1
$\begingroup$

The generating function of the Jacobi polynomials is given by $$ \sum_{n=0}^{\infty} P_{n}^{(\alpha, \beta)}(z) t^{n}=2^{\alpha+\beta} R^{-1}(1-t+R)^{-\alpha}(1+t+R)^{-\beta} $$ where $$ R=R(z, t)=\left(1-2 z t+t^{2}\right)^{\frac{1}{2}}. $$ However, we need a similar result or alternative representations for the square of the Jacobi polynomial given by $$ \sum_{n=0}^{\infty}( P_{n}^{(\alpha, \beta)}(z))^2 t^{n} $$ or more specifically, $$ \sum_{n=0}^{\infty}( P_{n}^{(1, 0)}(z))^2 t^{n}. $$ I have researched this in detail, but have not found an answer. If someone could give me some hints, that would be really helpful.

$\endgroup$
9
  • $\begingroup$ You can ("automatically") derive a (3rd order) recurrence relation, for example using a guessing package, but it does not look very nice. Would this be sufficient for you? $\endgroup$ Feb 11, 2022 at 8:17
  • $\begingroup$ Sage says ${}_3F_2\left( \begin{matrix} \tfrac32, \tfrac32, 1 \\ 2, 2 \end{matrix} \Big\vert\, 4tx^2 - 8tx + 4t \right)$. Comment, not answer, because I haven't sanity-checked it. $\endgroup$ Feb 11, 2022 at 8:50
  • $\begingroup$ @PeterTaylor I don't see how your formula could make sense. If you pick out the coefficient of $t^n$ you get a multiple of $(x-1)^{2n}$, not the square of a Jacobi polynomial. $\endgroup$ Feb 11, 2022 at 9:59
  • $\begingroup$ @MartinRubey Thank you for the comment. I am afraid I don't know much about what you told me, but the ideal expression is a simple mathematical expression, something like the Elliptic integrals or Hypergeometric function that Peter Taylor has provided. $\endgroup$
    – Kane
    Feb 11, 2022 at 12:10
  • $\begingroup$ @PeterTaylor Thank you for the comment. If possible, can you share the sage code with us? $\endgroup$
    – Kane
    Feb 11, 2022 at 12:16

3 Answers 3

2
$\begingroup$

I don't know if this is useful to you, but Bailey gave the identity \begin{multline*}\sum_{n=0}^\infty\frac{n!(\alpha+\beta+1)_n}{(\alpha+1)_n(\beta+1)_n}\,P_n^{(\alpha,\beta)}(x)P_n^{(\alpha,\beta)}(y)\,t^n\\ =\frac 1{(1+t)^{\alpha+\beta+1}}\sum_{m,n=0}^\infty\frac{(\alpha+\beta+1)_{2m+2n}}{m!n!(\beta+1)_m(\alpha+1)_n}\left(\frac t{4(1+t)^2}\right)^{m+n}((1+x)(1+y))^m((1-x)(1-y))^n.\end{multline*} The original reference is J. London Math. Soc. 13 (1938), 8-12, but I copied it from Stanton, Proc. Amer. Math. Soc. 80 (1980), 398–400. When $\beta=0$ and $x=y$, this gives an expression for $$\sum_{n=0}^\infty P_n^{(\alpha,0)}(x)^2\,t^n$$ as a double series.

$\endgroup$
2
$\begingroup$

Here is a way to guess an expression - it is a bit uneven, because fricas has no builtin jacobiP and sage has apparently trouble transferring a long list to fricas. But it works, at least if you fix $b=0$:

sage: S.<a,b,x> = QQ[]
sage: l = [S(jacobi_P(n, a, 0, x)) for n in range(30)]
sage: frec = fricas.guessPRec(l)[0]; frec
... # omitted to save space
sage: frec.name()
'sage74'
sage: fricas("guessPRec [(eval(sage74, ['n], [n])::FRAC POLY INT)^2 for n in 0..60]")
... # huge output of the form p3(n) f(n + 3) + p2(n) f(n + 2) + p1(n) f(n + 1) + p0(n) f(n)

If you fix $a = 1$, things become slightly nicer, but not much:

sage: l = [S(jacobi_P(n, 1, 0, x))^2 for n in range(50)]
sage: frec = fricas.guessPRec(l)[0].getEq().sage(); frec
(2*n^4 + 25*n^3 + 113*n^2 + (8*n^6 + 132*n^5 + 882*n^4 + 3047*n^3 + 5727*n^2 + 5544*n + 2160)*x + 216*n + 144)*f(n + 3) + (2*n^4 - (32*n^6 + 496*n^5 + 3136*n^4 + 10328*n^3 + 18642*n^2 + 17451*n + 6615)*x^3 + 19*n^3 - (24*n^4 + 244*n^3 + 902*n^2 + 1431*n + 819)*x^2 + 64*n^2 + (8*n^6 + 124*n^5 + 782*n^4 + 2561*n^3 + 4576*n^2 + 4212*n + 1557)*x + 90*n + 45)*f(n + 2) - (2*n^4 - (32*n^6 + 464*n^5 + 2736*n^4 + 8392*n^3 + 14122*n^2 + 12369*n + 4410)*x^3 + 21*n^3 - (24*n^4 + 236*n^3 + 842*n^2 + 1289*n + 714)*x^2 + 79*n^2 + (8*n^6 + 116*n^5 + 682*n^4 + 2079*n^3 + 3461*n^2 + 2973*n + 1022)*x + 125*n + 70)*f(n + 1) - (2*n^4 + 15*n^3 + 38*n^2 + (8*n^6 + 108*n^5 + 582*n^4 + 1593*n^3 + 2322*n^2 + 1701*n + 490)*x + 39*n + 14)*f(n)
sage: f = function("f")
sage: var("n")
n
sage: frec.coefficient(f(n+3)).factor()
(4*n^2*x + 16*n*x + 15*x + 1)*(2*n + 3)*(n + 4)^2*(n + 3)
sage: frec.coefficient(f(n+2)).factor()
-(4*n^2*x^2 + 20*n*x^2 - n^2 + 21*x^2 - 5*n + 2*x - 5)*(4*n^2*x + 24*n*x + 35*x + 1)*(2*n + 3)*(n + 3)
sage: frec.coefficient(f(n+1)).factor()
(4*n^2*x^2 + 20*n*x^2 - n^2 + 21*x^2 - 5*n + 2*x - 5)*(4*n^2*x + 16*n*x + 15*x + 1)*(2*n + 7)*(n + 2)
sage: frec.coefficient(f(n)).factor()
-(4*n^2*x + 24*n*x + 35*x + 1)*(2*n + 7)*(n + 2)*(n + 1)^2
$\endgroup$
1
$\begingroup$

Here is an expression that involves elliptic functions and a portion represented as an integral that (I think) cannot be reduced to elliptic functions.

$$ \sum_{n=0}^\infty P^{(1,0)}_n(x)^2 t^n = \frac{1}{(1-x)^2}\big( F(x,t)+\frac{1}{t} (F(x,t)-1) \big) -\frac{2x}{t}\big(F(x,t)-1\big) + \frac{1+x}{1-x}\frac{G(x,t)}{t} $$ where $$F(x,t) = \sum_{n=0}^\infty P_n(x)^2t^n$$ and has a closed-form in elliptic functions from Bailey's formula $$ \tag{B} \sum_{n=0}^\infty P_n(x) \, P_n(y) t^n = r \cdot {}_2F_{1}\big(1/2,1/2;1;-4\sqrt{(1-x^2)(1-y^2)} \,t \, r^2\big)$$ $$ r= \big(1+t(t-2\sqrt{(1-x^2)(1-y^2)} - 2xy)\big)^{-1/2} $$ I've found two integral representations for $G(x,t):$ $$G(x,t)=\frac{2x}{\pi (1-x^2)}\int_0^t \frac{du}{u(1-u)}\Big( \frac{(u-1)^2}{(1+u)^2-4ux^2} \,E\big(\frac{4u(x^2-1)}{(u-1)^2}\big) - K\big(\frac{4u(x^2-1)}{(u-1)^2}\big) \Big) $$ and $$G(x,t)=\frac{-4 t}{\pi \, x}\int_0^1 du\sqrt{\frac{u}{1-u}}\frac{1}{\rho(u;x,t)} \Big((1+t+ \rho(u;x,t))^{-1} + (1-t+ \rho(u;x,t))^{-1} \Big)$$ $$ \text{with } \rho(u,x,t)=((1-t)^2+4tu(1-x^2))^{1/2} $$ In the first integral, $E$ and $K$ are elliptic functions in Mathematica notation.

The following is a proof sketch: $$ P^{(1,0)}_n(x) = \frac{P_n(x)-P_{n+1}(x)}{1-x} $$ where the $P_n(x)$ are the ordinary Legendre polynomials. Square this expression within the original sum, and there are three sums involving summands with $(P_n)^2,$ $(P_{n+1})^2$ and $P_n P_{n+1}.$ The 'perfect squares' can be dealt with eq. (B), although an index shift is needed for the $(P_{n+1})^2.$ This gives two of the $F(x,t)$ terms in the answer. The cross-term series can be dealt with a relationship among Legendre polynomials $$\sum_{n=1}^\infty P_n(x)P_{n-1}(x)t^n= \sum_{n=1}^\infty P_n(x)\Big(xP_n(x) + \frac{1-x^2}{n} \frac{d}{dx}P_n(x) \Big)t^n $$ Again, we have a 'perfect square' case that can be dealt with eq. (B). The last sum to deal with is $$ \sum_{n=1}^\infty P_n(x) \frac{d}{dx} P_n(x) \frac{t^n}{n} =\frac{1}{2} \frac{d}{dx} \sum_{n=1}^\infty P_n(x)^2 \frac{t^n}{n} $$ The first integral representation I gave for $G(x,t)$ involves integrating eq. B (specialized to y=x) with respect to $t.$ Then do the derivative with respect to $x$ afterwards. The second integral relationship started from an integral expression for the square of a Legendre polynomial, Gradshteyn and Rhyzik 7.137.5. Again, after getting the integral, differentiate with repsect to $x.$

I doubt the integrals simplify to elliptic functions, but would love to be proved wrong.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.