{
  "id": "neumann-boundary-condition",
  "title": "Neumann Boundary Condition",
  "lang": "en",
  "source_lang": "en",
  "summary": "A Neumann boundary condition specifies the outward normal derivative, or an associated flux, of an unknown function on a domain’s boundary.",
  "category": "mathematics",
  "subcategory": "analysis",
  "type": "concept",
  "aliases": [
    "Boundary condition of the second kind",
    "Second-type boundary condition"
  ],
  "generated_at": "2026-11-06",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/neumann-boundary-condition/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/neumann-boundary-condition/",
      "json": "https://aiwiki.page/zh/neumann-boundary-condition.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "A Neumann boundary condition is a constraint in a boundary value problem that prescribes the derivative of an unknown function in the direction normal to the boundary. It is widely used with partial differential equations describing heat conduction, diffusion, electrostatics, and fluid flow. In divergence-form equations, the term also commonly includes prescribed conormal derivatives, which represent boundary fluxes. The condition is named after the nineteenth-century mathematician Carl Neumann."
    },
    {
      "heading": "Mathematical definition",
      "text": "Let \\(\\Omega\\subset\\mathbb{R}^n\\) be a domain with a sufficiently smooth boundary \\(\\partial\\Omega\\), and let \\(\\mathbf n\\) denote its outward unit normal. For a differentiable scalar function \\(u\\), a Neumann boundary condition takes the form\n\n\\[\\frac{\\partial u}{\\partial n} = \\nabla u\\cdot\\mathbf n = g \\qquad\\text{on }\\partial\\Omega,\\]\n\nwhere \\(\\nabla u\\) is the gradient of \\(u\\), and \\(g\\) is prescribed boundary data. The expression \\(\\partial u/\\partial n\\) is the directional derivative along the outward normal.\n\nThe condition is homogeneous if \\(g=0\\) and inhomogeneous otherwise. A homogeneous Neumann condition requires the normal derivative to vanish; it does not require the function itself to vanish or to be constant along the boundary. Tangential variation can remain unrestricted.\n\nThe outward-normal convention matters. On the interval \\((0,L)\\), the outward normal points left at \\(0\\) and right at \\(L\\), so\n\n\\[\\frac{\\partial u}{\\partial n}(0)=-u'(0), \\qquad \\frac{\\partial u}{\\partial n}(L)=u'(L).\\]\n\nThus endpoint derivatives and outward normal derivatives have different signs at the left endpoint."
    },
    {
      "heading": "Comparison with other boundary conditions",
      "text": "Boundary conditions specify different kinds of information:\n\n- A Dirichlet boundary condition prescribes the value of \\(u\\), such as \\(u=h\\).\n- A Neumann boundary condition prescribes its normal derivative or an associated flux.\n- A Robin boundary condition prescribes a linear combination, such as\n\n\\[\\alpha u+\\beta\\frac{\\partial u}{\\partial n}=r.\\]\n\nDifferent conditions may be imposed on different portions of the boundary. A problem with Dirichlet data on one portion and Neumann data on another is usually called a mixed boundary value problem. This differs from a Robin condition, which combines value and derivative information at the same boundary location."
    },
    {
      "heading": "Normal derivatives, conormal derivatives, and flux",
      "text": "The normal derivative and physical flux are related but are not generally identical. For isotropic heat conduction, Fourier’s law gives the heat-flux vector as\n\n\\[\\mathbf q=-k\\nabla T,\\]\n\nwhere \\(T\\) is temperature and \\(k>0\\) is thermal conductivity. The outward heat flux is therefore\n\n\\[q_n=\\mathbf q\\cdot\\mathbf n =-k\\frac{\\partial T}{\\partial n}.\\]\n\nPrescribing \\(q_n\\) is equivalent to prescribing the normal derivative when \\(k\\) is known and nonzero, but the sign and conductivity factor must be retained. In particular, positive outward heat flux corresponds to a negative outward temperature derivative.\n\nFor a divergence-form operator\n\n\\[Lu=-\\nabla\\cdot(A\\nabla u),\\]\n\nwhere \\(A(x)\\) is a coefficient matrix, the natural boundary quantity is the conormal derivative\n\n\\[(A\\nabla u)\\cdot\\mathbf n.\\]\n\nA condition prescribing this quantity is commonly called a Neumann condition. If the physical flux is \\(\\mathbf q=-A\\nabla u\\), its outward component is the negative of the conormal derivative. For anisotropic \\(A\\), a zero conormal derivative need not imply a zero ordinary normal derivative."
    },
    {
      "heading": "The Poisson problem: compatibility and uniqueness",
      "text": "An important example is Poisson’s equation with Neumann data:\n\n\\[-\\Delta u=f \\quad\\text{in }\\Omega, \\qquad \\frac{\\partial u}{\\partial n}=g \\quad\\text{on }\\partial\\Omega,\\]\n\nwhere \\(\\Delta\\) is the Laplace operator.\n\nCompatibility condition\n\nThe volume source and boundary data cannot be chosen independently. Integrating the equation and applying the divergence theorem gives\n\n\\[\\int_\\Omega f\\,dx = -\\int_\\Omega\\Delta u\\,dx = -\\int_{\\partial\\Omega}g\\,dS.\\]\n\nConsequently, a necessary compatibility condition is\n\n\\[\\int_\\Omega f\\,dx+ \\int_{\\partial\\Omega}g\\,dS=0.\\]\n\nThis expresses a global balance between interior sources and boundary transfer. For Laplace’s equation, where \\(f=0\\), the prescribed normal derivative must have zero total boundary integral.\n\nThe sign depends on how the equation and boundary data are defined. For example, writing \\(\\Delta u=f\\) instead gives \\(\\int_\\Omega f=\\int_{\\partial\\Omega}g\\).\n\nUniqueness up to an additive constant\n\nIf \\(u\\) solves the pure Neumann Poisson problem, then \\(u+C\\) also solves it for every constant \\(C\\). Neither the Laplacian nor the normal derivative detects an additive constant.\n\nOn a bounded, connected domain, under appropriate regularity assumptions, this is the only nonuniqueness. If two solutions have difference \\(w\\), then\n\n\\[-\\Delta w=0, \\qquad \\frac{\\partial w}{\\partial n}=0.\\]\n\nAn integration identity yields\n\n\\[\\int_\\Omega |\\nabla w|^2\\,dx=0,\\]\n\nso \\(w\\) is constant. Uniqueness can therefore be restored by imposing a normalization such as\n\n\\[\\int_\\Omega u\\,dx=0.\\]\n\nOn a disconnected domain, compatibility must hold separately on each connected component, and each component has its own additive constant.\n\nThese properties concern the pure Neumann Laplacian and related operators. They are not consequences of Neumann boundary conditions alone: adding a positive zeroth-order term, as in \\(-\\Delta u+cu=f\\) with \\(c>0\\), removes the constant ambiguity."
    },
    {
      "heading": "Weak and variational formulation",
      "text": "Neumann conditions arise naturally through integration by parts. Multiplying the Poisson equation by a test function \\(v\\) and integrating gives\n\n\\[\\int_\\Omega\\nabla u\\cdot\\nabla v\\,dx = \\int_\\Omega fv\\,dx + \\int_{\\partial\\Omega}gv\\,dS.\\]\n\nThis identity defines a weak solution in the Sobolev space \\(H^1(\\Omega)\\), with boundary pairings interpreted appropriately when the data are not ordinary integrable functions. It permits solutions whose normal derivatives do not exist pointwise.\n\nSetting \\(v=1\\) recovers the compatibility condition. Restricting to mean-zero functions removes the constant ambiguity. On a bounded, connected Lipschitz domain, the appropriate Poincaré inequality then supplies the estimate needed for existence and uniqueness for compatible data in suitable function spaces.\n\nFor sufficiently regular data, the corresponding energy functional is\n\n\\[J(u)= \\frac12\\int_\\Omega|\\nabla u|^2\\,dx -\\int_\\Omega fu\\,dx -\\int_{\\partial\\Omega}gu\\,dS.\\]\n\nIts stationary points satisfy the weak equation. In the calculus of variations, Neumann conditions are therefore called natural boundary conditions: they arise from boundary terms when variations are unrestricted there. By contrast, prescribed Dirichlet values are ordinarily built into the admissible functions."
    },
    {
      "heading": "Time-dependent equations and conservation",
      "text": "For the heat equation with constant diffusivity \\(\\kappa>0\\),\n\n\\[\\frac{\\partial u}{\\partial t}=\\kappa\\Delta u,\\]\n\nhomogeneous Neumann conditions imply\n\n\\[\\frac{d}{dt}\\int_\\Omega u\\,dx = \\kappa\\int_{\\partial\\Omega} \\frac{\\partial u}{\\partial n}\\,dS =0.\\]\n\nThe spatial integral of \\(u\\) is conserved. In a diffusion model this commonly represents conservation of total material in an impermeable container. For temperature, conserved thermal energy requires the corresponding heat-capacity weighting.\n\nOn a bounded, connected domain, solutions with homogeneous Neumann conditions approach their initial spatial average under the usual assumptions. The constant equilibrium is selected by the initial data, unlike the undetermined additive constant in the stationary Poisson problem.\n\nFor the scalar wave equation, a homogeneous Neumann condition can model a free end of a vibrating string or a reflecting boundary in an appropriate scalar-wave model. Its physical interpretation depends on the dependent variable and governing equations."
    },
    {
      "heading": "Spectral properties",
      "text": "The Neumann eigenvalue problem is\n\n\\[-\\Delta\\phi=\\lambda\\phi \\quad\\text{in }\\Omega, \\qquad \\frac{\\partial\\phi}{\\partial n}=0 \\quad\\text{on }\\partial\\Omega.\\]\n\nOn a bounded, connected domain, the smallest eigenvalue is zero, with constant eigenfunctions. This zero mode accounts both for the stationary constant ambiguity and for conservation of the mean in diffusion.\n\nOn the interval \\((0,L)\\), the eigenfunctions and eigenvalues are\n\n\\[\\phi_m(x)=\\cos\\left(\\frac{m\\pi x}{L}\\right), \\qquad \\lambda_m=\\left(\\frac{m\\pi}{L}\\right)^2, \\qquad m=0,1,2,\\ldots.\\]\n\nThese modes underlie cosine-series representations, a form of Fourier series, for heat and wave problems with homogeneous Neumann endpoints."
    },
    {
      "heading": "Numerical implementation",
      "text": "In the finite element method, Neumann data enter the boundary term of the weak formulation. Homogeneous data usually require no added boundary-load contribution, although the boundary still affects the mesh and integration domain.\n\nA pure Neumann discretization of the Poisson equation produces a singular stiffness matrix whose null space contains the constant mode. Solvers must handle this explicitly, for example by enforcing a mean-zero constraint or working in the subspace orthogonal to constants. The discrete load must satisfy the corresponding compatibility condition. Fixing a degree of freedom can select a representative solution, but does not make incompatible original data solvable.\n\nIn the finite difference method, Neumann conditions can be represented using one-sided derivatives, ghost points, or boundary flux formulas. Accuracy and conservation depend on both the boundary approximation and its consistency with the interior discretization."
    },
    {
      "heading": "Applications and limitations",
      "text": "Common applications include insulated thermal boundaries, impermeable diffusion barriers, prescribed electrical current through a conducting boundary, and specified flow across a porous-medium boundary.\n\nNeumann conditions describe only the selected derivative or flux. They do not by themselves prescribe boundary values, prohibit tangential transport, or guarantee existence and uniqueness. Their mathematical and physical meaning depends on the governing operator, coefficient properties, boundary regularity, and whether the condition is imposed on the entire boundary or only part of it."
    }
  ],
  "links": [
    "boundary-value-problem",
    "partial-differential-equation",
    "gradient",
    "directional-derivative",
    "dirichlet-boundary-condition",
    "matrix",
    "poisson-equation",
    "laplace-operator",
    "divergence-theorem",
    "laplace-equation",
    "integration-by-parts",
    "weak-solution",
    "sobolev-space",
    "calculus-of-variations",
    "heat-equation",
    "wave-equation",
    "eigenvalues-and-eigenvectors",
    "fourier-series",
    "finite-element-method",
    "null-space",
    "finite-difference-method"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}