Fact-checked by Grok 2 weeks ago

Functional derivative

In mathematics, the functional derivative is a fundamental concept in the calculus of variations and functional analysis, serving as the analog of the ordinary derivative for functionals—mappings from a space of functions to the real numbers.^[1] It quantifies the rate of change of a functional F with respect to infinitesimal variations \delta u in the input function u, formally defined via the Fréchet derivative such that F[u + \epsilon h] = F + \epsilon \int \frac{\delta F}{\delta u}(x) h(x) \, dx + o(\epsilon) as \epsilon \to 0, where h is a test function.^[1] Equivalently, in the Gâteaux sense, it is the directional derivative \frac{\delta F}{\delta u}[u](h) = \lim_{t \to 0} \frac{F[u + t h] - F}{t}.^[1] This notion arises naturally in optimization problems where one seeks to extremize a functional, such as finding curves of minimal length or surfaces of least area, by setting the functional derivative to zero, which yields the Euler-Lagrange equations.^[2] For a typical functional of the form F = \int_a^b L(x, u(x), u'(x)) \, dx, the functional derivative is given by \frac{\delta F}{\delta u}(x) = \frac{\partial L}{\partial u} - \frac{d}{dx} \left( \frac{\partial L}{\partial u'} \right), providing the condition for stationary points.^[2] In physics, functional derivatives play a central role in deriving equations of motion from the principle of least action, where the action functional S = \int L \, dt leads to \frac{\delta S}{\delta x}(t) = 0, recovering Newton's laws or field equations in Lagrangian mechanics.^[2] For instance, in the case of a harmonic oscillator, this yields m \ddot{x} + kx = 0.^[2] Beyond classical variational problems, functional derivatives extend to more abstract settings, including infinite-dimensional spaces where they facilitate the study of partial differential equations, quantum field theory, and statistical mechanics.^[3] Properties such as linearity, the product rule, and the chain rule hold analogously to ordinary calculus, enabling the construction of Taylor expansions for functionals and the analysis of higher-order variations like the Hessian, which is the second functional derivative used in stability assessments.^[1] In field theories, the functional derivative often includes a factor of $1/\mathrm{Vol} relative to partial derivatives when discretizing over volumes, reflecting the continuous nature of the underlying function space.^[3]

Fundamentals

Definition of functional derivative

In mathematics, particularly within the calculus of variations, a functional is a mapping that assigns a real number to each function in a specified class or vector space of functions.^[4] A prototypical example is the integral functional J = \int_a^b L(x, f(x), f'(x)) \, dx, where L is a given function known as the Lagrangian density, f is the input function, and the integral is taken over an interval [a, b].^[5] The functional derivative of such a functional J with respect to the input function f at a point x, denoted \frac{\delta J}{\delta f(x)}, is defined through its role in approximating the change in the functional value under an infinitesimal variation \delta f of the input function. Specifically, for small \delta f, the variation in J satisfies \delta J \approx \int \frac{\delta J}{\delta f(x)} \delta f(x) \, dx, where the integral is over the appropriate domain.^[1] This formulation generalizes the concept of the ordinary derivative to infinite-dimensional spaces, treating the functional as depending on "infinitely many variables" corresponding to the values of f at each point.^[6] Intuitively, \frac{\delta J}{\delta f(x)} functions as a local "density" of the derivative, quantifying the contribution to \delta J from the variation \delta f at each specific x; its units are those of J divided by the units of f times the inverse of the domain's measure (e.g., inverse length for one-dimensional domains).^[1] In variational problems, where one seeks to extremize J, the functional derivative plays a central role, with stationarity requiring \frac{\delta J}{\delta f(x)} = 0 for all x in the domain.^[5] As an analogy, it extends the finite-dimensional vector derivative, where the change is \delta J \approx \nabla J \cdot \delta \vec{v}, to the continuous case via an inner product-like integral.^[1] This is related to the Gateaux derivative, a directional variant in function space.^[1]

Relation to functional differential

The functional differential of a functional J represents the first-order infinitesimal change in J due to a variation \delta f in the input function f, expressed as \delta J = \left\langle \frac{\delta J}{\delta f}, \delta f \right\rangle, where \langle \cdot, \cdot \rangle denotes an inner product in the appropriate function space, such as the L^2 inner product \int \frac{\delta J}{\delta f}(x) \delta f(x) \, dx.^[7] This formulation captures the linear approximation to the change in the functional, analogous to the differential df = \frac{\partial f}{\partial x} dx in finite-dimensional calculus.^[8] The functional derivative \frac{\delta J}{\delta f} thus acts as the density or kernel that, when integrated against the variation, yields the total differential. In a more rigorous mathematical framework, the functional differential aligns with the Gâteaux derivative, which is defined as the directional derivative of J at f in the direction of a perturbation h, given by

D_J(f; h) = \lim_{\epsilon \to 0} \frac{J[f + \epsilon h] - J}{\epsilon},

provided the limit exists uniformly for small \epsilon.^[9] This derivative measures the rate of change along a specific path parameterized by \epsilon, without requiring uniformity over all directions, making it suitable for spaces like Banach spaces where full differentiability may not hold.^[10] The functional derivative then corresponds to the representer of this linear functional under an inner product, often involving a Dirac delta distribution in physics applications.^[1] A stronger notion is the Fréchet derivative, which generalizes the Gâteaux derivative by requiring the approximation to hold uniformly in the norm of the perturbation: J[f + h] = J + L(f)h + o(\|h\|), where L(f) is a bounded linear operator.^[9] In Hilbert or Banach spaces, the Riesz representation theorem ensures that this linear operator can be represented by an element in the dual space, which, under an L^2-like structure, takes the form of the functional derivative \frac{\delta J}{\delta f}.^[10] The Fréchet derivative thus provides a more robust foundation for local approximations in infinite-dimensional settings, such as those arising in the calculus of variations.^[11] The functional derivative typically presupposes an L^2-type inner product structure, enabling the identification of \frac{\delta J}{\delta f} as a pointwise density, whereas the functional differential is a more abstract concept that encompasses both Gâteaux and Fréchet variants without assuming such a specific representation.^[7] This distinction highlights how the derivative serves as a concrete tool in applications like physics, while the differential offers a general framework for differentiability in functional analysis.

Properties

Linearity and additivity

The functional derivative exhibits linearity with respect to linear combinations of functionals. Specifically, for two functionals J and K defined on the same function space, the functional derivative of their sum satisfies

\frac{\delta (J + K)}{\delta f(x)} = \frac{\delta J}{\delta f(x)} + \frac{\delta K}{\delta f(x)},

where the equality holds pointwise in x.^[12]^[5] This property follows directly from the linearity of the first variation \delta(J + K) = \delta J + \delta K, which is expressed as the integral \int \frac{\delta (J + K)}{\delta f(x)} \delta f(x) \, dx = \int \frac{\delta J}{\delta f(x)} \delta f(x) \, dx + \int \frac{\delta K}{\delta f(x)} \delta f(x) \, dx.^[13] Additionally, the functional derivative is homogeneous with respect to scalar multiplication of the functional. For a scalar \alpha and functional J,

\frac{\delta (\alpha J)}{\delta f(x)} = \alpha \frac{\delta J}{\delta f(x)}.

This homogeneity arises because the first variation scales linearly with \alpha, as \delta(\alpha J) = \alpha \delta J = \alpha \int \frac{\delta J}{\delta f(x)} \delta f(x) \, dx.^[12]^[5] In the context of functionals depending on multiple arguments, such as J[f^{(1)}, \dots, f^{(m)}], additivity manifests in the partial functional derivatives with respect to each argument. The total variation is the sum \delta J = \sum_{j=1}^m \int A_j(x) \delta f^{(j)}(x) \, dx, where A_j(x) = \frac{\delta J}{\delta f^{(j)}(x)} represents the partial influence of the j-th function, independent of the others under fixed boundary conditions.^[13]^[5] For a functional of the form J[f + g], the derivative \frac{\delta J}{\delta f(x)} isolates the contribution from variations in f while treating g as fixed, reflecting partial additivity in the arguments.^[12] A sketch of the proof for these properties relies on the integral representation of the first variation. Consider the increment \Delta J = J[f + \delta f] - J \approx \int \frac{\delta J}{\delta f(x)} \delta f(x) \, dx, where higher-order terms are neglected. For the sum J + K, \Delta (J + K) = \Delta J + \Delta K, so the linear approximations add, yielding the additivity of the kernels \frac{\delta J}{\delta f} + \frac{\delta K}{\delta f}. Similarly, for \alpha J, \Delta (\alpha J) = \alpha \Delta J, implying homogeneity. The linearity of the integral operator ensures these hold for arbitrary admissible \delta f.^[5]^[13] These linearity and additivity properties simplify the analysis of composite systems in variational calculus, allowing the decomposition of complex functionals into sums of simpler ones whose derivatives can be computed independently before recombination.^[12]^[5]

Product and chain rules

In the calculus of variations, the product rule for functional derivatives generalizes the Leibniz rule from ordinary differentiation to products involving functionals. For two functionals F and G depending on the function f, the functional derivative of their product satisfies

\frac{\delta (F G)}{\delta f(x)} = F \frac{\delta G}{\delta f(x)} + G \frac{\delta F}{\delta f(x)},

where the functionals are evaluated at f. This relation is derived by considering the first-order variation of the product under a perturbation f \to f + \epsilon \eta, expanding to linear order in \epsilon, and taking the limit as \epsilon \to 0.^[7] A particular instance of this rule, often referred to as the Leibniz form, applies to functionals expressed as integrals of a product, such as J = \int f(x) K[f](x) \, dx, where K is itself a functional of f. The functional derivative is then

\frac{\delta J}{\delta f(y)} = K[f](y) + \int f(x) \frac{\delta K[f](x)}{\delta f(y)} \, dx.

This follows directly from applying the general product rule to the integrand and interchanging the variation with the integral, assuming the necessary conditions for differentiation under the integral sign hold.^[7] The chain rule extends differentiation to compositions of functionals, enabling the computation of derivatives through intermediate mappings. For a functional J where g = g is a functional of f, the chain rule states

\frac{\delta J}{\delta f(x)} = \int \frac{\delta J}{\delta g(y)} \frac{\delta g(y)}{\delta f(x)} \, dy.

This formula arises from the variation of J induced by a variation in f, propagated through the variation in g, requiring that the mapping from variations in f to those in g be surjective to ensure all directions are covered.^[7] When applying these rules, particularly in explicit computations involving derivatives of the argument function, an integration-by-parts variant frequently appears to handle boundary contributions. For instance, in varying terms like \int u(f) \frac{\delta v(f)}{\delta f}(x) \, dx, integration by parts yields -\int v(f) \frac{\delta u(f)}{\delta f}(x) \, dx plus boundary terms evaluated at the domain endpoints; these boundary terms vanish if the test functions or variations are chosen to be zero at the boundaries.^[14] These rules presuppose that the functionals are Fréchet differentiable, with the underlying functions sufficiently smooth (e.g., continuously differentiable) over a domain where boundary effects can be controlled, such as compact intervals with fixed endpoints. Violations of these assumptions, like non-smoothness or infinite domains without decay conditions, may invalidate the formulas or introduce divergent terms.^[5]

Computation Methods

General formula for computation

The functional derivative provides a systematic way to compute the variation of an integral functional J = \int_a^b L(x, f(x), f'(x)) \, dx, where L is the Lagrangian density depending on the function f, its first derivative f', and the independent variable x. For such local functionals, the general formula for the functional derivative at a point x_0 in the interior of the interval is given by the Euler-Lagrange expression:

\frac{\delta J}{\delta f(x_0)} = \frac{\partial L}{\partial f}\bigg|_{x=x_0} - \frac{d}{dx} \left( \frac{\partial L}{\partial f'} \right)\bigg|_{x=x_0}.

This formula arises as the condition for stationary points where the first-order variation \delta J = 0, and it serves as the primary tool for explicit calculations in calculus of variations.^[15] To derive this formula, consider a small variation f(x) \to f(x) + \delta f(x), where \delta f(x) is an arbitrary smooth function vanishing at the endpoints x = a and x = b to satisfy fixed boundary conditions. The change in the functional is

\delta J = J[f + \delta f] - J = \int_a^b \left[ \frac{\partial L}{\partial f} \delta f + \frac{\partial L}{\partial f'} \delta f' \right] dx + O((\delta f)^2).

The second term requires integration by parts: \int_a^b \frac{\partial L}{\partial f'} \delta f' \, dx = \left[ \frac{\partial L}{\partial f'} \delta f \right]_a^b - \int_a^b \frac{d}{dx} \left( \frac{\partial L}{\partial f'} \right) \delta f \, dx. With fixed endpoints, the boundary term vanishes, yielding

\delta J = \int_a^b \left[ \frac{\partial L}{\partial f} - \frac{d}{dx} \left( \frac{\partial L}{\partial f'} \right) \right] \delta f \, dx + O((\delta f)^2).

By definition, \delta J = \int_a^b \frac{\delta J}{\delta f(x)} \delta f(x) \, dx, so equating coefficients identifies the functional derivative as the Euler-Lagrange operator above. This derivation assumes the variation is first-order and the functional is differentiable.^[14] For functionals depending on higher-order derivatives, such as J = \int_a^b L(x, f(x), f'(x), \dots, f^{(n)}(x)) \, dx, the formula generalizes to

\frac{\delta J}{\delta f(x_0)} = \sum_{k=0}^n (-1)^k \frac{d^k}{dx^k} \left( \frac{\partial L}{\partial f^{(k)}} \right)\bigg|_{x=x_0}.

The derivation follows analogously, with repeated integrations by parts up to order n, again requiring boundary conditions where variations and their derivatives up to order n-1 vanish at the endpoints to eliminate surface terms.^[16] In cases involving nonlocal kernels, such as the quadratic functional J = \iint K(x,y) f(x) f(y) \, dx \, dy, the first-order variation is \delta J = 2 \iint K(x,y) f(y) \delta f(x) \, dx \, dy assuming a symmetric kernel K(x,y) = K(y,x). Thus, the functional derivative is \frac{\delta J}{\delta f(x)} = 2 \int K(x,y) f(y) \, dy, computed via the variational expansion without local differential operators. Boundary conditions in nonlocal settings typically involve specifying f on the domain boundaries, ensuring the kernel integration respects the limits. For general g(f(x), f(y)), the expression involves distinct partial derivatives: \int K(x,y) \frac{\partial g}{\partial u}(f(x),f(y)) \, dy + \int K(y,x) \frac{\partial g}{\partial v}(f(y),f(x)) \, dy.^[7]

Test function approach with Dirac delta

The test function approach to computing functional derivatives utilizes the Dirac delta function as a localized probe to assess the variation of a functional J at a specific point y. This method defines the functional derivative as

\frac{\delta J}{\delta f(y)} = \lim_{\epsilon \to 0} \frac{J[f + \epsilon \delta(\cdot - y)] - J}{\epsilon},

where \delta(\cdot - y) is the Dirac delta distribution centered at y, and \epsilon represents an infinitesimal perturbation amplitude. This formulation replaces the more general arbitrary test function used in the Gâteaux derivative with a singular delta function, effectively isolating the response at the point of interest without requiring integration over an extended domain.^[17] This approach proves advantageous for functionals that are nonlocal or defined on infinite-dimensional spaces, such as those in quantum mechanics or field theory, where smooth test functions might fail to capture pointwise sensitivities due to the singular nature of the perturbation. By leveraging the delta function's property as a distribution, it accommodates variations in spaces of generalized functions, enabling computations for expectation values or density-based functionals that do not explicitly depend on local coordinates. In practice, approximate sequences of test functions (e.g., Gaussians narrowing to the delta) are often employed to ensure convergence in the limit.^[18]^[17] The method relates briefly to the Gelfand triple (rigged Hilbert space) framework, which embeds Hilbert spaces within dual spaces of distributions to rigorously handle objects like the Dirac delta, providing a mathematical structure for defining such derivatives in quantum contexts without violating domain constraints. Despite its utility, the Dirac delta approach remains formal and distribution-theoretic, often necessitating regularization techniques—such as smoothing the delta or introducing cutoffs—to mitigate divergences in physical implementations, particularly for normalized wave functions or densities where unconstrained variations are inadmissible.^[17]

Examples

Thomas–Fermi kinetic energy functional

The Thomas–Fermi kinetic energy functional arises in the semiclassical treatment of multi-electron atoms, approximating the kinetic energy of non-interacting fermions as a local function of the electron density \rho(\mathbf{r}). It is given by

T_{\mathrm{TF}}[\rho] = C_F \int \rho^{5/3}(\mathbf{r}) \, d^3\mathbf{r},

where C_F = \frac{3}{10} (3\pi^2)^{2/3} is a constant derived from the zero-temperature uniform electron gas energy in atomic units.^[19] This form captures the scaling of kinetic energy with density raised to the power of $5/3, reflecting the Fermi gas relation between kinetic energy and particle number density.^[20] To compute the functional derivative, consider the general formula for a local integral functional of the form \int f(\rho(\mathbf{r})) \, d^3\mathbf{r}, where the variation yields \frac{\delta}{\delta\rho(\mathbf{r})} \int f(\rho(\mathbf{r}')) \, d^3\mathbf{r}' = f'(\rho(\mathbf{r})).^[19] Here, f(u) = C_F u^{5/3}, so f'(u) = \frac{5}{3} C_F u^{2/3}. Thus, the functional derivative is

\frac{\delta T_{\mathrm{TF}}}{\delta\rho(\mathbf{r})} = \frac{5}{3} C_F \rho^{2/3}(\mathbf{r}).

^[19] This result follows directly from the power-law structure of the integrand, as the derivative with respect to \rho at a specific point \mathbf{r} depends only on the local density value. Physically, this functional derivative represents the local contribution to the effective potential from the kinetic energy in the Thomas–Fermi approximation, corresponding to the chemical potential of a uniform Fermi gas at density \rho(\mathbf{r}).^[19] In atomic physics, it enables self-consistent solutions for electron density under external potentials, providing a foundational semiclassical estimate for heavy atoms where quantum shell effects are averaged out.^[20]

von Weizsäcker kinetic energy functional

The von Weizsäcker kinetic energy functional provides a gradient-dependent correction to approximate the non-interacting kinetic energy of electrons in terms of the density, capturing effects of density inhomogeneity beyond local approximations. It is defined as

T_\mathrm{vW}[\rho] = \frac{\lambda}{8} \int \frac{|\nabla \rho(\mathbf{r})|^2}{\rho(\mathbf{r})} \, d^3\mathbf{r},

where \rho(\mathbf{r}) is the electron density and \lambda is a dimensionless parameter that equals 1 for the exact representation in single-orbital (Bohmian) systems.^[21] This form arises from expressing the kinetic energy operator for a real single-particle wave function \psi(\mathbf{r}) with \rho = \psi^2, leading to T = -\frac{1}{2} \int \psi \nabla^2 \psi \, d^3\mathbf{r}. Substituting \nabla^2 \rho = 2 \psi \nabla^2 \psi + 2 |\nabla \psi|^2 and applying the Green–Gauss theorem, with the surface integral vanishing due to exponential decay of \psi at infinity, yields \int \nabla^2 \rho \, d^3\mathbf{r} = 0, relating |\nabla \psi|^2 = \frac{1}{4} \frac{|\nabla \rho|^2}{\rho} and thus the functional for \lambda = 1.^[21] Originally proposed by Carl Friedrich von Weizsäcker in 1935, it serves as a quantum correction emphasizing shell structure and delocalization in atomic systems. The functional derivative, essential for variational applications, is obtained via the Euler–Lagrange equation for functionals depending on \rho and \nabla \rho. Let f(\rho, \nabla \rho) = \frac{\lambda}{8} \frac{|\nabla \rho|^2}{\rho}. The partial derivative with respect to \rho is \frac{\partial f}{\partial \rho} = -\frac{\lambda}{8} \frac{|\nabla \rho|^2}{\rho^2}, while \frac{\partial f}{\partial (\nabla \rho)} = \frac{\lambda}{4} \frac{\nabla \rho}{\rho}. The divergence term is \nabla \cdot \left( \frac{\partial f}{\partial (\nabla \rho)} \right) = \frac{\lambda}{4} \frac{\nabla^2 \rho}{\rho} - \frac{\lambda}{4} \frac{|\nabla \rho|^2}{\rho^2}. Thus,

\frac{\delta T_\mathrm{vW}[\rho]}{\delta \rho(\mathbf{r})} = \frac{\partial f}{\partial \rho} - \nabla \cdot \left( \frac{\partial f}{\partial (\nabla \rho)} \right) = -\frac{\lambda}{8} \frac{|\nabla \rho(\mathbf{r})|^2}{\rho(\mathbf{r})^2} - \frac{\lambda}{4} \nabla^2 \ln \rho(\mathbf{r}),

derived through integration by parts of the variation \delta T_\mathrm{vW} = \int \left[ \frac{\partial f}{\partial \rho} \delta \rho + \frac{\partial f}{\partial (\nabla \rho)} \cdot \nabla (\delta \rho) \right] d^3\mathbf{r}, with boundary terms neglected.^[22] This expression, equivalent to \frac{\lambda}{8} \frac{|\nabla \rho|^2}{\rho^2} - \frac{\lambda}{4} \frac{\nabla^2 \rho}{\rho} using \nabla^2 \ln \rho = \frac{\nabla^2 \rho}{\rho} - \frac{|\nabla \rho|^2}{\rho^2}, corresponds to the Bohm quantum potential -\frac{\lambda}{2} \frac{\nabla^2 \sqrt{\rho}}{\sqrt{\rho}} for \lambda = 1.^[22] In the context of gradient expansions, the von Weizsäcker term emerges as the second-order correction to the Thomas–Fermi functional, where the uniform-gas kinetic energy T_\mathrm{TF}[\rho] = \frac{3}{10} (3\pi^2)^{2/3} \int \rho^{5/3} \, d^3\mathbf{r} is augmented by inhomogeneity effects; the coefficient \lambda = \frac{1}{9} from this expansion balances slow variations in density, while \lambda = 1 exactly recovers the kinetic energy for rapidly varying or one-electron densities, thereby incorporating quantum delocalization absent in the local Thomas–Fermi approximation.^[22]

Entropy and exponential functionals

In information theory, the Shannon entropy functional quantifies the uncertainty associated with a probability density function p(x) over a continuous domain, defined as

S = -\int p(x) \ln p(x) \, dx,

where the integral is taken over the appropriate space and p(x) satisfies \int p(x) \, dx = 1 and p(x) \geq 0.^[23] This functional serves as a measure of information content and is central to deriving probability distributions under constraints. To compute the functional derivative \frac{\delta S}{\delta p(x)}, consider a small variation p(x) \to p(x) + \epsilon \eta(x), where \epsilon is infinitesimal and \eta(x) is a test function satisfying the normalization constraint \int \eta(x) \, dx = 0. The varied entropy is

S[p + \epsilon \eta] = -\int (p + \epsilon \eta) \ln (p + \epsilon \eta) \, dx.

Using the Taylor expansion \ln(p + \epsilon \eta) = \ln p + \ln(1 + \epsilon \eta / p) \approx \ln p + \epsilon \eta / p - \frac{1}{2} (\epsilon \eta / p)^2 + \cdots, the first-order term in \epsilon yields

S[p + \epsilon \eta] \approx S + \epsilon \int \eta(x) (-\ln p(x) - 1) \, dx.

By definition of the functional derivative, this implies

\frac{\delta S}{\delta p(x)} = -\ln p(x) - 1.[23]

This result highlights the logarithmic nature of the variation, arising from the expansion of the logarithm. The functional derivative of the entropy plays a key role in the principle of maximum entropy, where distributions are found by maximizing S subject to constraints such as fixed moments \int p(x) f_k(x) \, dx = a_k. The stationarity condition \frac{\delta S}{\delta p(x)} + \sum_k \lambda_k f_k(x) = 0 leads to p(x) \propto \exp\left(-\sum_k \lambda_k f_k(x)\right), yielding canonical distributions like the Gaussian or exponential under appropriate constraints.^[23] This approach ensures the selected distribution is the least informative consistent with the data, promoting applications in Bayesian inference and statistical modeling beyond physics. As an example of a nonlinear functional, consider J = \exp\left(\int f(x) \, dx\right), which depends exponentially on the integral of f. Varying f(x) \to f(x) + \epsilon \eta(x) gives

J[f + \epsilon \eta] = \exp\left(\int f \, dx + \epsilon \int \eta \, dx\right) = J \exp\left(\epsilon \int \eta \, dx\right) \approx J \left(1 + \epsilon \int \eta(y) \, dy\right).

The first-order variation is thus \epsilon J \int \eta(y) \, dy, implying

\frac{\delta J}{\delta f(x)} = J.

This derivative is independent of the specific point x and equals the full functional value, reflecting its nonlocal character: a local change in f at any point affects the global integral uniformly. Such forms appear in generating functions and path integral formulations, illustrating how functional derivatives capture multiplicative structure in exponential dependencies.

Applications

In density functional theory

In density functional theory (DFT), functional derivatives play a pivotal role in mapping the electron density \rho(\mathbf{r}) to the ground-state energy and properties of many-electron systems. The Hohenberg-Kohn theorems establish that the external potential v_{\text{ext}}(\mathbf{r}) is uniquely determined (up to an additive constant) by the ground-state density, implying a one-to-one correspondence between \rho and the total energy functional E[\rho]. This uniqueness relies on functional derivatives, as variations in density under constraints yield the Euler-Lagrange equation \frac{\delta E[\rho]}{\delta \rho(\mathbf{r})} = \mu, where \mu is the chemical potential, ensuring the density-to-energy mapping is invertible and variational. The second theorem provides a variational principle: the true ground-state energy is the minimum of E[\rho] over all densities yielding the correct particle number, with functional derivatives enforcing stationarity at the minimum.^[24] The Kohn-Sham formalism operationalizes these theorems by introducing a fictitious non-interacting system with the same density as the interacting one, leading to self-consistent equations for orbitals \psi_i. The effective potential in these equations is given by

v_{\text{eff}}(\mathbf{r}) = v_{\text{ext}}(\mathbf{r}) + \frac{\delta E_H[\rho]}{\delta \rho(\mathbf{r})} + \frac{\delta E_{\text{xc}}[\rho]}{\delta \rho(\mathbf{r})},

where E_H[\rho] is the Hartree electrostatic energy, and E_{\text{xc}}[\rho] is the exchange-correlation functional. The Hartree term is \frac{\delta E_H[\rho]}{\delta \rho(\mathbf{r})} = \int \frac{\rho(\mathbf{r}')}{|\mathbf{r} - \mathbf{r}'|} d\mathbf{r}'. This structure allows DFT to approximate many-body effects through density-dependent potentials derived via functional derivatives.^[25] Central to practical DFT is the exchange-correlation potential v_{\text{xc}}(\mathbf{r}) = \frac{\delta E_{\text{xc}}[\rho]}{\delta \rho(\mathbf{r})}, which captures all quantum mechanical exchange and correlation effects beyond the mean-field Hartree approximation. Approximations for E_{\text{xc}}[\rho] are essential, as its exact form is unknown; the local density approximation (LDA) assumes E_{\text{xc}}[\rho] \approx \int \rho(\mathbf{r}) \epsilon_{\text{xc}}(\rho(\mathbf{r})) d\mathbf{r}, where \epsilon_{\text{xc}}(\rho) is the exchange-correlation energy per particle in a uniform electron gas, yielding v_{\text{xc}}^{\text{LDA}}(\mathbf{r}) = \epsilon_{\text{xc}}(\rho(\mathbf{r})) + \rho(\mathbf{r}) \frac{d \epsilon_{\text{xc}}}{d \rho}(\rho(\mathbf{r})). LDA provides a foundational, computationally tractable starting point for v_{\text{xc}}, though it overestimates binding energies in molecules due to its local nature. More advanced functionals, such as generalized gradient approximations, build on this by incorporating density gradients, but all rely on functional derivatives to define the potential.^[25] In numerical implementations of Kohn-Sham DFT, functional derivatives appear in the Jacobians of self-consistent field (SCF) iterations, which solve the nonlinear eigenvalue problem iteratively. The SCF process updates the density \rho^{(n+1)}(\mathbf{r}) = \sum_i |\psi_i^{(n)}(\mathbf{r})|^2 from orbitals obtained in potential v_{\text{eff}}^{(n)}, converging when \rho^{(n+1)} \approx \rho^{(n)}. Preconditioning accelerates this by approximating the inverse Jacobian J = \frac{\delta \rho_{\text{out}}}{\delta \rho_{\text{in}}}, where the diagonal or elliptic approximations to J incorporate response kernels like \frac{\delta v_{\text{xc}}}{\delta \rho}, reducing iteration counts from hundreds to tens for large systems. This use of functional derivatives in Jacobians ensures robust convergence, particularly for challenging cases like transition metals. For instance, the Thomas-Fermi kinetic energy functional offers a simple semilocal approximation in early DFT applications.^[26]

In quantum field theory and statistical mechanics

In quantum field theory, the generating functional Z[J] serves as a central object for encoding vacuum expectation values and correlation functions through functional differentiation. It is defined via the path integral as

Z[J] = N \int \mathcal{D}\phi \, \exp\left( i \int d^4 x \left( \mathcal{L}[\phi] + J(x) \phi(x) \right) \right),

where N is a normalization constant, \mathcal{L}[\phi] is the Lagrangian density, and J(x) is an external source field coupled to the quantum field \phi(x). This formulation, introduced by Julian Schwinger, allows the expectation value of the field to be obtained as the first functional derivative: \langle \phi(x) \rangle_J = \frac{1}{i} \frac{\delta \ln Z[J]}{\delta J(x)}, with the subscript J indicating evaluation in the presence of the source. Setting J = 0 yields the vacuum expectation value in the absence of sources.^[27]^[28] Higher-order correlation functions, or Green's functions, are generated by successive functional derivatives of \ln Z[J]. Specifically, the connected n-point correlation function is given by

G^{(n)}_c(x_1, \dots, x_n) = (-i)^n \frac{\delta^n \ln Z[J]}{\delta J(x_1) \cdots \delta J(x_n)} \bigg|_{J=0},

which corresponds to the vacuum expectation value of the time-ordered product of n fields, \langle 0 | T \phi(x_1) \cdots \phi(x_n) | 0 \rangle. These functions capture the dynamical correlations in the theory and form the basis for perturbation theory, where Feynman diagrams represent expansions of these derivatives. This approach unifies the computation of scattering amplitudes and other observables in interacting quantum field theories.^[29]^[28] In statistical mechanics, particularly within the framework of statistical field theory, the partition function Z[J] plays an analogous role, defined as

Z[J] = \int \mathcal{D}\phi \, \exp\left( -\beta \int d^d x \left( F[\phi] - J(x) \phi(x) \right) \right),

where \beta = 1/(k_B T), F[\phi] is the free energy functional, and the source J(x) couples to the order parameter field \phi(x). Functional derivatives of \ln Z[J] yield response functions, such as susceptibilities. For instance, the linear response or susceptibility is \chi(x,y) = \beta \frac{\delta \langle \phi(x) \rangle}{\delta J(y)} \big|_{J=0} = -\beta^2 \frac{\delta^2 \ln Z[J]}{\delta J(x) \delta J(y)} \big|_{J=0}, which equals the connected two-point correlation function \langle \phi(x) \phi(y) \rangle_c. This connection links macroscopic susceptibilities, like magnetic susceptibility in Ising models, to microscopic fluctuations near critical points.^[30] Functional derivatives also play a crucial role in renormalization procedures through the effective action \Gamma[\phi_c], defined as the Legendre transform \Gamma[\phi_c] = -i \ln Z[J] - \int d^4 x \, J(x) \phi_c(x), where \phi_c(x) = \frac{\delta (i \ln Z)}{\delta J(x)} is the classical field. The n-point one-particle-irreducible (1PI) vertices, essential for renormalizing interactions, are obtained as \Gamma^{(n)}(x_1, \dots, x_n) = \frac{\delta^n \Gamma[\phi_c]}{\delta \phi_c(x_1) \cdots \delta \phi_c(x_n)} \big|_{\phi_c=0}. These vertices sum all 1PI diagrams and facilitate the renormalization group flow, allowing the absorption of ultraviolet divergences into renormalized parameters while preserving the structure of correlation functions. This framework, building on Schwinger's variational principles, enables non-perturbative treatments of quantum corrections in effective field theories.^[31]^[27]

History

Origins in calculus of variations

The concept of the functional derivative emerged from the foundational developments in the calculus of variations during the 18th century, where mathematicians sought to determine functions that extremize integrals representing physical or geometric quantities. Leonhard Euler laid the groundwork in his 1744 monograph Methodus inveniendi lineas curvas maximi minimive proprietate gaudentes sive solutio problematis isoperimetrici latissimo sensu accepti, in which he established that for a functional J = \int_{x_1}^{x_2} L(q, q', x) \, dx to achieve a maximum or minimum, its first variation must satisfy \delta J = 0. This variational principle yields the Euler-Lagrange equation \frac{\partial L}{\partial q} - \frac{d}{dx} \left( \frac{\partial L}{\partial q'} \right) = 0, which represents the condition for the functional derivative to vanish at the extremal function.^[32] Joseph-Louis Lagrange advanced this framework in his 1788 Mécanique Analytique, introducing a more rigorous and general method for handling variations. Lagrange employed the delta symbol \delta to denote infinitesimal changes in the function and its arguments, deriving the Euler-Lagrange equation systematically for problems in mechanics and geometry. His approach emphasized the analytical treatment of variations, transforming Euler's geometric insights into a calculus-based tool applicable to broader classes of functionals, such as those involving constraints.^[33] By the late 19th century, the study of variations evolved toward abstract functionals, with Vito Volterra providing key precursors to functional analysis in the 1880s. In his seminal 1887 paper "Sopra le funzioni che dipendono da altre funzioni," Volterra defined functionals as mappings from functions to real numbers—termed "functions of lines" or functions depending on infinite sets of values—and investigated their continuity and differentiability in variational contexts. This work bridged classical variational calculus with emerging ideas in integral equations, treating functions as elements in infinite-dimensional spaces and laying the analytical foundation for later functional derivatives.^[34] The early 20th century saw the rigorous formalization of derivatives in infinite-dimensional spaces. René-Louis Gâteaux introduced the directional derivative for functionals in his 1913–1919 works (published posthumously), while Maurice Fréchet developed the normed-space version in 1927, providing the mathematical basis for the Gâteaux and Fréchet derivatives that define the functional derivative in modern terms. Early uses of notation akin to the functional derivative \frac{\delta F}{\delta \phi} appeared in the variational analysis of classical problems, such as the brachistochrone (the curve of fastest descent under gravity) and isoperimetric problems (maximizing area for fixed perimeter). Euler applied delta variations to the brachistochrone in his 1744 work, computing changes in the time functional to identify the cycloid as the solution, while isoperimetric constraints were handled via auxiliary variations leading to analogous derivative conditions. David Hilbert's contributions, beginning with papers in 1904 and culminating in his 1912 book Grundzüge einer allgemeinen Theorie der linearen Integralgleichungen, further integrated these ideas, using integral equation theory to address existence and regularity in variational problems, solidifying the mathematical apparatus for functional derivatives.^[32]^[35]

Developments in 20th-century physics

In the early 20th century, Paul Dirac laid foundational groundwork for functional methods in quantum mechanics with his 1933 paper, where he proposed expressing the quantum mechanical transformation function between states as a functional of paths, weighted by the exponential of the classical action S. This approach implicitly incorporated functional variations to connect Lagrangian dynamics to quantum amplitudes, foreshadowing the use of functional derivatives in deriving equations of motion from path integrals.^[36] Building on Dirac's ideas, Richard Feynman in the 1940s formalized the path integral formulation, extending it to quantum electrodynamics (QED) through variational principles. In this framework, the stationarity of the action functional with respect to field variations yields the classical field equations via \frac{\delta S}{\delta \phi} = 0, where \phi represents field configurations, enabling the computation of quantum amplitudes and scattering processes in relativistic quantum theory.^[37] Julian Schwinger further advanced these concepts in the 1950s with his quantum action principle and source theory in quantum field theory (QFT). By introducing auxiliary source fields J(x) coupled to fields, Schwinger defined the generating functional Z[J] for Green's functions, from which correlation functions and propagators are extracted using functional derivatives such as \frac{\delta}{\delta J} applied to Z[J] or \log Z[J], providing a systematic variational approach to non-perturbative QFT calculations.^[38] The mid-1960s marked a pivotal development in many-body physics with the emergence of density functional theory (DFT). Pierre Hohenberg and Walter Kohn established that the ground-state energy E of an interacting electron system is a universal functional of the electron density \rho(\mathbf{r}), with the effective potential determined by the functional derivative v(\mathbf{r}) = \frac{\delta E[\rho]}{\delta \rho(\mathbf{r})}, proving the one-to-one mapping between densities and potentials.^[24] Walter Kohn and Lu Jeu Sham then provided a practical implementation by deriving self-consistent equations for a fictitious non-interacting system, where the exchange-correlation potential is v_{xc}(\mathbf{r}) = \frac{\delta E_{xc}[\rho]}{\delta \rho(\mathbf{r})}, revolutionizing computational quantum chemistry and solid-state physics.^[25]

References

[1]
[PDF] Waves and Imaging, Calculus of Variations, Functional Derivatives
It answers the question of how to differentiate with respect to functions, i.e., objects with an uncountable, infinite number of degrees of freedom.
[2]
[PDF] The Calculus of Variations - UC Davis Math
Thus, we may interpret the value of the functional derivative δS/δx at t as describ- ing the change in the values of the functional S(x) due to changes in the ...
[3]
[PDF] 17 calculus of
... functional derivative has a factor of one over volume relative to a partial derivative. An alternative, more formal, definition of a functional derivative is.
[4]
Functional -- from Wolfram MathWorld
A functional is a real-valued function on a vector space , usually of functions. For example, the energy functional on the unit disk assigns a number to any ...<|control11|><|separator|>
[5]
[PDF] Calculus of Variation
The present course is based on lectures given by I. M. Gelfand in the Mechanics and Mathematics Department of Moscow State University.
[6]
Functional Derivative -- from Wolfram MathWorld
The functional derivative is a generalization of the usual derivative that arises in the calculus of variations.
[7]
[PDF] Functionals and the Functional Derivative - Chimera
functional derivative corresponds to the second order total differential,. 2 More precisely, a functional F[f] which maps an open subset of some Banach space ...
[8]
[PDF] 13 Variational Calculus - MIT Mathematics
satisfies the functional differential equation. δF[f] δf(y). = c(y)F[f]. (15) ... This proposal marked the real beginning of general interest in the calculus of ...
[9]
[PDF] Introduction of Fréchet and Gâteaux Derivative - m-hikari.com
In this reviwe we contrast Fréchet derivative to the more general Gâteaux derivative. Both derivative are often used to formalize the functional derivative ...
[10]
[PDF] Gateaux differentials and Frechet derivatives - TTU Math
Mar 27, 2025 · The Frechet derivative Df of f : V → U is defined implicitly as the linear operator Df such that f(x + k) = f(x)+(Df) k + o(∥k∥). f(x + ch) = f ...
[11]
[PDF] II. Variation of functionals
Thus, it is not directly related to the continuity of a functional. For this purpose, another differential called Fréchet differential has been put forth.<|control11|><|separator|>
[12]
None
### Summary of Functional Derivatives from Field Theory Chapter 5 (Pages 3-12)
[13]
[PDF] Calculus of Variations
... Calculus of Variations. According to Eq. (20), the variation δF is a linear functional of δf. The linearity means that the sign of δF changes upon the ...
[14]
None
### Summary of Functionals and Functional Derivatives
[15]
[PDF] Functional Methods in Quantum Mechanics
May 3, 2013 · Functional methods are central in modern physics, mapping classes of functions to real numbers, like a function maps points to real numbers.Missing: ∫ | Show results with:∫
[16]
[PDF] 15 Functional Derivatives - Chimera
δf(y)= -2Z dx f//(x)δ(x - y) = -2f//(y). (15.10). Page 3. 15.2 Functional ... Thus y(x) is a straight line, y = mx + b. 15.3 Higher-Order Functional Derivatives.
[17]
[PDF] arXiv:1704.08943v2 [cond-mat.mtrl-sci] 13 Feb 2018
Feb 13, 2018 · We have formulated and implemented MGP, a new nonlocal Kinetic energy functional with a density independent kernel. In MGP, the inverse ...Missing: formula | Show results with:formula
[18]
https://www.researchgate.net/publication/308277471_On_the_functional_derivative_with_respect_to_the_Dirac_Delta
[19]
(PDF) On the functional derivative with respect to the Dirac Delta
Sep 22, 2016 · On the functional derivative with respect to the Dirac Delta. October ... test function sequence for the Dirac delta. with respect to the ...
[20]
[PDF] The role of the rigged Hilbert space in Quantum Mechanics - arXiv
Feb 9, 2005 · In particular, Dirac's bra-ket formalism is fully implemented by the rigged Hilbert space rather than just by the Hilbert space. In this paper, ...
[21]
[PDF] Testing the kinetic energy functional
Therefore the functional derivative of the kinetic energy is just the negative of the. Kohn–Sham potential, up to a constant. As we restrict our- selves here to ...
[22]
Pedagogic notes on Thomas-Fermi theory (and on some ...
Jan 1, 1991 · We will review the applications of the theory, from both qualitative and (Thomas-Fermi) quantitative viewpoints, to heavy atoms, where we are concerned with a ...
[23]
[PDF] Derivation of von Weizsäcker Equation Based on Green–Gauss ...
A simple and short derivation of von Weizsäcker equation for kinetic energy functional is presented. The derivation is based on the Green–Gauss theorem and is ...
[24]
https://link.aps.org/doi/10.1103/PhysRev.136.B864
[25]
[PDF] Information Theory and Statistical Mechanics
Reprinted from THE PHYSICAL REVIEW, Vol. 106, No. 4, 620-630, May 15, 1957. Printed in U. S. A.. Information Theory and Statistical Mechanics.
[26]
Inhomogeneous Electron Gas | Phys. Rev.
This paper deals with the ground state of an interacting electron gas in an external potential, and proves a universal functional of the density.
[27]
Self-Consistent Equations Including Exchange and Correlation Effects
From a theory of Hohenberg and Kohn, approximation methods for treating an inhomogeneous system of interacting electrons are developed.
[28]
Accelerating self-consistent field iterations in Kohn-Sham density ...
We present an efficient preconditioning technique for accelerating the fixed-point iteration in real-space Kohn-Sham density functional theory (DFT) ...
[29]
The Theory of Quantized Fields. I | Phys. Rev.
The surface generating operator also leads to generalized Schrödinger equations for the representative of an arbitrary state. Action integral variations which ...
[30]
[PDF] B1. The generating functional Z[J] - UBC Physics & Astronomy
The generating functional Z[J] is a key object in quantum field theory - as we shall see it reduces in ordinary quantum mechanics to a limiting form of the ...
[31]
[PDF] Correlation Functions and Diagrams - UMD Physics
That is, the correlation functions are defined by functional differentiation of the gener- ating functional with respect to the sources. We can also define ...
[32]
[PDF] Statistical Field Theory - DAMTP
Statistical Field Theory is about phase transitions, abrupt changes when matter is deformed, and is aimed at beginning graduate students with a background in ...
[33]
[PDF] Lecture 12 The Effective Action as a Generating Functional
Here we will rederive the effective potential by making use of the fact that the effective action is a generating functional of the one-particle-irreducible ...
[34]
[PDF] LEONHARD EULER, BOOK ON THE CALCULUS OF VARIATIONS ...
In this book Euler extended known methods of the calculus of variations to form and solve differential equations for the general problem of optimizing ...
[35]
[PDF] J. L. Lagrange's changing approach to the foundations of the ...
A central topic of this study concerns LAGRANGE'S changing derivation of the so-called EULER-LAGRANGE equations. Since the calculus of variations in its.
[36]
[PDF] The Establishment of Functional Analysis
This article surveys the evolution of functional analysis, from its origins to its establish- ment as an independent discipline around 1933.
[37]
[PDF] Grundzüge einer allgemeinen Theorie der linearen ...
Im vorliegenden Buche bringe ich meine sechs Mitteilungen „Grund- züge einer allgemeinen Theorie der linearen Integralgleichungen" im.
[38]
[PDF] THE LAGRANGIAN IN QUANTUM MECHANICS. By PAM Dirac.
The Lagrangian in Quantum Mechanics. 65 to take over the classical Lagrangian equations in any very. ~iirect way. These equations involve partial derivatives of.
[39]
Space-Time Approach to Non-Relativistic Quantum Mechanics
Space-Time Approach to Non-Relativistic Quantum Mechanics. R. P. Feynman ... J. A. Wheeler and R. P. Feynman, Rev. Mod. Phys. 17, 157 (1945) Ph.D ...Missing: URL | Show results with:URL