1: \begin{abstract}

2:     We study the convergence of the empirical spectral distribution of $\bm A \bm B \bm A$ for $N \times N$ orthogonal projection matrices $\bm A$ and $\bm B$, where $\frac{1}{N}\Tr(\bm A)$ and $\frac{1}{N}\Tr(\bm B)$ converge as $N \to \infty$, to Wachter's MANOVA law.

3:     Using free probability, we show mild sufficient conditions for convergence in moments and in probability, and use this to prove a conjecture of Haikin, Zamir, and Gavish (2017) on random subsets of unit-norm tight frames.

4:     This result generalizes previous ones of Farrell (2011) and Magsino, Mixon, and Parshall (2021).

5:     We also derive an explicit recursion for the difference between the empirical moments $\frac{1}{N}\Tr((\bm A \bm B \bm A)^k)$ and the limiting MANOVA moments, and use this to prove a sufficient condition for convergence in probability of the largest eigenvalue of $\bm A \bm B \bm A$ to the right edge of the support of the limiting law in the special case where that law belongs to the Kesten-McKay family.

6:     As an application, we give a new proof of convergence in probability of the largest eigenvalue when $\bm B$ is unitarily invariant; equivalently, this determines the limiting operator norm of a rectangular submatrix of size $\frac{1}{2}N \times \alpha N$ of a Haar-distributed $N \times N$ unitary matrix for any $\alpha \in (0, 1)$.

7:     Unlike previous proofs, we use only moment calculations and non-asymptotic bounds on the unitary Weingarten function, which we believe should pave the way to analyzing the largest eigenvalue for products of random projections having other distributions.

8: \end{abstract}

9: