{
  "id": "sturm-liouville-theory",
  "title": "Sturm–Liouville Theory",
  "lang": "en",
  "source_lang": "en",
  "summary": "Sturm–Liouville theory studies self-adjoint second-order differential eigenvalue problems, their spectra, eigenfunctions, and applications to mathematical physics.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Sturm–Liouville spectral theory"
  ],
  "generated_at": "2026-11-05",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/sturm-liouville-theory/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/sturm-liouville-theory/",
      "json": "https://aiwiki.page/zh/sturm-liouville-theory.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Sturm–Liouville theory is the study of a class of second-order linear ordinary differential equations together with boundary conditions that turn them into spectral problems. Its central equation is\n\n\\[-\\frac{d}{dx}\\left(p(x)\\frac{dy}{dx}\\right)+q(x)y =\\lambda w(x)y,\\]\n\nwhere the coefficient functions are prescribed, \\(w\\) is a positive weight, and the parameter \\(\\lambda\\) is to be determined. Values of \\(\\lambda\\) admitting nonzero admissible solutions are eigenvalues; the corresponding solutions are eigenfunctions. Under suitable assumptions, these eigenfunctions provide an orthogonal coordinate system for functions, much as eigenvectors do in finite-dimensional linear algebra.\n\nThe theory explains why Fourier-type expansions arise in vibration, heat conduction, wave propagation, and quantum mechanics. It connects local properties of a differential equation—such as the spacing of zeros—with global properties of its spectrum. Regular problems on finite intervals have discrete spectra and complete eigenfunction expansions; singular problems on infinite intervals or with degenerate coefficients can also have continuous spectra.\n\nThe subject emerged from work by Charles-François Sturm and Joseph Liouville in the 1830s, particularly their papers of 1836–1837. Their analysis extended the separation-of-variables methods associated with Joseph Fourier beyond explicitly solvable equations. During the twentieth century, functional analysis and operator theory placed these results within the general theory of self-adjoint operators, while preserving the special role of zeros, boundary conditions, and one-dimensional geometry."
    },
    {
      "heading": "Differential expression, weight, and regular problems",
      "text": "A Sturm–Liouville problem consists of a differential expression, an interval, a weight, and boundary conditions. A standard notation is\n\n\\[\\ell[y]=-(py')'+qy,\\qquad Ay=\\frac{1}{w}\\ell[y],\\]\n\nso that the eigenvalue equation becomes \\(Ay=\\lambda y\\). The minus sign in the principal term is conventional: with \\(p>0\\), it makes the derivative contribution to the associated energy nonnegative. The coefficient \\(q\\) acts as a potential or distributed restoring term, while \\(w\\) determines the measure used to compare functions.\n\nFor a classical regular problem, the interval \\([a,b]\\) is finite, \\(p\\) is continuously differentiable and positive, and \\(q,w\\) are continuous and real-valued, with \\(w>0\\). These assumptions are convenient rather than optimal. A broader regular theory allows measurable coefficients provided \\(1/p\\), \\(q\\), and \\(w\\) are integrable on the finite interval, with \\(p,w>0\\) almost everywhere. In that setting, the quasi-derivative \\(py'\\), rather than \\(y'\\) alone, is the natural quantity in endpoint conditions and differential identities.\n\nThe underlying space is the weighted Hilbert space\n\n\\[L^2_w(a,b)= \\left\\{f:\\int_a^b |f(x)|^2w(x)\\,dx<\\infty\\right\\},\\]\n\nwith inner product\n\n\\[\\langle f,g\\rangle_w =\\int_a^b f(x)\\overline{g(x)}w(x)\\,dx.\\]\n\nThe weight is therefore structural, not merely an extra coefficient in the equation. Two eigenfunctions may be orthogonal in this weighted space without being orthogonal for ordinary unweighted integration. In a vibrating string, for example, \\(w\\) represents linear mass density, so the weighted norm measures displacement relative to the distribution of mass.\n\nMany equations not initially written in divergence form can be converted into it. Given\n\n\\[a(x)y''+b(x)y'+[c(x)+\\lambda r(x)]y=0, \\qquad a(x)>0,\\]\n\nchoose\n\n\\[p(x)=\\exp\\left(\\int^x\\frac{b(s)}{a(s)}\\,ds\\right).\\]\n\nMultiplication by \\(p/a\\) yields\n\n\\[-(py')'-\\frac{pc}{a}y =\\lambda\\frac{pr}{a}y.\\]\n\nThus \\(q=-pc/a\\) and \\(w=pr/a\\). This is a positive-weight Sturm–Liouville equation when \\(r/a>0\\).\n\nThe differential equation alone does not determine the spectrum. On a finite interval, a second-order equation generally has a two-dimensional solution space for every fixed \\(\\lambda\\). Boundary conditions select the solutions that belong to the operator domain. Different choices at the same endpoints can change the lowest eigenvalue, introduce or remove a zero mode, and alter eigenvalue multiplicities."
    },
    {
      "heading": "Boundary conditions and self-adjointness",
      "text": "The mechanism behind Sturm–Liouville theory is integration by parts. For sufficiently regular functions \\(f\\) and \\(g\\), the Lagrange identity gives\n\n\\[\\langle Af,g\\rangle_w-\\langle f,Ag\\rangle_w = \\left[p\\left(f\\overline{g'}-f'\\overline g\\right)\\right]_a^b.\\]\n\nThe expression on the right is the boundary form. A differential expression with real coefficients has the appropriate formal symmetry, but a self-adjoint operator arises only after its domain has been specified so that the boundary form vanishes and the domain agrees with that of its adjoint.\n\nFor a regular problem, the most familiar choices are separated boundary conditions:\n\n\\[\\alpha_a y(a)+\\beta_a p(a)y'(a)=0,\\]\n\n\\[\\alpha_b y(b)+\\beta_b p(b)y'(b)=0,\\]\n\nwhere the coefficients are real and neither pair is identically zero. Each endpoint is treated independently. Dirichlet conditions prescribe \\(y=0\\); Neumann conditions prescribe \\(py'=0\\); Robin conditions impose a real linear relation between displacement and flux. At a regular endpoint with \\(p>0\\), \\(py'=0\\) is equivalent to \\(y'=0\\).\n\nBoundary conditions can also couple the two endpoints. Periodic conditions,\n\n\\[y(a)=y(b),\\qquad p(a)y'(a)=p(b)y'(b),\\]\n\nand antiperiodic conditions, with a minus sign in both relations, are self-adjoint. Their spectral behavior differs from that of separated conditions: periodic eigenvalues need not be simple. More generally, self-adjoint boundary conditions correspond to maximal subspaces of endpoint data on which the boundary form vanishes. This formulation classifies admissible domains without privileging a particular coordinate representation.\n\nA basic example illustrates why the boundary conditions matter. For \\(A=-d^2/dx^2\\) on \\([0,L]\\), Dirichlet conditions at both ends give\n\n\\[\\lambda_n=\\left(\\frac{n\\pi}{L}\\right)^2, \\qquad n=1,2,\\ldots.\\]\n\nNeumann conditions give the same positive eigenvalues but also \\(\\lambda_0=0\\), whose eigenfunctions are constant. A Dirichlet condition at one end and a Neumann condition at the other instead produce\n\n\\[\\lambda_n=\\left(\\frac{(n+\\tfrac12)\\pi}{L}\\right)^2, \\qquad n=0,1,\\ldots.\\]\n\nThe local equation is identical in all three cases.\n\nSelf-adjointness guarantees real eigenvalues and orthogonality of eigenspaces belonging to distinct eigenvalues. It also supplies the framework for spectral expansions and evolution equations. Formal symmetry alone is insufficient, particularly at singular endpoints, where square-integrability and endpoint behavior must be incorporated into the domain."
    },
    {
      "heading": "Discrete spectra and eigenfunction expansions",
      "text": "A regular self-adjoint Sturm–Liouville operator has a spectrum consisting entirely of real eigenvalues with finite multiplicity. They are bounded below, have no finite accumulation point, and tend to \\(+\\infty\\). With separated boundary conditions, every eigenvalue is simple, so they can be arranged as\n\n\\[\\lambda_0<\\lambda_1<\\lambda_2<\\cdots, \\qquad \\lambda_n\\longrightarrow+\\infty.\\]\n\nFor coupled conditions, the ordering must allow repeated eigenvalues. A key functional-analytic fact is that the regular problem has compact resolvent: away from the spectrum, its inverse maps bounded sets into relatively compact sets.\n\nOrthogonality follows directly from the differential identity. If \\(Ay_m=\\lambda_m y_m\\) and \\(Ay_n=\\lambda_n y_n\\), self-adjoint boundary conditions imply\n\n\\[(\\lambda_m-\\lambda_n) \\int_a^b y_m(x)\\overline{y_n(x)}w(x)\\,dx=0.\\]\n\nThus distinct eigenvalues yield weighted-orthogonal eigenfunctions. After normalization, these eigenfunctions form an orthonormal basis of \\(L^2_w(a,b)\\). This completeness statement is stronger than orthogonality: no nonzero square-integrable function is orthogonal to every eigenfunction.\n\nFor normalized eigenfunctions \\(\\phi_n\\), any \\(f\\in L^2_w(a,b)\\) has the expansion\n\n\\[f=\\sum_{n=0}^{\\infty}c_n\\phi_n, \\qquad c_n=\\int_a^b f(x)\\overline{\\phi_n(x)}w(x)\\,dx,\\]\n\nwith convergence in the weighted mean-square norm. Parseval’s identity reads\n\n\\[\\int_a^b |f(x)|^2w(x)\\,dx =\\sum_{n=0}^{\\infty}|c_n|^2.\\]\n\nThis generalizes ordinary Fourier series, whose sine and cosine functions arise from constant-coefficient Sturm–Liouville problems.\n\nMean-square convergence does not automatically imply pointwise convergence or uniform convergence. Stronger conclusions require additional regularity of the coefficients and the expanded function. Endpoint compatibility also matters. For instance, a uniformly convergent series of Dirichlet eigenfunctions must vanish at the endpoints, so a continuous function with nonzero endpoint values cannot have such a uniformly convergent expansion on the entire closed interval.\n\nThe expansion also solves nonhomogeneous boundary-value problems. If \\(\\lambda\\) is not an eigenvalue, then\n\n\\[(A-\\lambda)u=f\\]\n\nhas the spectral solution\n\n\\[u=\\sum_n \\frac{\\langle f,\\phi_n\\rangle_w}{\\lambda_n-\\lambda}\\phi_n.\\]\n\nAt an eigenvalue, solvability requires \\(f\\) to be orthogonal to the associated eigenspace, and any solution is determined only up to addition of an eigenfunction. This is the concrete Sturm–Liouville form of the Fredholm alternative and explains the appearance of resonance in forced systems."
    },
    {
      "heading": "Zeros, oscillation, and comparison",
      "text": "One distinctive feature of Sturm–Liouville theory is that spectral information can be read from zeros of solutions. Under regularity assumptions ensuring uniqueness, a nontrivial solution cannot satisfy both \\(y(x_0)=0\\) and \\(p(x_0)y'(x_0)=0\\) at an interior point. Consequently, its interior zeros are simple. They cannot accumulate at an interior point, because an accumulation would force both the solution and its derivative to vanish there.\n\nFor a regular problem with separated self-adjoint boundary conditions, the eigenfunction associated with \\(\\lambda_n\\), indexed from \\(n=0\\), has exactly \\(n\\) zeros in \\((a,b)\\). This is Sturm’s oscillation theorem. The first eigenfunction can therefore be chosen strictly positive in the interior, while higher eigenfunctions have successively more nodal intervals. The ordering by zeros is exact even when neither the eigenvalues nor eigenfunctions can be expressed using elementary functions.\n\nSturm’s separation theorem concerns two linearly independent real solutions of the same equation at the same spectral parameter: their zeros interlace. Between consecutive zeros of one solution lies exactly one zero of the other. A related comparison theorem concerns different equations. For\n\n\\[u''+Q_1(x)u=0,\\qquad v''+Q_2(x)v=0,\\]\n\nwith \\(Q_2\\ge Q_1\\), the second equation is at least as oscillatory as the first, with a strict comparison conclusion when the coefficient inequality is strict somewhere in the interval under consideration. Increasing the effective oscillatory coefficient generally decreases the distance between successive zeros.\n\nThe Prüfer transformation translates zero counting into the evolution of an angle. One may write\n\n\\[y=r\\sin\\theta,\\qquad py'=r\\cos\\theta,\\]\n\nwith a continuous phase \\(\\theta\\) and \\(r>0\\). Direct substitution gives\n\n\\[\\theta' =\\frac{1}{p}\\cos^2\\theta +(\\lambda w-q)\\sin^2\\theta.\\]\n\nZeros of \\(y\\) occur when \\(\\theta\\) is an integer multiple of \\(\\pi\\). At such points, \\(\\theta'=1/p>0\\), so the phase crosses the corresponding level in a definite direction. Boundary conditions become conditions on the initial and final phase.\n\nOscillation theory makes eigenvalue ordering both qualitative and computational. A computed solution with an incorrect number of zeros cannot represent the proposed eigenmode. Comparison with equations having constant or simpler coefficients gives bounds on zero spacing and eigenvalues. Extensions to singular problems relate the accumulation of zeros near an endpoint to spectral thresholds, although the elementary finite-interval statement must then be modified to account for endpoint behavior and possible continuous spectrum."
    },
    {
      "heading": "Variational principles and high-energy asymptotics",
      "text": "Eigenvalues also arise as stationary values of energy quotients. For Dirichlet conditions, the basic quadratic form is\n\n\\[\\mathfrak a[y] =\\int_a^b\\left(p|y'|^2+q|y|^2\\right)\\,dx,\\]\n\nand the corresponding Rayleigh quotient is\n\n\\[R[y]= \\frac{\\mathfrak a[y]} {\\displaystyle\\int_a^b w|y|^2\\,dx}.\\]\n\nIts admissible functions have sufficient weak differentiability and satisfy the prescribed Dirichlet conditions. The natural setting is an appropriate Sobolev space. Varying the quotient leads to the Sturm–Liouville equation through the Euler–Lagrange equations.\n\nFor the Robin conditions\n\n\\[p(a)y'(a)=h_a y(a),\\qquad p(b)y'(b)=-h_b y(b),\\]\n\nthe form contains endpoint terms:\n\n\\[\\mathfrak a[y] =\\int_a^b\\left(p|y'|^2+q|y|^2\\right)\\,dx +h_a|y(a)|^2+h_b|y(b)|^2.\\]\n\nThese signs follow from integration by parts. Robin conditions arise naturally from the form rather than being imposed on every trial function as derivative constraints. Dirichlet conditions, by contrast, are imposed directly on the form domain.\n\nThe lowest eigenvalue is the minimum of \\(R[y]\\) over nonzero admissible functions. More generally, the min–max principle states\n\n\\[\\lambda_n = \\min_{\\dim S=n+1}\\ \\max_{0\\ne y\\in S}R[y].\\]\n\nEquivalently, the \\(n\\)-th eigenvalue can be obtained by minimizing over functions orthogonal to the first \\(n\\) eigenfunctions. These principles provide bounds without solving the differential equation. If \\(q\\ge0\\) and the boundary terms are nonnegative, all eigenvalues are nonnegative. Raising \\(q\\) while keeping \\(p,w\\), and the boundary conditions fixed cannot lower any eigenvalue.\n\nAt high energy, the principal coefficients determine the leading spectral density. Define the effective length\n\n\\[L_{\\mathrm{eff}} =\\int_a^b\\sqrt{\\frac{w(x)}{p(x)}}\\,dx.\\]\n\nFor regular problems under standard smoothness assumptions,\n\n\\[N(\\Lambda)\\sim \\frac{L_{\\mathrm{eff}}}{\\pi}\\sqrt{\\Lambda}, \\qquad \\Lambda\\to+\\infty,\\]\n\nwhere \\(N(\\Lambda)\\) counts eigenvalues below \\(\\Lambda\\), including multiplicity. Equivalently, \\(\\lambda_n\\sim(n\\pi/L_{\\mathrm{eff}})^2\\), with the indexing shift irrelevant at leading order.\n\nThe Liouville transformation explains this effective length. With\n\n\\[t=\\int_a^x\\sqrt{\\frac{w(s)}{p(s)}}\\,ds, \\qquad u(t)=[p(x)w(x)]^{1/4}y(x),\\]\n\nsufficiently smooth positive coefficients transform the equation into\n\n\\[-u''(t)+Q(t)u(t)=\\lambda u(t).\\]\n\nThe transformation preserves the weighted \\(L^2\\) norm. It converts varying principal coefficients into a transformed potential and shows why the leading eigenvalue growth resembles that of a constant-coefficient problem on an interval of length \\(L_{\\mathrm{eff}}\\)."
    },
    {
      "heading": "Singular endpoints and continuous spectrum",
      "text": "A Sturm–Liouville problem is singular when an endpoint is infinite or fails the regular endpoint conditions. Singularity can result from a vanishing principal coefficient, a nonintegrable weight, or a singular potential. The principal issue is no longer merely which boundary condition is prescribed: it is whether a boundary condition is needed at all and how admissibility is determined by the weighted space.\n\nHermann Weyl’s limit-point/limit-circle classification addresses this question. For a nonreal spectral parameter, a singular endpoint is limit-circle if every local solution is square-integrable near it with respect to \\(w(x)\\,dx\\). It is limit-point otherwise. In the second-order positive-weight setting, a limit-circle endpoint requires one boundary condition for a self-adjoint realization; a limit-point endpoint requires none. At a singular endpoint, such conditions are generally expressed through limiting boundary forms, because ordinary endpoint values may not exist.\n\nConsider\n\n\\[-y''=\\lambda y\\]\n\non \\((0,\\infty)\\), with \\(y(0)=0\\). Infinity is limit-point, so there is no independent boundary condition there. The self-adjoint operator has continuous spectrum \\([0,\\infty)\\), not a sequence of square-integrable eigenfunctions. Oscillatory solutions such as \\(\\sin(kx)\\) are generalized eigenfunctions. The expansion is a Fourier sine integral, related to the Fourier transform, rather than a discrete series.\n\nSingularity does not itself imply continuous spectrum. The harmonic-oscillator operator\n\n\\[-\\frac{d^2}{dx^2}+x^2\\]\n\non the whole real line has purely discrete eigenvalues \\(2n+1\\), for \\(n=0,1,\\ldots\\). Its potential confines the states strongly enough to produce compact resolvent despite the unbounded interval. By contrast, the free operator on the real line has continuous spectrum. The behavior of the coefficients at infinity is therefore decisive.\n\nA useful local model is\n\n\\[-y''+\\frac{g}{x^2}y=\\lambda y\\]\n\nnear \\(x=0\\), with weight \\(1\\). For real \\(g\\), the endpoint is limit-point when \\(g\\ge3/4\\) and limit-circle when \\(g<3/4\\). The threshold follows from the square-integrability of the leading power-law solutions. It is distinct from the lower-boundedness threshold \\(g=-1/4\\): endpoint classification and boundedness below answer different questions.\n\nThe general spectral theorem replaces the purely discrete expansion by a spectral measure. A singular operator may have discrete eigenvalues, absolutely continuous spectrum, singular continuous spectrum, or combinations of these, depending on its coefficients and domain. Classical regular theory is the especially transparent case in which this measure consists entirely of point masses."
    },
    {
      "heading": "Special functions and orthogonal polynomials",
      "text": "Many classical special functions arise from Sturm–Liouville equations. Their differential equations, orthogonality relations, and expansion formulas are different aspects of the same spectral structure. The relevant weight can often be identified by rewriting the differential equation in divergence form. Endpoint behavior then distinguishes the admissible solutions from other members of the two-dimensional local solution space.\n\nLegendre’s equation has the form\n\n\\[-\\frac{d}{dx}\\left((1-x^2)y'\\right)=\\lambda y, \\qquad -1<x<1.\\]\n\nHere \\(p=1-x^2\\) and \\(w=1\\). The endpoints are singular because \\(p\\) vanishes. Selecting the regular, nonlogarithmic solution at each endpoint gives eigenvalues\n\n\\[\\lambda_n=n(n+1)\\]\n\nand eigenfunctions \\(P_n(x)\\), the Legendre polynomials. Their orthogonality is\n\n\\[\\int_{-1}^{1}P_m(x)P_n(x)\\,dx=0 \\qquad(m\\ne n).\\]\n\nIn spherical geometry, this equation describes the axisymmetric angular part of the Laplace equation.\n\nThe Bessel equation can be written\n\n\\[-(xy')'+\\frac{\\nu^2}{x}y=\\lambda x y.\\]\n\nThus \\(p=x\\), \\(q=\\nu^2/x\\), and \\(w=x\\). On a radial interval \\((0,R)\\), an admissibility condition at the origin together with \\(y(R)=0\\) selects Bessel functions \\(J_\\nu\\) and eigenvalues\n\n\\[\\lambda_n=\\left(\\frac{j_{\\nu,n}}{R}\\right)^2,\\]\n\nwhere \\(j_{\\nu,n}\\) is the \\(n\\)-th positive zero of \\(J_\\nu\\). The weight \\(x\\) is the radial area factor in two dimensions. This is why Bessel orthogonality uses \\(x\\,dx\\), rather than \\(dx\\).\n\nHermite and Laguerre polynomials provide examples on unbounded intervals. Hermite’s equation becomes\n\n\\[-\\left(e^{-x^2}y'\\right)' =\\lambda e^{-x^2}y, \\qquad -\\infty<x<\\infty,\\]\n\nwith polynomial eigenfunctions at \\(\\lambda_n=2n\\). For \\(\\alpha>-1\\), the Laguerre equation becomes\n\n\\[-\\left(x^{\\alpha+1}e^{-x}y'\\right)' =\\lambda x^\\alpha e^{-x}y, \\qquad 0<x<\\infty,\\]\n\nwith the classical endpoint realization giving \\(\\lambda_n=n\\). These are weighted polynomial problems; after a change of dependent variable, some can also be interpreted as Schrödinger equations.\n\nThe connection with orthogonal polynomials is precise but not universal. Classical families such as Jacobi, Laguerre, and Hermite polynomials satisfy second-order differential equations of Sturm–Liouville type. A general orthogonal-polynomial system need not do so. Orthogonality alone therefore does not imply a second-order differential eigenvalue equation. The classical families are distinguished by having both structures, which account for their recurrence relations, differential identities, and effectiveness in approximation methods."
    },
    {
      "heading": "Separation of variables and physical models",
      "text": "Sturm–Liouville problems commonly arise by separating variables in a partial differential equation. For a string with tension \\(T(x)>0\\), linear density \\(\\rho(x)>0\\), and a distributed restoring coefficient \\(k(x)\\), a linear transverse-displacement model is\n\n\\[\\rho(x)u_{tt} =\\frac{\\partial}{\\partial x}\\left(T(x)u_x\\right)-k(x)u.\\]\n\nSeeking a normal mode \\(u(x,t)=y(x)e^{i\\omega t}\\) yields\n\n\\[-(Ty')'+ky=\\omega^2\\rho y.\\]\n\nThe squared angular frequency \\(\\omega^2\\) is the spectral parameter, and mass density supplies the weight.\n\nFor a uniform string of length \\(L\\) fixed at both ends, \\(T\\) and \\(\\rho\\) are constant and \\(k=0\\). Its frequencies are\n\n\\[\\omega_n=\\frac{n\\pi}{L}\\sqrt{\\frac{T}{\\rho}}, \\qquad n=1,2,\\ldots.\\]\n\nThe integral-multiple pattern follows from the equally spaced wave numbers of the Dirichlet problem. With variable density or tension, the modes remain weighted-orthogonal under the appropriate assumptions, but the frequencies need not be integer multiples of the fundamental frequency. Oscillation theory still orders them by the number of interior nodes.\n\nFor diffusion, consider\n\n\\[w(x)u_t=(p(x)u_x)_x-q(x)u.\\]\n\nExpansion in normalized eigenfunctions reduces the equation to independent scalar equations:\n\n\\[u(x,t)=\\sum_n c_n e^{-\\lambda_n t}\\phi_n(x).\\]\n\nPositive eigenvalues produce decay; a zero eigenvalue gives a persistent component. In the insulated one-dimensional heat equation with no loss term, the constant eigenfunction represents the equilibrium temperature, and its coefficient is determined by the conserved weighted mean. Negative eigenvalues, when permitted by the coefficients or boundary conditions, instead produce growing modes.\n\nIn quantum mechanics, the one-dimensional stationary Schrödinger equation is\n\n\\[-\\frac{\\hbar^2}{2m}\\psi''+V(x)\\psi=E\\psi.\\]\n\nIt is a Sturm–Liouville problem with \\(p=\\hbar^2/(2m)\\), \\(q=V\\), and \\(w=1\\). Self-adjointness supplies real energies and unitary time evolution. In radial problems, changing from the physical radial wavefunction to a reduced radial function often removes the geometric weight while introducing an inverse-square centrifugal term. The condition at the origin must be consistent with the original physical domain.\n\nSeparation also explains why Bessel functions appear in cylindrical problems and Legendre functions in spherical problems involving the Laplace operator. The weight records the geometry of the coordinate system as well as material properties. Not all vibration models belong to this second-order theory: Euler–Bernoulli beam equations, for example, involve fourth-order spatial operators and require a broader self-adjoint boundary-value framework."
    },
    {
      "heading": "Green functions and numerical computation",
      "text": "For a regular self-adjoint problem and a spectral parameter \\(\\lambda\\) outside the spectrum, the resolvent can be represented by a Green function:\n\n\\[u(x)=\\int_a^b G_\\lambda(x,\\xi)f(\\xi)w(\\xi)\\,d\\xi.\\]\n\nThe kernel satisfies\n\n\\[\\bigl(\\ell_x-\\lambda w(x)\\bigr)G_\\lambda(x,\\xi) =\\delta(x-\\xi),\\]\n\ntogether with the prescribed boundary conditions in the \\(x\\)-variable. It is continuous across \\(x=\\xi\\), while its flux derivative has the jump\n\n\\[p(\\xi)\\left[ \\partial_xG_\\lambda(\\xi+,\\xi) -\\partial_xG_\\lambda(\\xi-,\\xi) \\right]=-1.\\]\n\nFor separated conditions, let \\(u_L\\) satisfy the left boundary condition and \\(u_R\\) the right one. Their modified Wronskian\n\n\\[W_p=p(u_Lu_R'-u_L'u_R)\\]\n\nis constant. When \\(W_p\\ne0\\),\n\n\\[G_\\lambda(x,\\xi) =-\\frac{u_L(x_<)u_R(x_>)}{W_p},\\]\n\nwhere \\(x_<=\\min(x,\\xi)\\) and \\(x_>=\\max(x,\\xi)\\). Its denominator vanishes at eigenvalues. For real \\(\\lambda\\) outside the spectrum, self-adjointness gives the symmetry \\(G_\\lambda(x,\\xi)=\\overline{G_\\lambda(\\xi,x)}\\).\n\nNumerically, shooting methods integrate a solution satisfying one endpoint condition and vary \\(\\lambda\\) until the other condition is met. They are straightforward but can become poorly scaled when solutions contain exponentially growing and decaying components. Prüfer-angle shooting replaces amplitude matching by phase matching and naturally counts nodes. This helps distinguish neighboring eigenvalues and identify the mode being computed.\n\nDiscretization produces a generalized matrix eigenvalue problem\n\n\\[K\\mathbf v=\\lambda M\\mathbf v,\\]\n\nwith a symmetric stiffness matrix \\(K\\) and positive-definite mass matrix \\(M\\) when the discretization preserves the self-adjoint structure. The finite element method derives these matrices from the variational form. For a conforming trial space and exact integration, the Rayleigh–Ritz eigenvalues are upper bounds for the corresponding exact eigenvalues. Increasing the trial space improves these bounds through the min–max principle.\n\nThe finite difference method often gives a symmetric tridiagonal problem in one dimension. Counts of negative pivots in an appropriate shifted factorization determine how many discrete eigenvalues lie below a trial value, enabling reliable bisection. Spectral methods expand the solution in global functions, frequently classical polynomials, and can converge rapidly for smooth data. In every approach, eigenvalue accuracy depends on resolving coefficient variation, endpoint singularities, and oscillation; preserving symmetry and the correct weight is important for both numerical stability and faithful mode orthogonality."
    },
    {
      "heading": "Inverse spectral theory",
      "text": "The direct Sturm–Liouville problem starts with coefficients and boundary conditions and asks for the spectrum. The corresponding inverse problem starts with spectral observations and asks what they determine about the operator. This question is especially transparent for\n\n\\[-y''+q(x)y=\\lambda y\\]\n\non a known finite interval with specified boundary conditions. Eigenvalues can represent measured resonance frequencies, while additional spectral data encode the relative strengths or shapes of the modes.\n\nA single spectrum generally does not determine an arbitrary potential uniquely. An elementary ambiguity occurs for Dirichlet conditions at both ends: the potentials \\(q(x)\\) and \\(q(a+b-x)\\) have identical spectra because reflection maps one problem to the other. More general isospectral families also exist. The leading eigenvalue asymptotics recover an effective interval length, and finer asymptotics constrain integral features of the potential, but such information alone does not usually fix its pointwise form.\n\nUniqueness can be restored by augmenting the data. For a fixed normalization at one endpoint, eigenvalues together with suitable norming constants determine the potential under standard hypotheses. A norming constant may be defined as\n\n\\[\\alpha_n=\\int_a^b |y_n(x)|^2\\,dx,\\]\n\nwhere the eigenfunction \\(y_n\\) has a prescribed endpoint normalization rather than unit \\(L^2\\) norm. Two appropriate spectra, obtained by changing one endpoint condition while keeping the other fixed, can also determine the potential. Borg’s 1946 work established central uniqueness results of this kind and clarified why one spectrum is insufficient without additional assumptions.\n\nThe Weyl–Titchmarsh function packages spectral information into an analytic function of the spectral parameter. In a common normalization, it is obtained from the endpoint derivative-to-value ratio of the solution satisfying the condition at the opposite endpoint. Its poles and residues encode eigenvalues and norming information in the discrete case, while its boundary behavior describes continuous spectral data in singular problems. It therefore links direct spectral analysis with reconstruction.\n\nGel'fand–Levitan and Marchenko methods reconstruct differential operators through integral equations derived from spectral or scattering data. The precise formulation depends on whether the interval is finite, a half-line, or the whole real line. Reconstruction also requires a clear choice of which coefficients are unknown: when \\(p\\), \\(q\\), and \\(w\\) may all vary, changes of coordinate and normalization can produce equivalent descriptions. Uniqueness for a Schrödinger potential with fixed principal coefficient is therefore different from simultaneous recovery of every coefficient in a general Sturm–Liouville expression."
    }
  ],
  "links": [
    "ordinary-differential-equation",
    "linear-algebra",
    "quantum-mechanics",
    "joseph-fourier",
    "functional-analysis",
    "hilbert-space",
    "inner-product",
    "integration-by-parts",
    "self-adjoint-operator",
    "orthonormal-basis",
    "fourier-series",
    "uniform-convergence",
    "boundary-value-problem",
    "quadratic-form",
    "sobolev-space",
    "euler-lagrange-equations",
    "fourier-transform",
    "spectral-theorem",
    "orthogonal-polynomials",
    "partial-differential-equation",
    "heat-equation",
    "laplace-operator",
    "greens-function",
    "eigenvalues-and-eigenvectors",
    "finite-element-method",
    "finite-difference-method",
    "numerical-stability",
    "inverse-problem"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}