transform 4 the discussion to the stability of the solution x (t) into that of y(tIr0. Therefore, in the fol4 lowing discussion on stability, we study only the trivial solution.
2.
The Stability of Solutions of Stationary Linear Equations
All solutions of the linear equations with a deviating argument L(x(t)) = f(t)
(3)
with fixed initial point t , as for a linear equation without a deviating apgument, are stable or unstable simultaneously. In fact, any solution x (t) of Eq. (3) by the 4 change of variable y(t)=x(t)-x (t) turns into the 4 trivial solution of the corresponding homogeneous equation L(Y(t)) = 0.
121
(4 1
DIFFERENTIAL EQUATIONS WITH DEVIATING ARGUMENTS
Consequently, all solutions of Eq. (31, in the sense of stability, will lead in the same way to the trivial solution of Eq. ( 4 ) . In particular, for f (t)-0 all solutions of the homogeneous equation, in the sense of stability, will lead in the same way to the trivial solution of this equation. Particularly simple is the study of the stability of the solution of linear equations with constant coefficients and constant deviating arguments:
where apj and
T
j
are constants, T m > T ~ - ~ > . . . > T ~ > T ~ = ~ .
..
In Chapter II,§3, it was shown that if in Eq. # 0 and a = 0, j = O,l,. ,m-1 (in this nj case Eq. (5) is of advanced type), then there always exists a solution of the form Cekt, where c is an arbitrary constant with Re k>O. Consequently, all solutions of Eq. ( 5 ) in this case are unstable, since,for arbitrarily small IC1,the solution of the form Cekt is either unboundedly increasing in modulus as t- or for complex k , the solutions CRe ekt and C Im ekt oscillate with unboundedly increasing amplitudes. (51, anm
If Eq. ( 5 ) is an equation with a retarded argument (a # 0 and ani= 0, i=l,2,...,m~, then any nO solution x (t) for t +T(tT may be decomposed into 0 0 an absolutely and uniformly convergent series of basic solutions.
where p.(t) is a polynomial of degree less than or 3 equal to c1 -1, c1 is the multiplicity of the root j j k . of the characteristic quasipolynomial 7
122
Ill. STABILITY THEORY
-? .z m c a zpe 3 p=o j=O pj Re kl > - Re k&. Re knl..
n
(7)
1
..)
.
If all roots of the characteristic quasipolynomial have negative real parts, then the remainder in the series may be represented in the form -knt e Rn(t) , where IRn(t) I < € for n>N(E). Thus, there follows the asymptotic stability of solutions of Eq. (5) t'tO+'
1
Moreover, it is possible to prove that for (Re kn+l+E)t c pj(t)ekjt/ < BRe I j=n+l
where B and E are constants, E > O and arbitrarily small (see [1.17]). Consequently, for Re kld; conse3 j quently, max Re k = Re kl exists and, if Re kl O , such that Re kl+E
.
An palogous theorem holds for equations of neutral type but in this case, because of the possibility of an unbounded approach of the roots to the complex axis, the sufficient condition for exponential asymptotic stability will not be the requirement "all Re ki
123
DIFFERENTIAL EOUATIONS WITH DEVIATING ARGUMENTS
Because of the possibility of the approach of the roots to the imaginary axis for equations of neutral type, the proof of exponential asymptotic stability of the solution for equations of neutral type under the condition "all Re k.<-a
k.t Im e
I
a>l,
the solution is unstable. Analogous theorems, besides the last proposition, hold for a system of linear equations with constant coefficients and constant deviating arguments (in the last assertion, the multiplicity of
124
Ill. STABILITY THEORY
the root was necessary to note the magnitude of the analogous degree of the elementary divisor). Finally, we consider peculiarities connected with the possibility of the appearance, for the equations of neutral type, of asymptotic/critical and supercritical cases of the location of the roots of the characteristic quasipolynomial. For Eq. (5) of neutral type with one deviating argument (m=l, ano#O, anl#O) , W. Hahn r158.11 showed that under the condition "all Re k.
the multiplicity k, satisfying the inequality (k-1)x (e-l)>n, then the solution of Eq. (5) is unstable. Instability of the solutions of Eq. (5) of neutral type is possible also in the supercritical case. For example, as shown in [23.1], for the equation
125
DIFFERENTIAL EQUATIONS WITH OEVlATlNG ARGUMENTS
2(t)-?(t-l)
+ a(x(t)+x(t-1))
= 0,
for any a>O the characteristic quasipolynomial has the supercrionly simple, purely imaginary roots tical case. For this problem, on the half-axis (O,-) there is situated an everywhere dense set of values of the parameter a, for which the solution of the considered equation is unstable.
-
3.
Conditions for Negativity of the Real Parts of All Roots of the Characteristic Quasipolynomial
A necessary and sufficient condition for asymptotic stability of the solutions of stationary linear equations, with the exception of the consideration in the preceding paragraph of the basic critical cases, and a sufficient condition for asymtotic stability of the trivial solution of a wide class of equations, stationary in the first approximation (cf. Ch. 111, 17) is the negativity of the real parts of all roots of the characteristic quasipolynomial.
Since approximate calculation of all roots of the quasipolynomial is a problem of great difficulty, there is great value for investigations in stability, to obtaining different tests of negativity of the real parts of all roots of the quasipolynomial. Among such tests, most often the following are employed:
1) the amplitude-phase method and its modifications; 2) the method of D-partitions and its modifications; 3 ) the method of Meiman and Chebotarev.
Below we study the basic idea of the amplitudephase method, details of the development of Ya. 2. Tsypkin (cf. [127.1]) and the method of D-partitions (cf. Yu. I. Naimark 185.11, Pinney [I.17]).
126
1 1 1 . STABILITY THEORY
For readers desiring acquaintance with the method of Meiman, Chebotarev, we refer to i I . 5 1 . 1. The Amplitude-Phase Method. We call to mind that if the function f(z) is analytic and different from zero on some simple, closed contour C, and the interior of the contour has only a finiteset of polar singularities, then
where NC is the number of zeroes of f(z) in the interior of C, counted according to their multiplicities, and Pc is the number of poles in the interior of C, counted according to their multiplicities. The geometric interpretation of this theorem about the logarithmic residue leads to the "Argument Principle 2 n1 Ac Arg f(z) = NC I'
-
Pc.
(9)
Arg f(z) is the total increase of the argument of the function f(z) under a single circuit of the point z, in a positive direction, around the contour C. On the other hand, the difference N -P c c equals the number of complete revolutions which the vector w performs in the plane, going from the point w=O to the point w = f (z), when the point z describes the contour C in the positive direction (the number of revolutions are regarded as positive, if the vector is rotated counterclockwise, and is considered negative under a clockwise rotation). A
c
The reader may find the proof of this theorem in any book in the theory of analytic functions in the chapter dealing with logarithmic residues. For obtaining a condition €or the absence in the characteristic quasipolynomial (o(z) of roots with positive real parts, we apply the argument principle to the contour CR, consisting of the segment of the imaginary axis [-iR,iR] and the semicircle of radius R with center at the origin, lying 127
DIFFERENTIAL EQUATIONS WITH DEVIATING ARGUMENTS
in the half-plane Re z>O (Fig. 7 ) ; as a preliminary, we check that the quasipolynomial does not have zeroes on the imaginary axis. We note that in this case Pc = 0 . Using the argument principle, we find from ( 9 ) NC and, if R lim NC = 0, then all roots zi of the quasipolynomial R+R satisfy the condition Re zi
For application of this general method to the quasipolynomial
corresponding to the nth order equation (and also to some system of n first order equations) with a s i n g l e retarded argument, where Pn(z) and Qn-,(z) are polynomials of degree n and not greater than n-I, respectively, it is possible to somewhat simplify the investigation. Instead of the function $ ( z ) , we consider the function
the zeroes of which coincide with the zeroes of the function $(z) (if Pn(z) and Qn-l(~)do not have common zeroes) and which has poles at the zeroes of the polynomial P (z). n
position as R- of the form of the contour CR under the mapping wT(z) is called the amplitude-phase characteristic. of the function
= 1
Since
@ correspond
which w T (z) = 1.
-
wT (z), the zeroes
to the points at
Therefore, applying the argument
128
Ill. STABILITY THEORY
principle to the function wT(z), it is necessary to calculate the number of circuits of the amplitudephase characteristic,not of the point z=O,but of the point z=1. The number of circuits of the amplitude-phase characteristic of the point z=l equals the difference NC-PC and, consequently, in order that NC = 0, it is necessary that the number of circuits of the amplitude-phase characteristic of the point z=l equal - P Again, we call to mind that, for this it is assu&ed that there are no zeroes of the function $(z) on the imaginary axis and that Pn(z) and Qn-l(~)do not have common zeroes, where both of these conditions are relatively easy to check.
.
We note that under the mapping wT(z), the semicircle entering into the structure of the contour contracts to a point (since the degree of as RcR’ Pn (z) is greater than the degree of Qn-1 (z)) and, consequently, it is necessary to construct only the form of the imaginary axis, transversed in the negative direction. For construction of the amplitude-phase characteristic, it is convenient at first to find the socalled limiting characteristic, appearing as the limiting form of the contour C R under the mapping
For construction of the form of the imaginary axis under the mapping
or
w (iy) = w (iy)e-Tiy I T 0
129
DIFFERENTIAL EQUATIONS WITH DEVIATING ARGUMENTS
knowing already the limiting characteristic, it suffices to consider the influence of the factor e-'Iiy of rotation without change of modulus, the radius vector of the point of the limiting characteristic corresponding to the value y at the angle
-TY. For construction of the amplitude-phase characteristic, particular attention should be given to the points of the limiting characteristic lying on the circle 1z1=1 since these points, under rotation by an angle -ry, may find themselves at the point z = 1 .
As an example, we find the domain of asymptotic stability in the space of coefficients a and b of the trivial solution of the equation
+ aX(t) + bx(t-'I)
*(t) where a r b rand
'I
= 0,
(10)
are constants, r>O.
In the case under consideration, the characteristic equation has the form
+
be-'IZ
wo(z) =
-b z+a -
z
+
a
=
0,
The limiting characteristic is the form of the imaginary axis under the fractional linear transformation (12). Under this mapping, the imaginary axis transforms into the circle of radius lb/2a] with center at the point z=-b/2a, the equation of which has the form
Let a>O, then the function wT(z) has no poles in the half-plane Re z>O and, if Ibl
Ill. STABILITY THEORY