JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS 178, 34-362 (1993) Global Geometry of the Stable Regions for Two Delay Differential Equations Jack K. HALe* anp WENZHANG HuaNG. Center for Dynamical Systems and Nonlinear Suis, Georgia nsate of Technology, Anta, Georgia 30892 Submited by 8. N.Busenbera Reocived January 28, 1991 1. InRoDUucTION The stable region for the two delay differential equations (I) + ax(t) + bx(t—1) + exo) 20,630,120 (11) has been investigated by many authors [1,3-9], but has not yet been globally solved. The numerical results [3] show that the boundary of a stable region may be a very complicated curve which has infinitely many kinks. So an interesting question is: What is the asymptotical behavior of the boundary of a stable region? Can it become chaotic? In this paper, We will give a complete geometrical description of the stable region'' for the equations (1.1) in the r-o plane. In particular, we prove that (i) Ifa stable region is unbounded, then its boundary will approach a straight line parallel to the r-axis or a-axis as r+ 0 ~+ 00 (i) Ifa half tine in the first quadrant of the r-o plane contains an unstable point, then the intersection of this line and the boundary of the stable region contains at most finitely many points and eventually the half line leaves the stable region. + Research partially supported by the National Science Foundation DMS-8507056 and DARPA TONANBSHO860, Here the stable region is defined as a maximal connected set D< [0 22) [0,:0) which contains the origin (0,0) such that for each (r, ¢)eD the zero solution of (1.1) is asymptotically stable. In this paper, we do not discuss the possiblity of existence of any other stable region in [0, 29)x [0, 20) which is disjoint with D. We conjecture that thee is only one connected stable region whenever it exists 344 (022-247X/93 $5.00 ‘Aigo peda any fo er‘TWO DELAY DIFFERENTIAL EQUATIONS. 345 2. PRELIMINARY It is clear that the investigation of the stable region will lead to studying the location of zeros of the characteristic equation of (1.1) 0. ay pdr 0) 2 Ata be + c6~# And, in particular, the study of the equation Pliv, 7.0) = iv +a+ bem" + ce-* =0, v20 2) will play a key role. Equation (2.2) implicitly defines a family of curves (r(v), (0) in the r-o plane which may have very complicated structures. But from the viewpoint of stability analysis, only those curves such that each point on these curves can be connected to the origin by a continuous path in the stable region will be of interest. Furthermore, since e (0 R) is a periodic function, as a first step, we will study Eq. (22) under the restriction of 00, 7 >0, the zero solution of (1.1) is asymptotically stable if a> |b +cl, stable but not asymptotically stable if a=|b +c] and unstable if a<|é+ cl. Therefore, without loss of generality, we suppose b>0 and divide the relation of the coefficients a, and c into the following cases according to their positions in the real line (see Fig. 2.1) 1. bta>lcl (cta>lbi can be considered as this case if we exchange r and 0). I. b+ad|el>b—a ML -cxatb0. W. atb+e<0 in ab atb ee A a arb Te : atb -¢ “eC we “e Froune 21346 HALE AND HUANG: Case IV is simple. We have Proposition 2.1. If a+b+e<0, then, for all delays r>0, 0>0, the zero solution of (1.1) is unstable The proof of Proposition 2.1 is trivial. In this paper, we will give a detailed analysis only for the case of b+a>|c| and then, by using the idea developed in discussion of this case, we can describe the stable regions for the other cases. 3. ANALYSIS OF THE STABLE REGION Throughout this section we suppose bta>lcl. Let © denote the complex plane and Fyq= {2eCst2l= lel} For r>0, f,: (0, 2] + € is defined by Ss)= 12 +a+ de® and for r>0, 720 iet As, 7,0) = f(s) +ce°"*, se [0, 2x]. GA) Itis apparent that (2.2) has a solution v € [0, 2x/r) for some r>0, ¢ <0 if and only if A(s,r,0)=0 for some se[0,2x) and if and only if FALO, 22)) VF), #D. So we will study the equivalent 4(s, r,¢)=0 rather than Eq. (2.2) Lemma 3.1. For each fixed r>0, f,([0,2r]) és @ simple and smooth curve in © and SAL 2 OT =. Proof. The first conclusion follows from the fact that f,: (0, 2n] + € is one-to-one and differentiable. Now for s¢ (x, 2r], we have Mrs? -bsins) + (a+ bcos 5)? st sb sin s +a? +6? + 2ab cos s >a?— 2a] b+57 That is, f,([x, 22) OT =D b—\al)?> lel?TWO DELAY DIFFERENTIAL EQUATIONS 347 Now let r’>r">0 and se (0, x]. It is clear that Ref.(s)=Re f(s) and Im f,(s) |cl,a—b< —|el, then there are O (7), re (rosa) and FAC AVF = (LSD LAM}, rE Uros Fa) Figure 1348 HALE AND HUANG: Figure 32 We prove this theorem in the Appendix. With the aid of Theorem 3.2, we are able to characterize the curve (r, 0) determined by Eq. (3.1). IA. Suppose b+a>e>0. Tt is clear that there is a unique rye(rasrs) (r= tan (=¥ a+ eW/B? = (at e)) such that Llslr=me (2(r=r JP a+), Now for each re (rg, r2), let p, = f,(s,(r)), i= 1, 2, and 8,(r)€ [0, 2x) be the angles from the negative real axis to the rays starting at the origin and passing through p,, i=1,2, in the clockwise sense respectively (see Fig. 3.3), It is clear that @, (r) is continuous on [ro, r3}, 8;(r) is continuous on (ro, r2]\{ry} and it has a jump at r=r, with lim 0,(r)=0, lim 03(r)= 2m (3.2) and Or) a3(r) on (r,,73)) IB. Suppose b+a>Icl, ¢<0. In_this_case, if we again let r, (@FeF, then we have sy(ry) =r ¥/ Ilse = —e Furthermore, for each re (ro, r2)s let 6,(r)€ £0, 2x] be the angles from the positive real axis to the rays starting at the origin and passing through FAs(r)), i= 1,2 im the clockwise sense (see Fig.35) and ¢,(r)= 70,(rVis,(r), i= 1, 2, respectively. By using analysis similar to case 1A, the curves o,(r) are shown as in Fig. 36 where alr =or(r), 10,2 a3(r) is continuous on [rp,r2]. 44 (r) is continuous on [rp, r3]\{ri} and it has a jump at r, with 2ary a lim o,(r)=0, lim o,(r)=—— = — fim ue) 0, fim u(r) = es and aur)> asl, re (ror) Fioure 34‘TWO DELAY DIFFERENTIAL EQUATIONS 351 Ficure 35 In both of the cases IA and IB, if we periodically extend o,(r) by oN(r)=a,(r) + 2nnr/s(r), re [rosroh i= 1,2 then the boundary of the stable region D (see first page for the definition of D) consists of the family {o"(r), i= 1,2, re [ro P2]},-o Which are bounded on the left by the line r= ry (see Figs. 3.7 and 3.8). Clearly, itis important to know how the curves a'(r) change as r varies near ry because it might be possible that the curves air) near ry become chaotic asm 20 o Ficure 36352 HALE AND HUANG: fo n 2 Fioure 37 Ficure 38TWO DELAY DIFFERENTIAL EQUATIONS 353 We now turn to study the properties of o(r) for r close to ro. First we try to reparameterize the curves (r,a,(r)) for r near ry as (H{s), 0(3)). We note that As, 7,0)=0 if and only if SiAs)+ce-”=0 (34) with 0 =Or/s. We may separate the real and imaginary parts of (3.4) to obtain the equivalent system Fis, 1,0) 25 —bsins—csin0=0 Gls, 7,0) a+b cos s+ 0s 6 Let Q=s$oo/ro (here o9=2,(ro)=02(ro) and sf is as defined in Theorem 3.2), It follows from the definition of s#, rq, and a9 that FUSS sro» 80) = Gls sro, 80) = and 0 ~csin A —ccos A (7D [trot which is invertible since sf /r3#0 and csin@)#0. Then the implicit function theorem yields that there are a neighborhood Ix Ux Vc R? of (sd, ro, 0) and continuously differentiable functions r:/—+U, 6:1-+V such that HS8)=Fo, — O(8$)= Oo, (3.5) F(s;r(s), 0(3))= G(s, r(s), O(s))=0, ser, (36) and for (s.r, 0)€1x UV, Fls,r,0)=G(s,r,0)=0 — if'and only if r=r(s), 0=O(s). (3.7) Now (3.6) is equivalent to fans) sel (See Fig. 3.9),394 HALE AND HUANG: Ficure 39 Since f,(+) is tangent to the circle I.) at —ce ™ and f,{-) moves upwards strictly as r decreases (see Fig. 3.10), thus for parameter values s, As )>r(s2rs8)=ro it sy>s\ 25h (8) and Msp>rsilersh) if sis$. So we have {<0 if sr(s¢)=ro. So 83(r(s)) > 5, (r(s)), ET \ {58}. It now follows from (3.8), (3.9) and (3.12) that s2(r(s)) > 5$ >, (s)) SEF\{S3}. By using (3.12) again we arrive at fairy) if ssa, = (il if s>st ed356 HALE AND HUANG: Now if we set o(s)= 0s) r(s)/s, then, for sel’, ssi, Qnnr(s) 2nnr(s) n os) + ME 69) + EHD —osies) (315) Note that F(s#)=0, r(s¢)>0 and 4(s) is continuous at sf, so there are e>0and 6>0, m>0 such that 1i(s) s| +00 ™ as n+ op uniformly for re (ro, ris —€)]. similarly, we have dose) ar —0 as n+00 uniformly for re (ro, ris¢ + 6)] (see Fig, 3.10). That is, the curves a4(r) and o4(r) for r near ro almost become a straight line as n becomes sufficiently large. As an immediate consequence we have THEOREM 3.3. Under the assumption of b>0, b+a>\el, the boundary of stable region approaches the straight line r= ro. Furthermore, the up half strip {(r,0); 00 atb>c,a—b=—c, ¢>0358 HALE AND HUANG: ID. atb=|c,a—b< e<0 atb>|d,a-b=—|e, <0 o stable not asy. st. TA. atb>c>a-b, c>0 atb=c>a-b, c>0‘TWO DELAY DIFFERENTIAL EQUATIONS 359 UB, atb>\l>a-bh cx UL c>ath>~ e>0 c>atb=-c, esd360 HALE AND HUANG: APPENDIX This appendix is devoted to proving Theorem 3.2. Let f, be defined as in Section 3 and let alr) PULP) y= 2bry sin y+ 2abr? cos y+ (a? +b7—c?). (Al) Then SACO, PIVOT B if and only if g(r, »)=0 has a solution ye [0 2}. Before proceeding to the proof, we need a few lemmas. Lemma I. Suppose H:[0,x]-+ R is twice differentiable and H(0)>0, H(x)>0. If H"(y) has at most one zero in (0, x), then H(y) has at most two zeros in [0,7], Proof. Since H"(y) has at most one zero in (0, x), H'(y) has at most two zeros in (0, n). If the lemma were false, then H(y’) would have at least three zeros in (0, x). Let y,0, H(y,)=0 and y, < yf it follows that H(y})<0. Similarly, we have H(y)<0. Note that H(y2)=0 and Yr€(Q yE) So H(y) is not monotone in (yf, yf). That is, H'(y) has a zero in (y#, y$). This is a contradiction and Lemma 1 follows. We let g,(r, ) = Cer, yW/dy, and for integer k>2, let Oates») ay a y= Lemaa 2. For each fixed r>0, g\2\r, y) has at most one zero in (0, m) Proof. We have 8, (r, 9) = 2y — (2br + 2abr?) sin y— bry cos y a(n, y) =2—2br{2+ ar) cos y+ 2bry sin y (a2) 27, y) = br(6 + 2ar) sin y + 2bry cos y (a3) (i) Suppose 2+ar>0, then gr, y)>0, ye [n/2, x]. Since 2+4ar>0 implies 6 +2ar>0, it follows from (A3) that gir, y)>0,‘TWO DELAY DIFFERENTIAL EQUATIONS 361 ye (0, m/2]. That is, g(r, y) is strictly increasing on [0, x/2]. Therefore, 2%, ») has at most one zero in (0, 2). (ii) Suppose 2+ar<0, 6+2ar>0. In this case (A2) yields that gir, y)>0, ye [0, 2/2] and g(r, »)= brl8 + 2ar) cos y— 2bry sin y <0, ve(§ | It follows that g(r, y) is strictly decreasing on (n/2, x]. Since g(r, 2/2) =1b(6 + 2ar)>0 and g(r, ) = —2brn <0, there is a unique y* € (m/2, x) such that 82M, )>0, ve[} BM y)<0, ye(y* x]. Hence g(r, y)>0 for ye [0, y*] and g(r, y) is strictly decreasing for ve(y*, x]. Therefore, g'?\(r, y) has at most one zero in [0, x] (iii) Suppose 6+2ar<0. In this case we have g@%(r, y)>0, ve [0, n/2] and gir, y) <0, ye (m/2, x). So it is clear that g!?'(r, y) has at most one zero in (0, x). Proof of Theorem 3.2. First, by the assumption of Theorem 3.2, we can easily check that, for each r>0, (1,0) >0, g(r, )>0. So it follows from Lemma 1 and Lemma 2 that g(r,-) has at most two zeros in [0, 2]. If we look at the function f,(-), itis clear that f,(-) is above Tq, ifr is sufficiently small and below 17, if r is large enough. Since f,(-) moves downward strictly as r increases, there are exactly two values rp and 7, such that f,([0, 1]) 0 Tq =D, re (0, ro) U (ra, 00) and f,,(-) (=0, 2) is tangential to T7,;. That is, g(r,,-) is tangential to the y-axis. Since g,(r,, ») has at most two zeros in (0, x), g(r.) is tangential to the y-axis at exactly one point, Hence, there is unique 5,< (0, x) such that FALO, AI) Tia = (Lrl50}p = 052. Furthermore, by the definition of ro, rz and the properties of f,(-), it is obvious that for each re (ro.r2) f-(L0, ®]) intersects F,, at least at two points. That is, g(r, y) has at least two zeros in [0, x]. But, since g(r, ») has at most two zeros in (0, x] for each r>0, we deduce that g(r, ») has362 HALE AND HUANG exactly two zeros in [0,7]. Equivalently, there are s,(r), s2(rh, s2(r) > s(r) such that SACO, AI 0 Te, = {FAS (PY LAS}, PE (ros Fa). The continuity of s,(r) easily follows from the continuity of g(r, y) The proof of Theorem 3.2 is completed, ACKNOWLEDGMENT We thank Professor K. L. Cooke for carefully reading this paper and his very useful comments REFERENCES 1, R. BELLMAN AND K. L. Cooke, "Dilferental-Diflerence Equations,” Academic Press, New York, 1963. 2. L. E, ELSGOL'TS AND S, B, NorKiN, “Introduction (0 the Theory and Application of Differential Equations with Deviating Arguments.” Academic Press, New York, 1973. 3. J, K. Hate, “Lectures in Applied Mathematics,” Vol. 17, pp. 157-185, 1975, 4.1. M. Manarey, “Geometry of the Stability Region for a Differential Equation with Two Delays,” Preliminary report in AMS meeting at Claremont, November, 1988. 5. C. MaRniorr, R. VALLEE, aND C. DeListe, Analysis of a first order delay dlferential-delay equation containing two delays, Phys. Rev. A 40, No.6 (1989), 3420-3428. 6. R. Nussuaum, Dillerential delay equations with two time lags, Mem. Amer. Math. Soc. 208, Providence, 1978 7. C.G. Racazz0 ano C. P. Matta, “Mode Selection on a Differential Equation with Two Delays Arising in Optics,” Preprint, 1990 8. J. Ruiz CLarrssen, Effet of delays on functional diferential equations, J. Differential Equations 20 (1976), 404-440, 9. H. W. Sri, The Hopf bifurcation: Stability result and application, J. Math, Anal. Appl 71 (1979), 525-546