Differential Equations › Section I · Partial differential equations › Topic I.3
PDE terminology
Before solving any PDE you need to know what kind it is. A few words, linear, homogeneous and the type of a second order equation, decide whether the method of the next topic can work at all and how solutions behave.
Why the labels matter
The heat and wave equations from the last two topics will be solved by building complicated solutions out of simple ones and adding them up. That only works for equations of a particular kind, and the same is true of the boundary conditions. So this topic sets out the vocabulary: operators, linear and nonlinear equations, homogeneous equations and conditions, the principle of superposition, and the three types of second order equation. Each idea is short, but every later solution depends on checking them first.
Operators
An operator takes a function as its input and returns another function, much as an ordinary function takes a number and returns a number. Differentiation is one: \(\frac{d}{dx}\) turns \(x^3\) into \(3x^2\). So is \(\frac{\partial}{\partial t}\), and so is integration over a fixed interval. A more interesting example comes from moving every term of the source-free heat equation to one side. The heat operator is
so that \(L(u) = u_t - k\,u_{xx}\), and the heat equation is simply \(L(u) = 0\).
Linear operators
An operator \(L\) is linear if, for any functions \(u_1, u_2\) and constants \(c_1, c_2\),
The heat operator passes this test, because the derivative of a sum is the sum of the derivatives and constants come outside derivatives. Every partial derivative, of any order, is linear, and so is multiplication by a known function such as \(x^2\). What breaks linearity is any operation on \(u\) itself that is not of this kind: squaring it, multiplying it by one of its own derivatives, or applying a function such as \(\sin u\).
Linear equations and homogeneous equations
A linear PDE can be written \(L(u) = f\), with \(L\) linear and the right-hand side \(f\) depending only on the independent variables, never on \(u\). The heat equation with a source, \(u_t - k\,u_{xx} = Q/(c\rho)\), is linear: the known term \(Q\) does not spoil linearity. The equation is homogeneous when \(f = 0\), so \(L(u) = 0\). There is a quick test: substitute \(u = 0\). For a linear equation, if every term vanishes, the equation is homogeneous, and \(u = 0\), the trivial solution, is one of its solutions. The order is the highest derivative that appears.
Worked example: classifying four equations
Classify each by order, linearity and homogeneity. 1. \(u_t = 4u_{xx} - 3u\): second order; every term is a constant times \(u\) or a derivative, so linear; \(u = 0\) works, so homogeneous. 2. \(u_{tt} = u_{xx} + xt\): second order and linear, but \(u = 0\) leaves \(0 = xt\), so nonhomogeneous. 3. \(u_t + u\,u_x = 0\): first order, nonlinear, because of the product \(u\,u_x\). 4. \(u_t = u_{xx} + u^2\): second order, nonlinear, because of \(u^2\). In this course the word homogeneous is kept for linear equations, so the last two are simply called nonlinear.
Linear and homogeneous boundary conditions
The same words apply to boundary conditions. A fixed temperature \(u(0, t) = 0\) is linear and homogeneous; \(u(0, t) = 25\) is linear but nonhomogeneous, since \(u = 0\) fails it. An insulated end, \(u_x(L, t) = 0\), is homogeneous. The Newton cooling condition from the heat equation topic is linear, and homogeneous only when the surrounding temperature \(g_1\) is zero. Periodic conditions are homogeneous, as rewriting them shows: \(u(-L, t) - u(L, t) = 0\) and \(u_x(-L, t) - u_x(L, t) = 0\). A condition involving \(u^4\), as in radiation, would be nonlinear.
The principle of superposition
If \(u_1\) and \(u_2\) solve a linear homogeneous equation, then so does \(c_1u_1 + c_2u_2\) for any constants. The proof is one line:
The same holds for linear homogeneous boundary conditions, so a combination of solutions that each satisfy them also satisfies them. Superposition fails for nonhomogeneous problems: if \(L(u_1) = f\) and \(L(u_2) = f\), then \(L(u_1 + u_2) = 2f\). It also fails for nonlinear equations, and it is the reason the solution methods ahead insist on linear homogeneous problems.
Worked example: superposition in action
Check that \(u_1 = e^{-kt}\sin x\) solves \(u_t = k\,u_{xx}\) with \(u(0, t) = u(\pi, t) = 0\), then combine. 1. \(u_{1,t} = -k\,e^{-kt}\sin x\), and \(u_{1,xx} = -e^{-kt}\sin x\), so \(u_{1,t} = k\,u_{1,xx}\). 2. \(\sin 0 = \sin\pi = 0\), so both conditions hold. 3. The same steps show \(u_2 = e^{-4kt}\sin 2x\) works. 4. By superposition, \(3u_1 - 2u_2\) solves the problem too. 5. Contrast a nonlinear equation: \(u = -1/t\) solves \(u_t = u_{xx} + u^2\), but \(v = -2/t\) gives \(v_t - v_{xx} - v^2 = -2/t^2\), not zero.
Three types of second order equation
A linear second order equation in \(x\) and \(y\) with constant coefficients, \(Au_{xx} + Bu_{xy} + Cu_{yy}\) plus lower order terms, is classified by its discriminant \(B^2 - 4AC\). If it is positive, the equation is hyperbolic; if zero, parabolic; if negative, elliptic. The heat equation, written in \(x\) and \(t\), has \(A = k\) and \(B = C = 0\), so it is parabolic. The wave equation, \(u_{tt} - c^2u_{xx} = 0\), gives \(0 - 4(-c^2)(1) = 4c^2 > 0\): hyperbolic. Laplace's equation, \(u_{xx} + u_{yy} = 0\), gives \(-4\): elliptic.
What the type tells you
The type predicts behaviour. Parabolic equations diffuse: a change anywhere is felt everywhere at once, however faintly, and profiles smooth out. Hyperbolic equations carry signals at a finite speed and keep sharp features, as the travelling pulses of the previous topic showed. Elliptic equations describe steady states, where the value at every point is a balance with the whole boundary, so they come with boundary conditions only, not initial conditions. For example, \(u_{xx} + 4u_{xy} + 3u_{yy} = 0\) has \(16 - 12 = 4 > 0\), so it is hyperbolic and behaves like a wave equation.
Common mistakes to avoid
First, calling the heat equation with a source nonlinear: a known function \(Q(x, t)\) makes it nonhomogeneous, not nonlinear. Second, missing a hidden nonlinearity such as \(u\,u_x\) or \(\sin u\); a nonconstant coefficient like \(x^2 u_{xx}\) is still linear. Third, applying superposition to a problem with a nonhomogeneous boundary condition, such as an end held at 25 °C. Fourth, forgetting that periodic conditions are homogeneous. Fifth, getting the discriminant wrong by writing \(B^2 - AC\) or by forgetting that the wave equation's coefficients have opposite signs.
Key takeaways
An operator acts on functions; it is linear if it respects sums and constant multiples. A linear PDE is \(L(u) = f\) with \(f\) known; it is homogeneous when \(f = 0\), which the test \(u = 0\) detects. Boundary conditions are classified the same way: zero temperature, insulated and periodic conditions are homogeneous. Superposition, any combination of solutions is a solution, holds for linear homogeneous problems only. With discriminant \(B^2 - 4AC\), wave equations are hyperbolic, heat equations parabolic and Laplace's equation elliptic.
Flashcards
Flashcard
A rule that takes a function and returns a function
Test your understanding
Question 1 of 5
Which equation is linear but nonhomogeneous?
Question 2 of 5
Which boundary condition is homogeneous?
Question 3 of 5
u₁ and u₂ both solve uₜ − uₓₓ = 4. What does u₁ + u₂ satisfy?
Question 4 of 5
Classify uₓₓ − 6uₓᵧ + 9uᵧᵧ = 0.
Question 5 of 5
Which statement about the wave equation uₜₜ = c² uₓₓ is correct?
