{
  "id": "boundary-condition",
  "title": "Boundary Condition",
  "lang": "en",
  "source_lang": "en",
  "summary": "A boundary condition specifies constraints at a domain’s edge that help determine a mathematical solution and its physical interpretation.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Boundary constraint"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/boundary-condition/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/boundary-condition/",
      "json": "https://aiwiki.page/zh/boundary-condition.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "A boundary condition is a requirement imposed on an unknown quantity at the boundary of the region in which a mathematical problem is posed. It may prescribe the quantity’s value, its rate of change, a flux, or a relation among several quantities. Boundary conditions are central to differential equations, but also occur in variational problems, stochastic models, and discrete systems. An equation describes behavior within a domain; boundary conditions describe how that domain interacts with its surroundings or how its edges are constrained.\n\nThe same interior equation can describe very different systems when supplied with different boundary conditions. A vibrating string may have fixed or movable ends; a conducting body may have prescribed surface temperature or prescribed heat flux; a fluid may encounter a solid wall, an inlet, or an open outlet. Mathematically, the conditions help determine whether a solution exists, whether it is unique, how sensitively it depends on data, and which oscillation frequencies or equilibrium states are possible.\n\nBoundary conditions are therefore part of the definition of a mathematical model, not merely supplementary instructions for solving it. Their appropriate form depends on the differential operator, the geometry of the domain, the regularity sought for the solution, and the physical processes represented at the boundary."
    },
    {
      "heading": "Mathematical formulation and historical development",
      "text": "Let \\(\\Omega\\) be a region in Euclidean space, with boundary \\(\\partial\\Omega\\). A typical boundary-value problem has the form\n\n\\[Lu=f\\quad\\text{in }\\Omega, \\qquad Bu=g\\quad\\text{on }\\partial\\Omega,\\]\n\nwhere \\(L\\) is an interior differential operator and \\(B\\) is a boundary operator. The unknown \\(u\\) may be a scalar, a vector, or a collection of coupled fields. The data \\(f\\) and \\(g\\) represent, for example, internal sources and externally imposed surface values. Different boundary operators may act on different portions of the boundary, and some conditions connect values at separate locations rather than acting pointwise.\n\nFor an ordinary differential equation on an interval, the boundary consists of its endpoints. The equation\n\n\\[-u''(x)=f(x),\\qquad 0<x<L,\\]\n\nmay be supplemented by \\(u(0)=a\\) and \\(u(L)=b\\). In contrast, specifying \\(u(0)\\) and \\(u'(0)\\) gives an initial-value problem, even if the independent variable represents position rather than time. The distinction concerns where the data are prescribed: at one starting location or across separated boundaries. For partial differential equations, entire surfaces may carry boundary data, so the geometry and differential character of those surfaces become essential.\n\nBoundary conditions need not be local. A periodic condition identifies values at opposite sides of a cell, while an integral condition may prescribe a total current, average displacement, or aggregate flux. Conditions may also depend on time or on the unknown solution itself. A boundary operator can involve tangential derivatives, normal derivatives, or auxiliary variables governed by their own surface equations.\n\nThe historical development of boundary conditions followed problems in mechanics, gravitation, and heat conduction. Eighteenth-century investigations of vibrating strings related endpoint constraints to standing waves. Joseph Fourier’s Théorie analytique de la chaleur of 1822 systematically connected heat flow, surface conditions, and trigonometric expansions. Nineteenth-century potential theory distinguished prescribed potential from prescribed normal derivative, leading to the terminology associated with Dirichlet and Neumann problems. Robin conditions are named after the French mathematician Victor Gustave Robin.\n\nIn modern analysis, an equation’s operator is often defined together with its admissible functions and boundary conditions. Thus the expression \\(-u''\\) alone does not specify a complete operator on a function space. Fixed-end, periodic, and zero-derivative conditions produce different operator domains and different spectral properties. This operator viewpoint makes precise why changing the boundary can change the mathematical problem even when its interior equation remains unchanged."
    },
    {
      "heading": "Dirichlet, Neumann, Robin, and related conditions",
      "text": "A Dirichlet boundary condition prescribes the unknown function itself:\n\n\\[u=g\\quad\\text{on }\\partial\\Omega.\\]\n\nExamples include a specified electric potential on a conductor and a specified displacement on a restrained elastic body. The homogeneous case is \\(g=0\\). “Homogeneous” refers to the boundary data, not necessarily to the interior equation: a problem with nonzero internal forcing can still have homogeneous boundary conditions. Conversely, a source-free equation can have nonhomogeneous boundary data.\n\nA Neumann boundary condition prescribes a normal derivative or, more generally, a conormal flux. If \\(\\mathbf n\\) denotes the outward unit normal, then\n\n\\[\\frac{\\partial u}{\\partial n} =\\nabla u\\cdot\\mathbf n=g\\]\n\nis the simplest form. Here \\(\\nabla u\\) is the gradient. For diffusion through an anisotropic material with conductivity matrix \\(K\\), the physically relevant quantity is instead \\(\\mathbf n\\cdot K\\nabla u\\). The sign convention matters: if the physical flux is \\(\\mathbf q=-K\\nabla u\\), an outward flux \\(h\\) corresponds to \\(-\\mathbf n\\cdot K\\nabla u=h\\), not to \\(\\partial_nu=h\\).\n\nA Robin boundary condition combines value and normal derivative:\n\n\\[\\alpha u+\\beta\\frac{\\partial u}{\\partial n}=g.\\]\n\nIts coefficients may vary along the boundary. A common heat-transfer example is\n\n\\[-k\\frac{\\partial T}{\\partial n} =h(T-T_\\infty),\\]\n\nwhere \\(k\\) is thermal conductivity, \\(h\\) is a surface heat-transfer coefficient, and \\(T_\\infty\\) is the ambient temperature. This relation models finite resistance to heat transfer between the surface and its surroundings. It is distinct from prescribing either the surface temperature or the heat flux independently.\n\nIn a mixed problem, separate portions of the boundary carry different types of conditions. For instance,\n\n\\[u=g_D\\quad\\text{on }\\Gamma_D, \\qquad \\partial_nu=g_N\\quad\\text{on }\\Gamma_N.\\]\n\nThe word “mixed” is sometimes also used for Robin conditions, so precise notation is preferable. At the junction between Dirichlet and Neumann portions, a solution may develop reduced smoothness even when the boundary and data are otherwise smooth.\n\nPeriodic conditions relate opposite boundaries. On an interval, a standard second-order periodic problem imposes\n\n\\[u(0)=u(L),\\qquad u'(0)=u'(L).\\]\n\nThese conditions represent a closed loop or a repeating cell rather than an externally constrained endpoint. Symmetry conditions provide another important reduction: an even scalar field across a symmetry plane has zero normal derivative there, whereas an odd scalar field vanishes there. Such conditions derive from the symmetry of the solution and must not be assumed merely because the geometry looks symmetric; forcing and material properties must respect the same symmetry."
    },
    {
      "heading": "Existence, uniqueness, compatibility, and stability",
      "text": "A well-posed problem, in the sense associated with Jacques Hadamard, has a solution, has at most one solution, and depends continuously on its data in specified function-space norms. Boundary conditions influence all three properties. Counting conditions is useful, but not sufficient: their location and relationship to the operator are equally important. For a nonsingular \\(m\\)th-order ordinary differential equation, \\(m\\) independent scalar conditions are generally needed, yet particular boundary arrangements may still produce nonexistence or nonuniqueness.\n\nFor example,\n\n\\[u''+u=0,\\qquad u(0)=u(\\pi)=0\\]\n\nhas every function \\(u(x)=A\\sin x\\) as a solution. Replacing the last condition by \\(u(\\pi)=1\\) makes the problem unsolvable. The number of conditions is unchanged; the failure arises because the homogeneous equation has a nontrivial solution satisfying the homogeneous endpoint constraints. This phenomenon is closely connected to eigenvalues and resonance.\n\nConservation laws impose compatibility requirements. Consider Poisson’s equation\n\n\\[-\\Delta u=f\\quad\\text{in }\\Omega, \\qquad \\partial_nu=g\\quad\\text{on }\\partial\\Omega.\\]\n\nThe divergence theorem gives\n\n\\[\\int_\\Omega f\\,dx =-\\int_{\\partial\\Omega}g\\,dS.\\]\n\nIf this relation fails, no sufficiently regular solution can exist. On a connected bounded domain, a solution to the compatible pure Neumann problem is determined only up to an additive constant. Prescribing its mean, such as \\(\\int_\\Omega u\\,dx=0\\), removes that ambiguity. In physical language, the source balance determines whether equilibrium is possible, while the arbitrary constant reflects the absence of a fixed reference level.\n\nToo much boundary information can also be problematic. Prescribing both \\(u\\) and \\(\\partial_nu\\) on an entire boundary for an elliptic equation usually imposes more conditions than arbitrary data can satisfy. Even where uniqueness holds, recovery may be unstable. For Laplace’s equation in a strip, the harmonic functions\n\n\\[u_n(x,y)=n^{-2}\\sin(nx)\\sinh(ny)\\]\n\nhave \\(u_n(x,0)=0\\) and \\(\\partial_yu_n(x,0)=n^{-1}\\sin(nx)\\), which tends uniformly to zero. At any fixed \\(y>0\\), however, their amplitudes grow exponentially with \\(n\\). Small errors in boundary measurements can therefore correspond to very large interior differences.\n\nWell-posedness is always relative to an equation, a domain, and a choice of spaces and norms. A condition that is appropriate for diffusion may be inappropriate for advection. Likewise, a boundary formulation stable in one regularity class may fail to control stronger derivatives. Energy estimates, maximum principles, and operator inequalities supply ways to establish continuous dependence rather than infer it from a plausible physical interpretation."
    },
    {
      "heading": "Weak formulations, traces, and variational boundary conditions",
      "text": "Classical boundary conditions assume that the required boundary values and derivatives exist pointwise. Solutions of realistic problems may not be that smooth. In a Sobolev space, functions are initially defined up to changes on sets of volume measure zero, so simply evaluating an arbitrary representative on the boundary is not meaningful. A trace operator supplies the appropriate generalized boundary value. On a bounded Lipschitz domain, the trace of an \\(H^1(\\Omega)\\) function belongs to the boundary space \\(H^{1/2}(\\partial\\Omega)\\).\n\nThe distinction between imposed and naturally arising conditions becomes clear through integration by parts. For\n\n\\[-\\nabla\\cdot(K\\nabla u)=f,\\]\n\nmultiplication by a test function \\(v\\) and integration yield\n\n\\[\\int_\\Omega K\\nabla u\\cdot\\nabla v\\,dx = \\int_\\Omega fv\\,dx +\\int_{\\partial\\Omega} v\\,\\mathbf n\\cdot K\\nabla u\\,dS,\\]\n\nwhenever the boundary integral is classically defined. In a fully weak formulation, it may instead be interpreted as a duality pairing. This identity transfers one derivative from the unknown to the test function, allowing solutions with fewer classical derivatives.\n\nSuppose \\(u=g_D\\) on \\(\\Gamma_D\\), while \\(\\mathbf n\\cdot K\\nabla u=g_N\\) on \\(\\Gamma_N\\). The admissible solutions have prescribed trace on \\(\\Gamma_D\\), and test functions have zero trace there. The weak equation becomes\n\n\\[\\int_\\Omega K\\nabla u\\cdot\\nabla v\\,dx = \\int_\\Omega fv\\,dx+\\int_{\\Gamma_N}g_Nv\\,dS.\\]\n\nDirichlet conditions are called essential because they restrict the admissible function space. Neumann conditions are called natural because they enter through the boundary term. These names depend on the chosen variational formulation; a different formulation can change which quantities are imposed directly.\n\nRobin data also fit this structure. For\n\n\\[\\mathbf n\\cdot K\\nabla u+\\alpha u=r,\\]\n\nthe weak equation contains\n\n\\[\\int_\\Omega K\\nabla u\\cdot\\nabla v\\,dx +\\int_{\\partial\\Omega}\\alpha uv\\,dS = \\int_\\Omega fv\\,dx+\\int_{\\partial\\Omega}rv\\,dS.\\]\n\nIf \\(K\\) is uniformly positive definite and \\(\\alpha\\) provides adequate nonnegative boundary control, the boundary term can help establish coercivity. In functional analysis, coercivity and continuity lead to existence and uniqueness results through the Lax–Milgram theorem.\n\nThe calculus of variations gives the same distinction an energetic meaning. Minimizing\n\n\\[E[u]= \\frac12\\int_\\Omega K\\nabla u\\cdot\\nabla u\\,dx -\\int_\\Omega fu\\,dx\\]\n\nover functions with fixed boundary values yields the interior equation and leaves no boundary variation on the fixed portion. Where variations are unrestricted, stationarity produces a natural zero-flux condition. Adding suitable surface-energy or boundary-load terms produces Robin or nonzero-flux conditions. Thus a weak solution incorporates boundary behavior through both its admissible space and its integral identity."
    },
    {
      "heading": "Time-dependent equations and characteristic boundaries",
      "text": "Evolution equations require a distinction between initial data and spatial boundary data. For the heat equation,\n\n\\[u_t-\\kappa\\Delta u=f,\\qquad \\kappa>0,\\]\n\none usually prescribes \\(u(x,0)=u_0(x)\\) together with boundary conditions for later times. A classical solution continuous up to the intersection of the initial surface and spatial boundary must satisfy corresponding compatibility relations. For time-dependent Dirichlet data \\(u=g\\), the first such relation is\n\n\\[u_0|_{\\partial\\Omega}=g(\\cdot,0).\\]\n\nHigher smoothness can require additional relations involving time derivatives of \\(g\\) and spatial derivatives of \\(u_0\\). Weaker solutions may exist without these pointwise corner compatibilities, but can exhibit an initial loss of regularity.\n\nFor the wave equation,\n\n\\[u_{tt}-c^2\\Delta u=f,\\]\n\ntwo initial fields are needed: displacement and velocity. Spatial boundary conditions describe reflection, restraint, or transmission. A fixed string end has \\(u=0\\), whereas a force-free endpoint in the simplest string model has \\(u_x=0\\). If the boundary is stationary and homogeneous, either condition removes the corresponding boundary contribution to the rate of mechanical energy, though the reflected wave has a different phase in the two cases.\n\nFor first-order transport, conditions belong primarily on inflow boundaries. Consider\n\n\\[u_t+a u_x=0,\\qquad 0<x<L,\\quad a>0.\\]\n\nIts solutions travel along characteristic curves \\(x-at=\\text{constant}\\). Initial data determine characteristics originating at \\(t=0\\); data at \\(x=0\\) determine those entering later. Arbitrary data should not also be prescribed at \\(x=L\\), because the outgoing value is already determined by the interior solution. If \\(a<0\\), the roles of the two endpoints reverse.\n\nFor a multidimensional transport field \\(\\mathbf b\\), the inflow region is\n\n\\[\\Gamma_-=\\{x\\in\\partial\\Omega:\\mathbf b\\cdot\\mathbf n<0\\}.\\]\n\nFor hyperbolic systems, the relevant count comes from characteristic modes entering the domain, commonly determined by the signs of eigenvalues of the normal flux matrix. A subsonic compressible-flow boundary and a supersonic one therefore require different numbers and kinds of conditions. Characteristic boundaries, where a mode travels tangentially or has zero normal speed, require particular care.\n\nThese distinctions also explain why boundary prescriptions cannot be transferred unchanged between models. Adding even a small diffusion term increases spatial differential order and typically creates a need for additional boundary data. As diffusion tends to zero, the solution may develop a thin boundary layer near a boundary whose imposed values cannot be followed by the limiting transport solution. Boundary conditions therefore play a central role in singular perturbation theory and in the transition between diffusive and wave-like descriptions."
    },
    {
      "heading": "Physical boundaries, interfaces, and conservation laws",
      "text": "At a physical boundary, conditions express a model of interaction rather than a purely geometrical fact. For the Navier–Stokes equations, a stationary impermeable wall imposes zero normal fluid velocity. The standard no-slip model additionally sets tangential velocity equal to that of the wall. A slip model instead relates tangential velocity differences to tangential stress. These alternatives represent different assumptions about momentum transfer, and their suitability depends on the physical scale and material interface being modeled.\n\nIn elasticity, a displacement condition fixes how part of a body moves, while a traction condition prescribes surface force per unit area:\n\n\\[\\boldsymbol{\\sigma}\\mathbf n=\\mathbf t,\\]\n\nwhere \\(\\boldsymbol{\\sigma}\\) is the stress tensor. A body with only traction conditions can retain rigid translations and rotations. For static equilibrium without other constraints, applied loads must also satisfy global force and moment balance. This is the mechanical analogue of both compatibility conditions and nonuniqueness in pure Neumann scalar problems. A clamped plate requires more boundary information than a membrane because its bending equation is fourth order; in a classical plate model, displacement and normal slope are both fixed.\n\nAn internal interface separates two materials or phases and carries transmission conditions. For perfect thermal contact without an interfacial source, temperature is continuous and normal heat flux is conserved. If a single normal \\(\\mathbf n\\) points from material 1 into material 2, these conditions can be written\n\n\\[T_1=T_2,\\qquad (-K_1\\nabla T_1)\\cdot\\mathbf n = (-K_2\\nabla T_2)\\cdot\\mathbf n.\\]\n\nA contact resistance allows a temperature jump related to the transmitted flux. Such an interface is not equivalent to independently prescribing temperature on each side.\n\nElectromagnetic interface conditions follow from Maxwell’s equations in integral form. In the standard setting without magnetic surface currents, the tangential electric field is continuous. The jump in normal electric displacement equals free surface-charge density, while the jump in tangential magnetic field is determined by free surface current. These are conditions on different components of different fields, not a single scalar boundary prescription. A perfect electric conductor is an idealization in which the tangential electric field vanishes at its surface.\n\nAcross all these examples, local conservation laws determine which fluxes must balance, while constitutive assumptions determine how fields and fluxes are related. A boundary can also store mass, charge, or energy, in which case the imbalance of normal fluxes drives a surface evolution equation. Boundary conditions then become coupled dynamical laws, rather than instantaneous fixed values. This distinction is important for membranes, reactive surfaces, thin coatings, and interfaces with their own thermal capacity."
    },
    {
      "heading": "Spectra, self-adjointness, and quantum mechanics",
      "text": "Boundary conditions determine the permitted modes of an oscillating or diffusing system. For a string of length \\(L\\), separation of variables leads to\n\n\\[-X''=\\lambda X.\\]\n\nFixed endpoints give\n\n\\[X_n(x)=\\sin(n\\pi x/L),\\qquad \\lambda_n=(n\\pi/L)^2,\\quad n=1,2,\\ldots.\\]\n\nZero-derivative conditions give cosine modes with the same positive eigenvalues and an additional constant mode with eigenvalue zero. One fixed end and one free end instead give\n\n\\[\\lambda_n=((n+\\tfrac12)\\pi/L)^2,\\quad n=0,1,\\ldots.\\]\n\nThe difference in boundary restraint therefore changes the entire set of natural frequencies, not just the shape of the solution near an endpoint.\n\nThis relationship is formalized in Sturm–Liouville theory. For a regular problem on a finite interval,\n\n\\[-(pX')'+qX=\\lambda wX,\\]\n\nsuitable endpoint conditions produce a self-adjoint eigenvalue problem in a weighted Hilbert space. Under standard real-coefficient and positivity assumptions, its eigenvalues are real and eigenfunctions associated with distinct eigenvalues are orthogonal. This structure supports expansions analogous to Fourier series and the solution of time-dependent equations by modal decomposition.\n\nThe role of the boundary follows from Green’s identity. For the differential expression \\(-d^2/dx^2\\),\n\n\\[\\langle -u'',v\\rangle-\\langle u,-v''\\rangle = \\left[u\\overline{v'}-u'\\overline v\\right]_0^L.\\]\n\nDirichlet, Neumann, real separated Robin, and appropriate periodic conditions make this boundary form vanish for functions in the chosen domain. A self-adjoint operator requires both the appropriate action and the appropriate domain; formal symmetry of a differential expression is not enough. Endpoint behavior becomes still more subtle for singular intervals or singular coefficients.\n\nIn quantum mechanics, boundary conditions help define the Hamiltonian. The infinite square well uses a wavefunction that vanishes at the walls. At an ordinary finite potential step, the constant-mass Schrödinger equation instead gives continuity of the wavefunction and its first derivative. A delta-function potential produces a derivative jump related to the wavefunction’s value. These differences follow from the operator and idealized potential, rather than from a universal rule that every quantum boundary must force the wavefunction to zero.\n\nOn a ring, periodic conditions identify the two endpoints of an interval. More generally, a consistent phase-twisted identification takes the form\n\n\\[\\psi(L)=e^{i\\theta}\\psi(0),\\qquad \\psi'(L)=e^{i\\theta}\\psi'(0).\\]\n\nIt changes the allowed wave numbers and illustrates how global identification conditions affect spectra. For closed quantum systems, self-adjointness ensures unitary time evolution; a boundary that absorbs probability instead represents an open-system approximation and requires a different interpretation."
    },
    {
      "heading": "Numerical enforcement and artificial boundaries",
      "text": "In computational models, boundary conditions must be translated into discrete equations without destroying the structure of the continuous problem. In the finite difference method, prescribed boundary values may be inserted directly into grid equations. Neumann conditions often require one-sided formulas or ghost points outside the physical domain. At a left endpoint with \\(u_x(0)=g\\), the centered approximation\n\n\\[\\frac{u_1-u_{-1}}{2h}=g\\]\n\ngives \\(u_{-1}=u_1-2hg\\). Substitution eliminates the ghost value. Because the outward normal at a left endpoint points in the negative coordinate direction, converting a specified \\(\\partial_nu\\) into \\(u_x\\) requires a sign change.\n\nIn the finite element method, the essential–natural distinction guides implementation. Dirichlet values commonly constrain nodal degrees of freedom, while flux data enter the load vector through boundary integrals. Robin terms generally contribute both a boundary matrix and a boundary load. Enforcing Dirichlet conditions weakly is also possible, using penalty methods, Lagrange multipliers, or Nitsche’s method. A simple penalty formulation trades more accurate enforcement against potentially poorer matrix conditioning as the penalty grows; consistent weak methods are designed to avoid treating this tradeoff as the only option.\n\nCompatibility and null spaces remain important after discretization. A pure Neumann Poisson problem produces a singular matrix whose null space contains the constant vector. Adding an arbitrary solver tolerance does not remove the underlying ambiguity. Common formulations constrain the discrete mean or explicitly project away the null space. Likewise, boundary data and interior sources must satisfy a discrete balance consistent with the numerical method. Failure to preserve that balance can produce misleading solver behavior even when the continuous data are compatible.\n\nUnbounded physical domains create another difficulty: computation normally uses a finite region with an artificial exterior boundary. An absorbing boundary condition attempts to let outgoing waves leave without significant reflection. For a one-dimensional right-traveling wave of speed \\(c\\), the exact local relation is\n\n\\[u_t+c\\,u_x=0.\\]\n\nIn several dimensions, a similarly simple condition is generally exact only for restricted directions of propagation. Accurate transparent conditions may be nonlocal in space, time, or both.\n\nPerfectly matched layers provide an alternative by surrounding the physical domain with an auxiliary absorbing region designed to have no reflection at its interface in the ideal continuous setting. Discretization, finite layer thickness, and the properties of the governing equations affect practical performance. Periodic computational boundaries are another modeling choice, but they represent repeating geometry rather than radiation into an infinite exterior.\n\nBoundary treatment can dominate numerical stability and error. A high-order interior approximation does not ensure a high-order or stable complete scheme if its boundary closure is inconsistent. Energy estimates, mesh refinement, conservation checks, and reflection measurements are therefore used to assess boundary implementations separately from interior accuracy."
    },
    {
      "heading": "Nonlinear, moving, and inferred boundary conditions",
      "text": "Many boundary conditions depend nonlinearly on the unknown field. Thermal radiation provides a standard example:\n\n\\[-k\\partial_nT =\\varepsilon\\sigma_{\\mathrm{SB}} \\left(T^4-T_{\\mathrm{env}}^4\\right),\\]\n\nunder the assumptions of a gray surface exchanging radiation with an effective surrounding environment. Here \\(\\varepsilon\\) is emissivity and \\(\\sigma_{\\mathrm{SB}}\\) is the Stefan–Boltzmann constant. The boundary contribution is nonlinear even if conduction inside the body is linear. Reactive transport has analogous conditions in which species flux depends on surface concentration, temperature, or electric potential.\n\nContact mechanics introduces inequality conditions rather than a fixed equality. A frictionless nonadhesive interface cannot interpenetrate and cannot transmit tensile normal contact force. With gap \\(g\\geq0\\) and compressive contact pressure \\(p\\geq0\\), the idealized complementarity condition is\n\n\\[gp=0.\\]\n\nWhere the gap is positive, pressure vanishes; where pressure is positive, the surfaces meet. The contact region is itself unknown. Such problems lead to variational inequalities and constrained optimization, rather than a boundary operator fixed in advance on a known active region.\n\nIn a free-boundary problem, part of the domain boundary must be found together with the solution. Melting and solidification provide a classical case. A phase interface is assigned an equilibrium phase-change temperature, while an energy-balance condition relates its normal velocity to the jump in heat flux and the latent heat. In free-surface fluid motion, a kinematic condition makes the interface move with the fluid, and a dynamic condition balances stresses, possibly including surface tension. The unknown geometry and the field equations are therefore coupled.\n\nConditions at infinity serve a related role in exterior-domain problems. An electrostatic potential may be required to decay far from a localized charge distribution. For time-harmonic outgoing waves in three dimensions, the Sommerfeld radiation condition is commonly expressed, under the corresponding time convention, as\n\n\\[r\\left(\\partial_ru-iku\\right)\\longrightarrow0 \\qquad (r\\to\\infty),\\]\n\nuniformly in direction in standard formulations. It selects outgoing rather than incoming wave behavior. Infinity is not an ordinary finite boundary, but an asymptotic requirement can be essential to uniqueness.\n\nBoundary quantities can also be unknown parameters in an inverse problem. Measurements inside a body may be used to infer an inaccessible surface temperature, heat-transfer coefficient, or flux. Such recovery can be far less stable than the corresponding forward problem, making regularization and explicit assumptions about measurement error important parts of the mathematical formulation. In coupled multiphysics models, the “boundary data” for one subsystem may instead be generated by another: fluid pressure loads a structure, while structural motion sets the fluid’s interface velocity. Solving the coupled problem requires consistent exchange of both motion and force across the interface."
    }
  ],
  "links": [
    "differential-equation",
    "euclidean-space",
    "boundary-value-problem",
    "ordinary-differential-equation",
    "initial-value-problem",
    "partial-differential-equation",
    "joseph-fourier",
    "dirichlet-boundary-condition",
    "neumann-boundary-condition",
    "gradient",
    "eigenvalues-and-eigenvectors",
    "poisson-equation",
    "divergence-theorem",
    "laplace-equation",
    "sobolev-space",
    "integration-by-parts",
    "functional-analysis",
    "calculus-of-variations",
    "weak-solution",
    "heat-equation",
    "wave-equation",
    "navier-stokes-equations",
    "tensor",
    "maxwells-equations",
    "sturm-liouville-theory",
    "hilbert-space",
    "fourier-series",
    "self-adjoint-operator",
    "quantum-mechanics",
    "finite-difference-method",
    "finite-element-method",
    "matrix",
    "numerical-stability",
    "mathematical-optimization",
    "inverse-problem",
    "regularization"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}