An Offprint from FIELDS INSTITUTE COMMUNICATIONS ‘Tue Fietps Instrrure ror Research IN MATHEMATICAL SCIENCES: Voune 21 On a Class of Nonlocal Problems with Applications to Mathematical Biology W. Allegretto P. NistriOn a Class of Nonlocal Problems with Applications to Mathematical Biology W. Allegretto Department of Mathematical Sciences University of Alberta Edmonton, Alberta, Canada TSG 2G P. Nistri Dipartimento di Sistemi ed Inf Universita di Firenze Vin di S. Marta 3 130 Firenze, Italy poistridingfft-ing.uaifs ie To our friends Lynn Brbe and Herb Freedman ‘Abstract, We consider the problem of finding positive solutions too family of honlocal portal diferentisl equations, with mixed boundary conditions. Teis a feoture of these problems that maxlium principle methods are not applicable and we employ topological methods in the analysis. A key role is played by elated eigenvalue extimate 1 Introduction Let Bs(c,u) denote a mollifier in RY, ic, Byles) = Bolle ~ yl) © Cis Sax Bolx,y)dy = 1 for any x, Bs(|x — yl) = 0 if |x — y| > 4, Belle — y|) bounded away from zero if |x — y| < < 6. Consider a smooth bounded domain 2c R™ and define for any u € L4(Q), J Pelenntvrar In this paper we consider the problem of finding positive solutions to the nonlocal partial differential equation =VlaVu + bul = (h = gt () 1001 Mathematics Sulject Classification, Primary 36356; Secondary 35325 Research supported in part by NSERC (Canada), CNR (Ital),2 W. Allegretto and P. Nistel subject to the mixed boundary conditions: on Mp; natural boundary conditions on Ay, ie., in this ease 2) oR + G-u where Ap, OM are disjoint parts of AA, BM = Ap U AN», AXp closed. We shall consider explicitly only eases where Ap and 6M are not trivial but simpler versions of our proofs work in the other cases. We also require that regularity conditions be obtained at points of 0 (resp. Ap ly) by means of biLipschitz ‘maps to hemispheres (resp. quarter spheres) Furthermore, given any constant K we ask that there exists a neighbourhood AN of A such that [atweit+ 6 vane> «fa? 8) for any nontrivial w € HE2(N' 11 (QU ANx)). ‘The reader interested in detailed descriptions of such conditions may find some of them given in, e.g., [18], [19], (22), [24] and elsewhere. As for the functions a, h, 9,6 in (1), we recall their possible physical meanings: a(2) represents the diffusion process; (2) = (by(z), x(x) is a possible drift; h(e) the growth condition and g(z) a factor to describe the magnitude of the “crowding effects”. We ask that a, 9 be nonnegative in £™() and piecewise smooth with a < a(x) < 9 for soine constants 0 & We do not require that the left-hand side of (1) be coercive. We feel that these conditions are moro realistic than many often used, and certainly more general. They do introduce new difficulties in the proofs We note that here 7 is caleulated using only values of y inside ©, i., the effects on u(a), for with dist (¢, 8M) <6, of values u(y) with y ¢ 9 are disregarded except through the boundary conditions. If such interaction effects were significant, then 2 could be enlarged. Tt may also be of interest to assume that w is given outside 2, This is actually what we do here, with the choice u(y) = 0if y ¢ ©, for most of the paper, but the small 6 case is treated differently.(On a Clase of Nonlocal Problems with Applications to Mathenstical Biology 3 We emphasize to the reader that order arguments (Le., upper/lower solutions) are in general not applicable to this family of problems, as was illustrated by explicit example in [5], even if g(z 0. The problem we consider is thus different from ‘those considered in (8}, 9), (12), [15]. where if g(z) was small enough then order brocedures were indeed applicable, and from those considered in, e.., [20], [21] Where order arguments can always be employed. We thus proceed by Topological Methods (Degree Theory). The paper is in final form and structured as follows: We shall first consider the ease where g is bounded (essentially) away from zero. Next, we will allow to be zero at some points of ©. The case of 6 small will then be considered. Some comments, related results and open problems will conclude the paper Finally, we point out that our proofs work without any changes in the case that the right-hand side of (1) is replaced by the more general expression f(x, 1, (t))u under suitable assumptions on f and the operator 7. More comments on this will be made in the last section 2 Existence for general § Note first that equation (1) and the maximum principle show that if 0< u € L%, nontrivial, then u > 0. Since most ofthe coefficients are only in L", to see this note that we must still have w © C°(3) for some ay determined by the problem data but not u, [24]. If u(P) = 0 at some P € 0 then lullzszs.¢ey) ~ 0 for small by the generalized Harnack’s Inequality, [14, p. 194], where $,() denotes the ball of radius 7, center P. It follows that w= 0 in S,(P) and thus u > 0 in since 0 is connected AAs may be expected, we shall formulate results in the sequel based on suitable eigenvalue estimates. Specifically here we consider the two parameter local problem: foe = Vlas + Be] + ng = ua(nphus «) subject to boundary conditions (2), where ju(7) denotes the eigenvalue with a Positive eigenvector. Problem (4) is classical, it and related problems have been studied by many authors, see, e., [4] (6). It is well known that w > 0 exists, and #4() is monotone increasing in 7. Our specific assumption is: 100) <1 < ai) (5) for large K, ic. ‘eigenvalue crossing” occurs for this related problem. It is fairly ‘easy to see that if this condition is dropped, then the nonlocal problem (I) - (2) may not have a positive solution We collect such preliminary facts as well as other results for the reader's con- venience in Lemma 0, and give brief proof. Lemma 1 For > 0 (8) Ha(n) exists and > 0 in QUAN. Moreover, wi (n) and w are wnigue (b) sala) Tn If g > a0 > 0 then p(n) © as 9 00 SS(c) 6(n) < mu(n) < 62(n) where 01(n), 83(n) are the first eigenvalues of the formally self-adjoint problems: Ht + ngior = Bi(n)heoy o 5.¥. [ova ria] +29 ond i ‘ 3 Von, fii? a ~v [avin + Jun] + Pee 4 [HE + ngjo2=es(ten respectively, with the associated conditions (2). Here the existence of 8:(1) is toned (A) If probleme (1) ~ (2) has a positive solution then condition (5) hotds. (e) Let 0 0 be given, and let Ss-(P) denote the sphere centered at P of radius 6°. Then there exists K, e9 > 0 independent of u, P such that ap weex(f a) Sas(PIO8 (8) Let € 0°) satisfy —VjaV2+Be]—Ahz = f with boundary conditions (2) and 02 FEL™. There exists Ay > 0 independent of f, 2 such that if 0 << dy and z>0 then z se (PIn Proof (a) That j1(n) exists and w > Os an immediate consequence of the fact {6} is a positive map for 7 > 0 (by the maximum principle applied to the formal E? adjoint, ([23], p. 97)) from oe to itself if the coefficients are smooth. See, e.g., ((16}, p. 67), and note that (f9)~'(hz) > cz for some c > 0,0 <2 € C®° andifw >0 then w > 0 in $1 by our earlier comments at the start ofthis seetion. ‘To obtain the result in our case we pass tothe limit over a normalized (in *) soquence, employing the estimates of part (c) to show that the saquence of approximating eigenvalies is bounded. That « > in Of follows from the extension process of part (e) “To see that there is only one such pa (7), note that by the Fredholm Alternative we also have B5(0") 2 wih” for some juj(7), w* > 0 where &} denotes the formal adj Han) (Io* we) = (00*, Low) = yj (1) (hue” 0. Since (his",w) > 0, this shows 41(7) is unique. Finally to see that w > 0 is also ‘unique (up to scalar multiples), note that if there were two such eigenfunctions a, v then fo(w — ex) ~ py(9)(w — £0) = 0. Set B= {e | w~ ev > 0 in 9}. Clearly O'€ B, Bis closed and if © € E then (wo — (¢ + 6)v)~ must have support near 80 for small 6 > 0, since given any connected compact set K C2 then, again by Harnack’s Inequality, either w — ev = 0 in K or itis bounded away from zero in K. But the form associated with fy is coercive over such functions, by inequality (3), whence ¢ +6 € F for 8 small, ie., Bis also open, aud the result follows. nt, Le,(On a Class of Nonlocal Problems with Applications to Mathematical Biology 5 (b) Again by the Fredholm Alternative (see also [2] for related concepts) we have for any 0 0 that: (0° fou — a(n) hu) = ie, for some ary € 9, foulxo) = pr (n)A(x0)u(ao) tat (fz) smo soup (92) For example, if —VlaVus + bus] + migui = (ms )hus with i = whence and 7 > m, then ee [=ze 4+ Bual + ne] Thur = gp [ere Fo lo < yom) Passing to a limit then gives the result for our case. That j1:(7)) -+ 00 a8 7 = oc for g > ao > 0 follows from jri(n) > 81(n) (shown in part (e)) and the Courant Min-Max Principle, (6) Observe 6;(7) < p(n) follows from: fyal¥uP + 6. Wo) + ng’ Soha? while 6)(1) is the infirnum of the right-hand side over w € H})?(QUAMy), hw #0. Next we obtain 82(7) > ya(n) by modifying the arguments In [1]. Specifically let # € C§(NUANy), andl observe the identity Loree (Sf seee-0(S) + Be a(n) = 2 I As in (1), we observe that the left-hand side is nonnegative definite and obtain the result by taking the infimum over y. (d) Suppose (1) ~ (2) has a solution 0 0 such that : -VlaVe + Bu] ~ Ohw =0 and w satisfies condition (2). In fact, 41(0) = @ < 1 since @ = 1 implies giu=0. Furthermore, if we choose K > w in 2 then -VaVu+ bu] + 9K'u > hw For example, 4(K) > 1. We can only have equality if u is actually the eigenvector, ie, K =i But in this ease, (K+ 1) > 1(ec) Note that for each point Q € GRy we have a neighbourhood A’ such that QAN can be mapped either to.a hemisphere or a quarter sphere if Q € Ay NAN of radius r = r(Q) by biLipschitz maps L. Without loss of generality, N'C Ss-(Q) Set M = L~!|S.,(L(Q))], 0 <¢ <1. First cover Ay 9 Mn in this way and then the remainder of 8x so that for each N we have either \’O8Mp = O or N is a neighbourhood of some Q € 32 NAN. By compactness of Ny we have and thus aay cma for some ¢, 0 < ¢ < 1. Clearly we have dist (0(U.M,), I) 2 pe > 0 for some yx Now if dist (P, fi) > p then the result is immediate by (14] for some co < j/2 On the other hand, if dist (P, 3)

ao > 0 ac. % and let an,gashn be smooth pointwise approximations to a,9,h with a < aq <9, a9 < gut 90 < hy 7h. Consider the regularized problem: Van Vin + bun] = [hn ~ gat ay(On & Class of Nonlocal Probleme with Applications to Mathematical Biology 7 subject to conditions (2). If assumption (5) holds, then problem (1*) ~ (2) has a solution 0 < um € O(G) with |ugllos Ki, lttallow > Ko, for some positive constants a, Ky, Kp, independent of wasn. Remark We recall that upper/lower solutions procedures cannot be applied to (1°) due to the presence of 7, Note also that by Lemma 1(b), for assumption (5) to hold it suffices to have 4 (0) <1 Proof We may assume that J+ > 0 on 60, for if not we need only make the variable change uy = ee with Yar >> 0 with w chosen suitably. ‘This will also change tho coefficients and. but it will not affect their relevant properties, 1m particular, we will still have j(0) < 1. Next, ad a large postive constant iy to both sides of (1*) so that the left-hand side becomes coercive over H#(91U Ay) We ths obtain the equation Luin = With which we associate family of problems: Ltin = Alin + Co ~ Gutinly ae Vids Vuln + Bn) + Cotin = [Pn + Co ~ Gntin|tin. for 0 < 1 < 1. A routine bootstrap argument shows that an vial solution tun © L™ of (1°) must be classical except at Ap : 2p NBNy, and furthermore up > Oy choice of Co. The earlier arguments then show tin > 0. Observe that there exists a Ay > 0 independent of n such that if A < Ap then uy = 0 by Lemma 1(e) and wn is thus bounded in this case. While, if o 0, by (5) we must have for any nontrivial solution ‘wsome point P € 9 such that t(P)2e>0 © With ¢ independent of t, and thus lltalloaq > Ko with Ko independent of nt Since O° is compactly embedded in C* for a > ao, we then have by a routine but lengthy argument using Degree Theory, [25], that (1**) — (2) has a positive solution in some annulus By ~ B, where By denotes the sphere of radius p in C* for some a>0. a ‘Theorem 1 Probie (1) — (2) has a positive solution wu. Proof Note that {i} are bounded in #1? 9 L> and thus we may assume ts > u weakly in Hand strongly in L¢ for any large & By (8), Un(P,) > ¢ and since we also may assume P, — P and thus %,(P,) —- U(P) by the molifer Propertios, we must have 0 < w nontrivial. Finally note that fj, aVu- Vg is an equivalent inner product to the usual one in H!? 0 {vlv = on 8p} and thus fovea ve~ J avn. te flee —aIn vet [ev (en —u)- Pp. ‘The first term converges to zero by Holder's Inequality since 0 < a < aq ta in Lf ‘for any large &, while the second term tends to zero by inner product equivalence. Since convergence of the other terms is immediate, we conclude that 1 is a positive (weak) solution of (1) ~ (2) in L®, and thus in C*(Q) for some a and classical where the coefficients are smooth, except on Ip MAN. o We now pass to the situation where g > 0 may actually vanish on parts of 2. We present the results assuming all coefficients to be smooth, since arguinents identical to Theorem 1 can then be used to extend the results to more general cases, Elementary examples show that now the existence of positive solutions depends on the relation betwoon the various coefficients in (1) - (2). We shall assume the following condition: Let K > 0 be given and consider the eigenvalue problem: ~VlaVe + bu] + 9K T(S5(2))w = e(K)hw @ with boundary conditions (2) and some é" < 6 where 6 is as given in the definition of U We recall that S5-(z) is the ball centered at 2 of radius &, 1(S»(2))(y) =1 if ¥ € Sy(2), = 0 otherwise, We then assume that there exists a K such that given any x € @ at which g(x) = 0, e(A) satisfies: e(0) <1 0, Ky > 0 independent of u such that fs, pyratt SK, at any maximum point P. Sot Z = (zlg(z) = 0}, Q = {x| dist (x, Z) > 60} with 6,6" such that 0 < 6) < & <6! + by <8" <6 and choose @ > 0 such that 9(z) > 0 for x in Q. Now if PQ then we proceed exactly as before. If P< %—@ then there exists y € Z such that ||P — yl| < &. Condition (8) implies the existence of some x € S(y) such that A> t{z). Since |j2— P| < 6 +69 <6” < 6 we obtain the result with = 6" ~ (6 +64). The rest of the proof is identical to the earlier one. o 8 The small & case We consitler here the ease of 6 small with g still allowed to vanish somewhere in 9, and smooth coefficients. Our starting points are the local problem: =¥laVu + by) Ih gulu o subject to conditions (2), and the fact that the degree is invariant under perturba- tion, We basically exploit the observation that T(z) and u(z) are near each other if 6 is sulficently small, The existence of a solution to (9) will then imply the existence of a solution to (1) We modify the definition of 7 as follows: If u € C*(@) then there exists @ C* extension of u, with [flonce*) < K'ulleagy- There are clearly infinitely many possible extensions, but here we select one such and then define & by we) = f Balenenay in this section, Note that this is different from the earlier W, unless Dirichlet con- ditions are chosen on the whole 89 and we set u* = 0 in IR” ~ 0. Unlike the earlier results, problem (9) - (2) can be considered by upper/lower solution arguments, but it is advantageous for our purposes to use only half of these arguments. Specifically, consider once again the related linear eigenvalue problem (4) with conditions (2), and again assume that condition (5) holds if K is large enough. It follows that we can find a value of K and a 8 > 0 such that for any constant z > 0 the problem: =VlaVur + bar] + Kaus = (1+ hey 94 (6-a=0 on Ay (10) on ®p0 W. Allegretto and P, Nistei has a positive solution w. This follows from the observation that for Cp sufficiently large the problem —VlaVu2 + ba] + Dog — (1+ 6)A + Colun =0 w= on 6p 8246-0 on ay has a solution 22 > 0 by coercivity of the left-hand side, Indeed w» > 0 in 7 since it cannot have a minimuun of zero in the interior by the Generalized Harnack’s Inequality while on the boundary either wy = 1 or without loss of generality bit > 0 On the other hand the equation ‘VlaV2n + Bzo] + og ~ (1 + Bhlzo = 0 with the same boundary conditions has a solution zy by our eigenvalue assumptions, But then we have ¢a(z —w2) > 0 and thus — wy > 0 by the same arguments as in Lemma O(a). We then need only note that wo; = 229. By the maximum principle, if z is large enough we have for the solution w; of (10): bm =VlaVu + bin] + gw? > hurr a9 +m =0 on Oy a >0 on Mp i.e, a1 is an upper solution to (9) — (2) ‘Theorem 8 If (5) holds then problem (1) - (2) has for 6 small enough a positive solution. Proof Note that (9) may be rewritten as ~VlaVu + bu] + cu = f(x,u) (uu) with a ¢ > 0 chosen large enough so that the left-hand side is coercive and f is monotone in t for u in the interval 0 > 0, such that w* > max(1n,u). By the usual Ioeal problem iteration procedure we construct a sequence {4n} such that wy =w* and un > int > max(us, ua). Furthermore, dy converges to a solution pointwise, so that we may conclude the existence of three solutions u1,tt2,13 with us 2 uy. But then we have from (9) that gus > gay and by the same arguments as in the Proof of Lemma 1(b) it follows that gus — gua somewhere in © and thus it follows that us =u and hence up = uy. If need be wwe replace w by w” and thus conclude without loss of generality that the (unique) for some 0 2¢ as K + 00 in (7), e(IC) converges to a value at least equal to the smaller of the three eigenvalues obtained by consicering, the three segments: (02-8), (fe, £2), (2+6, L).By symmetry and eigenvalue ‘monotonicity, this will be least for 2* = (0, £ +e —6). If we choose § > 5+e > 32 then the least eigenvalue willbe no less than wea sufficient requirement for the existence of a positive solution to (1) - (2) thus becomes: Lx Pele Observe that g does not appear in this condition. Its magnitude will have an effect con the size/shape of the solution u but not on its existence. Example 2 Assume now g > 0 in , that the coefficients are smooth and the boundary conditions are purely Dirichlet. To apply our results we need only show that the eigenvalue problem ~Vlavw + bi = Hi (0)hw with conditions (2) has least eigenvalue j11(0) <1, since a(K") > 1 for K large by Lemma 1(b). Wo apply Lemma 1(c) by comparing ,11(0) with 02(0) where this is given by: =VleVen] + [- Fav G+ La] = 00)m02 doB W, Allegretto and P. Nistel ‘That 11(0) <1 is thus ensured by showing the existence of a ws such that [atv lo where we recall 0 < a 0 solve in = Aw hw =0 1 on (69,)N2 w=0 on af. This is possible if 00 is regular enough and ¢ is small. Since Au — hu < 0 in 2, we then have: u c fp, ,¢p) for some constants 66> 0. Indeed, v exceeds —Az = u in Bg(P), 2 = 0 on Bs(P), for some small é independent of uP, ie. Pra focal zal a We can thus bound |jullz~ as before. Note also that if the coefficients are assumed regular near @{2, then the neigh- bourhood AV of the introduction need only be a neighbourhood of Op Oy. IF B/a is a gradient field, then this condition can be dropped altogether. ‘As a final observation, we point out that if the coefficients in (1) are subject to a finite jump at some point 25 € I’, where T' is a smooth surface in © then besides the C® continuity of w we may also conclude by (17, p. 222] that, the “flux matching” condition holds. For example, if a(z) has a jump at x9 then @ 8 |ay is continuous, where n here denotes the normal to T'in a chosen direction References [1] Allgretto, W. (2972) A comparison theorem for nonlinear operators, Annall Scuola Normale ‘Superiore Pisa XXV, 41-46 [2] Allegreto, W. [1987] On the principal eigenvalues of indefinite eliptic problemas, Math. Z. 105, 2095, |p] Allegretto, W. and Barabanov, A. [1996] Positisity of solutions of elliptic equations with nonlocal terms, Proc. Royal Soe. Edin. 126A, 643-63. [a] Allogreteo, W. and Mingareli, A. [1989] Boundary problems of the second order with an Indefinite weight-fanction, J. Reine Agnew. Math. 398, 1-24 [b] Allegretto, W. and Nistl,P. to appest] Zzistence ond optimal control for periadte porabolic ‘uations with nonlocal terms, LMLA. Journal of Mathematical Control and Information,“ W. Allogrotto and P. Nistsi 6] Binding, P, and Browne, PV. [1987] Spectral properties of two parameter eigenoalue problems, Proc. Royal Soe. Edin, 106A, 30.51 7 Calaina, A., Perelio, C. and Saldana, J. (1994] Non-toa! reaction-diffusion equations modeltiag predator-prey cocvolation, Publicacions Matennatiques 38, 315-425. {| Calsina, A. and Perello, C. [1005] Equations for biological evolution, Proc. Reyal Soc. Fain 125A, 999-958, 18] DeFigueiredo, D. G. and Mitdieri, F. [1986] 4 marimum principle for on eliptic aystem and ‘applications to semilinear problems, SIAM Journal Math, Ans. 17, 85-849, (10) Freeman, #. 1. and Takeuchi, Y. [1980] Globe stability and predator dynamics in a madet of prey dispersal in a patehy environment, Nonl. Anal TMA 18, 909-1002 [Aa] Freedman, H. 1 and Wu J. (1992) Steady state analysis in a model for population diffusion. in a malt-potch environment, Nonl. Anal. TMA 18, 517-542, [12] Freitas, P. and Sweers, G. [1904] Pocitiaty raults for a nonlocal elliptic equation, prepa [a3] Furter, J. and Grineld, M, [1989] Local ws. nonlocal énteractions sn population dynamics, J Math, Biol 27, 65-80, [14] Gilbarg, D. and ‘Trudinger, N, (1983) EUiptic Partie! Differential Equations of Second Onder, 2" Edition, Springer-Vetlag, New York. [15] Hemandes, J. 1986] Continuation and eomparison metheds for some nonlinear elliptic sys tems, in: Contributions to nonlinear partial diferential equations, Val IT (J. Dias and P. L. Lions, eds.) Longman, Essex, 137-14. [18] Krosnoselski, M.A. [1964] Positive Solutions of Operator Bguations, Noordboff, Groningen [17] Ladyzhenstaya, O. A. and Uraltseva, N.N. [1968] Linear and Quasiliear Elliptic Equations, Academie Press, New York and London, [18] Lieberman, G. ME [1992] Intermediate Sehauder Theory for second onder parabolic emuations IV: time irregularity and repadarity, Dit. Int. Eq, 8, 1219-1286, 19] Murty, MK. V. and Stampacchia, G. [1972] 4 earietionat equality with mized boundary conditions, laael J, Math. 18, 189-204, (20) Pao, C. V. [1992] Nontinear Parebolie and Eliptic Eguations Plenum Press, New York Lenhart, 8. and Protopopescn, V, [1994] Optimal contro! for pardbolc systems with competic tive interactions, Math, Meth, Appl Sci 17, 501-524, 22] Shamir, F. [1968 Reguarisation of mized second order boundary value problems, Iarael Math. 6, 151-168, 23] Troinniello, GME. {1987| Bliptic Diferential Equations end Obstacle Problems, Plenum Press, New York [24] Xie, H, [1995] L%-estimate to the mized boundary value problem for second onder elliptic equations and applications tn thermistor problems, Noul. Anal. TMA 24, 9-28 [25] Zeidler, E, [1986] Nonlinear Functional Analysis and sts Applications, Vol. Springer-Verlag, New York