Last modified: July 06, 2022

This article is written in: 🇺🇸

Partial Differential Equations (PDEs)

A partial differential equation (PDE) is an equation that involves:

I. Multiple independent variables, typically denoted $x, y, z$ (or $x_1, x_2, \ldots, x_d$ in $d$-dimensional space), and often $t$ if time is also included.

II. One (or more) dependent variable(s), which we denote by $u(x_1, x_2, \ldots, x_d)$ or $u(\mathbf{x}, t)$.

III. The partial derivatives of $u$ with respect to these independent variables.

Formally, a PDE can be written as:

$$F\bigl( x_1, \ldots, x_d, u, u_{x_1}, u_{x_2},\ldots, u_{x_1 x_1}, u_{x_1 x_2},\ldots \bigr) = 0$$

where $u_{x_i}$ denotes the partial derivative of $u$ with respect to $x_i$, and $u_{x_i x_j}$ denotes second partial derivatives, and so on.

In ODEs, there is only one independent variable, whereas in PDEs, there are multiple independent variables.

Understanding Partial Differential Equations

PDEs capture how a function $u(\mathbf{x}, t)$ varies in multiple directions or with time. They arise in virtually all areas of physics, engineering, finance, and many other fields:

• In Physics, continuum mechanics uses stress-strain analysis, electromagnetics employs Maxwell’s equations, fluid dynamics applies the Navier–Stokes equations, and quantum mechanics is based on the Schrödinger equation.
• In Engineering, heat transfer is modeled by the heat equation, wave propagation is described using the wave equation, and elasticity alongside structural analysis is approached through mathematical models.
• In Economics and Finance, the Black–Scholes equation is used for option pricing while dynamic optimization handles multiple variables to analyze decision-making processes.
• In Biology, reaction-diffusion systems are applied to population genetics and pattern formation in developmental biology is explored to understand the emergence of structures.

Boundary and Initial Conditions

To uniquely solve a PDE, one typically needs to specify boundary conditions (BCs) and/or initial conditions (ICs), depending on the PDE’s type and physical context:

Boundary conditions:

I. Dirichlet BC:

This condition requires specifying the value of the function $u$ directly on the boundary; mathematically, it is expressed as

$$u\big|_{\partial \Omega} = f$$

where $f$ is a prescribed function on the boundary $\partial \Omega$. This is typically used in scenarios where the state of the system is fixed at the boundary.

II. Neumann BC:

This condition involves setting the normal derivative of the function $u$ on the boundary; it is given by

$$\frac{\partial u}{\partial n}\bigg|_{\partial \Omega} = g$$

where $g$ is a known function defined on $\partial \Omega$ and $\frac{\partial u}{\partial n}$ denotes the derivative in the direction normal to the boundary. This is useful when the flux across the boundary is specified.

III. Robin (or mixed) BC:

This condition combines both the function value and its normal derivative on the boundary, and is formulated as

$$\alpha u + \beta \frac{\partial u}{\partial n}\bigg|_{\partial \Omega} = h$$

where $\alpha$, $\beta$, and $h$ are given functions (or constants) on $\partial \Omega$. This type of condition is applied when both the state of the system and its flux at the boundary are influenced by external factors.

Initial conditions:

When time $t$ is involved (often in parabolic or hyperbolic PDEs), one typically specifies the initial state of $u$ (and possibly some derivatives) at $t = t_0$. For instance, for the heat equation:

$$u(\mathbf{x}, t_0) = \phi(\mathbf{x}).$$

The combination of a PDE with its boundary/initial conditions is referred to as a boundary value problem (BVP) or an initial-boundary value problem (IBVP) or initial value problem (IVP) if the domain is unbounded in space.

Main Concepts in PDEs

Order of a PDE

As with ODEs, the order of a PDE is determined by the highest order partial derivative that appears. For instance:

• A first-order PDE involves only first partial derivatives (e.g., $u_x$, $u_t$).
• A second-order PDE can have second partial derivatives like $u_{xx}$, $u_{xy}$, $u_{tt}$, etc.

Linearity vs. Nonlinearity

A PDE is linear if $u$ and its partial derivatives appear only in the first power (i.e., no products of partial derivatives or higher powers/sin/log of $u$) and if each coefficient depends at most on the independent variables (but not on $u$ or its derivatives).

For a second-order PDE in two variables $(x,t)$, a linear PDE can often be expressed as:

$$a(x,t)u_{xx} + 2b(x,t)u_{xt} + c(x,t)u_{tt} + d(x,t)u_x + e(x,t)u_t + f(x,t)u = g(x,t)$$

• Homogeneous if $g(x,t) \equiv 0$.
• Nonhomogeneous if $g(x,t) \neq 0$.

A PDE that is not linear is nonlinear. Examples include the Navier–Stokes equations, the nonlinear Schrödinger equation, or the Fisher–KPP equation in biology.

Semilinear, Quasilinear, and Fully Nonlinear

When discussing PDEs of order $n$, we often distinguish:

• Semilinear: The highest-order derivatives appear linearly, but lower-order terms may be nonlinear in $u$.
• Quasilinear: The highest-order derivatives appear linearly in each of them but the coefficients may depend on $u$ and its lower-order derivatives.
• Fully nonlinear: The PDE cannot be put in a form in which the highest-order partial derivatives appear linearly and separate from one another.

For instance, a quasilinear second-order PDE in two variables might look like:

$$a\bigl(x,t,u,u_x,u_t\bigr)u_{xx} + 2b\bigl(x,t,u,u_x,u_t\bigr)u_{xt} + c\bigl(x,t,u,u_x,u_t\bigr)u_{tt} = g\bigl(x,t,u,u_x,u_t\bigr)$$

Classification of Second-Order PDEs

For a second-order PDE in two independent variables $x$ and $t$ (or $x$ and $y$), one often writes it in the form

$$A(x,t)u_{xx} + 2B(x,t)u_{xt} + C(x,t)u_{tt} + \cdots = 0$$

(Plus lower-order terms omitted for brevity.) The discriminant $\Delta$ is given by:

$$\Delta = B^2 - AC$$

Elliptic if $\Delta < 0$

• Classic example: Laplace’s equation, $\nabla^2 u = u_{xx} + u_{yy} = 0$.
• Elliptic PDEs often describe steady-state phenomena (e.g., electric potential, steady heat distribution).

Parabolic if $\Delta = 0$.

• Classic example: Heat (or diffusion) equation, $u_t = \alpha u_{xx}$.
• Parabolic PDEs often describe time-evolving diffusion-type phenomena.

Hyperbolic if $\Delta > 0$.

• Classic example: Wave equation, $u_{tt} = c^2u_{xx}$.
• Hyperbolic PDEs typically model wave propagation or signals at finite speed.

This classification extends to higher dimensions and helps determine the nature of the PDE (well-posedness, appropriate boundary conditions, solution methods, etc.).

Forms of Common PDEs

Elliptic PDEs

Laplace’s Equation:

$$\nabla^2 u = 0 \quad \Leftrightarrow \quad u_{xx} + u_{yy} = 0 (\text{in 2D}), \quad u_{xx} + u_{yy} + u_{zz} = 0 (\text{in 3D}), \dots$$

Describes steady-state temperature distribution, gravitational/electrostatic potential.

Poisson’s Equation:

$$\nabla^2 u = f(\mathbf{x})$

A nonhomogeneous extension of Laplace’s equation.

Parabolic PDEs

Heat (Diffusion) Equation:

$$u_t = \alpha\nabla^2 u \quad \text{(e.g., in 1D: } u_t = \alpha u_{xx}\text{).}$$

Models heat diffusion, particle diffusion in fluids, etc. It evolves in time toward an equilibrium (steady state).

Hyperbolic PDEs

Wave Equation:

$$u_{tt} = c^2 \nabla^2 u \quad \text{(in 1D: } u_{tt} = c^2u_{xx}\text{).}$$

Models vibrations of a string, sound waves, electromagnetic waves in simplified settings.

Nonlinear PDEs

Navier–Stokes Equations for fluid flow:

$$\rho \left(\frac{\partial \mathbf{v}}{\partial t} + \mathbf{v}\cdot \nabla \mathbf{v}\right) = -\nabla p + \mu \nabla^2 \mathbf{v} + \mathbf{F}$$

where $\mathbf{v}$ is velocity field, $p$ is pressure, $\rho$ density, $\mu$ viscosity. Highly nonlinear.

Nonlinear Schrödinger Equation:

$$i\psi_t + \alpha\nabla^2 \psi + \beta|\psi|^2 \psi = 0$$

arises in optics, quantum mechanics for certain approximations.

Reaction-Diffusion Equations:

$$u_t = D\Delta u + R(u)$$

model chemical reactions combined with diffusion. Nonlinear if $R(u)$ is nonlinear.

Methods of Solving PDEs

Analytical Methods

I. Separation of Variables:

• Assume $u(x,t) = X(x)T(t)$ (or in higher dimensions, products of single-variable functions).
• Transform the PDE into ODEs for $X(x)$ and $T(t)$.
• Typically used for linear PDEs with nice boundary/initial conditions (heat, wave, Laplace equations).

II. Fourier and Laplace Transforms:

• Useful for PDEs on infinite or semi-infinite domains.
• Transform w.r.t. space and/or time reduces PDE to ODE or algebraic equation in transform space.

III. Method of Characteristics:

• Commonly used for first-order PDEs, such as transport equations, or for certain quasilinear PDEs.
• Convert PDE into a set of ODEs describing characteristic curves along which PDE becomes an ODE.

IV. Green’s Functions:

• Integral operator method primarily for linear, inhomogeneous PDEs.
• Builds solutions from fundamental solutions of simpler PDEs (like the Dirac delta response).

Numerical Methods

For more general PDEs—especially nonlinear PDEs or higher-dimensional problems—analytical solutions may be either impossible or extremely difficult to obtain in closed form. In such cases, numerical approximation is crucial: - Finite Difference Methods (FDM): Approximate derivatives via differences on a grid. - Finite Element Methods (FEM): Approximate $u$ by basis functions on a mesh, widely used in engineering (structural analysis, fluid flow, etc.). - Finite Volume Methods (FVM): Common in computational fluid dynamics, conserves fluxes across cell boundaries. - Spectral Methods: Approximate $u$ using trigonometric (Fourier) or polynomial expansions (e.g., Chebyshev polynomials).

Existence and Uniqueness Theorems

Unlike ODEs, where Picard–Lindelöf gives a neat existence and uniqueness result, PDE theory is much richer and more nuanced. A few highlights:

• Elliptic PDEs often rely on tools like the Lax–Milgram theorem (for linear, elliptic PDEs in weak form), Schauder estimates, Sobolev space theory, etc.
• Parabolic PDEs: Existence and uniqueness often shown via semigroup theory (fractional step methods), Galerkin methods, or energy estimates.
• Hyperbolic PDEs: Typically rely on finite speed of propagation arguments, energy estimates, method of characteristics (for first-order or specific second-order problems). Shock formation in nonlinear hyperbolic PDEs complicates uniqueness.

Examples of Partial Differential Equations

I. Laplace’s Equation $\nabla^2 u = 0$:

• Governs steady heat distribution or electrostatic potential in a region $\Omega\subseteq \mathbb{R}^n$.
• Usually accompanied by boundary conditions like $u|_{\partial \Omega} = f(\mathbf{x})$.

II. Heat Equation $u_t = \alpha u_{xx}$:

• Models temperature evolution in a rod.
• Often with initial condition $u(x,0) = \phi(x)$ and boundary conditions (Dirichlet or Neumann).

III. Wave Equation $u_{tt} = c^2u_{xx}$:

• Models vibrations of a string or waves on a membrane (in 2D).
• Typically has initial conditions for displacement and velocity, plus boundary conditions if the domain is finite.

IV. Navier–Stokes for incompressible fluid flow:

$$\begin{cases} \frac{\partial \mathbf{v}}{\partial t} + (\mathbf{v}\cdot \nabla)\mathbf{v} = -\frac{1}{\rho}\nabla p + \nu\nabla^2 \mathbf{v} + \mathbf{f}, \ \nabla \cdot \mathbf{v} = 0, \end{cases}$$

where $\mathbf{v}$ is velocity, $p$ pressure, $\nu$ kinematic viscosity, $\mathbf{f}$ body forces (like gravity).

V. Black–Scholes Equation (Finance)

$$\frac{\partial V}{\partial t} + \tfrac{1}{2}\sigma^2 S^2 \frac{\partial^2 V}{\partial S^2} + rS\frac{\partial V}{\partial S} - rV = 0$$

where $V(S,t)$ is the price of an option, $S$ is the underlying asset price, $r$ is risk-free rate, $\sigma$ volatility.

Applications of PDEs

• In physics, partial differential equations describe phenomena such as material elasticity, electromagnetic fields using Maxwell’s equations, gravitational fields through Einstein's equations, and quantum states via Schrödinger or Dirac equations.
• In engineering, they are essential for analyzing structures, designing efficient heat exchangers, modeling fluid flows (whether compressible or not), and studying aerodynamic forces.
• Biology uses these equations to simulate reaction-diffusion systems that explain pattern formation like Turing patterns, predict population genetics trends, and model tumor growth dynamics.
• In finance, partial differential equations help price derivatives using models like Black–Scholes or Heston and assess risks in financial systems.
• For geosciences, these equations support geological modeling, simulate the propagation of seismic waves, and guide reservoir simulation for resource management.

Limitations and Complexities

• Existence and regularity of solutions to PDEs can be challenging to prove, particularly for nonlinear equations, requiring tools from advanced functional analysis such as Sobolev spaces and distributions.
• Nonlinear phenomena like shock waves, turbulence, and pattern formation add complexity to PDE theory, with some aspects remaining unresolved, such as the Navier–Stokes regularity problem in 3D.
• Boundary and initial conditions must be carefully chosen to align with the PDE classification (elliptic, parabolic, or hyperbolic) to ensure the problem is well-posed and avoids issues like non-unique or highly sensitive solutions.
• Dimensionality in real-world scenarios often leads to PDEs in 4D or higher (e.g., (x, y, z, t)), making accurate solutions computationally intensive due to the curse of dimensionality.
• Parameter sensitivity in PDEs means that small changes in physical or material parameters can lead to significant shifts in behavior, such as the onset of turbulence in fluid dynamics.