Example: tourism industry

LECTURE 4 - math.ucdavis.edu

94 LECTURE 4 sturm -Liouville Eigenvalue ProblemsPossibly one of the most useful facts in mathematics is that a symmetric matrichas real eigenvalues and a set of eigenvectors that form an orthonormal basis. Thisproperty of symmetric matrices has a wide-ranging generalization to the spectralproperties of self-adjoint operators in a Hilbert space, of which the sturm -Liouvilleordinary differential operators are fundamental equations arise throughout applied mathematics. For example,they describe the vibrational modes of various systems, such as the vibrations ofa string or the energy eigenfunctions of a quantum mechanical oscillator, in whichcase the eigenvalues correspond to the resonant frequencies of vibration or energylevels. It was, in part, the idea that the discrete energy levels observed in atomicsystems could be obtained as the eigenvalues of a differential operator which ledSchr odinger to propose his wave problems arise directly as eigenvalue problems in one spacedimension.

LECTURE 4 Sturm-Liouville Eigenvalue Problems Possibly one of the most useful facts in mathematics is that a symmetric matric has real eigenvalues and a set of eigenvectors that form an orthonormal basis.

Tags:

  Lecture, Lecture 4, Sturm, Lecture 4 sturm

Information

Domain:

Source:

Link to this page:

Please notify us if you found a problem with this document:

Other abuse

Advertisement

Transcription of LECTURE 4 - math.ucdavis.edu

1 94 LECTURE 4 sturm -Liouville Eigenvalue ProblemsPossibly one of the most useful facts in mathematics is that a symmetric matrichas real eigenvalues and a set of eigenvectors that form an orthonormal basis. Thisproperty of symmetric matrices has a wide-ranging generalization to the spectralproperties of self-adjoint operators in a Hilbert space, of which the sturm -Liouvilleordinary differential operators are fundamental equations arise throughout applied mathematics. For example,they describe the vibrational modes of various systems, such as the vibrations ofa string or the energy eigenfunctions of a quantum mechanical oscillator, in whichcase the eigenvalues correspond to the resonant frequencies of vibration or energylevels. It was, in part, the idea that the discrete energy levels observed in atomicsystems could be obtained as the eigenvalues of a differential operator which ledSchr odinger to propose his wave problems arise directly as eigenvalue problems in one spacedimension.

2 They also commonly arise from linear PDEs in several space dimensionswhen the equations are separable in some coordinate system, such as cylindrical orspherical general form of the sturm -Liouville equation is an ODE foru(x) of theform( ) (pu ) +qu= ,p(x),q(x) are coefficient functions,r(x) is a weighting function (equal to onein the simplest case) and is an eigenvalue, or spectral, parameter. The ODE issupplemented by appropriate self-adjoint boundary simplest example of a sturm -Liouville operator is the constant-coefficientsecond-derivative operator, whose eigenfunctions are trigonometric functions. Manyother important special functions, such as Airy functions and Bessel functions, areassociated with variable-coefficient sturm -Liouville as we may expand a vector with respect to the eigenvectors of a symmet-ric matrix, we may expand functions in terms of the eigenfunctions of a regularSturm-Liouville operator; the expansion of periodic functions in Fourier series is feature that occurs for sturm -Liouville operators, which does not occurfor matrices, is the possibility of an absolutely continuous (or, for short, contin-uous) spectrum.

3 Instead of eigenfunction expansions, we then then get integraltransforms, of which the Fourier transform is an , more complicated spectral phenomena can also occur. For example,eigenvalues embedded in a continuous spectrum, singular continuous spectrum,and pure point spectrum consisting of eigenvalues that are dense in an interval (seeSection on the Anderson localization of waves in random media for an example).95961. Vibrating stringsConsider the vibrations of a string such as a violin string. We label material pointson the string by a Lagrangian coordinatea R; for example, we can defineaasthe distance of the point from some fixed point in a given reference configurationof the string. We denote the position of the material pointaon the string at timetby~r(a,t).Let 0(a) denote the mass-density of the string in the reference configuration,meaning that the mass of the part of the string withc a dis given by dc 0(a) assume that the mass of the string is conserved as it vibrates, in which case thedensity 0(a) in the reference configuration is independent of suppose that the only force exerted by one part of the string on another is atension force tangent to the string.

4 This assumption distinguishes a string from anelastic rod that resists bending. We also suppose, for simplicity, that no externalforces act on the contact force~Fexerted by the part of the string withb > aon the partwithb < ais then given by~F(a,t) =T(a,t)~t(a,t)whereT(a,t) is the tension in the string and~t=~ra|~ra|is the unit tangent vector. We assume that~ranever vanishes. The part of thestring withb < aexerts an equal and opposite force ~Fon the part of the stringwithb > s second law, applied to the part of the string withc a d, givesddt dc 0(a)~rt(a,t)da=~F(d,t) ~F(c,t).For smooth solutions, we may rewrite this equation as dc{ 0(a)~rtt(a,t) ~Fa(a,t)}da= this holds for arbitrary intervals [c,d], and since we assume that all functionsare smooth, we conclude that 0(a)~rtt(a,t) =~Fa(a,t).This equation expresses conservation of momentum for motions of the close the equation, we require a constitutive relation that relates the tensionin the string to the stretching of the string.

5 The local extension of the string fromits reference configuration is given bye(a,t) =|~ra(a,t)|.We assume thatT(a,t) =f(e(a,t),a)wheref(e,a) is a given function of the extensioneand the material 4. sturm -LIOUVILLE EIGENVALUE PROBLEMS97It follows that the position-vector~r(a,t) of the string satisfies the nonlinearwave equation( ) 0(a)~rtt(a,t) = a{f(|~ra(a,t)|,a)~ra(a,t)|~ra(a,t)|}. Equilibrium solutionsA function~r=~r0(a) is an exact, time-independent solution of ( ) if( )dda{f(|~r0a|,a)~r0a|~r0a|}= consider a solution such that the tangent vector of the string is in a constantdirection, say the~ may then use as a material coordinate the distanceaalong the string inthe equilibrium configuration, in which case( )~r0(a) =a~ ( ) and the corresponding extensione= 1, in ( ), we find that the tensionf(1,a) =T0is constant in equilibrium, as required by the balance of longitudinal Linearized equationFor small vibrations of the string about an equilibrium state, we may linearize theequations of motion.

6 We look for solutions of the form( )~r(a,t) =a~i+~r (a,t),where~r is a small perturbation of the equilibrium solution ( ). We decompose~r into longitudinal and transverse components~r (a,t) =x (a,t)~i+~r (a,t),where~i ~r = use ( ) in ( ), withe= 1 +x a+.., and Taylor expand the resultingequation with respect to~r . This gives 0~r tt= a{(T0+kx a) (1 x a)[(1 +x a)~i+~r a]}+..wherek(a) =fe(1,a). Linearizing the equation, and separating it into longitudinaland transverse components, we find that( ) 0x tt= (kx a)a, 0~r tt=T0~r we obtain decoupled equations for the longitudinal and transverse motions ofthe longitudinal displacement satisfies a one-dimensional wave equation of theform ( ). The density is given by the density in the reference configuration,and the stiffness by the derivative of the tension with respect to the extension; thestiffness is positive, and the equation is a wave equation provided that the tension inthe string increases when it is stretched.

7 In general, both coefficients are functionsofa, but for a uniform string they are the longitudinal mode, the stiffness constantT0for the transverse modeis necessarily constant. IfT0>0, meaning that the string is stretched, the trans-verse displacement satisfies a wave equation, but ifT0<0 it satisfies an elliptic98equation. As we explain in the following section, the initial value problem for suchPDEs is subject to an extreme instability. This is consistent with our experiencethat one needs to stretch a string to pluck Hadamard instabilityAs a short aside, we consider the instability of the initial value problem for an ellipticPDE, such as the one that arises above for transverse vibrations of a compressedelastic string. This type of instability arises in other physical problems, such as theKelvin-Helmholtz instability of a vortex sheet in fluid simplest case of the transverse equation in ( ) with constant coefficients,normalized to 0= 1,T0= 1, and planar motionsx=aandy=u(x,t), is theLaplace equation( )utt= ( ) has solutions( )u(x,t) =Aeinx+|n|tfor arbitraryn RandA C.

8 Since the equation is linear with real-coefficients,we may obtain real-valued solutions by taking the real or imaginary parts of anycomplex-valued solution, and we consider complex-valued solutions for solution in ( ) has modulus|u(x,t)|=|A|e|k|t. Thus, these solutionsgrow exponentially in time with arbitrarily large rates. (The solutions proportionaltoeinx |n|tgrow arbitrarily fast backward in time.)This behavior is a consequence of the invariance of ( ) under the rescalingsx7 ,t7 t. This scale-invariance implies that if there is one solution withbounded initial data and a nonzero growth rate, then we can obtain solutions witharbitrarily fast growth rates by rescaling the initial a result, solutions do not depend continuously on the initial data in anynorm that involves only finitely many derivatives, and the resulting initial valueproblem for ( ) is ill-posed with respect to such norms. For example, if u K(t) = max0 k Ksupx R kxu(x,t) andun(x,t) =e |n|1/2{einx nt+einx+nt},then for everyK Nwe have un K(0) 0asn ,but|un(x,t)| asn ift6= failure of continuous dependence leads to a loss of existence of example, the Fourier seriesf(x) = k= e |k|1/2eikxconverges to aC -function, but there is no solution of ( ) with initial datau(x,0) =f(x), ut(x,0) = 0in any time interval about 0, however 4.

9 sturm -LIOUVILLE EIGENVALUE PROBLEMS99It is possible to obtain solutions of ( ) for sufficiently good initial data, suchas analytic functions (which, from the Cauchy-Kovalewsky theorem, are given bylocally convergent power series). Such assumptions, however, are almost alwaystoo restrictive in applications. The occurrence of Hadamard instability typicallysignals a failure of the model, and means that additional stabilizing effects must beincluded at sufficiently short The one-dimensional wave equationConsider a uniform string with constant density 0and and constant , from ( ), longitudinal vibrations of the string satisfy the one-dimensionalwave equation( )utt=c20uxxPlanar transverse vibrations of a stretched string satisfy the same equation withc20=T0/ The d Alembert solutionThe general solution of ( ) is given by the d Alembert solution( )u(x,t) =F(x c0t) +G(x+c0t)whereF,Gare arbitrary functions. This solution represents a superposition of aright-moving traveling wave with profileFand a left-moving traveling wave follows that the solution of the Cauchy problem for ( ) with initial data( )u(x,0) =f(x), ut(x,0) =g(x)is given by ( ) withF(x) =12f(x) 12c0 xx0g( )d , G(x) =12f(x) +12c0 xx0g( )d.

10 Here,x0is an arbitrary constant; changingx0does not change the solution, itsimply transformsF(x)7 F(x) +c,G(x)7 G(x) cfor some Normal modesNext, consider a boundary value problem (BVP) for ( ) in 0 x Lwithboundary conditions( )u(0,t) = 0, u(L,t) = BVP describes the vibration of a uniform string of lengthLthat is pinned atits look for separable solutions of the form( )u(x,t) = (x)e i twhere is a constant frequency and (x) is a function of the spatial variable real and imaginary parts of a complex-valued solution of a linear equation withreal coefficients are also solutions, so we may recover the real-valued solutions fromthese complex-valued functionu(x,t) in ( ) satisfies ( ), ( ) if (x) satisfies +k2 = 0, (0) = 0, (L) = 0100where the prime denotes a derivative with respect tox, andk2= spectral problem = , (0) = 0, (L) = 0has a point spectrum consisting entirely of eigenvalues( ) n= 2n2L2forn= 1,2,3,..Up to an arbitrary constant factor, the corresponding eigenfunctions n L2[0,L]are given by( ) n(x) = sin(n xL).


Related search queries