Continued fraction formulae involving ratios of three gamma functions

For n\in\mathbb{N}, let

See [AS, p. 255, (6.1.11) and (6.1.13)]. Mortici et al. [MCL] first investigated the continued fraction approximations for T_1 and T_2. Bai et al. [BLW] showed some inequalities for T_1 and T_2. Corollary 1 and 2 confirm two questions proposed in [CTZ, CW].

We may compare the above two corollaries with the following Bauer’s continued fraction in 1872, which was communicated by Ramanujan [Ram62, p. xxvii] in his first letter to Hardy:

The idea of Lemma 2 is originated from Mortici [Mor], which will be called a Mortici transformation. In its applications, it should be stressed that we always choose a suitable f(x) such that f(x)/f(x+1) is a rational function in x by using the recurrence relation \Gamma(x+1)=x\Gamma(x) and variable transformation.

To guess continued fraction formula of the form \frac{1-P}{1+P}, we often use the following simple transformation (see Lorentzen and Waadeland [LW, p. 560, (1.1.2)])

For instance, Entries 34 and 35 in Chapter 12 of Ramanujan’s second notebook [Berndt], Theorem 5 and three conjectures in Section 9 below. With the aid of the MC-algorithm, we can guess almost all continued fraction formulae in Chapter 12 of Ramanujan’s notebooks (see [Berndt, BLW]) involving products and quotients of gamma functions apart from his Entry 40.

This transformation is in particular useful if b_0+\mathop{\mbox{\bf\large K}}\big(a_m/b_m\big) and its Bauer-Muir transformation converge to the same value. This was proved to be the case for positive continued fractions and positive modifying factors r_m by Perron [Perron, p. 27], but of course it holds whenever \mathop{\mbox{\bf\large K}}\big(a_m/b_m\big) converges with exceptional sequence \{r_m^\dag\} and \lim\inf \mathfrak{m}(r_m,r_m^\dag)>0, see [LW1, p. 83] or [Loren94c]. Berndt [Berndt98, p. 37-38] describes this transformation in details.

Although the Bauer-Muir transformation is classical, there is still some room for further interesting applications. The new idea of this paper is that we always choose suitable nonlinear modifying factors \left\{r_m\right\}_{m=0}^{\infty} so that \varphi_{m+1}/\varphi_m equals to -1 for m\in \mathbb{N}. In other words, for each proof of five theorems \varphi_m is not constant, which will be called as the adjoint factors of \{r_m\}_{m=0}^{\infty} for b_0+\mathop{\mbox{\bf\large K}}\big(a_m/b_m\big), or the adjoint factors \left\{\varphi_m\right\}_{m=1}^{\infty} for short. In addition, we shall employ the Bauer-Muir transformation twice with different modifying factors. To the best knowledge of authors, such an approach is used for the first time in this paper. So our theorems may be viewed as new applications of the Bauer-Muir transformation. For such approaches, we will show more examples elsewhere.

Via the MC-algorithm, we can only guess that (1.5) holds. However, we can not prove such a claim directly by the Bauer-Muir transformation, we have to find some “simpler" continued fractions such that the continued fraction on the right side of (1.5) is the even part of it. Therefore, firstly we should discuss 20 possibilities, and then pick out 3 “uncomplicated" expressions among them. Secondly, write each of them into the polynomial continued fraction. Thirdly, try to check whether each of them satisfies the identity (1.5). It is eventually shown by the Bauer-Muir transformation that only one among them is workable. The other two continued fractions can be are expressed as follows: for j=1,2

where b_0(1;x)=x^2-(1-\alpha)^2-\alpha\beta(\alpha+\beta)/4, and for k\in\mathbb{N}

and b_0(2;x)=x^2-(1-\alpha-\beta)^2+\alpha\beta(\alpha+\beta)/4, and for k\in\mathbb{N}

respectively.

Note that there are four parameters in Conjecture 4. Setting \eta=0 , we obtain Entry 34 of Ramanujan on continued fractions involving gamma functions in his notebooks directly, see [Berndt, p. 156]. Applications for Theorem 5, Conjectures 1 to 3 will be discussed elsewhere.

Of course, it is an interesting question to prove our theorems by the theory of hypergeometric functions or orthogonal polynomials (for instance, see  [KLS, KS]). Moreover, it certainly would be a very interesting work to extend our results to q-analogue (see [GM1993, GM1998, Masson1995]). Finally, solving these conjectures should help progress the theory of continued fractions, the hypergeometric function, orthogonal polynomials, q-series, etc.