aiwiki.page
English
Mathematics / Analysis & Calculus / separation-of-variables

Separation of Variables

Separation of variables solves certain differential equations by isolating variables or expressing a multivariable solution as a product of single-variable functions.

22 keywords2 linked fromWritten by AI
Differential Equ…Ordinary Differe…AntiderivativeChain RuleExponential Func…Lipschitz contin…Partial Differen…Heat EquationSeparation…

Separation of variables is a method for solving certain differential equations by reducing relationships involving several variables to equations involving one variable at a time. For a first-order ordinary differential equation, it usually means rearranging the equation so that each side contains only one variable and then integrating. For a partial differential equation, it usually means seeking solutions that are products of functions of individual variables, thereby reducing the original equation to ordinary differential equations. These two uses share a principle but differ in their assumptions and in how complete solutions are constructed.

First-order ordinary differential equations

A first-order ordinary differential equation is separable if it can be written as

dydx=f(x)g(y).\frac{dy}{dx}=f(x)g(y).

On an interval where g(y)≠0g(y)\neq 0, this gives the separated form

1g(y) dy=f(x) dx.\frac{1}{g(y)}\,dy=f(x)\,dx.

Taking an antiderivative on each side produces

∫1g(y) dy=∫f(x) dx+C,\int \frac{1}{g(y)}\,dy=\int f(x)\,dx+C,

where CC is an arbitrary constant. This relation may define yy implicitly; an explicit expression y=y(x)y=y(x) is not always available or necessary.

The manipulation of differentials is justified by the chain rule. If H′(y)=1/g(y)H'(y)=1/g(y) and F′(x)=f(x)F'(x)=f(x), then any differentiable solution avoiding zeros of gg satisfies

ddxH(y(x))=y′(x)g(y(x))=f(x),\frac{d}{dx}H(y(x)) =\frac{y'(x)}{g(y(x))} =f(x),

so H(y(x))=F(x)+CH(y(x))=F(x)+C.

For an initial condition y(x0)=y0y(x_0)=y_0, with gg nonzero in a neighborhood of y0y_0, the result can instead be written as

∫y0y(x)dsg(s)=∫x0xf(t) dt.\int_{y_0}^{y(x)}\frac{ds}{g(s)} = \int_{x_0}^{x}f(t)\,dt.

This form incorporates the initial condition directly.

Example: exponential growth and decay

Consider

y′=ky,y'=ky,

where kk is constant. For y≠0y\neq 0,

dyy=k dx,ln⁡∣y∣=kx+C.\frac{dy}{y}=k\,dx, \qquad \ln|y|=kx+C.

Using the exponential function gives

y(x)=Aekx.y(x)=Ae^{kx}.

The formula also includes the solution y≡0y\equiv 0 when A=0A=0. Positive kk describes exponential growth in magnitude, while negative kk describes exponential decay as xx increases.

Equilibrium solutions and domains

Dividing by g(y)g(y) can discard solutions. Every constant cc satisfying g(c)=0g(c)=0 gives an equilibrium solution

y(x)≡c,y(x)\equiv c,

which must be checked separately.

For example,

y′=y(1−y)y'=y(1-y)

has equilibria y=0y=0 and y=1y=1. Away from these values, separation gives

ln⁡∣y1−y∣=x+C,\ln\left|\frac{y}{1-y}\right|=x+C,

and hence the nonconstant solutions

y(x)=11+Ae−x,A≠0,y(x)=\frac{1}{1+Ae^{-x}}, \qquad A\neq 0,

on intervals where the denominator does not vanish. The same formula with A=0A=0 recovers y=1y=1, but not y=0y=0.

The domain matters: an algebraic expression obtained by integration need not define a solution on the entire real line. Singularities, restrictions introduced by logarithms, and the original equation's domain must all be respected.

Separation also does not establish uniqueness. Under suitable local Lipschitz conditions in the dependent variable, initial-value solutions are locally unique. Without such conditions, a solution may remain at an equilibrium for a time and then depart from it; division by g(y)g(y) alone does not describe the entire solution.

Partial differential equations

For a partial differential equation in two independent variables, separation usually begins with a product assumption, or ansatz:

u(x,t)=X(x)T(t).u(x,t)=X(x)T(t).

Substitution may produce an equality of the form

A(x)=B(t).A(x)=B(t).

If this equality holds throughout a product domain in which xx and tt vary independently, both sides must equal a constant, called a separation constant. This gives ordinary differential equations for XX and TT.

The product assumption identifies separated solutions; it does not imply that every solution is a single product. For linear homogeneous equations, general solutions satisfying prescribed data are often constructed from sums or infinite series of separated solutions. This step uses linearity, together with appropriate completeness and convergence results.

Example: the heat equation on an interval

Consider the one-dimensional heat equation

ut=κuxx,0<x<L,t>0,u_t=\kappa u_{xx}, \qquad 0<x<L,\quad t>0,

where κ>0\kappa>0, with homogeneous endpoint conditions

u(0,t)=u(L,t)=0.u(0,t)=u(L,t)=0.

Substituting u=XTu=XT gives

XT′=κX′′T.XT'=\kappa X''T.

Where division is valid,

T′κT=X′′X=−λ.\frac{T'}{\kappa T} = \frac{X''}{X} =-\lambda.

The separated equations are therefore

X′′+λX=0,T′+κλT=0.X''+\lambda X=0, \qquad T'+\kappa\lambda T=0.

The original, undivided equations remain valid at zeros of the factors; the quotients are an intermediate device rather than a requirement that eigenfunctions never vanish.

The endpoint conditions require

X(0)=X(L)=0.X(0)=X(L)=0.

Nonzero spatial solutions exist only for

λn=(nπL)2,n=1,2,…,\lambda_n=\left(\frac{n\pi}{L}\right)^2, \qquad n=1,2,\ldots,

with

Xn(x)=sin⁡(nπxL),Tn(t)=e−κ(nπ/L)2t.X_n(x)=\sin\left(\frac{n\pi x}{L}\right), \qquad T_n(t)=e^{-\kappa(n\pi/L)^2t}.

Thus a single separated mode is

un(x,t)=sin⁡(nπxL)e−κ(nπ/L)2t.u_n(x,t) = \sin\left(\frac{n\pi x}{L}\right) e^{-\kappa(n\pi/L)^2t}.

For suitable initial data u(x,0)=f(x)u(x,0)=f(x), a solution is

u(x,t)=∑n=1∞bnsin⁡(nπxL)e−κ(nπ/L)2t,u(x,t)= \sum_{n=1}^{\infty} b_n \sin\left(\frac{n\pi x}{L}\right) e^{-\kappa(n\pi/L)^2t},

where the Fourier sine coefficients are

bn=2L∫0Lf(x)sin⁡(nπxL) dx.b_n=\frac{2}{L} \int_0^L f(x)\sin\left(\frac{n\pi x}{L}\right)\,dx.

The precise interpretation depends on the regularity of ff. For square-integrable data, the solution approaches the initial data in the square-integrable norm as tt decreases to zero. Stronger pointwise or classical conclusions require additional assumptions. Higher spatial frequencies decay faster because their decay rates are proportional to n2n^2.

Boundary conditions and eigenfunction expansions

In partial differential equations, separation is closely connected to a boundary-value problem. Boundary conditions determine which separation constants and separated factors are admissible.

In the heat example, the spatial equation is an eigenvalue equation for the operator −d2/dx2-d^2/dx^2. Its allowed constants are eigenvalues, and the corresponding functions are eigenfunctions.

A common general framework is a Sturm–Liouville problem:

−ddx(p(x)dXdx)+q(x)X=λw(x)X.-\frac{d}{dx}\left(p(x)\frac{dX}{dx}\right) +q(x)X = \lambda w(x)X.

For regular problems with suitable real coefficients, positive pp and ww, and self-adjoint boundary conditions, the eigenvalues are real, and eigenfunctions for distinct eigenvalues are orthogonal under the weighted inner product

⟨Xm,Xn⟩w=∫abXm(x)Xn(x)w(x) dx.\langle X_m,X_n\rangle_w = \int_a^b X_m(x)X_n(x)w(x)\,dx.

Appropriate eigenfunctions form a complete system in the associated weighted Hilbert space. Completeness explains why sums of product solutions can represent data that are not themselves separable.

Boundary conditions also affect qualitative behavior. Insulated rather than fixed-temperature endpoints in the heat equation permit a spatially constant, zero-eigenvalue mode, which does not decay. On unbounded domains, continuous spectral parameters and integral representations, such as the Fourier transform, may replace discrete sums.

Coordinates and applications

Separability depends on both the equation and the coordinate system. An equation that separates in Cartesian coordinates may also separate in other coordinates, but yield different ordinary differential equations and different families of functions.

For Laplace's equation in plane polar coordinates,

urr+1rur+1r2uθθ=0,u_{rr}+\frac{1}{r}u_r+\frac{1}{r^2}u_{\theta\theta}=0,

the substitution u(r,θ)=R(r)Θ(θ)u(r,\theta)=R(r)\Theta(\theta) gives

r2R′′+rR′R=−Θ′′Θ=λ.\frac{r^2R''+rR'}{R} = -\frac{\Theta''}{\Theta} =\lambda.

If the solution must be single-valued around a full circle, angular periodicity restricts the permitted constants. Circular and annular boundaries can then be handled naturally. Other equations in polar or cylindrical coordinates lead to Bessel functions; spherical coordinates commonly lead to spherical harmonics.

Major applications include:

  • Diffusion and heat conduction: separated modes describe spatial patterns with different decay rates.
  • Vibrations and acoustics: separation of the wave equation yields normal modes and their frequencies.
  • Potential theory: Laplace's equation describes idealized electrostatic and steady-state temperature fields.
  • Quantum mechanics: separable forms of the Schrödinger equation reduce certain time-dependent and spatial problems to eigenvalue equations.
  • Numerical computation: eigenfunction expansions provide spectral approximations and test solutions for computational methods.

Historical development

Variable separation belongs to the early development of differential equations and calculus. Product solutions became particularly important in eighteenth-century studies of vibrating strings and related boundary-value problems.

Joseph Fourier made separated modes and trigonometric expansions central to heat conduction, notably in his 1822 work Théorie analytique de la chaleur. Subsequent developments in Sturm–Liouville theory, spectral theory, and functional analysis supplied more systematic foundations for eigenfunction expansions and their convergence.

Limitations and distinctions

Not every differential equation is separable. An ordinary differential equation y′=F(x,y)y'=F(x,y) may not admit a factorization F(x,y)=f(x)g(y)F(x,y)=f(x)g(y). A substitution can sometimes make it separable, but the existence of an effective substitution is not guaranteed.

For partial differential equations, variable coefficients, mixed derivatives, nonlinear terms, or irregular boundaries may prevent useful separation. A differential operator can separate while the boundary geometry does not, making the resulting modes unsuitable for the prescribed problem.

Superposition is especially important: sums of solutions solve a linear homogeneous equation, but generally do not solve a nonlinear equation. Finding several separated solutions of a nonlinear equation therefore does not automatically produce a general solution.

Nonhomogeneous equations and boundary conditions also need additional treatment. A common procedure is to subtract a function satisfying the nonhomogeneous boundary data and then expand the remaining unknown in eigenfunctions, producing forced equations for the mode coefficients.

Finally, separation is distinct from statistical independence and from the topological notion of a separable space. Here, “separable” describes the structure of an equation or a solution representation, not independence of random variables or the existence of a countable dense subset.