Series solutions of second order linear differential equations We start with Definition 1. A function f of a complex variable z is called analytic at z = z0 if there exists a convergent Taylor series for f at z = z0 with positive radius of convergence: f (z) = ∞ X an (z − z0 )n with an = n=0 f (n) (z0 ) , n! n = 0, 1, 2, . . . for |z − z0 | < R with R > 0. Now we consider second order linear differential equations of the form y 00 + p(z)y 0 + q(z)y = 0 d2 y dy + p(z) + q(z)y = 0. 2 dz dz or (1) Then we have the following definition: Definition 2. A point z0 ∈ C is called a regular point of the differential equation (1) if both p(z) and q(z) are analytic in z0 . Otherwise z0 is called a singular point of (1). For singular points we distinguish between two cases: Definition 3. If z0 ∈ C is a singular point of (1) and both (z − z0 )p(z) and (z − z0 )2 q(z) are analytic in z0 , then z0 is called a regular singular point of (1). Otherwise z0 is called an irregular singular point of (1). For regular points we have: Theorem 1. If z0 ∈ C is a regular point of the differential equation (1), then p(z) and q(z) are both analytic at z = z0 . This implies that there exist positive constants R1 and R2 such that p(z) = ∞ X n pn (z − z0 ) , |z − z0 | < R1 and q(z) = n=0 ∞ X qn (z − z0 )n , |z − z0 | < R2 . n=0 Then there exist solutions y of (1) which are analytic at z = z0 . In fact, if R = min(R1 , R2 ), then the solution y can be written as y(z) = ∞ X cn (z − z0 )n for |z − z0 | < R. n=0 Moreover, c0 = y(z0 ) and c1 = y 0 (z0 ) can be chosen arbitrary. Then y(z) = c0 y1 (z) + c1 y2 (z) with {y1 (z), y2 (z)} For regular singular points we have Frobenius’ method: 1 linearly independent. Theorem 2. If z0 ∈ C is a regular singular point of the differential equation (1), then (z − z0 )p(z) and (z − z0 )2 q(z) are both analytic at z = z0 . This implies that there exist positive constants R1 and R2 such that (z − z0 )p(z) = ∞ X pn (z − z0 )n , |z − z0 | < R1 (2) qn (z − z0 )n , |z − z0 | < R2 . (3) n=0 and (z − z0 )2 q(z) = ∞ X n=0 Then there exist solutions y of (1) which can be written in the form y(z) = (z − z0 )r ∞ X cn (z − z0 )n , c0 6= 0. (4) n=0 In fact, if R = min(R1 , R2 ), then we have r y(z) = (z − z0 ) ∞ X cn (z − z0 )n for 0 < |z − z0 | < R. n=0 Here r ∈ C denotes a solution of the indicial equation r(r − 1) + p0 r + q0 = 0. (5) Moreover, if r1 , r2 ∈ C denote the solutions of the indicial equation (5) with Re r1 ≥ Re r2 , then we have: 1. If r1 − r2 ∈ / {0, 1, 2, . . .}, then we have r1 y1 (z) = (z − z0 ) ∞ X an (z − z0 )n , a0 6= 0 bn (z − z0 )n , b0 6= 0 n=0 and y2 (z) = (z − z0 )r2 ∞ X n=0 are two linearly independent solutions of the differential equation (1). 2. If r1 − r2 = 0, then we have y1 (z) = (z − z0 )r1 ∞ X an (z − z0 )n , a0 6= 0 n=0 and y2 (z) = y1 (z) ln(z − z0 ) + (z − z0 )r2 ∞ X bn (z − z0 )n n=0 are two linearly independent solutions of the differential equation (1). 2 3. If r1 − r2 ∈ {1, 2, 3, . . .}, then we have y1 (z) = (z − z0 ) r1 ∞ X an (z − z0 )n , a0 6= 0 n=0 and y2 (z) = Ay1 (z) ln(z − z0 ) + (z − z0 )r2 ∞ X bn (z − z0 )n , b0 6= 0 n=0 are two linearly independent solutions of the differential equation (1), where A is a constant that might be zero or not. Some remarks: • A series of the form (4) is called a generalized power series. Only for r ∈ {0, 1, 2, . . .} this generalized power series reduces to a normal power series. • The solutions r1 and r2 of the indicial equation (5) are called indices. • In the second case it is always possible to choose b0 = 0 since r1 = r2 . In that case the second solution always has a logarithmic singularity at z = z0 . • In the third case the second solution might have a logarithmic singularity at z = z0 (in the case that A 6= 0) or not (in the case that A = 0). The proof of this theorem due to Frobenius is constructive. We multiply the differential equation (1) by (z − z0 )2 to obtain (z − z0 )2 y 00 + (z − z0 )2 p(z)y 0 + (z − z0 )2 q(z)y = 0. Since (z − z0 )p(z) and (z − z0 )2 q(z) are analytic at z = z0 we use (2) and (3) to find ! ! ∞ ∞ X X 2 00 0 n n (z − z0 ) y + (z − z0 ) pn (z − z0 ) y + qn (z − z0 ) y = 0. n=0 n=0 Now we set y(z) = (z − z0 )r ∞ X cn (z − z0 )n = n=0 cn (z − z0 )n+r , c0 6= 0. n=0 Then we have 0 ∞ X y (z) = ∞ X (n + r)cn (z − z0 )n+r−1 n=0 and y 00 (z) = ∞ X (n + r)(n + r − 1)cn (z − z0 )n+r−2 . n=0 Hence we have ∞ X (n + r)(n + r − 1)cn (z − z0 )n+r + n=0 ∞ X ! pn (z − z0 )n n=0 + ∞ X n=0 ! qn (z − z0 )n ∞ X n=0 3 ! (n + r)cn (z − z0 )n+r n=0 ! cn (z − z0 )n+r ∞ X = 0. This can only be true if the coefficients of all powers of z −z0 are equal to zero. The coefficient of the lowest power of z − z0 , id est (z − z0 )r , is {r(r − 1) + p0 r + q0 } c0 . Since c0 6= 0 this leads to the indicial equation (5). More details are left out here. Frobenius’ method is based on the theory of Euler’s differential equation: x2 y 00 + αxy 0 + βy = 0, α, β ∈ R. (6) By using the change of variables x = et or t = ln x we find that dy dy dt 1 dy = · = · dx dt dx x dt and d d2 y = 2 dx dx 1 dy · x dt = 1 d2 y dt 1 dy 1 d2 y 1 dy · 2 · − 2· = 2· 2 − 2· . x dt dx x dt x dt x dt This implies that dy d2 y d2 y dy dy = and x2 · 2 = 2 − . dx dt dx dt dt Hence, with y(x) = Y (t) Euler’s differential equation (6) turns into x· Y 00 − Y 0 + αY 0 + βY = 0 ⇐⇒ Y 00 + (α − 1)Y 0 + βY = 0, which is a second order linear differential equation with constant coefficients. The characteristic equation for this differential equation is r2 + (α − 1)r + β = 0 ⇐⇒ r(r − 1) + αr + β = 0. (7) If r1 and r2 denote the roots of this characteristic equation, then we have 1. If r1 , r2 ∈ R and r1 6= r2 , then Y1 (t) = er1 t and Y2 (t) = er2 t are two linearly independent solutions. This leads to y1 (x) = xr1 and y2 (x) = xr2 being two linearly independent solutions of Euler’s differential equation (6). 2. If r1 , r2 ∈ R and r1 = r2 , then Y1 (t) = er1 t and Y2 (t) = ter1 t are two linearly independent solutions. This leads to y1 (x) = xr1 and y2 (x) = xr2 ln x being two linearly independent solutions of Euler’s differential equation (6). 3. If r1 , r2 ∈ / R, say r1,2 = λ ± µi with λ, µ ∈ R and µ 6= 0, then Y1 (t) = eλt cos µt and Y2 (t) = eλt sin µt are two linearly independent solutions. This leads to y1 (x) = xλ cos(µ ln x) and y2 (x) = xλ sin(µ ln x) being two linearly independent solutions of Euler’s differential equation (6). This can also be obtained by looking for solutions of the form y(x) = xr of Euler’s differential equation (6) directly. Then we have y 0 (x) = rxr−1 and y 00 (x) = r(r − 1)xr−2 which leads to r(r − 1)xr + αrxr + βxr = 0 ⇐⇒ This leads to the characteristic equation (7). 4 {r(r − 1) + αr + β} xr = 0. Examples 1. The hypergeometric differential equation is given by z(1 − z)y 00 + {c − (a + b + 1)z} y 0 − aby = 0, a, b, c ∈ C. (8) It is clear that z = 0 and z = 1 are the only singular points of this differential equation. Since c − (a + b + 1)z abz zp(z) = and z 2 q(z) = − 1−z 1−z are both analytic at z = 0, we conclude that z = 0 is a regular singular point of (8). Note that p0 = c and q0 = 0 which leads to the indicial equation r(r − 1) + cr = 0 ⇐⇒ r(r − 1 + c) = 0. We also have that (z − 1)p(z) = − c − (a + b + 1)z z (z − 1)2 q(z) = and ab(z − 1) z are both analytic at z = 1, which implies that z = 1 is a regular singular point of (8) too. In that case we have p0 = a + b − c + 1 and q0 = 0 which leads to the indicial equation r(r − 1) + (a + b − c + 1)r = 0 ⇐⇒ r(r + a + b − c) = 0. This implies that there exists solutions of the form y1 (z) = ∞ X an z n and y2 (z) = z 1−c n=0 ∞ X bn z n n=0 for the hypergeometric differential equation (8). These two solutions are linearly independent at least for c ∈ / Z. We also have solutions of the form y3 (z) = ∞ X n cn (z − 1) c−a−b and y4 (z) = (z − 1) n=0 ∞ X dn (z − 1)n n=0 for the hypergeometric differential equation (8). These two solutions are linearly independent at least for c − a − b ∈ / Z. Note that we have y(z) = ∞ X an z n =⇒ 0 y (z) = n=0 ∞ X nan z n−1 00 and y (z) = n=0 ∞ X n(n − 1)an z n−2 . n=0 Substitution into the differential equation (8) leads to ∞ X n(n − 1)an z n−1 − n=0 ∞ X n(n − 1)an z n n=0 +c ∞ X nan z n−1 − (a + b + 1) n=0 ∞ X n=0 5 nan z n − ab ∞ X n=0 an z n = 0. Hence ∞ X [{n(n + 1) + c(n + 1)} an+1 − {n(n − 1) + (a + b + 1)n + ab} an ] z n = 0. n=0 This leads to the recurrence relation (n + 1)(n + c)an+1 = (n + a)(n + b)an , n = 0, 1, 2, . . . for the coefficients {an }∞ n=0 . The solution can be written as an = Hence y(z) = ∞ X (a)n (b)n a0 , (c)n n! an z n = a0 n=0 ∞ X (a)n (b)n z n (c)n n=0 Similarly, for y(z) = z n = 0, 1, 2, . . . . 1−c ∞ X n bn z = n=0 n! = a0 2 F1 ∞ X a, b ;z . c bn z n+1−c n=0 we have y 0 (z) = ∞ X (n + 1 − c)bn z n−c and y 00 (z) = n=0 ∞ X (n + 1 − c)(n − c)bn z n−1−c . n=0 Substitution into the differential equation (8) leads to ∞ X (n + 1 − c)(n − c)bn z n−c − n=0 ∞ X (n + 1 − c)(n − c)bn z n+1−c n=0 +c ∞ X (n + 1 − c)bn z n−c − (a + b + 1) n=0 − ab ∞ X (n + 1 − c)bn z n+1−c n=0 ∞ X bn z n+1−c = 0. n=0 Hence ∞ X [{(n + 2 − c)(n + 1 − c) + c(n + 2 − c)} bn+1 n=0 − {(n + 1 − c)(n − c) + (a + b + 1)(n + 1 − c) + ab} bn ] z n+1−c = 0. This leads to the recurrence relation (n + 1)(n + 2 − c)bn+1 = (n + a + 1 − c)(n + b + 1 − c)bn , n = 0, 1, 2, . . . for the coefficients {bn }∞ n=0 . The solution can be written as bn = (a + 1 − c)n (b + 1 − c)n b0 , (2 − c)n n! 6 n = 0, 1, 2, . . . . Hence y(z) = z 1−c ∞ X bn z n = b0 z 1−c n=0 ∞ X (a + 1 − c)n (b + 1 − c)n z n (2 − c)n a + 1 − c, b + 1 − c ;z . 2−c n! n=0 = b0 z 1−c 2 F1 2. The confluent hypergeometric differential equation is given by zy 00 + (c − z)y 0 − ay = 0, a, c ∈ C. (9) Note that z = 0 is the only singular point of this differential equation. Further we have in this case zp(z) = c − z and z 2 q(z) = −az, which are both analytic at z = 0. Note that p0 = c and q0 = 0, which implies that z = 0 is a regular singular point of the differential equation (9) with indicial equation r(r − 1) + cr = 0 ⇐⇒ r(r − 1 + c) = 0. Hence, the indices are r1 = 0 and r2 = 1 − c. This implies that for c ∈ / Z we have two linearly independent solutions of the form y1 (z) = ∞ X an z n and y2 (z) = z 1−c n=0 ∞ X bn z n . n=0 If we substitute y(z) = ∞ X an z n , y 0 (z) = n=0 ∞ X nan z n−1 and y 00 (z) = n=0 ∞ X n(n − 1)an z n−2 n=0 into the differential equation (9) we obtain ∞ X n(n − 1)an z n−1 + c n=0 ∞ X nan z n−1 − n=0 ∞ X nan z n − a n=0 ∞ X an z n = 0 n=0 or equivalently ∞ X [{(n + 1)n + c(n + 1)} an+1 − (n + a)an ] z n = 0. n=0 This leads to the recurrence relation (n + 1)(n + c)an+1 = (n + a)an , with solution an = (a)n a0 , (c)n n! 7 n = 0, 1, 2, . . . n = 0, 1, 2, . . . . Hence we have a solution of the form y(z) = ∞ X an z n = a0 ∞ X (a)n z n n=0 n=0 If we substitute ∞ X y(z) = z 1−c (c)n n! bn z n = n=0 y 0 (z) = ∞ X (n + 1 − c)bn z n−c = a0 1 F1 ∞ X a c ;z . bn z n+1−c , n=0 and y 00 (z) = n=0 ∞ X (n + 1 − c)(n − c)bn z n−1−c n=0 into the differential equation (9) we obtain ∞ X (n + 1 − c)(n − c)bn z n−c +c n=0 ∞ X (n + 1 − c)bn z n−c n=0 − ∞ X (n + 1 − c)bn z n+1−c −a n=0 ∞ X bn z n+1−c = 0 n=0 or equivalently ∞ X [{(n + 2 − c)(n + 1 − c) + c(n + 2 − c)} bn+1 − (n + a + 1 − c)bn ] z n+1−c = 0. n=0 This leads to the recurrence relation (n + 2 − c)(n + 1)bn+1 = (n + a + 1 − c)bn , with solution bn = (a + 1 − c)n b0 , (2 − c)n n! n = 0, 1, 2, . . . n = 0, 1, 2, . . . . Hence we have a solution of the form y(z) = z 1−c ∞ X n=0 n bn z = b0 z 1−c ∞ X (a + 1 − c)n z n n=0 (2 − c)n n! = b0 z 1−c 1 F1 a+1−c ;z . 2−c 3. The Bessel differential equation is given by z 2 y 00 + zy 0 + (z 2 − ν 2 )y = 0, ν ≥ 0. (10) In this case z = 0 is the only singular point of the differential equation. Note that we have zp(z) = 1 and z 2 q(z) = z 2 − ν 2 , which are both analytic at z = 0. This implies, since p0 = 1 and q0 = −ν 2 , that z = 0 is a regular singular point of the Bessel differential equation (10) with indicial equation r(r − 1) + r − ν 2 8 ⇐⇒ r2 − ν 2 = 0. This implies that the indices are r1 = ν and r2 = −ν. Hence, for 2ν ∈ / {0, 1, 2, . . .} we have two linearly independent solutions of the form y1 (z) = z ν ∞ X an z n and y2 (z) = z −ν n=0 ∞ X bn z n . n=0 For ν = 0 we have two linearly independent solutions of the form y1 (z) = ∞ X an z n and y2 (z) = y1 (z) ln z + n=0 ∞ X bn z n . n=0 Finally, for 2ν ∈ {1, 2, 3, . . .} we have two linearly independent solutions of the form y1 (z) = z ν ∞ X an z n and y2 (z) = Ay1 (z) ln z + z −ν n=0 ∞ X bn z n , n=0 where the constant A might be zero or not. For instance, in the case ν 2 = 1/4 the indices are r1 = 1/2 and r2 = −1/2. In that case we have two linearly independent solutions y1 (z) = z 1/2 ∞ X (−1)n z 2n n=0 sin z = √ (2n + 1)! z and y2 (z) = z −1/2 ∞ X (−1)n z 2n n=0 (2n)! cos z = √ . z This can easily be seen by setting y(z) = z −1/2 u(z) into the differential equation (10) with ν 2 = 1/4. Then we have 1 y 0 (z) = z −1/2 u0 (z) − z −3/2 u(z) 2 3 and y 00 (z) = z −1/2 u00 (z) − z −3/2 u0 (z) + z −5/2 u(z), 4 which leads to 3 1 1 z 3/2 u00 (z) − z 1/2 u0 (z) + z −1/2 u(z) + z 1/2 u0 (z) − z −1/2 u(z) + z 3/2 u(z) − z −1/2 u(z) = 0 4 2 4 or z 3/2 u00 (z) + u(z) = 0 =⇒ u00 (z) + u(z) = 0. This leads to the linearly independent solutions u1 (z) = sin z and u2 (z) = cos z. In the case ν = 0 we substitute y(z) = ∞ X an z n , n=0 y 0 (z) = ∞ X nan z n and y 00 (z) = n=0 ∞ X n(n − 1)an z n−2 n=0 into the differential equation (10) with ν = 0 to find ∞ X n(n − 1)an z n + n=0 ∞ X n=0 nan z n + ∞ X an z n+2 = 0, n=0 which leads to a1 z + ∞ X [{(n + 2)(n + 1) + (n + 2)} an+2 + an ] z n+2 = 0. n=0 9 This implies that a1 = 0 (n + 2)2 an+2 = −an , and n = 0, 1, 2, . . . . The solution can be written as a2k+1 = 0 and a2k = (−1)k a0 , 22k (k!)2 k = 0, 1, 2, . . . . This leads to the solution y(z) = ∞ X n an z = a0 n=0 ∞ X (−1)k z 2k 22k (k!)2 k=0 = a0 0 F1 − z2 ;− 1 4 = a0 J0 (z), where J0 (z) denotes the Bessel function of the first kind of order zero. In the general case with ν > 0 we substitute y(z) = z ν ∞ X an z n = n=0 ∞ X an z n+ν , y 0 (z) = n=0 and y 00 (z) = ∞ X ∞ X (n + ν)an z n+ν−1 n=0 (n + ν)(n + ν − 1)an z n+ν−2 n=0 into the differential equation (10) to obtain ν(ν − 1) + ν − ν 2 a0 z ν + (ν + 1)ν + (ν + 1) − ν 2 a1 z ν+1 ∞ X + (n + ν + 2)(n + ν + 1) + (n + ν + 2) − ν 2 an+2 + an z n+ν+2 = 0. n=0 This leads to the recurrence relation (2ν + 1)a1 = 0 and (n + ν + 2)2 − ν 2 an+2 = −an , n = 0, 1, 2, . . . or equivalently, since ν > 0, a1 = 0 (n + 2)(n + 2ν + 2)an+2 = −an , and n = 0, 1, 2, . . . . The solution can be written as a2k+1 = 0 and a2k = (−1)k a0 , 22k (ν + 1)k k! k = 0, 1, 2, . . . . This leads to the solution y(z) = z ν ∞ X n=0 n an z = a0 z ν ∞ X k=0 (−1)k z 2k = a0 z ν 0 F1 22k (ν + 1)k k! − z2 ;− ν+1 4 The Bessel function Jν (z) of the first kind of order ν is defined by (z/2)ν − z2 Jν (z) = F ; − . 0 1 Γ(ν + 1) ν+1 4 10 .