The Brown measure of the free multiplicative Brownian motion
Abstract

The free multiplicative Brownian motion$$b_{t}$$${b}_{t}$is the large-Nlimit of the Brownian motion on$$\mathsf {GL}(N;\mathbb {C}),$$$\mathrm{GL}\left(N;C\right),$in the sense of$$*$$$\ast$-distributions. The natural candidate for the large-Nlimit of the empirical distribution of eigenvalues is thus the Brown measure of$$b_{t}$$${b}_{t}$. In previous work, the second and third authors showed that this Brown measure is supported in the closure of a region$$\Sigma _{t}$$${\Sigma }_{t}$that appeared in the work of Biane. In the present paper, we compute the Brown measure completely. It has a continuous density$$W_{t}$$${W}_{t}$on$$\overline{\Sigma }_{t},$$${\overline{\Sigma }}_{t},$which is strictly positive and real analytic on$$\Sigma _{t}$$${\Sigma }_{t}$. This density has a simple form in polar coordinates:\begin{aligned} W_{t}(r,\theta )=\frac{1}{r^{2}}w_{t}(\theta ), \end{aligned}$\begin{array}{c}{W}_{t}\left(r,\theta \right)=\frac{1}{{r}^{2}}{w}_{t}\left(\theta \right),\end{array}$where$$w_{t}$$${w}_{t}$is an analytic function determined by the geometry of the region$$\Sigma _{t}$$${\Sigma }_{t}$. We show also that the spectral measure of free unitary Brownian motion$$u_{t}$$${u}_{t}$is a “shadow” of the Brown measure of$$b_{t}$$${b}_{t}$, precisely mirroring the relationship between the circular and semicircular laws. We develop several new methods, based on stochastic differential equations and PDE, to prove these results.

more » « less
Award ID(s):
NSF-PAR ID:
10372851
Author(s) / Creator(s):
; ;
Publisher / Repository:
Date Published:
Journal Name:
Probability Theory and Related Fields
Volume:
184
Issue:
1-2
ISSN:
0178-8051
Page Range / eLocation ID:
p. 209-273
Format(s):
Medium: X
National Science Foundation
##### More Like this
1. Abstract

Ramanujan’s partition congruences modulo$$\ell \in \{5, 7, 11\}$$$\ell \in \left\{5,7,11\right\}$assert that\begin{aligned} p(\ell n+\delta _{\ell })\equiv 0\pmod {\ell }, \end{aligned}$\begin{array}{c}p\left(\ell n+{\delta }_{\ell }\right)\equiv 0\phantom{\rule{0ex}{0ex}}\left(mod\phantom{\rule{0ex}{0ex}}\ell \right),\end{array}$where$$0<\delta _{\ell }<\ell$$$0<{\delta }_{\ell }<\ell$satisfies$$24\delta _{\ell }\equiv 1\pmod {\ell }.$$$24{\delta }_{\ell }\equiv 1\phantom{\rule{0ex}{0ex}}\left(mod\phantom{\rule{0ex}{0ex}}\ell \right).$By proving Subbarao’s conjecture, Radu showed that there are no such congruences when it comes to parity. There are infinitely many odd (resp. even) partition numbers in every arithmetic progression. For primes$$\ell \ge 5,$$$\ell \ge 5,$we give a new proof of the conclusion that there are infinitely manymfor which$$p(\ell m+\delta _{\ell })$$$p\left(\ell m+{\delta }_{\ell }\right)$is odd. This proof uses a generalization, due to the second author and Ramsey, of a result of Mazur in his classic paper on the Eisenstein ideal. We also refine a classical criterion of Sturm for modular form congruences, which allows us to show that the smallest suchmsatisfies$$m<(\ell ^2-1)/24,$$$m<\left({\ell }^{2}-1\right)/24,$representing a significant improvement to the previous bound.

more » « less
2. Abstract

The numerical analysis of stochastic parabolic partial differential equations of the form\begin{aligned} du + A(u)\, dt = f \,dt + g \, dW, \end{aligned}$\begin{array}{c}du+A\left(u\right)\phantom{\rule{0ex}{0ex}}dt=f\phantom{\rule{0ex}{0ex}}dt+g\phantom{\rule{0ex}{0ex}}dW,\end{array}$is surveyed, whereAis a nonlinear partial operator andWa Brownian motion. This manuscript unifies much of the theory developed over the last decade into a cohesive framework which integrates techniques for the approximation of deterministic partial differential equations with methods for the approximation of stochastic ordinary differential equations. The manuscript is intended to be accessible to audiences versed in either of these disciplines, and examples are presented to illustrate the applicability of the theory.

more » « less
3. Abstract

Motivated by the Rudnick-Sarnak theorem we study limiting distribution of smoothed local correlations of the form\begin{aligned} \sum _{j_1, j_2, \ldots , j_n} f(N(\theta _{j_2}-\theta _{j_1}), N(\theta _{j_3}-\theta _{j_1}), \ldots , N(\theta _{j_n}-\theta _{j_1})) \end{aligned}$\begin{array}{c}\sum _{{j}_{1},{j}_{2},\dots ,{j}_{n}}f\left(N\left({\theta }_{{j}_{2}}-{\theta }_{{j}_{1}}\right),N\left({\theta }_{{j}_{3}}-{\theta }_{{j}_{1}}\right),\dots ,N\left({\theta }_{{j}_{n}}-{\theta }_{{j}_{1}}\right)\right)\end{array}$for the Circular United Ensemble of random matrices for sufficiently smooth test functions.

more » « less
4. Abstract

Let$$(h_I)$$$\left({h}_{I}\right)$denote the standard Haar system on [0, 1], indexed by$$I\in \mathcal {D}$$$I\in D$, the set of dyadic intervals and$$h_I\otimes h_J$$${h}_{I}\otimes {h}_{J}$denote the tensor product$$(s,t)\mapsto h_I(s) h_J(t)$$$\left(s,t\right)↦{h}_{I}\left(s\right){h}_{J}\left(t\right)$,$$I,J\in \mathcal {D}$$$I,J\in D$. We consider a class of two-parameter function spaces which are completions of the linear span$$\mathcal {V}(\delta ^2)$$$V\left({\delta }^{2}\right)$of$$h_I\otimes h_J$$${h}_{I}\otimes {h}_{J}$,$$I,J\in \mathcal {D}$$$I,J\in D$. This class contains all the spaces of the formX(Y), whereXandYare either the Lebesgue spaces$$L^p[0,1]$$${L}^{p}\left[0,1\right]$or the Hardy spaces$$H^p[0,1]$$${H}^{p}\left[0,1\right]$,$$1\le p < \infty$$$1\le p<\infty$. We say that$$D:X(Y)\rightarrow X(Y)$$$D:X\left(Y\right)\to X\left(Y\right)$is a Haar multiplier if$$D(h_I\otimes h_J) = d_{I,J} h_I\otimes h_J$$$D\left({h}_{I}\otimes {h}_{J}\right)={d}_{I,J}{h}_{I}\otimes {h}_{J}$, where$$d_{I,J}\in \mathbb {R}$$${d}_{I,J}\in R$, and ask which more elementary operators factor throughD. A decisive role is played by theCapon projection$$\mathcal {C}:\mathcal {V}(\delta ^2)\rightarrow \mathcal {V}(\delta ^2)$$$C:V\left({\delta }^{2}\right)\to V\left({\delta }^{2}\right)$given by$$\mathcal {C} h_I\otimes h_J = h_I\otimes h_J$$$C{h}_{I}\otimes {h}_{J}={h}_{I}\otimes {h}_{J}$if$$|I|\le |J|$$$|I|\le |J|$, and$$\mathcal {C} h_I\otimes h_J = 0$$$C{h}_{I}\otimes {h}_{J}=0$if$$|I| > |J|$$$|I|>|J|$, as our main result highlights: Given any bounded Haar multiplier$$D:X(Y)\rightarrow X(Y)$$$D:X\left(Y\right)\to X\left(Y\right)$, there exist$$\lambda ,\mu \in \mathbb {R}$$$\lambda ,\mu \in R$such that\begin{aligned} \lambda \mathcal {C} + \mu ({{\,\textrm{Id}\,}}-\mathcal {C})\text { approximately 1-projectionally factors through }D, \end{aligned}$\begin{array}{c}\lambda C+\mu \left(\phantom{\rule{0ex}{0ex}}\text{Id}\phantom{\rule{0ex}{0ex}}-C\right)\phantom{\rule{0ex}{0ex}}\text{approximately 1-projectionally factors through}\phantom{\rule{0ex}{0ex}}D,\end{array}$i.e., for all$$\eta > 0$$$\eta >0$, there exist bounded operatorsABso thatABis the identity operator$${{\,\textrm{Id}\,}}$$$\phantom{\rule{0ex}{0ex}}\text{Id}\phantom{\rule{0ex}{0ex}}$,$$\Vert A\Vert \cdot \Vert B\Vert = 1$$$‖A‖·‖B‖=1$and$$\Vert \lambda \mathcal {C} + \mu ({{\,\textrm{Id}\,}}-\mathcal {C}) - ADB\Vert < \eta$$$‖\lambda C+\mu \left(\phantom{\rule{0ex}{0ex}}\text{Id}\phantom{\rule{0ex}{0ex}}-C\right)-ADB‖<\eta$. Additionally, if$$\mathcal {C}$$$C$is unbounded onX(Y), then$$\lambda = \mu$$$\lambda =\mu$and then$${{\,\textrm{Id}\,}}$$$\phantom{\rule{0ex}{0ex}}\text{Id}\phantom{\rule{0ex}{0ex}}$either factors throughDor$${{\,\textrm{Id}\,}}-D$$$\phantom{\rule{0ex}{0ex}}\text{Id}\phantom{\rule{0ex}{0ex}}-D$.

more » « less
5. Abstract

The softmax policy gradient (PG) method, which performs gradient ascent under softmax policy parameterization, is arguably one of the de facto implementations of policy optimization in modern reinforcement learning. For$$\gamma$$$\gamma$-discounted infinite-horizon tabular Markov decision processes (MDPs), remarkable progress has recently been achieved towards establishing global convergence of softmax PG methods in finding a near-optimal policy. However, prior results fall short of delineating clear dependencies of convergence rates on salient parameters such as the cardinality of the state space$${\mathcal {S}}$$$S$and the effective horizon$$\frac{1}{1-\gamma }$$$\frac{1}{1-\gamma }$, both of which could be excessively large. In this paper, we deliver a pessimistic message regarding the iteration complexity of softmax PG methods, despite assuming access to exact gradient computation. Specifically, we demonstrate that the softmax PG method with stepsize$$\eta$$$\eta$can take\begin{aligned} \frac{1}{\eta } |{\mathcal {S}}|^{2^{\Omega \big (\frac{1}{1-\gamma }\big )}} ~\text {iterations} \end{aligned}$\begin{array}{c}\frac{1}{\eta }{|S|}^{{2}^{\Omega \left(\frac{1}{1-\gamma }\right)}}\phantom{\rule{0ex}{0ex}}\text{iterations}\end{array}$to converge, even in the presence of a benign policy initialization and an initial state distribution amenable to exploration (so that the distribution mismatch coefficient is not exceedingly large). This is accomplished by characterizing the algorithmic dynamics over a carefully-constructed MDP containing only three actions. Our exponential lower bound hints at the necessity of carefully adjusting update rules or enforcing proper regularization in accelerating PG methods.

more » « less