Plancherel formula for \(\mathrm{GL}_n(F)\setminus \mathrm{GL}_n(E)\) and applications to the Ichino-Ikeda and formal degree conjectures for unitary groups (Q2041859)

Let \(E|F\) be a quadratic extension of local fields of characteristic zero, and let \(Y_n\) be the homogeneous space \(\mathrm{GL}_n(F) \backslash \mathrm{GL}_n(E)\). The main goal of the long paper under review is to obtain an explicit decomposition of \(L^2(Y_n)\) as a unitary representation of \(\mathrm{GL}_n(E)\). Such a decomposition is called a Plancherel formula in harmonic analysis. Fix an additive character \(\psi'\) of \(F\) to define Haar measures and an invariant measure on \(Y_n\). Denote by \(\mathrm{Temp}(\mathrm{GL}_n(E))\) the set of equivalence classes of tempered irreducible representations of \(\mathrm{GL}_n(E)\), and denote by \(\mathrm{Temp}(U(n))/\mathrm{st}\) the set of tempered \(L\)-parameters (i.e.\ tempered \(L\)-packets) of the quasisplit unitary group \(U(n)\) of rank \(n\) relative to \(E|F\). One endows the space \(\mathrm{Temp}(U(n))/\mathrm{st}\) with a measure that basically comes from unramified twists of parabolic inducing data. For each \(\sigma\) in \(\mathrm{Temp}(U(n))/\mathrm{st}\), let \(\mathrm{BC}_n(\sigma)\) denote its stable (resp.\ unstable) base change to \(\mathrm{GL}_n(E)\) if \(n\) is odd (resp.\ even). In its abstract and weak form (Theorem 1), the decomposition reads \[ L^2(Y_n) \simeq \int^{\oplus}_{\mathrm{Temp}(U(n))/\mathrm{st}} \mathcal{H}_\sigma d\sigma \] where \(\sigma \mapsto \mathcal{H}_\sigma\) is a measurable field of unitary irreducible representations of \(\mathrm{GL}_n(E)\); in fact, \(\mathcal{H}_\sigma \simeq \mathrm{BC}_n(\sigma)\). This is in accordance with a general prediction of Sakellaridis and Venkatesh. In order to obtain the explicit Plancherel formula, one follows the idea of Bernstein to consider the Schwartz spaces \(\mathcal{S}(\cdot)\), which have been defined in general, and seek an explicit spectral decomposition of the \(L^2\)-inner product \((\varphi_1, \varphi_2)_{Y_n}\) for \(\varphi_1, \varphi_2 \in \mathcal{S}(Y_n)\). The answer is given as follows. Fix an additive character \(\psi\) of \(E\) with \(\psi|_F = 1\), giving rise to the standard Whittaker datum \((N_n(E), \psi_n)\) for \(\mathrm{GL}_n(E)\). For all \(\pi \in \mathrm{Temp}(\mathrm{GL}_n(E))\), denote its Whittaker model as \(\mathcal{W}(\pi, \psi_n)\) and define an Hermitian form \((\cdot, \cdot)_{Y_n, \pi} \geq 0\) on \(\mathcal{S}(\mathrm{GL}_n(E))\) by \[ (f_1, f_2)_{Y_n, \pi} = \sum_{W} \beta(\pi(f_1) W) \overline{\beta(\pi(f_2)W)} \] where \(W\) ranges over an orthonormal basis of \(\mathcal{W}(\pi, \psi_n)\), with the scalar product induced from the Kirillov model (i.e. from \(L^2(N_n(E) \backslash P_n(E))\)), and \(\beta(W) = \int_{N_n(F) \backslash P_n(F)} W\), which turns out to converge. One shows that \((\cdot, \cdot)_{Y_n, \mathrm{BC}_n(\sigma)}\) factors through \(\mathcal{S}(Y_n)\). The explicit Plancherel formula (Theorem 2) asserts that \[ (\varphi_1, \varphi_2)_{Y_n} = \int_{\mathrm{Temp}(U(n))/\mathrm{st} } (\varphi_1, \varphi_2)_{Y_n, \mathrm{BC}_n(\sigma)} \cdot \frac{|\gamma^*(0, \sigma \mathrm{Ad}, \psi')|}{|\mathcal{S}_\sigma|} d\sigma. \] Here \(\mathcal{S}_\sigma\) is the usual component group in local Langlands correspondence, and \(\gamma^*(0, \cdots)\) is the value at \(s=0\) of \(\zeta_F(s)^{n_\sigma} \gamma(s, \cdots)\) where \(n_\sigma\) is the order of vanishing of \(\gamma(s, \cdots)\). The overall strategy goes as follows. We may suppose that \(f_i \mapsto \varphi_i\) where \(f_i \in \mathcal{S}(\mathrm{GL}_n(E))\). Then \((\varphi_1, \varphi_2)_{Y_n} = \int_{\mathrm{GL}_n(F)} \overline{f_2} \star f_1^\vee\). We are thus led to the spectral decomposition of the form \(\int_{\mathrm{GL}_n(F)}\) on \(\mathcal{S}(\mathrm{GL}_n(E))\). The subsequent manipulations are arguably similar to the global ones in the factorization of Flicker--Rallis periods: a local unfolding leads to an integral \[ \int_{N_n(F) \backslash P_n(F)} dp \int_{N_n(F) \backslash \mathrm{GL}_n(F)} dh W_f(p, h) \] where \(W_f\) is a Whittaker function associated with \(f\), defined precisely in Section 2.14. Due to convergence issues, one cannot simply use the available spectral decomposition for \(W_f\). The workaround is to express the inner integral as the residue at \(s=0\) of some local Rankin-Selberg zeta integral for a test function \(\phi \in \mathcal{S}(F^n)\), perform the spectral decomposition in \(\Re(s) > 0\), and then use a local functional equation in terms of Asai \(\gamma\)-factors to reduce to a residue computabke via the available explicit formulas for Plancherel measures for \(\mathrm{GL}_n\). The author gives three main applications. The first (Theorem 3) describes the Planchrel density for unitary groups in terms of adjoint \(\gamma\)-factors. The case \(F=\mathbb{R}\) is known by Harish-Chandra, whilst the special case of discrete series reduces to the formal degree conjecture of Hiraga-Ichino-Ikeda. The second application (Theorem 4) establishes the matching of relative characters in the local unitary Gan-Gross-Prasad conjecture (codimension-one case). Compared to the earlier works, a remarkable feature is that the arguments here work uniformly for all \(F\). This confirms a conjecture of Wei Zhang. The final application (Theorem 5) is that the Ichino-Ikeda conjecture in the local unitary Gan-Gross-Prasad case (codimension-one) holds when the representation \(\pi\) of \(G(\mathbb{A})\) is everywhere tempered and \(\mathrm{BC}(\pi)\) is supercuspidal at some place. These proofs are intertwined in a comparison of local Jacquet-Rallis trace formulas and their limits as \(\gamma \to 1\), and this is where the explicit Plancherel formula is applied.