Está en la página 1de 15
Chapter 4 Qualitative Behavior of Solutions to Evolution Equations 41 INITIAL VALUE AND INITIAL-BOUNDARY VALUE PROBLEMS. Unlike elliptic PDEs, which describe a steady state, parabolic or hyperbolic evolution equations describe processes that are developing in time. For such an equation, thc initial state of the system is part of the auxiliary data for awell-posed problem. If the equation contains time derivatives up to order k, the initial state can be characterized by specifying the initial values of the unknown function and its time derivatives through order k ~ 1. EXAMPLE 4.1 ‘The heat equation serves as the canonical example of a parabolic evolution equation, Problems ‘which aro well-posod for the heat equstion will be well-posed for more general parabolic equations. (a) Well-Posed Initial Valve Problem (Cauchy Problem) =x Mule) in RY, 2>0 uCe0)= fe) in RY lu ISM in RY, 0 ‘The boundedness condition at infinity (which is not the most general condition possible) is independent of the spatial dimension (@) WellPosed Initial-Bonndary Value Problem uae Put) in, 0 u,Q=/@) ina teu) £0) Hs) en8 0000 Special values of a and # Iead to boundary conditions of Dirichlet or Neumann type (Section 1.3). If 0 is not bounded (e.g, a half-space), then g(x, 4) must be specified over the accessible portion of S and additional behavior-at-infinty conditions may be needed. EXAMPLE 4.2 “The wave equation serves asthe prototype for hyperbolic evolution equations. (@) WelhPosed Initial Value Problem (Cauchy Problem) m= eu) in, 1>0 (s,0)= f@) in” (8, 0)= 266) ia RY No bchavior-atnfnity conditions ace necessary in order to obtain a unique solution tothe Cauehy problems forthe wave equation. (©) Welt-Posed Initiat-Boundary Value Problem maa) Inf, 20 x, 0)= f() and w(x, 0)= 262) in au extn, 9+ £8) (m9 (31) on S (> 0; af =0 8 may be unbounded, with no condition at infinity required. 36 See seat cuar. 4] SOLUTIONS TO EVOLUTION EQUATIONS 37 If the initial conditions in a well-posed initial value or initial-boundary value problom for an evolution equation are replaced by conditions on the solution at other than the initial time, the resulting problem may not be well-posed, even when the total number of auxiliary conditions is unchanged. EXAMPLE 4.3 (a) Backward Heat Equation Wa nuns) O05 G0 tang yl ton mek tata} EdD Results analogous to'Theorems 4.1 and 4,2 hold for parabolic equations more general than the heat equation Definition: Ifthe linear differential operator = S (1) a U1, S a, d i aa; is uniformly elliptic (Section 3.2) in 9 for each + in " 7, then the operator a] Fare a 2s esa ] (4.1) is said to be uniformly parabolic in 0x (0, T). ‘Theorem 43: Let © be a bounded region in R" with smooth boundary S, and suppose that &/4¢~L- uniformly parabolic in 2x0, T], with coefficients a, and }, continuous in {1x [0, T] and coefficient c=0. Suppose also that u(x,t) is continuous in Ox (0, 7). ‘Then the conclusions (), (i), Gil) of Theorem 4.1 bold, with VJ replaced by L{}. ‘Theorem 4.4: Theorem 4.2 remains valid when the operator 2/at— V" is replaced by the uniformly parabolic operator 4/at~L of Theorem 4.3. ‘The conclusions of Theorems 4:3 and 4.4 regarding the solution to u— L{u] = continue to hold if e(x, 50 and M=0—and even in another case (see Problem 4.21). 43. DIFFUSIONLIKE EVOLUTION (PARABOLIC PDEs) ‘Two propeities characterize the time-behavior of systems modeled by parabolic PDEs. To describe the second of these, we introduce the notion of an “evolution operator” that takes the’in state u(x, 0) of the system into the evolved state u(x, 1), Infinite speed of propagation. At any time t>0 (no matter how small), the solution to a parabolic initial value problem at an arbitrary location x depends on all of the initial data, (See Problem 4.7.) ‘As a consequence, the problem is well-posed only if behavior-at-infinity conditions are imposed. Smoothing action of the evolution operator. A solution u(x, 1) to the Cauchy problem for the heat equation is, for each x and all :>-0, infinitely differentiable with respect to both x and & (See Problem 48) ‘There is an interesting consequence of the smoothing property of the evolution operator for the heat equation, A sectionally continuous initial state w(x, 0) can always evolve forward in time in accordance with the heat equation. However, if it is not infinitely differentiable with respect to both x and 1, then it cannot have originated from an earlier state u(x, ), ¢<0, Thus the heat equation is, irreversible in the mathematical sense that “forward” time is distinguishable from “backward” time, Correspondingly, any physical process for which the heat equation is a mathematical model is inreversible in the sense of the Second Law of Thermodynamics | [Bae soit cama ea ha eR Ne ARR CHAP. 4) SOLUTIONS TO EVOLUTION EQUATIONS 39 In an initial-boundary value problem for a parabolic PDE, the solution will be smooth for all ¢>0 and all x inside the domain. In order for the solution to be smooth on the boundary and at , it is necessary to impose smoothness and compatibility conditions on the data. 1 EXAMPLE 4.4 The problem 3, )= tals) OS ESL, 10 QO) fla) O0, the solution to a hyperbolic initial value problem depends on only a portion of the initial data, The set of locations x such that the solution value assumed at x at time ¢= 0 affects the value of the solution at (x, f.) constitutes the domain of dependence for (Ky f). Foreach (xy, f) the domain of dependence is of finite extent; consequently, the initial value problem is well-posed without the specification of behavior at infinity. ‘The lack of simoothing action in wavelike evolution is related to the fact that physical processes modeled by the wave equation are thermodynamically reversible. An initial state u(x, 0) that is Jacking in smoothness can evolve forward in time in accordance with the wave equation and can, as, well as not, have originated from an carlier state, ‘Thus we can solve forward in time to find subsequent states into which u(x, 0) will evolve, or we can solve backward in time to find earlier states from which u(x, 0) has evolved. 40 41 42 SOLUTIONS TO EVOLUTION EQUATIONS (CHAP. 4 Solved Problems Prove Theorem 4.1. Suppose that u(x, y, 2,4) is continuous in £1 {0, 7] and satisfies wu Pus0 in OX Q,7) For €20, let ol 42 = uly ys 4 Ob e(A+ y+ 2%), Then, w-Fee-6e<0 in Ox (0,7) wo ‘Suppose now that v assumes its maximum value at (xo, oy 20 fi), Where (to, Yon 20) is an interior point of Q and 0< 55 7. Then, at (8, Yo, 20 fae w= and Vos which contradicts (1), lence, 6 saust assume its maximum value for (x, 5 2) on S and O=: £57, or else =O and (% y,2) in 0. ‘The definition of v implies that usv0 is arbitrary, i follows that 1 = M in & 0, 7]. Thus Theorem 4.19 is proved. “Theorem 4.1Gi) is proved by applying Theorem 4.1(i to the function —u; then ‘Theorems 4.1(2) and 4.1Gi) together imply Theorem 4-1(i). Let O, S, T be as previously described and consider the problem 4-Fu=O% 2) in Ox0,7) UG 920)" Fy 2) in us 292/029 in $XI0,T] ‘Show that if this problem has a solution 1 that is continuous ip & x [0, 7), then this solution is unique, Method 1 ‘The difference v(x, yz 8) 1% ¥ 2, Male Wz 4) of continuous solution tinuous and satisfies : 8 Hs itself con- a-Vu=0 i OxO.7) @ vGy%20)-0 in @ oles )20 in x0, 7) ° ‘Ms=0, and (3) implies ms=Mg=0. Hence, m= M=0, and, by For the function o, (2) iy ‘Theorem 4.1(ii), »=0. Method 2 For 05547, define the energy integral sepa oran heros hss tenon Met Chely,9*9 oy 2) 0-0. ton som paaiea | eee [e2tesa] oran 43 44 schiap. 4) SOLUTIONS TO EVOLUTION EQUATIONS rn where we have used (1) and (1.7). By (3), the boundary integral vanishes, leaving sig=-2f RoPan<0 6) “Thus J()is noninereasing, which fac, along with J()= 0 and J(@)= 0, implies that J(2)=0. But then, the integrand o(% 2 in (4) being nonnegative and continuous wth respect toallargumentsitollows that» is identically zero in 0 for 1=0. “The energy integral method may be extended to theeasc of Neumann or mixed boundary conditions, to which ‘Theorem 4.1 is not divecly applicable. See Problem 4.16. For the initielboundary value problem of Problem 4.2, lett and 1 denote solutions corresponding to data {%,, Ff} and {®,, Ff}, respectively. Suppose that @,=@, in 2x (0, T), F,=Fyin , ff; in 8X0, T}. Prove that u,= win Gx (0, 7). Letting = ik~ ta, We have a V v= -&s0 in 2X ,T) In addition, for the function v, My=max(Fi-R}=0 | Mg~ max (fim fi SO = sean whence M=0, Then Theorem 4.1@) implies that v0, or w= in x [0, 7}. For x, T' positive constants, suppose that v(x, 1) satisfies BAK, OSL OTST @ a )=0 O0 for £0 (whence g(t)> 0 for > 0), show that for 0= x= 1, 05 ¢ 7, () of )30 — &) 4G N=0 (©) Hale NSO (2) Apply Theorem 4:3 (or, after changing the time variable, Theorem 4.1) to (1) in order to conclude that - s&N=M 05x51, 05157 ‘The outward normal derivative of v at x=0 is ~v4(0, )=—g(0) <0; hence, by Theorem 44 (or ‘Theorem 42), the maximum M cannot occur at x= 0. The outward normal derivative at x™ 1 is tx(1, )=0; 80 the maximum cannot occur at x= 1 either. It follows that M = Ma=0, whence v@,)50 Osxslosr=T (©) Let ula, = oss, #), Differentiate (1) with respect to x, to find tur OSES, OST. o te fv is suficiontly smooth, 4, = Yr Wile OS KSI, O<1ET ©) tty and (5) becomes which is the seme PDE as is satisfied by v. ‘Theorem: 1f a function satisies a linear PDB with constant eveficien same PDE, auations (2), (7), and (4) imply the following auxiliary conditions for (6) its derivatives satisfy that 2 45 SOLUTIONS TO EVOLUTION EQUATIONS [cHap. 4 Hy 0)=0 OT ” uO,)=s) O0, 845 DSK YC NSO, for OSes) and Os 1s T: Making use of the results of Problem 44, plot v(x, 1) versus x for several values of 4 “Profiles” for three time values aro ploited in Fig. 4-1. the curves are bolow the x-axis, in accordance with Problem 44(a). On x0 we have 0, )=0 (Problem 4.4(@) and v4 (0, s(0); thus, the starting value is a negative, decreasing function of time, while the starting slope is a positive, inczeasing function of time. The curves are concave (Problem 4.4(¢), and have final slope 2010 a0 Inspection of Fig. 4-1 suggests that, for each fixed in (0, 7), 90) 50% 9s0,) Osxs1 \leed, this follows at once from Problem 4.4(0). and, 266 8) O<4shsHeT Fig. 4 CHAP. 4) SOLUTIONS TO EVOLUTION EQUATIONS B 46 — Let u(x,t) be a solution of the nonlinear problem UG, )= alu) O<¥<1, 0<00 for #>0. In addition, ‘assume that a(u) is continuous and satisfies 0e o rey={ (3 )>0 for all x and every postive & no matter how stall the positive umber, (a) For each #>0, »y exp >0 forall x sad y 44 48 SOLUTIONS TO EVOLUTION EQUATIONS ICHAP. It folfows from (1) that for cach 130 and all x the value u(x 0) incoxporates every F(y) ~wayee, (2) The change of variable (2 0,-20 for all real z, i follows that u(x, #) >0 for all x and & That is, the solution u(x, 1) is immediately positive everywhere, even though F is zero everywhere except in the arbitrarily small interval (6 ¢). "Note, however, that because eis monotone decreasing, (6) implies that wo aden eel ‘even though the influence of the initial state propagates with infinite speed, the strength of this influence dies out very rapidly (as ¢~") as the distance (r= |x~ el) from the set where F 0 increases. We ate therefore able to claim that although solutions to the heat equation exhibit a nonphysical property (infinite speed of propagation), they do behave in a manner that is an acceptable approximation of reality. Practically speaking, effects governed by the heat equation propagate with finite speed. For more on this matter, see Problem 7.8, Let f(z) denote a sectionally continuous function in (0, 7). Then, using separation of variables (Chapter 8; sce also Problem 4.18), one shows that ue = SfeMsinme 00 @) ux, )=f0) 00 @) provided =2f fe)sinnede (0 =1,2,...) © Demonstrate the smoothing action of the evolution operator in this case, AC #=0, (1) reduees to u(x, 0)= & fa sin x CHAP. 4) SOLUTIONS TO EVOLUTION EQUATIONS 45 and this series converges pointwise to F(z), the odd 2r-periodie extension of fle), provided F, and are sectionaily continuous (see Problem 6.1(6)), For each n= 1,2,... and for each positive & Use" sin mx] (constant) "=< (constant) (e~"" and the geometric series is convergent. It follows from the Weierstrass 24-test thal for cach fixed £>0, the series in (1) converges absolute and uniformly to» continzous function of The same can be sid ofthe series obtained from (1) by term-by-torm diferentiation aay numbcr of times with respect fo x andor t. We conclude thatthe series in (1) represents function which is not just continuous but is infinitely diferentiable with respect to both x and # for ¢>0 and 0~ 0, even it u(s,O)is not particularly smooth For parabolic equations having vibiable coefficients, the smoothing action ofthe evolution operator may be limited by a lack of smoothness inthe coefficients Show that the backward heat problem is ill-posed, as asserted in Example 43(a). For simplicity, choose « = 1 Write w(x) = 1,0), the intial state (temperature). Then (cf. Problem 4.8), the function ua =F uPersinnm Orel, Ore T a) where wpa u(s)sin neds (0=1,2,..,) @ will solve the problem, provided the tf? are such that fo)= Sue in nn 0x1 6 But the series in (3) converges uniformly to an infinitely differentiable function of x, whatever the uf. It follows that no solution exists when f(s) isnot infinitely diferentabe In the case where f(x) is infinitely diflerentiable, the solution does not depend continuously on the data. If, for instance, sin Nex N fe)= (v= integer) the unique solution to the problem is, eA Sin Nex 08 <1, OLS T uly For large N, on the one hand, f(x} becomes uniformly small; chat is, the data function differs by as litle 4s we wish from the data function f=0, to which corresponds the solution #=0, On the other hand, |uls, | grows with N; i., the solution docs not remain close to w=0. Thus, there is no continuity of dependence on the data. 46 SOLUTIONS TO EVOLUTION EQUATIONS [cHAP, 4 | 4.10 Show that the solution to WO D= au (1) >< x<, 10 a) U5 0)= FQ) ~e0; (b) if F and G bath vanish for jx]> 1, show that udtaQ=0 fordsts1 and interpret this result, (a) By (4) of Problem 4.10, ula, ) depends on the values of F for the two arguments x4 afp and and on the values of G over the interval (xy~ at e+ at). Thus, the domain of ce of (t,t) s the closed interval [29~ at, xo at, which is precisely the portion ofthe seaxis cut off by the two characteristics xtal=f x atm a that pass through (o,f). See Fig, 4-2, cHaP. 4] SOLUTIONS TO EVOLUTION EQUATIONS 47 ie eee ‘Domain of dependence Big. 42 (Because F and G each vanish outside [x] <1, u(l+ 1) must remain zero s0 long asthe domain of dependence ofthe point (1+ a, 1) [L+a~ at, +a + all remains disjoint from (~1, 1); thats so long tta-ar=l or = Ost Now, the distance from the point (14+ 4,0) to the interval (~4, 1) is just a units, Consequently, , our result may be interpreted to mean that the influence of the initial data requires just 1 unit of, {ime to traverse this distance; i. the propagation speed is a units of distance per unit of time, 4.12 Consider the following modification of the n-dimensional wave equation: 4 Tyg w ee ¢, denote real constants. Show that for an arbitrary function F in C* and an 2G )y Hi where ¢,. arbitrary unit vector a= (a, z us, )= Flax pt) @) satisfies (1), provided » satisfies wD ehal @) Substitute (2) in (1), to fine (@- Seiai)rten ayo Byldently, i 4 sutistes (3), then u(x, 1) as given by (2) isa solution of (1), with no further restrictions on. Forona, For each fixed ¢ ax j1(+ cons. is the equation of a plane in R* having normal vector ex For this reason, (2) i ealled a plane wave solution to (1), The function F is called the waveform and w represents the direction in which the wave progresses. While both F and e are arbitrary, the wave velocity ju depends on et via (3), Evidently, (1) models wave propagation in & nonisotropie medium. pose nr nmin SPARE ee eemnmnmtnaee noun 4B 4.13 414 SOLUTIONS TO EVOLUTION EQUATIONS [cHAP, 4 Consider the initial-boundary value problem gl = g(t) O<¥0 u(y 0)= FG) 00 Show by the energy integral method (cf. Problem 4.2) that if this problem has a solution, then the solution is unique. Let ms(x 0) and u(x) denote two C! solutions and let u(x £) = (x £)~ la) for OS ¥ 0- ‘Then v(4, 4) satisfies the initial-boundary value problem with F(x)= G(z)~0 for 0O w 0)" fa) O0 ) is given by Mos )= 1A G+ ath Fox al +3 i Gls) ds 6) (6) Relate the smoothness of the solution u(x, #) to the smoothness of the data f, g and to the compatibility between the initial data and the boundary conditions (4), CHAP. 4) SOLUTIONS TO EVOLUTION RQUATIONS 49 (@) According to Problem 4.10, (5) isthe solution of the following initial value problem: WO D= ims) eo x0. Thus ‘Cx £) 8 given by (5) satisfies (4) as well, In Problem 4.13 we proved that the problem (1) through (4 as at most one solution; therefore, (5) is the solution. (8) Differentiation of (5) gives ontu arar PPM et a) IONE all} (m,n=0,1,2,..). Evidently, the continuity of u(x 1) and its derivatives is determined by the smoothness of f, and G,, which, in turn, is dependent on the smoothness of f and g in [0 3), and the compatibility of f und g with the boundary conditions (4), In Problem 64 it will be shown that F.(x) and all its derivatives through order Mare for all x if and only if: {J43) and all its derivatives through order M are continuous on [0, x}; for all nonnegative integers m such that 2n 0, (i) or Gi) is not satisfied by both f and g, the solution ux, ¢) will ‘experience some sort of discontinuity slong a characteristic. For example, if g(0) #0, then G(x) is discontinuous at every integer multiple of m which means, by (6), that u(x, 1) and was 0) ‘experience discontinuities for (x, such that x2 al= kar (k= integer) continu Supplementary Problems 4.45 Determine the most general spherically symmetric solution to the three-dimensional wave equation, ya? Wu=0.[Hint: Find the PDE satisfied by o(y 2) re(y, 2.) 4.16 In Problem 42, let the boundary soncition be replaced by AG. DMs 75D PU KOTO YL D= fH 2) ia $x10,7) where the continuous functions a end f satisfy op=0 ah pP>0 in Sx {0, 7]. Prove uniqueness by the energy integral method, Hint 50 4a 4as 419 420 aan an 423 44 SOLUTIONS ‘TO EVOLUTION EQUATIONS IcHAP. 4 (a) Differentiate under the integral sign to verify that us =a Rope ay Viet satisfies = tes (6) Infer from (5) of Problem 4.7 that it F is continuous, n ula f)= FC). For N a positive integer, let uy Seer dnne d0 (0, )=u(i,)=0 120 {or all choices of the constants Cy. (@) Find plane wave solutions for Ug Gite + ality + Bu 0. For 7 a fixed positive parameter, let ve(x, 7) denote the solution of vee, N= Vols, xin > -@ v%2)=0 xin RY oy (6,7) Fy 7) xia RY @ Show that "owls, tr) dr @ satistes ls, = Pula DEF!) xin RS, 10 © tug, 0) = (8, 0)=0 io R™ o ‘This observation is known as Duhamel's principle. Derive a yersion of Duhsamel’s principle for the heat equation, ‘Use Duhamet's principle to solve unt fe) xin RY, 1>0 (x50) (2,0)=0 x in RE

También podría gustarte