GENERALIZEDRODRIGUES FORMULA SOLUTIONS

FOR CERTAINLINEAR DIFFERENTIALEQUATIONS

BY
JAMES M.HORNER

In [l] and [2] solutions were given in terms of generalized Rodrigues

formulas for the second order differential equation
(1) PÁx)y" + Px(x)y' + P0y = R(x),
where P¿(x) is a polynomial of degree not exceeding i. These results de-
pended on roots of the associated quadratic equation

(2) ^t(t+l)P'2'-tP'x + P0 = 0.

It was found that, when this equation has a positive integer root n, (1)
has solutions which can be expressed in terms of a generalized Rodrigues
formula having n — 1 for the index of differentiation. For nonpositive integer
roots of (2) a general solution for (1) can be given by an iterated indefinite
integral and for nonintegral roots the corresponding solution is a contour
integral which reduces to the Rodrigues formula for integer roots. The con-
tour integral result was obtained only for the homogeneous form of (1).
The purpose of this paper is to present the remaining results obtained
by the author in [3], namely the extension of the above results for second-
order equations to certain nth-order linear differential equations. The equa-
tion to be discussed is

(3) L[y] = ¿ Pi(x)y« - R(x), n^2,

;=o

where P,(x) is a polynomial of degree ^ i and, except for Theorem 6, x is

a real variable. It is assumed that P0 ^ 0 and that P„(x) fá 0.
Consider now the system of equations

(4) FM {TT^iy p-+1) " p»w-1}+ p-r-1 ■* °>

where FAt) is the polynomial of degree r given by

(5) FAt) =t(t-l)(t-2)-.-(t-r+ l)/r!
and r takes the values r= 1,2,3, •••,» —1.

Received by the editors December 9, 1963.

31
32 J. M. HORNER [March

Notice that the identity in (4) is with respect to x. That is, the system
of equations (4) consists of polynomials in t arising by equating the co-
efficient of each power of x to zero. For each value of r the system (4) pro-
duces n — r polynomials of degree r + 1 in t. So the total system consists
of \n(n — 1) polynomials; one of degree n, two of degree n — 1, three of
degree n — 2, ■•-, and finally n — 1 polynomials of degree two.
In the following discussion only those differential equations (3) whose
polynomial coefficients satisfy (4) are considered. The system (4) of poly-
nomials in t is called the associated system and it is assumed that not all
of the polynomials in the associated system are identically zero. If those
equations of the associated system which are not identically zero have a
common root it is called a root of the associated system. We will consider
only those differential equations for which the associated system has at
least one root. Notice that for n = 2 the associated system reduces to the
single equation (2).
The restriction P0 7e 0 prohibits the roots t = 0,1,2, ■■-,n — 2 and also
requires that either Pn(x) or xP„_i(x) be exactly of degree n. It is important
to note that it is always possible to exhibit a differential equation of the
form (3) whose associated system has a prescribed root (with the exceptions
t = 0,1,2, •••,n — 2). A particular method is to select the desired root,
choose Pn(x) arbitrarily (but of degree n), take Pn-X(x) = 0; and define
the remaining polynomial coefficients by (4).
Actually the homogeneous form of the differential equation (3) with the
restrictive condition that the associated system has at least one root yields
precisely the general Pochhammer equation [4, pp. 109-113]. To see this,
put R(x) = 0 in (3), suppose that the associated system (4) has the root
t= - u - 1, and let G(x) = - P„_i(x) - uP'nix). Then (4) becomes

p„_r_l(x) = (_1)r+^^+i);-\(,"+r)p^(x)
(6) (r+1)!
, , 1(u + l)(u + 2)---(u + r)
+ ( u (r+1)! ^ w.
for r— 1,2,3, • ••,» — 1, and so (3) can be written

Lb] = V - upy-11 + "te+Upiy*-*

(7)
_ Gy1"-1»+ {U + 1} G'y(n-2]-= 0.

(See also [5, pp. 454-465].)

It is well known that this equation can be solved using the Euler trans-
formation [4, pp. 97-99], [5, pp. 191-193], [6, pp. 333-338]. This trans-
1965] GENERALIZED RODRIGUES FORMULA SOLUTIONS 33

formation gives rise to the so-called Jordan-Pochhammer contour integrals

from which it is possible to obtain n distinct solutions for (7) [7, pp. 240-276],
[8], [9], [10].
However, in the discussion which follows the form given by (3) and
(4) is more convenient and solutions (including the Jordan-Pochhammer
integrals) can be obtained for both (3) and the homogeneous form of (3)
without appealing directly to the Euler transformation.
For convenience, let

(8) W(x)= exp[ -J (/W P„)dx]

and
(9) g(p,x) = [PAx)}pW(x),
where (except in Theorem 6) p is an integer. W(x) is, of course, a constant
multiple of the Wronskian of any fundamental system of solutions for
(3) [11, p. 328].
Theorem 1. // R(x) = 0 and the associated system has a positive integer
root j ^ (n — 1) then (3) is satisfied by the j + 1 functions

do) yito-££k{8U,*)\
and

(11) y2+1(x)= ^j^+i [gU.x) J [xl/g(j + l,x)]dxj ,

where i = 0,1,2, ■■-,j — 1.

Proof. Construct the first-order differential equation
(12) PnZ' + (Pn-l - jP'n)z = Qj-lix),
where Qj x(x) is an arbitrary polynomial of degree ^ (j — 1).
By elementary methods (e.g. [12, p. 19]) or by direct substitution (12)
is seen to have the general solution

(13) 2 = g(j,x) I k + j [Qj-JgQ+ 1,x)]dxj

where k is an arbitrary constant.
Now, using the Leibnitz rule for product differentiation [13, p. 19], dif-
ferentiate (12) j times. This yields

U4) hk\(j-k)\F"z +hk\(j-k)\{Pn-l-jPn »2 °-

34 J. M. HORNER [March

In the first sum remove the first term and in the remaining sum replace
k by r + 1. In the second sum replace k by r and remove the last term.
The two resulting sums can then be combined and (14) becomes
P^+l> + (P?-i-jI*!+1>)z

Now, in (15) remove the first term of the sum and replace the term
corresponding to r = j. This gives
PnzÍJ+l) + P„_l2w

7^xr\i] -r)\\_ ir+1) J

Since the degree of the polynomials P, does not exceed the subscript,
any terms in (16) for which r > in — 1) are zero. So the upper limit on
the sum can be changed to n — 1. Also, from (5), Fr(j) =j\/[r\(J — r)\]
so that, with the aid of (4), the coefficient of z{J~r)in (16) is seen to be Pn-r-X.
Hence, (16) is

(17) Z Pn-kZU+l-k)= 0.

But this is exactly (3) with y replaced by *°'-"+u when Ä(x) = 0. Thus
when Rix) = 0 (3) is satisfied by

(18) y = -^j^î [g(j,x) { k + j [Qj.r/gU+ 1,x)]dx}J

for every constant k and every polynomial Q>_i(x) whose degree is i= (/ — !)•
To obtain yxix) take k = 1 and Qj-X = 0. The remaining solutions y2+¿(x)
are obtained by taking k = 0 and Q>_i= x'.
In some cases it is convenient to replace gij, x) by some equivalent solu-
tion to the homogeneous form of (12).
For the particular case n = 2 results somewhat similar to the first solution
of Theorem 1 have been obtained by Brenke [14]. The method used by
Brenke consisted essentially of showing that certain orthogonal polynomials,
given in terms of a Rodrigues formula by Abramescu [15], satisfied a second-
order equation (see also [4, pp. 272-273]).
Corollary 2. // P„(x) is a polynomial of degree n then the polynomials

(19) PÍ(n_1)(x)=^{[P„(x)]*¡

satisfy the nth-order differential equation

1965] GENERALIZED RODRIGUES FORMULA SOLUTIONS 35

(20) ¿ (t+ n-!)![(.-1)-(■-!)(* + !)] 0

£J (k + n-i)\i\
Proof. The first term of (20) is P„yw and, in the notation of (3), P„_i
= (n- l)PUnd
(k + n-l)\[n-l-r(k+l)} X)
(21) Pn-r-!
(k + n-r-l)\(r+l)\ " "
Substituting these values into (4), using (5), and removing the nonzero
factor Pir+1,/(r+1)!, we obtain the associated system
t(t - 1) ... (i- r+ 1) [r(t+ 1) - (n - l)(r+ 1)]
-(* + »- l)(k + n-2) ■-■(k + n-r)[r(n + k) - (n - l)(r+l)],
which obviously has the root t = n + k — 1.
Now put j = n + k — 1 in Theorem 1 and replace g(j,x) by the equiva-
lent form [P„(x) ]J n+1 = [P„(x) ]*. The conclusion then follows using yx(x)
in Theorem 1.
Example 1. The Legendre polynomials [16, pp. 302-336]

(23) Pl(j0-¿ £[(*_!)•]

satisfy the second-order differential equation

(24) (x2- l)y" + 2xy' - k (k + l)y = 0.
Example 2. The Appell polynomials [15, p. 209]

(25) p2/fe= --r[x*(l-x2)*]

k\dx
satisfy the third-order differential equation
x(l - x2)y'" + 2(1 - 3x2)y"
(26)
+ 3(k¿ - 1)(k + 2)xy' + 2k\k + 1)(k + 2)y = 0.
Corollary 3. // Qn(x) is a polynomial of degree n and in the differential
equation

(27) y» + t (" + *~).},!Q.V-« = 0,

¡_i (n + ß - i)\

we put y = u(x)exp[— Q„(x)], the resulting differential equation is satisfied

by the polynomials
dh
(28) "*(„-«(*)= exp^M^jexpf- QAx)]].
36 J. M. HORNER [March

Proof. The first term of (20) is yln) and in the notation of (3) P„ = 1,
Pn-i = Qn, and
(29) ^-,(Ii*T1)!i.q—
(ra+ ß — 1 — r)!
As in Corollary 2, substitution of these values into (4) shows that the
associated system has the root j = n + k — 1. In this case
(30) gU,x) = exp[-Qn(x)],
so (27) is satisfied by

(31) y(x)=^Ljexp[-Q„(x)]},

and the conclusion follows.

Example 3. If in Corollary 3 Q„(x) = x2, we obtain the Hermite equa-
tion [17, pp. 187-199]
(32) u" - 2xu' + 2ku = 0
satisfied by

(33) Hkix) = ( - l)"exp(x2) ¿[exp( - x2)].

The method employed in Theorem 1 can also be used to obtain a parti-

cular solution to (3) when i2(x) ^ 0.
Theorem 4. // the associated system has a positive integer root / ^ (n — 1)
then a particular solution to (3) is

(34) yp(x) = ^jZfl {g(>, x) j [Ijix)/gij + 1,x)]dx} ,

where

(35) IJix)=jj...JRix)idx)J.

Proof. The proof is exactly the same as that of Theorem 1 after replacing
Qj-Xix) by Ijix) in (12) and replacing the general solution (13) by

(36) gU,x)j[IJ/g<J+l,x)]dx.

Differentiation of the left member of (12) j times will, as before, give

the left member of (3) while differentiation of Ijix) j times yields i?(x).
Because of Theorem 1 it is sufficient to neglect any polynomial terms
of degree ^ (j: — 1) occurring in Ijix) before computing yp(x). In particu-
1965] GENERALIZED RODRIGUES FORMULA SOLUTIONS 37

lar, the constants of integration may be taken as zero when computing Ij(x).
If not, the effect is to combine with a particular solution some of the solu-
tions in (10) and (11) for the homogeneous form of (3).
Example 4. The nonhomogeneous Legendre equation
(37) (x2 - l)y" + 2xy' - k(k + l)y = ex
yields an associated system consisting of a single quadratic equation whose
roots are t = — k and t = k + 1. So when k is a non-negative integer a
particular solution of (37) is

(38) yp(x)= ^[ (x2- l)*JV(x2 - i)-k-ldx]

In some instances it is desirable to replace the indefinite integrals by

definite integrals. For example, W(x) might be replaced by W(x, x0) where

(39) W(x,xQ)= exp{- fX[Pn_x(v)/PAv)]dv} ,

and x0 is any point for which the integral exists and satisfies the necessary
differentiability and integrability conditions implied in the conclusions of
the theorems. In particular, I Ax) of the previous theorem may be replaced by

(40) /,(x)=JajfXJaj_x
f •••JaxCR(v)(dv)J
or, taking ax = a2= • • • = a¡ = a, by

(41) IAx)= 7j4t)! X (*- o)J~Wdv.

The idea of interpreting iterated integrals as derivatives of negative order
suggests the following result.
Theorem 5. // the associated system has a negative integer root t = —j
and R(x) £ CJ then every solution of (3) can be obtained from

(42) y(x)=J f •■■fg(-J,x)[k+f[Ru>/g(-j+l,x)]dx}

(dx)^-1
for appropriate choices of k and the constants of integration.
Proof. By reversing the sum in (3) it becomes

(43) tPn^y{n-k) = R(x).

Suppose that y(x) is any solution to (43) and differentiate (43) j times.
This gives
38 J. M. HORNER [March

(44) ¿ ¿ ../' Pt*y{n+1-k-l) = Ä0).

k= Q í_0 l\\J — l)\

Collecting coefficients of y(n+;,) we have

(45) Zl Z ,,J'
z(Z ,J' ^,Pn-k) y{n+J~q)
= RiJ>, i£j,k¿n.
„Ptk\>*+'-*-A«,
Now remove the terms corresponding to q = 0 and q = 1 from the sum
in (45). Also since P{nlk= 0 if i + k > n the upper limit on the first sum
can be changed to n. Then (45) becomes

Pny^ + iPn-i + jP'Jy^'1*

(46) n I
+ ZE .,/• .,,P¿'i*y("+^
«=2 i+k=q V.(J — l)\
= A^, i ay.
So we are led to consider

(47) fq= Z u-'' •\ipn-'» Oá» a;, 9-2,3,...,».

There are two cases to consider: those q for which q s¡ /' and those for
which q > ;'. If q^j then in (47) put i = a — ß in the summand and sum
on ß from 0 to a. After a shift in summation index, (47) then becomes

f Jm pW i_i:_ pn-u
/? <?!(>-<?)!" + <ç-l)!(j + l-«)l -1
(48)
1-1 it

^^iq-r-DHj + r+l-qV/"-'-1'
where 2 ^q ^j, qfkn.
Now t = —j is by hypothesis a root of the associated system. So put
t = —y in (4) and after differentiation q — r — 1 times substitute the re-
sulting expression in (48), and simplify. This gives, for qikj, 2¿¡q^n,

¿i_pi«) _i_¿l_]
qlU-qV- " (9-DÎ0 + 1-9)!

+ yu ni, ¿-(9_r_1)!0 + r+1_g)!(r+1)

£í (9-r-l)!(j + r+l-9)!r!

Now,
1965] GENERALIZED RODRIGUES FORMULA SOLUTIONS 39

If (50) is differentiated q — 2 times using the series the result is

tKU d"~2 « (_l)V0' + *-2)!

du«"2' v ' '"-1 £,(,-*)!*!(/ +A-ç)!-

On the other hand if (50) is differentiated q — 2 times using the left

member and Leibnitz rule, the resulting expression vanishes for u = 1. So

fci (a-Ä)!Ä!0 + Ä-9)!

Also

(53) ^-l[-9(l-«),-l]-L
*-i (q-k)\k\
As before, differentiating (53) o — 2 times by the two methods and putting
u = 1 yields

, . ± {-l)kqlk(j + k-2)l ^
k(q-k)\klU + k-q)\
Subtraction of (52) from (54) gives

i55v f (-l)kgl(k-l)(j + k-2)\ = Q-2)!

h (q-k)\k\(j + k-q)l 0-9)!'
Putting k = r+1 in this sum shows that the coefficient of P„] in (49)
is zero, while removal of the first term in (54) followed by the same index
shift shows that the coefficient of Pn-1] in (49) is also zero. Thus /, = 0
ifqèj.
Now suppose o > j. In (47) again put i = q — k in the summand. In this
case, however, the sum is from k = q —j to k = q. Then put k = m + q —j
in the resulting sum to obtain

(56) /, = ± ,,- •/',,m, «S&— j<q,2^q<n.

m=o (J - m)\m\

Since t = —jis a root of the associated system, (4) becomes

(57) n-r-i = 0_1)!r! - (''-i + j-^Fn ) '

which in this form is valid also for r = 0. In (57) put r = m + q —j — 1,

40 J. M. HORNER [March

differentiate j — m times, and substitute the resulting expression into (56).

This will give, by rearranging,

_ (_ p«-;-* J jl(-ir (m + q-2)\

0-D! "-1 ^(j-m)\m\ (m + q-j-l)\
(58)
(-lr^Q-l) p(„ + j!(-Dm (m + q-j-l)(m + g-2)l
+ " 0-D! • ¿?00-m)!m! (m + g-;)!
Paralleling the previous case,

du; M „=o 0 - ™)!m! (m + o-j-l)!

so the coefficient of Pi-Y' in (58) is zero. If ; = 1 the second term of (58)

is obviously zero. For j > 1,

n dJ~2 i •-«/, wl V^ >!(-l)m(m + ç-2)!

(60 0 = T-F2
du1
u« 2(l-u)J „_!= ^o(j-m)\m\(m
X r-vr-fT-¡-Tr + q-j)\

and

^o.^K-i-y.i -)>-]|-,-¿ff(-"££?."»,
au; m=0 0 - m)!m!(/n + q — ;)!

If (60) is multiplied by (q —j — 1) and added to (61) it is seen that the

coefficient of P^ in (58) is also zero. Hence for all g = 2,3, •••,«, /,= 0.
So from (46), if y(x) is any solution of (3) and the associated system has
the negative integer root t = —j then y(n+'~v> satisfies

(62) Pnz' + (Pn_x + jP'n)z = R(J>.

Thus for some constant k and appropriate choices of constants of integration,

(63) d^T? = g(-j,x){k+f [R^/g(-j+l,x)]dx} ,

and the proof is complete. It is interesting to compare (62) with (12) and
(63) with (18) and (34).
For the case of the second-order equation some results similar to the
conclusion of Theorem 5 have been obtained by Abbé Laine [18], though
his results are in a considerably different form.
Example 5. The Hermite equation (32) has the associated root t = — k
so every solution can be obtained from

(64) y(x) = C ••• exp(x2)(dx) \k+i

-«//-/
1965] GENERALIZED RODRIGUES FORMULA SOLUTIONS 41

for some choice of C and the constants of integration.

Though it has been assumed throughout that x is a real variable the
results can be easily extended to a complex variable setting.
In conclusion we state without proof the well-known result of applying
the Euler transformation to (3) to obtain the Jordan-Pochhammer integrals.
The theorem is given here in slightly different (but equivalent) form from
the usual statement for reasons of comparison with the first solution of
Theorem 1 and with Theorem 5 (see [19, pp. 104-111]).
Theorem 6. // fi(x) = 0 and a is a root of the associated system, then

(65) y(x) fjicgia,t)dt

Jcit- x) a —n+2
is a solution to (3) provided that C is any contour on any Riemann surface
of the integrand for which
iaa\ ui a g(a+l,t)
(66) »tefl-(t_xri
resumes its original value after describing C, and provided that differentiation
with respect to x under the integral sign is valid for C.
Proof. The proof of this theorem in an equivalent form is given in many
of the standard references for differential equations. The usual approach
is to use (7), instead of (3) with the additional condition (4).
It is, however, an interesting exercise in Taylor's formula and series
manipulations to prove Theorem 6 by substituting (65) into (3) and using
(4) to arrive at

(67) L[y]= -a(o-l)...(o-n + 2) f Uh(x,t)\dt.

Jc at
A proof of Theorem 6 by this method is given in detail in [3].

References
1. James M. Homer, A note on the derivation of Rodrigues' formulae, Amer. Math. Monthly
70 (1963), 81-82.
2. _, Generalizations of the formulas of Rodrigues and Schläfli, Amer. Math.
Monthly 71 (1964), 870-876.
3. _, Generalizations of the formulas of Rodrigues and Schläfli, MA Thesis,
University of Alabama, University, Alabama, August, 1962.
4. E. Kamke, Differentialgleichungen: Lösungsmethoden und Lösungen, Vol. 1, 4th ed.,
Akademische Verlagsgesellschaft, Leipzig, 1951.
5. E. L. Ince, Ordinary differential equations, Dover, New York, 1956.
6. A. R. Forsyth, Theory of differential equations. Ill {Ordinary linear equations), Vol. 4,
Cambridge University Press, Cambridge, 1902.
7. M. C. Jordan, Cours d'analyse, Vol. Ill, 2nd ed., Gauthier-Villars, Paris, 1896.
42 J. M. HORNER

8. L. Pochhammer, Ueber ein Integral mit doppeltem Umlauf, Math. Ann. 35 (1890),
470-494.
9. _, Zier Theorie der Euler'schen Integrale, Math. Ann. 35 (1890), 495-526.
10._, Ueber eine Klasse von Integralen mit geschlossener Integrationscurve, Math.
Ann. 37 (1890), 500-543.
11. Ralph Palmer Agnew, Differential equations, McGraw-Hill, New York, 1942.
12. Murray R. Spiegel, Applied differential equations, Prentice-Hall, Englewood Cüffs,
N. J., 1958.
13. Wilfred Kaplan, Advanced calculus, Addison-Wesley, Reading, Mass., 1952.
14. W. C. Brenke, On polynomial solutions of a class of linear differential equations of
second order, Bull. Amer. Math. Soc. 36 (1930), 77-84.
15. N. Abramescu, Suite serie di polinomi di una variable complessa. Le serie di Darboux,
Ann. Mat. Pura Appl. (3) 31 (1922), 207-249.
16. E. T. Whittaker and G. N. Watson, Modern analysis, 4th ed., Cambridge University
Press, Cambridge, 1927.
17. Earl D. Rainville, Special functions, Macmillan, New York, 1960.
18. Abbé Laine, Sur l'intégration de quelques équations différentielles de second ordre,
Enseignement Math. 23 (1923), 163-173.
19. E. G. C. Poole, Linear differential equations, Oxford University Press, Oxford, 1936.

University of Alabama,
University, Alabama