On the cubic Shimura lift for \(PGL_3\) (Q5957373)

This paper introduces a conjecture on irreducible, automorphic, cuspidal representations \(\pi\) on \(PGL_3\). Namely, the period \[ \int_{PGL_3(F)\setminus PGL_3(\mathbb{A})} \varphi(g) \theta( \text{Ad}(g)) dg\tag{1} \] is not identically zero if and only if \(\pi\) is in the image of the cubic Shimura correspondence. Here \(\varphi\) is a cusp form in the space of \(\pi\), \(\theta\) is an automorphic form in the space of the theta representation \(\theta_{SO_8}\) of (split) \(SO_8\), and Ad is the adjoint representation of \(PGL_3\) inside \(SO_8\). Finally, \(F\) is a global field and \(\mathbb{A}\) is its adele ring. The authors give two types of evidence for the validity of the conjecture. The first is a consideration of an analogous question over a finite field \(\mathbb{F}_q\). For this they take an irreducible representation \(\pi\) of \(PGL_3 (\mathbb{F}_q)\), assumed to be in general position, and consider the inner product of its character with that of the minimal (nontrivial) representation \(\theta\) of \(SO_8 (\mathbb{F}_q)\), composed with the adjoint representation. Using the Deligne-Lusztig parametrization and character tables, they show (assuming that \(q\cong 1\pmod 6\)) that if \(\pi\) is in the principal series, the inner product is nontrivial if and only if the character appearing in the Deligne-Lusztig parametrization is a cube (and then this inner product equals 3). If \(\pi\) is cuspidal, then the inner product is nontrivial (it equals 1). Recall again that this is for \(\pi\) in general position. A similar related property, which is a known theorem [see \textit{D. Ginzburg, S. Rallis} and \textit{D. Soudry}, Am. J. Math. 119, No. 2, 251-335 (1997; Zbl 0877.11031); \textit{Z. Mao} and \textit{S. Rallis}, Isr. J. Math. 113, 95-124 (1999; Zbl 0954.11017)] is that for an irreducible, automorphic, cuspidal representation \(\pi\) of \(SL_2(\mathbb{A})\), \(\pi\) is in the image of the cubic Shimura correspondence, if and only if the following period \[ \int_{SL_2(F)\setminus SL_2(\mathbb{A})} \varphi(g) \theta(s^2(g)) dg\tag{2} \] is not identically zero. The authors verify the similar analogous statement over a finite field, for a representation in general position. The second type of evidence that the authors bring is over a global field, where they examine the conjecture not for a cuspidal representation \(\pi\), but rather for Eisenstein series, coming off the \((2,1)\) parabolic subgroup and a cuspidal representation \(\tau\) on \(GL_2\). They compute the period (1), and after unfolding it, they get two terms, following a Fourier expansion of \(\theta_{SO_8}\) along the highest root. The first term is the constant coefficient in the Fourier expansion, and the second term contains as an inner integral the period (2). This then implies that \(\tau\) is in the image of the Shimura cubic lift, and hence so is the corresponding Eisenstein series.