McGill University Math 319B: Partial Dierential Equations Linear Second Order PDEs The general second order

linear PDE in two independent variables x, y is A 2u 2u 2u u u +B +C 2 +D +E + F u = G, 2 x xy y x y

where A, B, C, D, E, F, G are functions of x, y and at least one of A, B, C, D = 0. The PDE is said to be homogeneous if G = 0. If A, B, C, D, E, F are constants then the PDE is said to be a constant coecient PDE. Linear second order PDEs are classied according to the discriminant = B 2 AC. The PDE is said to be hyperbolic if > 0, elliptic if < 0 and parabolic if = 0. For example, the PDEs 2u 2u 2u 2u 2u u 2 + cu = d, + 2 + cu = d, k + cu = d 2 2 x y x y x2 y are respectively, hyperbolic, elliptic and parabolic. Conversely, the general second order linear constant coecient PDE can be brought to this standard form by a change of variables, dependent and independent. This result is known as the Classication Theorem and we now outline its proof. The change of independent variables will be the linear change of variable = ax + by = cx + dy. Using the chain rule, we get u u u u u = + =a +c x x x u u u u u = + =b +d . y y y Notice that the coecient matrix for u , u in terms of u , u is the transpose of the coecient x y matrix for , in terms of x, y. Using such a change of variable we can always make B = 0 in the new equation. To see how, we let X = x , Y = u . Then y A 2 2 2 +B + C 2 = AX 2 + BXY + CY 2 . x2 xy y

If A = 0 we complete the square in X AX 2 + BXY + CY 2 = A(X + BY /2A)2 + (C B 2 /4A)Y 2 = A(X + BY /2A)2 Y 2 /4A. After possibly multiplying the original equation by 1, we can assume A > 0. We now make a change of variable = ax + by = cx + dy.

so that B = A(X + BY /2A) = A + x 2 A y = /AY = /A y in the case > 0, B = A(X + BY /2A) = A + x 2 A y = /4AY = /A y in the case < 0 and B = A(X + BY /2A) = A + x 2 A y = y in the case = 0. With this change of variable A becomes 2u 2u 2u +B +C 2 x2 xy y 2u 2u + 2, 2 2u 2

2u 2u 2, 2

according as > 0, < 0, = 0. If A = 0 but C = 0 we complete the square in y. If A = C = 0, we use the fact that 4XY = (X + Y )2 (X Y )2 . We are therefore reduced to the case A = 1, B = 0, C = 1 or C = 0. We can get rid of a rst order derivative, say X = , using X 2 + DX = (X + D/2)2 D2 /4 which can be simplied using X + a = ea Xea where a = D/2. This can be also be done for the rst order derivative in in the elliptic or hyperbolic case. Using this, we get in the elliptic case X 2 u + Y 2 u + DXu + EY u = ea X 2 ea u + eb Y 2 eb u a2 b2 , where b = E/2. Multiplying on the left by ea+b and setting v = ea+b u, we get 2v 2v + 2 + dv 2 with d = a2 b2 . The case < 0 is handled similarly. In the case = 0 we can get rid of Xu but not Y u. Then, if Y u appears, the standard form can be achieved with k = 1, c = 0. The proof of this is left to the reader.

Example Consider the PDE 2u 2 u u u 2u + 2 + + =0 2 x xy y x y which can be written as L(u) = 0 with L = X 2 + XY Y 2 + X + Y = (X + Y /2)2 5Y 2 /4 + X + Y. We now set 1 = + x 2 y 5 = 2 y which can be achieved by the change of variables x= 1 5 y= + . 2 2

Our dierential equation becomes 2 u 2 u u 1 u 2 + + = 0. 2 5 If we now make the change of variable v = e/2/2
5

u,
5

the above equation becomes, after multiplying by e/2/2

2v 2v 2 4v/5 = 0. 2