You are on page 1of 28

Chapter III

Classification of
partial differential equations into
elliptic, parabolic and hyperbolic
types

The previous chapters have displayed examples of partial differential equations in various fields
of mathematical physics. Attention has been paid to the interpretation of these equations in
the specific contexts they were presented. 1
In fact, we have delineated three types of field equations, namely hyperbolic, parabolic and
elliptic. The basic idea that the mathematical nature of these equations was fundamental to
their physical significance has been creeping throughout.
Still, the formats in which these three types were presented correspond to their canonical
forms, that is, a form that one recognizes at first glance. Such is not the general case. For
example, it is not obvious (to this author at least!) that the following second order equation,

∂2u ∂2u ∂ 2 u ∂u
2 − 4 − 6 + = 0,
∂x2 ∂x∂t ∂t2 ∂x
is of hyperbolic type. In other words, it shares essential physical properties with the wave
equation,
∂2u ∂2u
− 2 = 0.
∂x2 ∂t
Indeed, this is the aim of the present chapter to show that all equations of mathematical
physics can be recast in these three fundamental types. By the same token, we introduce a new
notion, that of a characteristic curve. A method to solve IBVPs based on characteristics will
be exposed in the next chapter.
The terminology used to coin the three types of PDEs borrows from geometry, as the
criterion will be seen to rely on the nature of the roots of quadratic equations.
We envisage in turn first of order equations, sets of first order equations, and second order
equations. The use of a common terminology to class first and second order equations is
challenged by the fact that a set of two first order equation may be transformed into a second

1
Posted, December 05, 2008; updated, December 12, 2008

57

58 Classification of PDEs
order equation, and conversely. The point will not be developed throughout, but rather treated
via examples.
Since we are concerned in this chapter with the nature of partial differential equtions, we
will not specify the domain in which they assume to hold. On the other hand, the issue surfaces
when we intend to solve IBVPs, as considered in Chapters I, II and IV.

III.1 First order partial differential equations
III.1.1 A single equation
We consider first a single first order partial differential equation for the unknown function
u = u(x, y),
u = u(x, y) unknown,
(III.1.1)
(x, y) variables ,
that can be cast in the format,
∂u ∂u
a +b + c = 0. (III.1.2)
∂x ∂y
This equation is said to be (please think a little bit to this terminology),
- linear if a = a(x, y), b = b(x, y), and c constant;
- quasi-linear if these coefficients depend in addition on the unknown u;
- nonlinear if these coefficients depend further on the derivatives of the unknown u.
Let " #
1 a
s= √ , (III.1.3)
a2 + b2 b
be the unit vector that makes it possible to recast the PDE (III.1.2) into the format,

s · ∇u + d = 0 , (III.1.4)

with d = c/ a2 + b2 .
The curves, starting from an initial curve I 0 , and with a slope,
dy b
= , (III.1.5)
dx a
are called characteristic curves. A point on these curves is reckoned by the curvilinear
abscissa σ,
(dσ)2 = (dx)2 + (dy)2 . (III.1.6)
Typically, σ is set to 0 on the initial curve I 0 .
Then " #
dx/dσ
s= , (III.1.7)
dy/dσ
and the partial differential equation (PDE) (III.1.4) for u(x, y),
∂u dx ∂u dy du
+ +d= + d = 0, (III.1.8)
∂x dσ ∂y dσ dσ
magically becomes an ordinary differential equation (ODE) for u(σ) along a characteristic
dy/dx = b/a. Hum· · · puzzling, how is that possible? There should be a trick here · · · My mum

1.1.8). Analytical and/or Numerical solution The above observations provide the basics to a method for solving a partial differential equation.the characteristics and curvilinear abscissa are obtained by (III. the transformation of a PDE to an ODE is a phenomenon that we have already encountered. this is in fact the basic principle of Laplace or Fourier transforms. the subsequent curves I1 · · · would be I0 itself.the three steps above are repeated. Indeed. .8): .1.5) and (III. j ∈ [1. Each characteristic is endowed with a curvilinear abscissa σ. step by step. characteristic network I2 abscissa s dy s dy dx I1 dy abscissa s ds dx initial data on I0 dx (s=0. “my little boy. the characteristic network and so- lution are built simultaneously. (III. one may obtain and draw the characteristic using (III.1 Given data on a non characteristic initial curve I0 .1. except AIDS perhaps”. .9) (x.Benjamin LORET 59 warned me.1. Indeed. If the PDE is linear. starting from I 1 . nothing comes for free in this world. uj = uj (x.at each point of I0 . .5) and (III. then . The price to pay here is the inverse transforma- tion.1.the solution u is deduced from (III.1.s) Figure III. which are not known beforehand.1.1. The initial PDE is transformed into an ODE where the variable associated to the transform is temporarily seen as a parameter.2 A system of quasi-linear equations The concept of a characteristic curve is now extended to a quasi-linear system of first order partial differential equations for the n unknown functions u 0 s. t). III. and so on. which provides also dσ by (III. Indeed. otherwise. t) variables . .8). whence the solution on the new curve I 1 .6). Taking a step backward. It is now clear why the initial curve I 0 should not be a characteristic.5). n]. which is required not to be a characteristic. . If the PDE is quasi-linear. y) other than on I0 . while points on the initial curve I0 are reckoned by a curvilinear abscissa s. there is a price to pay.6).du results from (III.assume u to be known along a curve I0 . a numerical scheme is developed to solve simultaneously (III. and the price is to find the characteristic curves. so that the solution could not be obtained at points (x. unknowns.1.

To each characteristic is associated a curvilinear abscissa σ.15) dt This characteristic equation should be seen as a polynomial equation of degree n for dx/dt. and vector c = (ci ). or if the eigenvalues are real and the system is not defective. the associated coefficient matrix should be singular.1. may depend on the variables and unknowns.1. we devise a linear combination of these n partial differential equations. dt To prove this property. Let us recall that a system of size n is said non defective if its eigenvectors generate n.1. that is.if the eigenvalues are real and distinct. defined by its differential. dt (III. . In order to form an ordinary differential equation in terms of a (yet) unknown curvilinear abscissa σ. the system is said hyperbolic. dσ ∂t dt ∂x .12) dx λi (aij − bij ) = 0.10) ∂uj ∂uj Lij uj = aij + bij + ci = 0 .1.1.1. ∂t ∂x where the coefficient matrices a = (aij ) and b = (bij ) with (i.10) by λ. j ∈ [1. For the eigenvector λ not to vanish. dx det (a − b) = 0 . with i ∈ [1.if the nb of real eigenvalues is 0. n] . ∂u ∂u L·U = a· + b· +c =0 ∂t ∂x (III. namely. d dt ∂ dx ∂ = + dσ dσ ∂t dσ ∂x (III.13) ∂t ∂x which can be of the form (III. n]. but the system is defective. j) ∈ [1. dσ The vector λ will appear to be a left eigenvector of the matrix a dx/dt − b. we pre-multiply (III. (III.12).1. t) called characteristic. namely dx λ · (a − b) = 0.11) only if λ·a λ·b p = = . but not of their derivatives. n]2 . du λ·L·u = p· +r =0 dσ (III.60 Classification of PDEs that can be cast in the format. the algebraic and geometric multiplicities of each eigenvector are identical. Classification of first order linear PDEs . Characteristic curves and Riemann invariants Each eigenvalue dx/dt defines a curve in the plane (x. . i ∈ [1. the system is said to be parabolic. ∂u ∂u λ·a· +λ·b· + λ · c = 0.1.1.if the eigenvalues are real. (III.14) dt dx dσ Elimination of the vector p in this relation yields the generalized eigenvalue problem (III. the system is said elliptic. The classification of first order partial differential equations is based on the above spectral analysis. n] .11) duj λi Lij uj = pj + r = 0.1.16) dt  ∂ dx ∂  = + . (III.

an immediate question comes to mind: are the eigenvalue problems λ · (a dx/dt − b) = 0 and λ · (a − b dt/dx) = 0 equivalent? The answer is not so straightforward.20) ∂t ∂x it is termed a conservation law. du dt λ·a· + λ · c = 0. bT ). Please remind that the left and right eigenvalues of an arbitrary square matrix are identical.1. III. (III.2.1) ∂x2 c2 ∂t2 the change of coordinates. ∂F(u) ∂G(u) + = 0. Some further terminology If the system of PDEs. 2. i.18) ∂t ∂x can be cast in the format. if the matrix is not symmetric. The left eigenvectors of a matrix a are the right eigenvectors of its transpose a T . ξ = x − c t. and the variables (t.1. delineating the case of constant coefficients from that of variable coefficients.1 A single equation with constant coefficients Let us start with an example.2 Second order partial differential equations The analysis addresses a single equation. That temerity might not be without consequence. b) are also the right eigenvectors of the pencil (a T . The left eigenvectors of the pencil (a. but the left and right eigenvectors do not.2. (III. but subtle and tricky issue 1. Note the subtle interplay between the sets of matrices (a.1. The above writing has made use of the ratio dx/dt.2.2) transforms the canonical form (III.12) into (III.19) ∂t ∂x it is said to be of divergence type.2.1.Benjamin LORET 61 Inserting (III. (III. λ · (a dx/dt − I) = 0. III.1.e. and not of dt/dx: we have broken symmetry without care.17) dσ dσ Quantities that are constant along a characteristic are called Riemann invariants. ∂u ∂G(u) + = 0. η = x + ct. A simple. The generalized (left) eigenvalue problem λ · (a dx/dt − b) = 0 becomes a standard (left) eigenvalue problem when b = I.1. For the homogeneous wave equation.2. ∂2u Lu = = 0. as will be illustrated in Exercise III. (III. x).2. ∂2u 1 ∂2u Lu = − = 0.1) into another canonical form. ∂u ∂u a· +b· + c = 0. (III. Indeed. b). (III. In the special case where the system can be cast in the format.3) ∂ξ∂η . (III.13) yields. 3.

so as to cast this PDE into an ODE? The answer was positive for the wave equation.10) ∂ξ∂η ∂u ∂u +(−α1 D + E) + (−α2 D + E) + F u + G = 0. (III.2.2.9) ∂y 2 ∂ξ 2 ∂ξ∂η ∂η 2 ∂2u ∂2u ∂2u ∂2u = −α1 − (α1 + α2 ) − α2 .2.6) yields the PDE in terms of the new coordinates. (III. Now come some tedious algebras. u = u(x. ξ = −α1 x + y. ∂x∂y ∂ξ 2 ∂ξ∂η ∂η 2 Inserting these relations into (III.2. with constant coefficients. simply.2.2.6) ∂x ∂x∂y ∂y ∂x ∂y The question is the following: can we find characteristic curves. they are the lines ξ constant and η constant. ∂2u ∂2u ∂2u ∂u ∂u Lu = A 2 + 2 B + C 2 +D +E + F u + G = 0. η) = f (ξ) + g(η) . y) variables . they are to be discovered. So we bet on a change of coordinates. Let us try to generalize this result to a second order partial differential equation for the unknown u(x. But where are the characteristics here? Well. To be safe. that is. the solution expresses in terms of two arbitrary functions. (III. (III. we are on a moving ground here.4) which should be prescribed along a non characteristic curve. y).7) where the coefficients α1 and α1 are left free. we should keep some degrees of freedom. ∂y ∂ξ ∂y ∂η ∂y ∂ξ ∂η and ∂2u ∂2u ∂2u ∂2u = α21 + 2 α 1 α2 + α22 ∂x2 ∂ξ 2 ∂ξ∂η ∂η 2 ∂2u ∂2u ∂2u ∂2u = + 2 + (III. u(ξ.2. What do we get in this more general case? Well.62 Classification of PDEs Therefore.2.5) (x. ∂u ∂u ∂ξ ∂u ∂η ∂u ∂u = + = −α1 − α2 ∂x ∂ξ ∂x ∂η ∂x ∂ξ ∂η (III. η = −α2 x + y . y) unknown. ∂ξ ∂η . ∂2u ∂2u Lu = (A α21 − 2 B α1 + C) + (A α 2 2 − 2 B α 2 + C) ∂ξ 2 ∂η 2 ∂2u +2 (α1 α2 A − (α1 + α2 ) B + C) (III.8) ∂u ∂u ∂ξ ∂u ∂η ∂u ∂u = + = + .

g. β 6= α .19) ∂η 2 ∂ξ ∂η where the superscript ’ indicates a modification of the original coefficients.1.11) namely. But before we enter this classification. ∂2u ∂2u ∂u ∂u (H) 2 − 2 + D 00 + E 00 + F 00 u + G00 = 0 .11) is strictly positive.15) ∂σ ∂τ ∂σ ∂τ is obtained by the new set of coordinates. (III.1 Hyperbolic equation B 2 − A C > 0.g. Another equivalent canonical form.2.15) indicates another modification of the original coefficients.13). First order terms and zero order terms do not play a role here. we can make a very important observation: the nature of the equation depends only on the coefficients of the second order terms. ∂2u ∂u ∂u (H) + D0 + E0 + F 0 u + G0 = 0 .2. (III.13) A The equation can then be cast in the canonical form. depending on the nature of these roots.2. ξ = −α x + y. the two roots are real distinct.18) which allows to cast the equation in the canonical form.2.17) A A second arbitrary coordinate is introduced. (III. ∂2u ∂u ∂u (P) + D0 + E0 + F 0 u + G0 = 0 . (III.Benjamin LORET 63 Let us choose the coefficients α to be the roots of A α2 − 2 B α + C = 0.14) ∂ξ∂η ∂ξ ∂η where the superscript ’ indicates that the original coefficients have been divided by the non zero term (III. defined by B α1 = α 2 = . 4p 2 2 (α1 α2 A − (α1 + α2 ) B + C) = − B − A C 6= 0 .2.12) A A Therefore we are led to distinguish three cases.2. the wave equation If the discriminant of the quadratic equation (III. .2. e.2 = ± B − AC .2. III.2.2. heat diffusion A single family of characteristics exists. (III.2. 1 1 σ= (ξ + η). (III. e.16) 2 2 The superscript 00 in (III.2.1. (III. B 1 p 2 α1. III.2.2 Parabolic equation B 2 − A C = 0. (III. and the equation is said hyperbolic.2. η = −β x + y. (III. The coefficient of the mixed second derivative of the equation does not vanish. τ= (ξ − η) .

τ= (ξ − η) = −b x .23) whence ∂u ∂u ∂ξ ∂u ∂η = + ∂x ∂ξ ∂x ∂η ∂x (III. ξ = ξ(x. 1 1 σ= (ξ + η) = −a x + y. ∂y ∂ξ ∂y ∂η ∂y and ∂2u  ∂ξ 2 ∂ 2 u ∂ξ ∂η ∂ 2 u  ∂η 2 ∂ 2 u ∂ 2 ξ ∂u ∂ 2 η ∂u = + 2 + + + ∂x2 ∂x ∂ξ 2 ∂x ∂x ∂ξ∂η ∂x ∂η 2 ∂x2 ∂ξ ∂x2 ∂η ∂2u  ∂ξ 2 ∂ 2 u ∂ξ ∂η ∂ 2 u  ∂η 2 ∂ 2 u ∂ 2 ξ ∂u ∂ 2 η ∂u = + 2 + + + ∂y 2 ∂y ∂ξ 2 ∂y ∂y ∂ξ∂η ∂y ∂η 2 ∂y 2 ∂ξ ∂y 2 ∂η ∂2u ∂ξ ∂ξ ∂ 2 u  ∂ξ ∂η ∂ξ ∂η  ∂ 2 u ∂η ∂η ∂ 2 u ∂ 2 ξ ∂u ∂ 2 η ∂u = + + + + + .2. a= . A0 = Q(ξ. (III. ξ). (III. like in the constant coefficient equation. one may introduce the real coordinates. C 0 = Q(η.2.2.2. (III. (III.2. III.2. ∂x∂y ∂x ∂y ∂ξ 2 ∂y ∂x ∂x ∂y ∂ξ∂η ∂x ∂y ∂η 2 ∂x∂y ∂ξ ∂x∂y ∂η (III. η = η(x.26) ∂ξ ∂ξ∂η ∂η ∂ξ ∂η The coefficients of higher order. Moreover. The characteristics are sought in the more general format.21) A A so as to cast the equation in the canonical form.24) ∂u ∂u ∂ξ ∂u ∂η = + .2 A single equation with variable coefficients When the coefficients of the second order equation are variable.25) yielding finally. b= A C − B2 . y). fortunately.64 Classification of PDEs III. the analysis becomes more complex.3 Elliptic equation B 2 − A C < 0. e. B 0 = Q(ξ.2. the laplacian There are no real characteristics.1. ∂2u 2 0 ∂ u 2 0 ∂ u ∂u ∂u Lu = A0 2 + 2 B + C 2 + D0 + E0 + F 0 u + G0 = 0 .2. η) . B Bp α1.2 = a ± i b. Still. (III. (III.2.22) ∂σ 2 ∂τ ∂σ ∂τ where the superscript ” indicates yet another modification of the original coefficients.2.20) 2 2i with real coefficients a and b.27) . but. ∂2u ∂2u ∂u ∂u (E) + 2 + D 00 + E 00 + F 00 u + G00 = 0 . the main features of the constant case remain. the anal- ysis below shows that this nature relies entirely on the sign of B 2 − A C. y) .g. η).

one real characteristic defined by A 0 = B 0 = 0. (III.2.Benjamin LORET 65 are defined via the bilinear form Q. III. arbitrary but not parallel to the curves ξ constant.2. (III. (III. Let us therefore consider the case A 6= 0. ∂ξ ∂η  ∂ξ ∂η ∂ξ ∂η  ∂ξ ∂η Q(ξ.31) ∂f /∂y A A and. The roots f = ξ and η of A0 = 0 and C 0 = 0 are.  ∂ξ ∂η ∂ξ ∂η 2 B 02 − A0 C 0 = (B 2 − A C) − . (III. ∂ξ ∂ξ ∂2ξ ∂2ξ ∂2ξ D0 = D +E + A 2 +2B +C 2 ∂x ∂y ∂x ∂x∂y ∂y ∂η ∂η ∂2η ∂2η ∂2η E0 = D +E + A 2 + 2B +C 2 (III.34) ∂ξ/∂y A A second family of curves η(x.2 Parabolic equation B 2 − A C = 0.1 Hyperbolic equation B 2 − A C > 0.2.32) dx ∂f /∂y f is constant.2. C 0 6= 0 A single family of characteristics exists.35) ∂η/∂y ∂ξ/∂y . ∂η/∂x ∂ξ/∂x 6= = a. the original equation is already in the canonical format.2.2. b= B − A C 6= 0 . defined by A 0 = 0.2. like for the constant coefficient equation. these curves are the characteristics we were looking for. depending on the nature of the roots of a quadratic equation. Crucially. along the curves of slope dy ∂f /∂x =− = −(a ± b) . (III. y) is introduced.2. ∂f /∂x B 1p 2 = a ± b. a=− . (III. η) = A +B + +C . ∂f ∂f df = dx + dy = 0 . III.30) ∂x ∂y ∂y ∂x Therefore.2.28) ∂x ∂x ∂x ∂y ∂y ∂x ∂y ∂y The remaining coefficients are. (III. we are led to distinguish three cases.33) ∂x ∂y Consequently.2.2. B 0 6= 0 If A = C = 0. ∂ξ/∂x B =a=− .29) ∂x ∂y ∂x ∂x∂y ∂y F0 = F G0 = G. two real characteristics defined by A 0 = C 0 = 0.2.

τ ) = C 0 6= 0. σ) = Q(τ.2.2.2.2. y (E) (P) x (H) Figure III. .2 Tricomi equation of transonic flow provides a conspicuous example of second order equa- tion with variable coefficients where the type varies pointwise.2. = (III.2.3 Elliptic equation B 2 − A C < 0. ∂ξ/∂x B Bp = a + i b.66 Classification of PDEs On the other hand.2. A0 = C 0 6= 0 There are no real characteristics.36) ∂ξ  ∂η AC  ∂ξ ∂ξ  ∂η = A −a + −a .2. (III. η = σ − iτ . since B/A = C/B = −a.40) into B 0 . The roots of Q(f.27)-(III. yields. III.27)-(III.28). (III.2. A0 = Q(σ. B 0 defined by eqns (III.2.42) The reals A0 = C 0 do not vanish because the roots ξ and η of Q(f.2. 1 1 σ= (ξ + η). Q(σ. (III.34). two complex characteristics defined by B 0 = 0.2.2. f ) = 0 are complex. (III.2. (III. σ) − Q(τ. τ ) + 2 i Q(σ. a= .2.39) 2 2i Inserting ξ = σ + i τ. τ ) = 0 .41) and therefore.28).35). as a consequence of the inequality (III.  ∂ξ ∂ξ  ∂η  ∂ξ ∂ξ  ∂η B0 = A +B + B +C ∂x ∂y ∂x ∂x ∂y ∂y | {z } | {z } B/A=−a C/B=−a  ∂ξ (III. ∂x ∂y ∂x B ∂x ∂y ∂y | {z } | {z } =0 =0 vanishes. eqns (III. τ ) = 0 . f ) = 0 are complex. due to (III.38) ∂ξ/∂y A A Still.  ∂η C ∂η   ∂η ∂η  C0 = + A +B 6 0. τ= (ξ − η) .37) ∂x B ∂y ∂x ∂y does not vanish. but. one may introduce the real coordinates. B 0 = Q(σ. b= A C − B2 .

we have shown that the existence of real characteristics corresponds to either A0 = 0. for example. Even if. III. Thus. III. in this equation. The prototype that illustrates best this issue is the Tricomi equation of transonic flow. a set of two first order hyperbolic equations is considered in Exercise III.2. man. and it is therefore hyperbolic as well! III. Ex- amples are provided in Exercises III.3.6. equations convey the same physical information. that of Sect. dy ∂f /∂x =− . on the interpretations exposed at length in the previous chapters. we have defined two classifications for second order equations. y) = constant be the analytical expression of such a real characteristic.45) into Q(f. Indeed. we should emit a warning. ∂f ∂f df = dx + dy = 0 . B = 0. this is by no means a general situation.2. However. therefore.2. and conversely. whether first order or second order. y) between the (H) and (E) types is of (P) type.45) dx ∂f /∂y Inserting (III. (III. The associated second order equation turns out to propagate disturbances at the same speed as the first order set. along a characteristic. III. At this point.2. As they say in French.2.43) ∂x2 ∂y 2 corresponding to A = 1. or both.1 The equation of the characteristics In the previous section.28) yields the equation that provides the slope(s) of the real characteristic(s). A (dy)2 − 2 B dy dx + C (dx)2 = 0 .2.Benjamin LORET 67 Remark 1: the Tricomi equation of fluid dynamics The type of a nonlinear equation may change pointwise.1. (III. C = y.1.2. (III. f ) defined by (III. This was perhaps a bit too presumptuous.3 Properties of real characteristics III. whence the types displayed on Fig. .2. the boundary in the plane (x. Whether written in one form or the other.46) Please pay attention to the sign in front of the mixed term. Let f = f (x.2.1. we are safe! This is because the classifications were built on physical grounds. and III. we looked for the stick to be beaten. a second order equation can be written in the format of two first order equations. ∂2u ∂2u + y = 0. or C 0 = 0. no worry. and that associated to the set of two first order equations exposed in Sect. so that the nature of the equation depends on B 2 − A C = −y.44) ∂x ∂y and. Remark 2: are the classifications of first and second order equations compatible? Note that we have used the same terminology to class the types of equations.2. As an illustration. that is. (III. Therefore.

Z: nb.j=1 ∂xi The coefficient matrix in (III. III. and ∂u/∂y.(E) for (Z = 0 and P = n) or (Z = 0 and P = 0) . . For example.consider a function u that satisfies the equation (III. (III.3.(ultraH) for (Z = 0 and 1 < P < n − 1) where . . The classification is as follows: .46) is easily checked. Another way to express the indeterminacy of the Cauchy problem is to state that charac- teristics are the sole curves along which discontinuities can be propagated.2.(H) for (Z = 0 and P = 1) or (Z = 0 and P = n − 1) . . n n X ∂2u X ∂u aij + bi + cu + g = 0. of zero eigenvalues of a .1) should be symmetrized because we have tacitly assumed the partial derivatives to commute.47)   0 dx dy ∂ 2 u/∂y 2 d(∂u/∂y) That the matrix displayed here is singular along the characteristics curves defined by (III. and ∂u/∂y.prescribe u. (III.3.68 Classification of PDEs III. namely ∂ 2 /∂xi ∂xj = ∂ 2 /∂xj ∂xi . and depending on more than 2 variables. let us consider the second order equation depending on n variables.6).j=1 ∂xi ∂xj i.1) i. n]. ∂u/∂x.P : nb.2.2. for any i and j in [1. ∂u/∂x.2. ∂ 2 u/∂x2      A 2B C −D ∂u/∂x − E ∂u/∂y − F u − G       dx dy 0   2   ∂ u/∂x∂y  =  d(∂u/∂x)   .obtain the three second order derivatives of u in terms of u.3.(P) for Z > 0 (⇔ det a = 0) .3 Extension to more than two variables The classification can be extended to equations of order higher than 2. The 3 × 3 linear system to be solved is. of strictly positive eigenvalues of a The alternatives in the (H) and (P) definitions are due to the fact that multiplication by -1 of the equation leaves it unchanged.2 Indeterminacy of the Cauchy problem The characteristics may be given another definition: these are the curves along which the Cauchy problem is indeterminate or impossible The issue is the following: .

3.2) X ∂2u (E) i=1 ∂ξi2 P n X ∂2u X ∂2u (uH) − i=1 ∂ξi2 i=P +1 ∂ξi2 To make the link with the analysis of the previous section. z).4) ⇔ λ1 λ2 > 0 ⇔ A C − B 2 > 0 (E)    λ=0 ⇔ A C − B 2 = 0 (P) Example: consider the second order equation for the unknown u(x.5) ∂x2 ∂y 2 ∂y∂z ∂z 2 Its nature is obtained by inspecting the spectral properties of the symmetric matrix   3 0 0   a=  0 1 2   ⇒ det (a − λ I) = (3 − λ) λ (λ − 5) .3.Benjamin LORET 69 The canonical forms in the characteristic coordinates ξ 0 s generalize the previous expressions for two variables: n ∂2u X ∂2u (H) − ∂ξ12 i=2 ∂ξi2 n−Z X ∂2u (P) ± i=1 ∂ξi2 n (III.3. det (a − λ I) = λ2 − (A + C) λ + A C − B 2 = 0   λ1 λ2 < 0 ⇔ A C − B 2 < 0 (H)   (III. set " # A = a11 B = a12 n = 2. ∂2u ∂2u ∂2u ∂2u 3 + + 4 + 4 = 0. (III. (III.3.3. .3) B = a21 C = a22 whence.6) 0 2 4 which turns out to have a zero eigenvalue so that the equation is parabolic. y. . (III.

1.t) fluid perturbed surface y=-h(x) bedrock Figure III. ∂ζ ∂(uH) + = 0. t) . Mass conservation. 4. involving the gravitational acceleration g. t) be the horizontal particle velocity. Define the Riemann invariants. surface at rest y y=0 y=z(x.t) H(x. Then H(x.20). t) and fluid height H(x. To simplify further the problem. horizontal wavelengths are much larger than the depth. However. t) = h(x) + ζ(x. Give an interpretation to your finding. and h(x) the vertical position of the fixed bedrock. 1. . (2) ∂t ∂x and horizontal balance of momentum. filled with an incompressible fluid with low viscosity. and find its nature.70 Classification of PDEs Exercise III. t) the position of the perturbed free surface.3 In shallow water channels. in the sense of (III. 3. In a shallow channel. 2. (3) ∂t ∂x ∂x are the two coupled nonlinear equations governing the unknown velocity u(x. ζ(x. ∂u ∂u ∂ζ +u +g = 0. Disturbances at the surface of a fluid surface may give rise to waves because gravity tends to restore equilibrium. we should prescribe bound- ary conditions and initial conditions. t). In this circumstance. the horizontal flow is restricted to one direction along the x-axis. and water flows essentially in the horizontal directions. horizontal wavelengths are longer than the depth. If we were interested in solving completely the associated IBVP. Let u(x. show that the system of equations remains coupled. we are only interested in checking the nature of the field equations (FE). Hint: linearize the equations. (1) is the water height.1: 1D-waves in shallow water. the equations that govern the motion of the fluid take a simplified form. here. In the case of an horizontal bedrock h(x) = h constant. Show that the set of the two equations is a conservation law.

λ+ =  p . Inserting the expressions (10) in (7) shows that indeed these waves can propagate for arbitrary u+ and u− and for c c ζ+ (x + c t) = − u+ (x + c t). ∂t ∂x Applying the operator −g ∂/∂x to the first line. let us seek if first order waves of the form.12). (6) dt gH dt − gH so that the system is hyperbolic.17). and ∂/∂t to the second line.1. it is expected to be able to propagate disturbances at finite speed. (9) is the wave-speed at which infinitesimal second order disturbances propagate. (12) dσ . that is. which specializes here to. (5) dt yields two real distinct eigenvalues. u(x. the Riemann invariants are defined via (III.10). ζ− (x − c t) = u− (x − c t) . ∂ζ ∂u +H =0 ∂t ∂x (7) ∂u ∂ζ +g = 0. can propagate to the right and to the left at the very same speed c. and adding the results yields the second order wave equation. We now come back to the nonlinear system. (11) g g 3.Benjamin LORET 71 Solution: The system of equations is first cast into the standard format (III.1. but we are not totally satisfied because we started from first order equa- tions. For each characteristic.     dx+ p g dx− p g = u + gH. 2. (8) ∂t2 ∂x2 which shows that p c= gH .1. that is fine. Indeed. du λ· = 0. 0 1 ∂t u g u ∂x u 0 The resulting eigenvalue problem (III. = u − gH. Well. let us linearize the equations around u = 0. dx λ · (a − b) = 0 . ζ = 0. and associated independent eigenvectors. t) = ζ+ (x + c t) + ζ− (x − c t) . as second order waves. Can we define this speed? To clarify this issue. ∂2u p 2 2∂ u − ( gH) = 0. a u b z" }| #{ z" }| #{ z" }| #{ " # " # 1 0 ∂ H u H ∂ H 0 (4) + = . t) = u+ (x + c t) + u− (x − c t) (10) ζ(x. λ− =  p .

the vertical component of the momentum balance is dominated by the pressure gradient and gravity terms. (16) ∂t simplifies to equation (2) since the particle velocity v is essentially horizontal. Momentum balance expresses in terms of the gradient of pressure ∇p. yielding the hydrostatic pressure p = ρ g (ζ − y).  ∂v  −∇p + ρ g − − v · ∇v = 0 . . and requiring the time rate of change of its mass M = ρSH to be equal to the flux M v traversing the column. u G(u) z" }| #{ z" }| #{ " # ∂ H ∂ uH 0 (14) + = . ∂t u ∂x g H + u2 /2 0 For those who want to know more.1. (17) ∂t For shallow waters. (13) dσ± dt The interpretation is as follows: u + 2c is constant along the characteristic dx/dt = u + c.20). mass conservation. (15) ∂t Since the density ρ is constant. Indeed.72 Classification of PDEs Substituting c for H = c2 /g yields d dx (u ± 2 c) = 0 along the characteristic = u ± c. and u − 2c is constant along the characteristic dx/dt = u − c. ∂ζ + div(Hv) = 0 . vertical gravitational acceleration g. 4. and particle acceleration dv/dt = ∂v/∂t + v · ∇v. once again under the assumption of an essentially horizontal flow. ∂M + div(M v) = 0 . Inserting this expression in the horizontal component of the momentum balance yields equation (3). the system can be recast into the format (III. Mass conservation is obtained by considering a vertical column of height H and constant horizontal area S.

we recognize the Cauchy Riemann equations. + = 0. Define the nature of this equation. λ · (a − b λ) = 0. according to the terminology of Sect. (1) ∂x ∂y ∂y ∂x Obtain the equivalent second order equation. assuming sufficient smoothness. 2. y). ∂  ∂u ∂v  ∂  ∂u ∂v  − + + = ∆u = 0 .2: Switching from first and to second order equations. (3) c ∂t ∂x ∂t where a. Obtain an equivalent first order system of partial differential equations.12) implies det(a λ − b) = λ 2 + 1 = 0 with λ = dy/dx. The system of equations can be cast into the standard format (III. b and c are positive quantities. possibly dependent on position. 1 ∂2u ∂2u ∂u 2 2 − 2 +a + bu = 0. and therefore solution of an elliptic second order equation. (u.1. so that the eigenvalues are complex. if the set ((x. 0 0 ∂t v 1 0 ∂x v −v 0 The resulting eigenvalue problem. a u b z" }| #{ z" }| #{ z" }| #{ " # " # 1 0 ∂ u 0 −1 ∂ u 0 (4) + = . a basic manipulation of the equations.Benjamin LORET 73 Exercise III. ∂u ∂v ∂u ∂v − = 0. v)) satisfies the Cauchy Riemann equations. det (a − b λ) = λ2 . 1. namely ∆v = 0.10). Was the nature of the system unexpected? 2. ∂u ∂ 2 u − 2 = 0. 3. Let us change the angle of attack. and conversely. x). So is v. and the system is therefore elliptic. For example. (5) ∂x ∂x ∂y ∂y ∂y ∂x shows. Now. due to the fact that. with λ = dx/dt. so does the set ((y.1. that u is harmonic. and consider the eigenvalue problem. we may set v = ∂u/∂x.2. with associated characteristic polynomial. (2) ∂t ∂x which is the prototype of a parabolic equation. (7) . III. u)).1. Analyze its nature. Define the nature of the set of first order equations. 0 1 ∂x v 1 0 ∂y v 0 The resulting eigenvalue problem (III. (v. Solution: 1. The first order equivalent system becomes. a u b c z" }| #{ z" }| #{ z" }| #{ " # z" }| #{ " # 1 0 ∂ u 0 −1 ∂ u 0 0 (6) + + = . Consider the heat equation. Consider the telegraph equation. Obtain an equivalent set of two first order equations. implies det(a λ−b) = 1 · · · strange · · · Never mind! We should not be deterred at the first difficulty. Of course.

generated by the vectors λ such that. The associated eigenspace. Therefore. ± c are real and distinct. With u1 = u. λ = 0 is an eigenvalue of algebraic multiplicity 2. λ · (a − b λ) = λ · a = 0 . and c > 0 is the wave speed. a u b c z }| { z }| { z }| { z }| {             1 0 0 u1 0 0 0 u1 −u3 0   ∂     ∂        0 1  0  ∂t  u2  +  0    0 −1   ∂x  u2  +     0  = 0 .74 Classification of PDEs where now λ = dt/dx. the telegraph equation may be equivalently expressed as a first order system of PDEs.1. The telegraph equation is clearly hyperbolic. according to the terminology of Sect. the generalized eigenvalue problem (8) is defective.2. (8) is in fact spanned by the sole eigenvector λ = [0. Therefore.    2 2 0 0 1 u3 0 −c 0 u3 c (a u3 + b u1 ) 0 (9) The generalized eigenvalues dx/dt = 0. u2 = ∂u/∂x and u3 = ∂u/∂t. 3. . 1]. and the set of the two first order equations is parabolic. III. Thus we have another example where the terminologies used to class first and second order equations are consistent.

its pressure p and density ρ are linked by the constitutive relation p/p 0 = (ρ/ρ0 )γ . It is instrumental to define a quantity c. (2) γ −1 c A driving gradients B pressure pA pressure pB density rA density rB x flow u Figure III. Define the nature of this set of first order equations. ∂t ∂x ∂x 1. Solution: 1. 2.4 The flow of ideal gas in a tube is triggered by differences of pressure and density at the ends of the tube. The one-dimensional flow u of an ideal gas in a tube of axis x is triggered by gradients of pressure p > 0 and density ρ > 0. The field equations governing these three unknown functions of space x and time t are the three coupled nonlinear partial differential equations. and the Riemann invariants along each characteristic. Air compressibility can not be neglected in high-speed aerodynamics.10). Show that this system is of divergence type along the definition (III. dp p0 p c2 = = γ γ ργ−1 = γ . ρ0 ) being reference values and γ > 0 is a material constant. 2 ρ dp = c2 dρ. dρ = dc .1. equal to the ratio of heat capacities.3: Air compressibility in high-speed aerodynamics. The system of equations can be cast into the standard format (III. the eigenvalues and eigenvectors associated to the characteristic problem.     (c) 0 0 1 p 0 ρ c2 u p 0 .19). (1) dρ ρ0 ρ and therefore. (p0 .Benjamin LORET 75 Exercise III. If air is assumed to be a perfect gas. ∂ρ ∂ mass conservation : + (ρu) = 0 ∂t ∂x ∂u ∂u 1 ∂p momentum conservation : +u + =0 (3) ∂t ∂x ρ ∂x ∂p ∂p ∂u constitutive equation : +u + ρ c2 = 0. a u b z }| { z }| { z }| {           (a) 1 0 0 ρ u ρ 0 ρ 0   ∂     ∂     (4) (b)  0 1  0   ∂t  u  +  0    u 1/ρ   ∂x  u  =  0  . that can be interpreted as a wave-speed.1.

3 ∂p γ − 1 ∂ (c)0 = + (ρ u2 ) ∂t 2 ∂x ∂p γ − 1 2 ∂ρ ∂u = + u + (γ − 1) ρ u ∂t 2 ∂x ∂x (c) (a) (b) (9) z }| { z }| { z }| { ∂p ∂u γ − 1 2 ∂(ρu) ∂u 1 ∂p = −u − γp − u − (γ − 1) ρ u (u + ) ∂x ∂x 2 ∂x ∂x ρ ∂x ∂(up) ∂ u3 = −γ − (γ − 1) (ρ ) ∂x ∂x 2 2. The Riemann invariants along each characteristic are obtained by (III. λ1 = 0 .2 (b)’=ρ×(b)+u×(a). Indeed.  2  d u± c = 0. λ2. (7) γ−1 2.3 = 1 . (6)       0 ±1/(ρc) so that the system is hyperbolic.3 = u ± c .     1   0   λ1 = u . namely. (5) provides three real distinct eigenvalues and independent (left) eigenvectors. the system can be cast in the format. λ2.17). F(u) G(u) z }| { z }| { (a)0 ρ ρu  0      ∂  ρu  ∂  p + ρu2    (8) (b)0  +  = 0 . along the second and third characteristics.76 Classification of PDEs The characteristic equation.1.   det (a λ − b) = (λ − u) (λ − u)2 − c2 = 0 . 2. 2. . ∂t   ∂x      ρu2    ρu2   (c)0 p + (γ − 1) p + (γ − 1) (p + ) u 0 2 2 The proof is as follows: 2.1 (a)’=(a). and. dρ = 0 along the first characteristic.

∂2u ∂2u ∂2u ∂u 3 2 + 2 + 5 2 x +2 = 0.2. Define the nature of the second order equation. (2) ∂x ∂x∂y ∂y 4. define the characteristics and cast into canonical form the second order equation. Same questions for the second order equation. so that B 2 − A C = −14 < 0. so that the conclusion above is retrieved ! 2.3. and therefore the equation is elliptic. (7)  (dx)2 + (dy)2 = 0  3.Benjamin LORET 77 Exercise III. ∂2u ∂2u ∂ 2 u ∂u 2 − 4 − 6 + = 0. III.  2 ∂2u ∂2u  c 2 − 2 =0  ∂x ∂t x ± a t = constant. Thus. we identify A = 3.2. whose roots are real. (1) ∂x ∂x∂y ∂y ∂y 2.1 With the method developed in Sect.2 With the method exposed in Sect. 1. Along (III. 1. III. 3. (4) 1 5 The characteristic equation becomes det (a − λ I) = λ 2 − 8 λ + 14 = 0. the slope(s) of the characteristic(s) is(are) defined by the equation. (6)  D (dt)2 = 0  for the Laplacian. " # 3 1 a= . the symmetric matrix a is identified as.  ∂2u ∂2u  + 2 =0  ∂x2 ∂y no real characteristics .46) gives the slope of the real characteristics of a second order equation. Equation (III.46). Indicate the nature.4: Second order equations. (3) ∂x2 ∂x∂y ∂y 2 ∂x Solution: 1. (8) . respectively for the wave equation. e2x (dy)2 − 2 ex+y dy dx + e2y (dx)2 = (ex dy − ey dx)2 = 0 . (5)  c (dt) − (dx)2 = 0  2 2 for the heat equation. Define the respective characteristic curves. positive and distinct. ∂2u ∂2u ∂2u e2x 2 + 2 ex+y + e2y 2 = 0 . Consider the wave equation c2 ∂ 2 u/∂t2 − ∂ 2 u/∂x2 = 0.2. the characteristics are.  ∂ 2 u ∂u  D 2− =0  ∂x ∂t t = constant. and the Laplacian ∂ 2 u/∂x2 + ∂ 2 u/∂y 2 = 0. B = 1 C = 5. the heat equation ∂u/∂t − ∂ 2 u/∂x2 = 0.

Using the relations (III. η). y) = e −x dx − e−y dy = constant. the equation becomes ∂2u ∂u ∂u −32 − +3 = 0. Along (III.2. 4.25) between partial derivatives.78 Classification of PDEs and therefore there is a single characteristic ξ(x. Using the relations (III. (10) and therefore there are two characteristic ξ(x.g. 2 (dy)2 +4 dy dx − 6 (dx)2 = 2 (dy + 3 dx) (dy − dx) = 0 . y) = 3 x + y constant. the slope(s) of the characteristic(s) is(are) defined by the equation. (11) ∂ξ∂η ∂ξ ∂η in terms of the coordinates (ξ.25) between partial derivatives. η(x. y) = x 6= ξ(x. y) = −x + y constant. A second arbitrary coordinate may be defined. and the equation is hyperbolic. . η(x.46). y). the equation becomes ∂2u ∂u e2 η 2 + ··· = 0. η). e.2. (9) ∂ξ ∂η in terms of the coordinates (ξ. and the equation is parabolic.2.

b). The unknown vector u = u(x. t). solve the system of partial differential equations. so is d. if one admits that. (1) ∂t ∂x where a. such that. with c = 0. dt dt det (a − b ) = det (d − I ) det b = 0 . b are non singular constant matrices and c is a vector. b= . 1-b Conversely. v(x. The nature of the system is defined by the spectral properties of the pencil (a. for any non singular matrices a and b.Benjamin LORET 79 Exercise III. ∂t ∂x with b̃ = t · b. of size n. a= . (2) Show that a normal system is hyperbolic. 1-a The system is said under normal form if the matrix a can be decomposed into a product of a diagonal matrix d times the matrix b . dt dt dt det (a − b ) = − (2 + ) (3 − ) = 0. i. ∂u ∂u a· +b· + c = 0. a = d · b. " # " # " # " # u 2 −2 1 −3 −v u= . t· a = d · t· b. Given initial data. 1-b. (3) show that any hyperbolic system can be written in normal form. 2-a Consider now the particular matrices and vectors. Solution: 1-a. with characteristic polynomial. ∂u ∂u t·a· + t·b· + t·c = 0 ∂t ∂x (6) ∂u ∂u d · b̃ · + b̃ · + c̃ = 0. Note that since a and b are non singular. 2-a. t = 0) = v0 (x). (5) dx dx so that the eigenspace generates n . and write it in normal form. there exist a non singular matrix t and a non singular diagonal matrix d. (7) dx dx dx . Pre-multiplication of (6) by t yields. n]. (4) v 1 −4 0 1 −u Show that the system is hyperbolic. and the i-th vector is associated to the eigenvalue (dt/dx) i = di . 2-b Consider now the above homogeneous system.e. t = 0) = u0 (x). c= . namely u(x. obeys the first order differential system. c̃ = t · c. The eigenvalues. define the characteristics. i ∈ [1.5: Normal form of a hyperbolic system.

(8) Let d = diag[−2. 0) the points of the x-axis from which the characteristics that meet at point P emanates. namely t 12 = −4 t11 . (u − 2 v) = 0 . The characteristics are the lines. ∂ξ 2-b. ξ = t + 2 x const. (9) 2 t21 + t22 −2 t21 − 4 t22 3 t21 −9 t21 + 3 t22 The matrix t can be defined to within two arbitrary degrees of freedom. One may take. The solution at an arbitrary point (x. t) an arbitrary point. " # 1 −4 t= . " # " # " # " # " # −2 14 ∂ u 1 −7 ∂ u 4u − v · + · + = 0. 5 5 . Then the matrix t has to satisfy the equations. (10) 1 1 and the system (6) then writes.  ∂ ∂  −2 + (u − 7 v) +4 u − v = 0 ∂t ∂x (12)  ∂ ∂  3 + (u − 2 v) −u − v = 0. E0 (x − t/3. η). 7 2 u(x. and X 0 (x + t/2. The homogeneous system. t) = (u0 − 2 v0 )(X0 ) − (u0 − 7 v0 )(E0 ) . t) = (u0 − 2 v0 )(X0 ) − (u0 − 7 v0 )(E0 ) 5 5 (16) 1 1 v(x. (15) u − 2 v = (u − 2 v)(X0 ) constant along the characteristic ξ = const . " # " # 2 t11 + t12 −2 t11 − 4 t12 −2 t11 6 t11 − 2 t12 a·a=d·d·b ⇔ = . u − 7 v = (u − 7 v)(E0 ) constant along the characteristic η = const . Let P (x. η = t − 3 x const . 0). (14) ∂η ∂ξ displays the Riemann invariants in explicit form. t) reads. 3] be the diagonal matrix of the eigenvalues. ∂ ∂ (u − 7 v) = 0. t21 = t22 . (11) 3 −6 ∂t v 1 −2 ∂x v −u − v or equivalently. ∂t ∂x or in terms of the coordinates (ξ.80 Classification of PDEs are real and distinct. ∂ −5 (u − 7 v) +4 u − v = 0 ∂η (13) ∂ 5 (u − 2 v) −u − v = 0.

the solution is built from the characteristic network.5 Given initial data on the x-axis. t): indeed.Benjamin LORET 81 t x characteristic h characteristic t P 1 -1 3 2 E0 x X0 x x-t/3 domain x+t/2 of dependence at (x. . change of the initial conditions outside the interval E0 X0 will not modify this solution. The region E0 PX0 is referred to as the domain of depen- dence of the solution at point P(x.t) Figure III. playing with the Riemann invariants.

6. This equation is referred to as telegraph equation. It involves an inductance L = L(x). t) and potential V (x. Let q = C dx V be the charge across the capacitor. The variation of the current dI is due to the capacitance C dx dV /dt and to the conductance G dx V . t ) conduc tan ce capaci tan ce ¶V V( x. the equations governing the current I(x. t )dx ¶x + V( x. Therefore. a capacitance to ground C = C(x). The line properties are henceforth assumed to be uniform in space. Show that tR/L and V (t) etR/L satisfy √ I(t) e a canonical form of the wave equation. t) in a transmission line of axis x can be cast in the format of a linear system of two partial differential equations. The variation of the potential dV over a segment of length dx is due to the resistance R dx I and to the inductance L dx dI/dt. Observe that this second order equation involves a single variable. 0 C ∂t V 1 0 ∂x V GE 0 Show that the system is hyperbolic. resis tan ce induc tan ce Rdx Ldx ¶I I( x . a u b z " }| #{ z" }| #{ z" }| #{ " # " # " # L 0 ∂ I 0 1 ∂ I RI 0 (1) + + = . and define its characteristics. obtain the equivalent second order equation that the unknowns I and V satisfy. So far we have manipulated the equations assuming all line coefficients to be different from . An electrical circuit representing a transmission line is shown on Fig. t )dx Gdx Cdx ¶x - x x + dx Figure III. III. find a modified function in the same mood as above that satisfies the non homogeneous wave equation. t ) + ( x . 2-c. 2-b.6: Transmission lines and the telegraph equation.6 Elementary circuit of length dx used as a model of the transmission lines. t ) I( x . Find the nature of this system. with wave speed 1/ LC. In a first step. at the price of a higher order operator. a resistance R = R(x). Consider now a distortionless line RC = LG.82 Classification of PDEs Exercise III. t ) + ( x . and comment. a conductance to ground G = G(x). 2-a. we have obtained a decoupled system. To substantiate the nature of the system. They may vary along the line. 1-b Write the system in normal form. When RC 6= LG. 1-a Therefore. all these material properties are assumed to be strictly positive.

t 21 = −t 22 C/L. that is. The function Y (t) satisfies the equation. and therefore  √ √   √ √  1  C L 1 L C t= √ √ . (3) dx L dx − L so that the system is hyperbolic.  ∂2 1 ∂2  − √ Y = 0. V . Consider now Heaviside’s ideal line with L = G = 0. (6) ∂t v LC 0 −1 ∂x v RC − LG RC + LG v 0 2-a. X = I. We take t11 = 1/(2L C). √ √ 1-b Let d = diag[ LC. and associated independent eigenvectors λ. (8) ∂t2 ∂x2 ∂t The coefficients of the zero and first order terms vanish simultaneously only if RC = LG and then α = R/L. Let Y (t) = X(t) eα t with α an unknown exponent. = √ √  . − LC] be the diagonal matrix of the eigenvalues. the system (1) becomes " # " # " # " #" # " # ∂ u 1 1 0 ∂ u RC + LG RC − LG u 0 +√ + = . We look for a matrix t such that t · a = d · t · b. p The matrix t is p of course not unique. λ+ =  √  . (2) dx yields two real and distinct eigenvalues dt/dx. (5) v V 2 LC L − C V V L − L v Upon pre-multiplication by t. (7) ∂t2 ∂x2 ∂t 2-b. (9) ∂t2 ( LC)2 ∂x2 . t · a = √  √ √ .  √   √  dt+ √ C dt − √ C = LC. λ− =  √  . What is its nature? Solution: 1-a. adding the results and using again the first line to eliminate the undesirable unknown.  ∂2 ∂2 ∂  LC − + (RC + LG) + GR X = 0. = − LC.Benjamin LORET 83 zero. it is expected to be able to propagate disturbances at finite speed. as explained in Exercise III. and to the second line the operator ∂/∂x. t 12 = t 11 L/C. The resulting eigenvalue problem (III. (4) 2 LC C − L 2 LC L − C Let us introduce the new unknowns.5. Applying the operator −C∂/∂t to the first line of (1). we get the telegraph equation. t22 = −1/(2C L). and Y (t) satisfies the wave equation. and√ in fact there is a double √ indeterminacy. dt λ · (a − b ) = 0.12). " # " #  √ √ " # " #  √ √ " # u I 1 L C I I C C u = t·a = √  √ √  .1.  ∂2 ∂2 ∂ 2  LC − + (RC + LG − 2 α LC) + (α LC − (RC + LG) α + GR) Y = 0.

(10) ∂t2 ∂x2 4 LC The solution is then a wave followed by a residual wave due to the source term. When L = G = 0. For a voltage shock V0 applied at the end of the line. V . Therefore in these circumstances. if α = (RC + LG)/(2 LC). and an arrival time when the voltage is say 50% of V0 . (11) ∂x2 ∂t with a diffusion coefficient equal to 1/(RC). one might define qualitatively a beginning of arrival time at a point x when the voltage is equal to say 10% of V 0 . the solution has the form of the complementary error function. X = I. the first order term vanishes.  ∂2 ∂2 (RC − LG)2  LC − − Y = 0. and the characteristic time is in proportion to RC x 2 . More generally. but it looses its hyperbolic character and becomes a diffusion equation.84 Classification of PDEs √ where c = 1/ LC appears clearly as the wave speed. the mode of propagation of the electrical signal is quite different from the general analysis above. A typical value is 3 × 10 8 m/s.  ∂2 ∂ − + RC X = 0. the telegraph equation (7) is still valid. As indicated in Chapter I. This second order analysis is of course consistent with the hyperbolic nature of the initial first order system. 2-c. and we have an inhomogeneous wave equation. .