1. Introduction
A decade ago the author [
1] (under influence of Odake and Sasaki’s celebrated paper [
2] and its extension in [
3]) demonstrated the existence of the three isospectral families of the rational SUSY partners of the hyperbolic Pöschl-Teller (
h-PT) potential [
4] quantized by the rational Darboux transforms (
) of the Romanovski-Jacobi (R-Jacobi) polynomials [
5,
6,
7]. In all the cases (labeled
,
, and
below for the reasons specified later) the quasi-rational transformation functions (q-RTFs) were represented by the principal Frobenius solutions (PFSs) of the Jacobi-reference (JRef) canonical Sturm-Liouville equation (CSLE) nonvanishing in the quantization interval (+1, ∞). The finite exceptional orthogonal polynomial EOP sequence reported in [
3] was identified by us as type
. As pointed to in [
8], the infinitely many finite polynomial sequences of type
had been constructed by Grandati [
9] a year earlier; however, he did not recognized the fact that, in contrast with other potentials discussed in the paper, the aforementioned polynomial sequences are X-orthogonal with a positive weight.
In the same year as [
1], Yadav et al. [
10] calculated the scattering amplitude for the rationally extended
h-PT potential solvable by the finite sequence of the EOPs of type
, referring to the latter simply as ‘X
m-Jacobi EOPs’. The epithet (repeatedly used by these authors in the more recent papers [
11,
12,
13,
14]) seemed confusing since the EOPs in question do not belong to the X
m-Jacobi orthogonal polynomial system (OPS) [
15,
16,
17] and therefore cannot form its finite subset.
Before continuing our discussion, let us first point to the dubious use of the term ‘EOP’ in the literature, similar to the slang use of the term ‘orthogonal Jacobi polynomials’, instead of ‘classical Jacobi polynomials’, which disregards the existence of the finite orthogonal subsets formed by the R-Jacobi polynomials. Similarly, Gómez-Ullate, Milson et al. [
15,
16,
18,
19] use the term ‘EOPs’ as the synonym for ‘X-OPS’, disregarding the existence of the finite EOP sequences [
1,
2,
3,
8,
10,
11,
12,
13,
20,
21,
22,
23,
24,
25,
26,
27,
28,
29,
30] represented by the rational Darboux transforms (
) of Romanovski polynomials [
1,
8,
21,
24,
25,
26,
27,
28,
29]. (As a puzzling exception, Gómez-Ullate, Grandati, and Milson [
31], when citing the studies on the EOPs, did mention the papers [
9] and [
20], which deal solely with the problems solved by the finite EOP sequences).
On the contrary, Yadav et al. [
12], after giving credit to the three pioneering studies [
15,
32,
33] on the X
m-Jacobi and X
m-Laguerre OPSs, made reference to a mixture of papers covering both infinite and finite EOP sequences. This statement (in contrast with their earlier remark [
11] viewing the finite EOP sequences as subsets of the X-OPSs) is mostly consistent with our use of the term ‘EOP’.
The rigorous analysis of the aforementioned finite EOP sequences of types
,
, and
in [
24] revealed that the EOP sequences of types
and
are formally generated by the same shift operators as the cases J1 and J2 case in [
3], except that the indexes of the seed Jacobi polynomials in our case were independent of the polynomial degrees. This observation brought the author [
24] to the concept of the exceptional differential polynomial systems (X-DPSs), with the term ‘DPS’ used in exactly the same sense as it was suggested by Everitt et al. [
34,
35] for conventional sequences of polynomials obeying the Bochner theorem [
36]. In following the commonly accepted terminology suggested by Gómez-Ullate et al. [
33] for the X
1-OPSs, we call the given DPS exceptional, because the polynomial sequence in question either does not start from a constant or lacks the first-degree polynomial and therefore do not obey the Bochner theorem. Indeed, as stressed by Kwon and Littlewood [
37], Bochner himself “did not mention the orthogonality of the polynomial systems that he found. The problem of classifying all classical orthogonal polynomials was handled by many authors thereafter” based on his analysis of possible polynomial solutions of
complex second-order differential eigenequations.
Compared with the rigorous mathematical analysis of the X-OPSs in [
18,
19], the concept of the X-DPSs put forward by us in [
24] represents the parallel direction dealing with the solvable rational CSLEs (RCSLEs) and related X-Bochner ordinary differential equations (ODEs), instead of the (irregular) exceptional Bochner (X-Bochner) operators in [
18] and related polynomial Sturm-Liouville problems (PSLPs) sketched in [
15,
18,
32].
The interrelation between the two approaches is closely related to the dual use of the term ’Darboux transformation’ (DT), following the discovery by Andrianov et al. [
38,
39] that the renowned transformation of the Schrödinger equation initially suggested by Darboux [
40] for the generic second-order canonical differential eigenequation (long before the birth of the quantum mechanics) is equivalent to its intertwining factorization. We refer the reader to the excellent overview of this issue in [
41].
More recently Gómez-Ullate et al. [
42] initiated the new direction in the theory of the rational Sturm-Liouville equations (RSLEs) by applying the intertwining factorization to the second-order differential eigenoperator. This operation was termed ’Darboux transformation’, based on the dualism existent in the particular case of the Schrödinger operator. This innovation followed by its extension to the X-OPSs [
15,
32,
33] laid the foundation for their rigorous theory, which was advanced to the higher level in [
18,
19].
The author (being accustomed [
43] to the strict use of the mentioned term) took the different turn [
1] in the extension of the DTs to the SLEs. As we understand now, our original intuitive idea was based on the three-step operation:
- (i)
the Liouville transformation from the CSLE to the Schrödinger equation;
- (ii)
the Darboux deformation of the corresponding Liouville potential;
- (iii)
the reverse Liouville transformation from the Schrödinger equation to the new CSLE using the same change of variable as at Step (i).
This was referred to by us years later [
44] as ‘Liouville-Darboux transformation’. Note that we invented the term ‘Darboux deformation’ (DD), used instead of ‘DT’, simply to avoid the multiple repetition of the word ‘transformation’. It is worth stressing again that we give to this term its original meaning implied by Darboux [
40].
The extensive exploration of the literature revealed that the transformation of the CSLE sketched above has been introduced by Rudyak and Zakhariev [
45] in the late eighties. Schnizer and Leeb [
46,
47] named it ‘generalized Darboux transformation’ (GDT); this name was also accepted in some later studies on this subject. However, since various authors give to this widely used name completely different meanings in both physics and mathematics (see [
44] for numerous examples), we suggested the aforementioned name ‘Liouville Darboux transformation’ as an alternative. Our current perception is that the latter is slightly misleading because it relates the definition of the transformation to the DD of the Liouville potential, which is absolutely irrelevant to the problem under consideration unless we are interested in quantum-mechanical applications.
Below we simply refer to the above operations as ‘Rudjak-Zakhariev transformations’ (RZTs) and consider its three-step decomposition suggested in [
44] just as one of its realizations, but not as its definition (see
Appendix A for more details). We term the RZT of the RCSLE as ‘rational RZT’ (RRZT) if it uses a q-RTF. The use of the RRZTs for constructing the RCSLEs with infinite sequences of quasi-rational solutions (q-RSs) constitutes the key element of the developed-by-us [
8,
24,
25,
26,
27,
28,
29] theory of the X-DPSs formed by the polynomial components of the q-RSs of the given series.
In this paper we, based on the results of our previous studies [
8,
24], scrupulously analyze the manifold of the RCSLEs obtained by RRZTs of the Jacobi-reference (JRef) CSLE, which is defined via (1)–(3) in
Section 2. Each q-RS formed by a m-degree Jacobi polynomial can be used as the q-RTF giving rise to the RCSLE with m + 2 poles in the finite complex plane. The Jacobi indexes of the seed polynomials are defined in the segments carved by the three vertical and three horizontal lines with the abscissas and respectively ordinates equal to −1, 0, and +1.
Each transformed CSLE is then converted to the Bochner-type ODE with polynomial coefficients, taking advantage of the fact that the density function of our interest has only simple poles in the finite plane and as a results the mentioned gauge transformation is energy-independent [
1]. Consequently, the linear coefficient function of the resultant differential equation does not depend on the degrees of the sought-for polynomials.
It has been proven in [
24] that each transformed RCSLE constructed in such a way has a quartet of infinite sequences of q-RSs with polynomial components forming the four X
m-Jacobi DPSs of series J1, J2, D, and W, as they were termed by us. The shift operators for the cases J1 and J2 in [
3] match our equations for the X
m-Jacobi DPSs of series J1 and J2 accordingly, though, as pointed to in [
8], the indexes of the seed Jacobi polynomials in our scheme are independent of the polynomials, in a sharp contrast with [
3]. The innovative concept of the X
m-Jacobi DPSs containing both infinite and finite EOP sequences constitutes the cornerstone of our formalism.
Another fundamentally important part of our approach is the formulation of the rational Sturm-Liouville problem (RSLP) to detect infinite or finite orthogonal subsequences in each of the four X
m-Jacobi DPSs. Note that, until now, we have not mentioned any restriction on the zeros of the seed Jacobi polynomials. However, to select the RSLPs solvable by either infinite or finite EOP sequences, we have to focus solely on the seed Jacobi polynomials with no zeros inside the quantization interval. At this point we come to the main advantage of our approach, compared with the general theory of X-OPSs advanced by Garcia-Ferrero et al. [
18,
19]. Namely we formulate the RSLP for both finite and infinite quantization intervals. In contrast, the X
m-Jacobi OPSs appear only if the OBCs for the given SLP are imposed at the ends of the finite interval (−1, +1). Since the cited authors were interested only in the X-OPSs they term the X-Bochner operator ‘regular’, if the q-RTF used to generate the RCSLE in question does not have nodes inside the interval (−1, +1). On the contrary, we have to additionally specify the open interval, where the RCSLE of our interest may not have any singularities. For the purposes of this paper (unless explicitly specified otherwise) the q-RTF is termed regular if does not have real nodes larger than 1. Similarly we refer to the RRZT as regular (reg-RRZT) if the corresponding transform of the JRef CSLE does not have poles in the interval (1, ∞).
A certain deficiency of our Sturm-Liouville approach, compared with the intertwining technique advanced in [
18,
19], is that we [
48] require that the RRZT of the RSLE preserves both the leading and weight coefficient functions. As a result, the RRZTs, as they define here, represent only a narrow subset of the rational Darboux transformations (RDTs) in the terminology of Garcia-Ferrero et al. [
18,
19].
As it has been already done above in our references to the of the Romanovsky polynomials, we will often use the commonly accepted term ‘’, instead of ‘’, despite the fact that we have to introduce the more restrictive requirement for both indexes of the classical Jacobi polynomial or the first index of the R-Jacobi polynomial to be positive.
Our next step is to find all quasi-rational solutions (q-RSs) of the JRef CSLE, which do not have nodes inside the selected quantization interval, which assures that the transformed RCSLE does not have poles inside the interval of our interest. It was taken for granted in our earlier works [
1,
8,
24] that any PFS below the lowest eigenvalue is necessarily nodeless. As one can see from the proof presented in
Appendix B, this is not by any means a trivial (though wide-spread) presumption.
Using Klein’s formulas (see §6.72 in [
49]) for the numbers of zeros of a Jacobi polynomial in the intervals (−∞, −1), (−1, +1), and (+1, ∞), we showed [
1] that the JRef CSLE also has irregular-at-both-ends q-RSs (type
) with no nodes in the interval (+1, ∞). Keeping in mind the RRZT using this q-RS as q-RTF inserts the extra energy level below the lowest eigenvalue, it can be called [
41] ‘dressing’ transformation.
To find the orthogonal
of the classical and R-Jacobi polynomials, we convert the transformed RCSLE to its prime form defined in such a way [
48] that the Dirichlet boundary condition (DBC) unambiguously selects the PFS near the singular end in question. As a result, the DBCs at the ends of the given quantization interval select the solutions representing the PFSs near both singular ends. This is the brand new approach, which has allowed us to treat both infinite and finite EOP sequences in parallel, in contrast with the conventional technique [
15,
16,
17] paying attention solely to the X-OPSs. The technique advanced in our papers [
8,
24,
25,
26,
27] made it possible to develop the uniform approach to the
of the three families of the Romanovski polynomials, with the current paper focusing exclusively on the
of the R-Jacobi polynomials.
In the limit point (LP) region the Dirichlet problem mentioned above unequivocally specifies the eigenfunctions square integrable with the weight function of the SLE under consideration. Our procedure also provides the prescription for constructing at least some of the EOP sequence existent in the limit circle (LC) region, where the corresponding Liouville potential has only continuously degenerate bound energy states (CDBESs).
We found that the X
m-Jacobi DPSs of both series J1 and J2 contain X
m-Jacobi OPSs, which are interrelated via the reflection of their argument accompanied by the interchange of the Jacobi indexes. For this reason Gómez-Ullate et al. [
15,
16] restricted their analysis to the properties of one of them (which happened to be the X
m-Jacobi OPS of series J2 in our terms), referring to the latter simply as ‘X
m-Jacobi polynomials’ and dropping the X
m-Jacobi OPS of series J1 from any future consideration. To be consistent with the commonly adopted terminology in [
15,
16], we will term polynomials from the X
m-Jacobi DPS of series J2 and the X
m-Jacobi OPS of series J2 as ‘X
m-Jacobi polynomials’ and ‘orthogonal X
m-Jacobi polynomials’ accordingly, by analogy with the conventional terms ‘Jacobi polynomials’ and ‘orthogonal (classical) Jacobi polynomials’. Skipping the epithet ‘orthogonal’ in [
15,
16] can be thus treated as a slang term, similar to the widespread practice of omitting it in references to orthogonal Jacobi polynomials.
On other hand, our analysis revealed that the infinitely many finite EOP sequences of type and limitedly many the finite EOP sequences of type (constructed using the PFSs near the origin and at infinity respectively) belong to the Xm-Jacobi DPS of series J1 and accordingly J2, which makes it necessary to analyze both X-DPSs in parallel.
In particular, as discussed in
Section 4.2 below, the X
m-Jacobi DPS of series J2 contains the X
m-Jacobi OPS (in terms of [
15,
16]), the finite EOP sequence orthogonal on the negative interval (−∞, −1), and another finite EOP sequence orthogonal on the positive interval (1, ∞). The latter EOP sequence of type
is composed of the
of the R-Jacobi polynomials, which represent the polynomial components of the eigenfunctions [
10,
11,
12] of the rationally extended
h-PT potential [
11,
12].
Regretfully, both our works [
8] and [
24] overlooked the important modification in the definition of the ‘X
m-Jacobi polynomials’ by Yadav et al. in [
11]. Namely the indexes α and β appearing in the expression for X
m-Jacobi polynomials’ in [
10] were defined independently of the degree of the seed Jacobi polynomials and thereby (as illuminated in
Section 4.2 below) made the mentioned expression fully consistent with our definition of the X
m-Jacobi DPS of series J2.
As for the Xm-Jacobi DPS of series J1, it contains the OPS (‘Xm-Jacobi OPS of series J1 in our terms), the finite EOP sequence orthogonal on the negative interval (−∞, −1), and another finite EOP sequence (this time of type ) orthogonal on the positive interval (1, ∞). Obviously, the finite EOP sequence orthogonal on the negative interval (−∞, −1) can be obtained from the EOP sequence of type by the reflection of the argument, followed by the interchange of the indexes; i.e., the two finite EOP sequences in question are interrelated in exactly the same way as the DPSs of series J1 and J2 to which they belong.
The third EOP sequence of type
generated using the PFSs near origin (similarly to EOP sequence of type
) constitutes the orthogonal subset of the X
m-Jacobi DPS of series W composed of the Wronskians of two Jacobi polynomials with common pairs of the indexes. As it has been already pointed to by us in [
8], the Wronskian transforms of the R-Jacobi polynomials have been already brought to light in the cited article [
9] by Grandati. However, he did not realized that the constructed polynomial Wronskians are X-orthogonal. Indeed, the rational realization of the
h-PT potential (using the variable
cosh 2x) belongs to group A in Odake and Sasaki’s classification scheme [
22] of the rational translationally shape-invariant (RTSI) potentials (contrary to the other RTSI potentials discussed in [
9], which all belong to Group B). As a result, any admissible RDT of the R-Jacobi polynomials results in a finite EOP sequence.
Ironically the most obvious rational extension of the h-PT potential, using the TS with the polynomial component represented by the classical Jacobi polynomials and therefore necessarily nodeless on the interval (1, ∞), has been never discussed in the literature (apart from our works). The very remarkable feature of the finite EOP sequences of type
discovered in [
24] is that they can be arranged into the rectilinear polynomial matrix with a finite number of rows and an infinite number of columns composed of the X-Jacobi orthogonal polynomial system (X-Jacobi OPS) and
of the R-Jacobi polynomials.
Finally, the EOP sequences of type
belong to the X
m-Jacobi DPS of series D. We introduced this label in [
24] to stress that the DPS in question is composed of the so-called [
48] ‘polynomial determinants’ (PDs). However, after realizing [
8] that this X-DPS contains the X
m-Jacobi OPS of series J3 discovered by Grandati and Bérard [
50] we switched to the term ‘X
m-Jacobi DPS of series J3′, i.e., the X
m-Jacobi OPSs of series j1, j2, and J3 constitute infinite orthogonal subsets of the X
m-Jacobi DPSs of series J1, J2, and J3 respectively.
To summarize, let us note that, compared with the PSLPs roughed out in [
15,
18,
33], the DBCs at the ends of the quantization interval [−1,+1] cover only the rational Rudjak-Zakhariev transforms (
) of the classical Jacobi polynomials with positive indexes as it has been already demonstrated in [
29] for m = 1. On the other hand, our technique is sufficient to find all the rational Liouville potentials solvable in terms of either infinite and finite X
m-Jacobi EOP sequences, making it possible to construct the new finite EOP sequences.
One of the most important achievements of this paper (in addition to the systematic description and more precise summary of the earlier results spread between the three preprints [
1,
8,
24]) is the representation of the X
m-Jacobi DPSs of series J1, J2, and J3 (and therefore the corresponding X
m-Jacobi OPSs) via the ‘pseudo-Wronskian polynomials’ (
) of two Jacobi polynomials [
51]. As outlined in
Section 6, this fresh development opens a promising new direction in the theory of both infinite and finite EOP sequences composed of ‘simple’
of several seed Jacobi polynomials with common indexes and a single either classical Jacobi or R-Jacobi polynomial.
3. Use of RZTs for Constructing Xm-Jacobi DPSs
The very important new element of our formalism is the use of RZTs for constructing new RCSLEs which have four infinite sequences of q-RSs. As demonstrated in
Section 4.2 below, the polynomial components of these q-RSs obey the Bochner-type ODEs and therefore form four X-Jacobi DPSs.
Let us consider the RRZT with the TF
where
stands for the monomial product
with m simple zeros
, or, in other words, iff the TF has the following rational logarithmic derivative:
By definition of the RZT (see
Appendix A for details), the quasi-rational function
is the solution of the transformed CSLE
at the energy (25), with the density function defined via (7) for the interval (−1, +1), or at the energy
if the density function is defined via (2) for the interval (1, ∞). Each of the nodeless PFSs below the lowest eigenvalue can be used as the TF to generate the
of the CSLE (1).
Let us now take advantage of the fact that JRef CSL (1) is TFI [
72], namely, that quasi-rational function
is the solution of the JRef CSLE (1) with
replaced for |
, namely,
Let us also point to another remarkable feature of the JS q-RSs (23)—the Jacobi polynomials in the given sequence are multiplied by the same quasi-rational function and moreover the Jacobi indexes are independent of polynomial degrees [
1]. This is the direct consequence of the fact that this CSLE belongs to group A, which is also true for the corresponding Liouville potentials [
22].
Let us re-write both RCSLE (71) and
in the Riccati form:
and
respectively, where the symbolic expression
ld f[η] denotes the logarithmic derivative of the function f[η]. If the density function is identically equal to 1 then the derived expression turns into the standard supersymmetric representation of the quantum mechanical potential in terms of the superpotential represented by the logarithmic derivative of the TF
. In
Section 3 we will use a similar representation for the RefPFs of the
of the JRef CSLE (1) using the quasi-rational TFs (23).
Substituting the q-RS
into (75), coupled with (76), one finds
where [
1,
44]
It is worth mentioning that the derived expression (78) is valid on the both intervals (1, ∞) and (−1, +1). It can be also trivially extended to two other CSLEs of group A with the Liouville potentials represented by the isotonic oscillator and by Morse potential (assuming that the Schrödinger equation in the latter case is converted to the Bessel-reference CSLE [
26]). Making use of (25) and (76), coupled with
and
we can re-write (78) as
On other hand, comparing the PF (80) with the Quesne PF
(see, e.g., (4.11) in [
21], with
standing for
here), one finds
In our earlier works [
1,
44] we overlooked that the Quesne PF (81) is nothing more than
and therefore does not have simple poles. Taking into account that the trivial identity
coupled with the Jacobi equation (26), we can simplify the RefPF in question as follows:
One can easily verify that the derived expression turns into (27) in [
29] if we choose m = 1.
The quartet of the Liouville potentials for the RCSLE (71) defined in the four quadrant of the vector parameter
can be thus represented as follows:
or, using (82),
with
For future references, we made the above expression to be applicable to the Liouville transformations on both intervals (1, ∞) and (−1, +1), which constitutes the essence of the unified approach put forward in [
30] for m = 1.
Taking into account that
and
One can verify that each of the potentials (87) vanishes at infinity, as expected.
Keeping in mind that
for m = 1, we can re-write the corresponding Liouville potential as
If we set
by analogy with (20), and take into account that
we find that
and therefore
Substituting (98) into (94) for
then gives
The change of variable (15) on the interval (1, ∞) then turns (99), with
, into the potential function (9) in [
20], while the change of variable η =
sin x converts (99), with
, into (3.5) in [
71], with A and B replaced in both cases for *A and *B respectively.
Let us draw reader’s attention to the fact that the two generally distinct branches of the potential (88) with
collapse into the same Bagchi-Quesne-Roychoudhury (BQR) potential for m = 1, as expected from the rigorous analysis of the latter case in [
29]. The similar collapse takes place for the two other branches of this potential (
) but the resultant potential does not have discrete energy spectrum [
29] and therefore cannot be linked to any EOP sequence.
Coming back to the general case m ≥ 1, let us re-write the potential (88) for
as
While our expression (100) for the BQR potential (m = 1) fully agrees with (15) in [
11], we detected some discrepancies between (100), with
, and the corresponding expression for this potential in [
11,
12].
First, both the potential (19) in [
11] and the following expression (37) for the rationally-extended
t-PT (Scarf I) potential (after being converted by the change of variable η =
sin x (|η| < 1) to its rational form (100) above) lack the term associated with the second derivative of the Jacobi polynomial in the right-hand side of (100). Since this derivative vanishes for the first-degree polynomial, one cannot detect the missed term simply by setting m = 1 in the general formula. The mentioned term was also missed in (81) in [
12] or in the preceding expression (64) for the rationally-extended
t-PT (Scarf I) potential.
Disregarding the missed term, the rest of both cited expressions in [
12] match (100) for
, if we set
or, taking into account (97),
in agreement with the definition of these indexes in [
12]. It has been proven above that the potential (88) vanishes at infinity and therefore the function (81) in [
12] does not, keeping in mind that the omitted term tends to −m(m − 1) as η → ∞. (The reason for introducing the second pair of the potentials (82) and (83) following (81) in [
12] is unclear to me).
As already pointed to in [
29], the parameter swap
does not result in the new potential. The potentials (19) in [
11] and (81) in [
12] differ only by notation, with the constraint
reversed for
. This also true for he two potentials (12) and (15) listed in [
11] for m = 1.
5. Isospectral Triplet of RCSLEs Solved via of R-Jacobi Polynomials
Starting from this point, we discuss only the admissible RRZTs using the TFs
with no nodes in the interval (1, ∞) for the specified ranges of the parameters
. In this Section we will consider only the TFs with the vector
lying in the first three quadrants. As demonstrated in [
1] using the Klein formula [
49], the admissible TFs also exist for certain segments of the vector
in the fourth quadrant, but this family of the finite EOP sequences lies beyond the scope of this paper.
The innovatory part of this subsection is based on our extensive finding [
48] that the DBC imposed on solutions of the
p-SLE unambiguously selects the PFS near the given endpoint. After converting the RCSLE (103) to its prime forms on the intervals (−1, +1) and (1, ∞), we then prove that the
of the PFSs of the JRef CSLE (1) are themselves the PFSs of the transformed RCSLE (103).
To formulate the SLP on the interval (1, ∞) we (by analogy with the analysis) presented in
Section 2.3 for the JRef CSLE) first convert the RCSLE (103) to its prime form
and then solve it under the DBCs:
The zero-energy free term in the SLE (159) is related to the RefPF of the RCSLE via the generic elementary Formula (31), i.e.,
Taking into account that
we confirm that the ChExps of the Frobenius solutions for the pole at +1 have the same absolute value, while differing by their sign. The ExpDiff for the pole of the RCSLE (103) at infinity turned out to be energy-dependent. By analogy with (34) we find that
We thus assert that the ChExps of the Frobenius solutions for the pole at ∞ are real only at negative energies and have in this case the same non-zero absolute value
for
or
for
, while differing by their sign.
Combining (163) with the similar limit
for the pole at +1, we conclude that the PFSs of the
p-SLE (159) near both singular endpoints are unambiguously determined by the DBCs.
By applying the RRZT with the admissible TF
to the eigenfunction
of the JRef CSLE (1) with 0 ≤ j ≤ j
max and then converting the resultant function (124) to its prime form, we come to the following expression for the
of the (j + 1)-th eigenfunction of the
p-SLE (30) solved under the DBCs (35):
with 0 ≤ j ≤ j
max and
The q-RS (165) obeys the DBCs
iff
and
We thus assert that the given q-RS represents an eigenfunction of the
p-SLE (159) solved under the DBCs provided both conditions (168) and (169) hold We will analyze these conditions on the case-by-case basis in
Section 5.1,
Section 5.2 and
Section 5.3 for the vector
lying in the first, second and third quadrant accordingly. We delay the discussion of the fourth quadrant
for a separate study.
It has been proven in [
56] that the eigenfunction of the generic SLE solved under the DBCs must be mutually orthogonal with the weight (32) on the infinite interval in question:
Consequently, the polynomial components of the quasi-rational eigenfunctions (165) must be mutually orthogonal with the m-dependent weight
namely,
Let us now demonstrate the power of the developed formalism by proving that the q-RSs (165) form the complete set of the eigenfunctions for the formulated Dirichlet problem.
Theorem 2. The Dirichlet problem for the p-SLE (159) defined on the interval (1, ∞) with
does not have any solutions other than the eigenfunctions (165), assuming that the constraints (168) and (169) hold and that the pole of the RCSLE (103) at +1 lies within the LP region.
Proof. Suppose that the given Dirichlet problem on the interval (1, ∞) has a solution
at an energy
The RRZT of the RCSLE (103) with the TF
converts the extraneous eigenfunction into the following q-RS of the
p-SLE (30):
i.e.,
keeping in mind that
Let us now recall that the extraneous eigenfunction (like any other) must decay as
in the limit
. Examination of the sum (176) then shows that it vanishes at the lower end of the interval (1, ∞) as far as the pole the RCSLE (103) at +1 lies in the LP region (
).
Furthermore, taking into account that
and
one can verify that the solution in question also vanishes in the limit η → ∞ and therefore represents an eigenfunction of the
p-SLE (30) with an eigenvalue differing from any of the eigenvalues (40) with j ≤ j
max. This result contradicts to the fact that the cited energies represent the complete discrete energy spectrum of the given Dirichlet problem. We thus confirmed that the
p-SLE (159) solved under the DBCs (160) may not have eigenvalues other than (40) in the LP range of the parameter
. □
The direct consequence of the proven theorem is that the Wronskians (116) or (119) forming the eigenfunctions (165) have exactly j + 1 real zeros larger than 1.
In the following subsections we discuss three distinguished finite EOP subsets of the Xm-DPSs of series J1, W, and J2. While each of the X-DPSs of series J1 and J2 also contain Xm-OPSs, the Xm-DPS of series W may not comprise any, because the RSLP on the interval (−1, +1) has infinitely many eigenfunctions. On other hand, the RSLP on the infinite interval has only a finite number of the eigenfunctions and as a result the sequence of Jacobi polynomials with the indexes in the second quadrant contains an infinite subsequence of Jacobi polynomials with no real zeros larger than 1.
5.1. Infinitely Many EOP Sequences of Series
Let us start from the vector
lying in the first quadrant:
i.e., the ExpDiff for the pole of the CSLE (103) necessarily lies within the LP range. As the direct consequence of (183), the seed polynomial in the denominator of the weight (169) turns into the classical Jacobi polynomial with positive indexes and therefore all the poles of the RCSLE (103) are located in the closed interval [−1,+1]. Ironically, this most obvious case has never been discussed in the literature and our study of these EOP sequences in [
24] did not receive a proper response.
The finite EOP sequence under consideration exists in any X
m-Jacobi DPS of series J1 specified by the upper sign in (127), i.e.,
is allowed to take any positive value m. Bearing in mind (166), we can represent this X-orthogonal polynomial subset as
Setting z = −
,
, we come to the polynomials (72) in [
16]:
with the interchanged degrees m and j, i.e.,
We thus proved that the polynomial (183) converted to its monic form coincides with the monic (j + m)-th degree polynomial from the X
j-Jacobi OPS of series J1 [
24]. This implies that the polynomials (187) arranged into the infinite row (m = 1, 2, …) at fixed j are mutually orthogonal with the weight
(cf. (87) in [
16]). This brings us to the (j
max+ 1)
∞ rectilinear polynomial matrix mentioned in Introduction.
Making use of (183) and (184), one finds
so the q-RS (165) takes the form:
Keeping in mind that
we conclude that that the function
is the eigenfunction of the
p-SLE (30), satisfying by definition the DBC (36). (As expected, the absolute values of the power exponents
of η
coincide with halves of the ExpDiffs
for the poles of the RCSLE (103) at
1). The condition (169) also holds since
This confirms that the q-RSs (190) are the eigenfunctions of the
p-SLE (159) solved under the DBCs (160) and therefore their polynomial components are mutually orthogonal with the weight
Since all the prerequisites for Theorem 2 are valid, the analytical Formula (40) with j ≤ j
max determines all the eigenvalues of the
p-SLE (159) solved under the DBCs is exactly solvable.
From the perspective of the quantum mechanical applications, the corresponding isospectral Liouville potential (100) with 1 < < 2 constitutes the very special case, because its SUSY partner, h-PT potential with the positive parameter smaller than 1 (1/2 < < 3/2) has only the CDBESs. This implies that the discrete energy spectrum for the Liouville potential (100) with 1 < < 2 cannot be found using the conventional rules of the SUSY quantum mechanics.
5.2. Infinitely Many EOP Sequences of Series
Let us now consider the peculiar case when both vector
and
lie in the same (second) quadrant (
). The admissible TFs of this type exist only for finite EOP sequences. According to the inequality (65), the q-RS
is admissible for any
from the infinite sequence of the positive integers starting from
. Substituting (115) into (123) with
=
gives
As expected, the absolute values of the power exponents
of η
coincide with halves of the ExpDiffs
for the poles of the RCSLE (103) at
. Since
,
, and therefore
we confirm that the q-RS (195) necessarily vanishes at +1.
To confirm the DBCs (167) at infinity, first note that the PF in the right-hand side of (195) grows at large η as ηj−1. Taking into account that the q-RSs and have the same asymptotics at infinity, while the eigenfunction with j ≤ jmax satisfies the DBC at this endpoint, we conclude that the q-RS (195) vanishes at infinity, as required.
Since
, Theorem 2 again assures the exact solvability of the Dirichlet problem in question. Once more, the corresponding Liouville potential (100) with 1 <
< 2 and
is not covered by the conventional rules of the SUSY quantum mechanics. In Grandati’s notation [
9]:
so his analysis is applicable solely to the parameter range
On the contrary, the technique developed here made it possible to extend the discussion of the rationally-extended
h-PT potentials to the parameter range
The border case
requires a more cautious analysis.
I doubt that the extension of the domain definition for the parameter α to negative values (α > −
in (31) in [
9]) makes any sense in the quantum-mechanical studies of the
h-PT potential and its rational extensions. Though it is possible that formulating the appropriate PSLPs would allow one to construct finite EOP sequences of types
and
for −1 < α < 0 the q-RSs composed of these polynomials could be hardly applied in physics.
5.3. Finitely Many EOP Sequences of Series
By placing the vector
into the third quadrant, we finally come to the case which so far has attracted much more attention from the physicists [
10,
11,
12,
13]—the rationally extended
h-PT potential of type
constructed using the finite number of the nodeless TFs
under the constraint (63). In the sharp contrast with the two cases discussed above, the RRZTs in question decreases by 1 the ExpDiff for the pole at +1. As a result, the
p-SLE (159) can be analytically solved within the LC range of the ExpDiff
. However, we were unable to prove that the SLP is exactly solvable in this case. The most serious setback of this failure is that our formalism cannot anymore guarantee that the
(119) forming the eigenfunctions (165) with
and
have exactly j + 1 real zeros larger than 1 if
< 1.
Keeping in mind that
and
one finds
As expected, the absolute values of the power exponents
1/
2 of η
in (165) thus coincide with halves of the ExpDiffs for the poles of the RCSLE (103) at
, provided that the latter lay in the LP region.
Since
the q-RS (165) obeys the DBC at +1 iff the pole of the CSLE (1) at this end is LP.
Furthermore,
so the second prerequisite (169) for Theorem 2 also holds.
It is interesting to mention that the basic PFS of the given type (
= 0) have been used in [
84] to study the DTs between the LP and LC regions of the h-PT potential. Keeping in mind that
in our terms, we find the close connection between the analysis presented by Gangopadhyaya et al. [
84] and the current discussion. Namely, the cited authors pick up the
of each eigenfunction (well-defined in the LP region) as the special representative of the CDBESs for the eigenvalue. As more recently proven in [
48], the
of the eigenfunction of the p-SLE with the LP singularities at the ends necessarily obeys the DBCs under consideration. In other words, the cited authors select the bound energy states described by the PFSs at the origin, with a clear resemblance to our prescriptions.
6. Discussion
The paper is based on the three core notions advanced by the author in the aforementioned publications. One of them is the scrupulous analysis of the RCSLEs obtained by the RRZTs of the JRef CSLE (1). As pointed to in Introduction, any RRZT of the RCSLE is directly related to a DT of the corresponding Liouville potential and as a result this part of our studies lays the rigorous foundation for the SUSY theory of the quantum-mechanical potentials solvable by polynomials.
Another important element of our approach is the concept of the ‘prime’ SLE (
p-SLE) chosen in such a way that the two ChExps for the poles at the endpoints differ only by sign. As a result, the energy spectrum of the given Sturm-Liouville problem can be obtained by solving the
p-SLE under the DBCs. This in turn allows one to take advantage of the rigorous theorems proven in [
56] for eigenfunctions of the generic SLE solved under the DBCs.
Finally, we put forward the concept of the X-Jacobi DPSs formed by polynomial solutions of the Bochner-type ODEs. Since the X-DPSs either do not start from a constant or lack a first-degree polynomial, they do not satisfy the Bochner theorem [
36], as originally noticed by Gómez-Ullate et al. [
33] in the context of the discovered by them X
1-Jacobi OPS. In general, each X-Jacobi OPS belongs to one of the X-Jacobi DPSs. If the given X-Jacobi DPS does contain a X-Jacobi OPS, then we say that they belong to the same series. This is why we [
8] started referring to the X
m-Jacobi DPS of series D [
24] as being of series J3, after becoming aware of Grandati and Bérard’s [
50] discovery of the X
m-Jacobi OPS of series J3 for even m. (We still continue to refer the X
m-Jacobi DPS of series D as being of series J3 for odd m even though the latter do not contain any OPSs). (As proven in [
28], the X
1-Jacobi DPS of series D does not contain any finite EOP sequences either).
In addition to the two finite EOP sequences identified in [
9] and in [
10,
11] accordingly, our original research [
24] discovered the (undeniably new to our knowledge) finite EOP sequences of type
, which were constructed using the q-RTFs composed of the classical Jacobi polynomials. Since all the zeros of the latter polynomials lie between −1 and +1, the q-RTFs of this kind (representing the PFSs at the origin) do not have nodes within the quantization interval (+1, ∞). The very remarkable feature of the finite EOP sequences constructed in such a way [
24] is that they can be arranged into the rectilinear polynomial matrix with a finite number of rows and an infinite number of columns representing the X-Jacobi OPSs of series J1 and
of the R-Jacobi polynomials accordingly.
Compared with the cited preprints [
1,
8,
24], the brand new element of the current analysis, is the proof that the X
m-Jacobi DPSs of series J1 and J2 are composed of the
of two Jacobi polynomials [
51], and therefore this is also true for their infinite and finite X-orthogonal subsets. In particular the mentioned rectilinear polynomial matrix is formed by the
of the classical Jacobi and R-Jacobi polynomials, respectively.
The simple
representation of the X-orthogonal subsets of the X
m-Jacobi DPS of series J1 has very interesting far-reaching implications, making it possible to obtain analytical expressions for both infinite and finite
of the classical Jacobi and R-Jacobi polynomials using the PFSs of the same type
as seed functions. In [
25] we have discussed the RCSLEs obtained from the JRef CSLE (1) using the seed Jacobi polynomials with the same indexes. Since the RDCTs using the seed functions of types
and
are specified by same series of the Maya diagrams [
88], any RCSLE using an arbitrary combination of these seed functions can be alternatively obtained by considering only infinitely many combinations
of the PFSs of the same type
[
51,
72,
89]. In particular, the Liouville potentials constructed by means of the TFs
discussed above can be alternatively obtained by means of the
th-order RDCT with the seed functions
[
72,
89].
One can easily verify that the quasi-rational functions defined via the relations
where
, constitute polynomials of degree
, where [
90]
The explicit expression for the ‘simple’
polynomials (201) can be easily obtained using (92) in [
51] and will be examined in a separate paper. As expected, the ‘simple’
polynomials (201) turn into the Xm-Jacobi DPSs of series J1 for p = 1.
It has been proven in [
25] that the ExpDiffs for the poles of the RCSLE
at
are given by the simple formula
The corresponding eigenfunctions have the form [
91]:
where
Representing (206) as
we find that the absolute values of the power exponents of η ± 1 coincide with halves of the ExpDiffs (204) and therefore the simple
polynomials (194) form a X-Jacobi DPS, which will be referred to by us as being of series
. A similar X-Jacobi DPS of series
simply constitutes another representation of one of the X-DPSs mentioned above and might be dropped though it would come with a catch: the dropped net of the X-Jacobi OPSs (
,
) starts from the infinite manifold of the X
m-Jacobi OPSs in the conventional sense [
15,
16,
17].
If we choose and restrict j from above by the constraint j ≤ jmax, then the resultant of the R-Jacobi polynomials form the finite EOP sequences of series , where are the the admissible sets of the JS solutions assuring that the corresponding Jacobi polynomial Wronskians (JPWs) do not have nodes within the given quantization interval for the specified ranges of the parameters . We thus come to the very broad brand families of both infinite and finite EOP sequences, which will be examined in detail in a separate publication.
One can also combine the nodeless PFSs of type
with the juxtaposed pairs of the eigenfunctions (type
) to construct the RCSLEs quantized by the finite EOP sequences composed of the Wronskian transforms of the R-Jacobi polynomials, similar to the finite EOP sequences formed by the Wronskian transforms of the R-Bessel and R-Jacobi polynomials [
26,
28].
The general case using the seed functions of all the four types , , , and represent a much more challenging problem.