Abstract
In this paper, we compare the solutions of the Dyson Brownian motion for general \(\beta \) and potential V and the associated McKean–Vlasov equation near the edge. Under suitable conditions on the initial data and the potential V, we obtain optimal rigidity estimates of particle locations near the edge after a short time \(t={{\,\mathrm{o}\,}}(1)\). Our argument uses the method of characteristics along with a careful estimate involving an equation of the edge. With the rigidity estimates as an input, we prove a central limit theorem for mesoscopic statistics near the edge, which, as far as we know, has been done for the first time in this paper. Additionally, combining our results with Landon and Yau (Edge statistics of Dyson Brownian motion. arXiv:1712.03881, 2017), we give a proof of the local ergodicity of the Dyson Brownian motion for general \(\beta \) and potential at the edge, i.e., we show the distribution of extreme particles converges to the Tracy–Widom \(\beta \) distribution in a short time.
This is a preview of subscription content, access via your institution.

References
Adhikari, A., Che, Z.: Edge universality of correlated Gaussians. Electron. J. Probab. 24(44), 25 (2019)
Akemann, G., Atkin, M.R.: Higher order analogues of Tracy–Widom distributions via the Lax method. J. Phys. A 46(1), 015202 (2013). 19
Alt, J., Erdős, L., Krüger, T.: The Dyson equation with linear self-energy: spectral bands, edges and cusps. preprint, arXiv:1804.07752 (2018)
Alt, J., Erdős, L., Krüger, T., Schroder, D.: Correlated random matrices: band rigidity and edge universality. preprint, arXiv:1804.07744 (2018)
Bauerschmidt, R., Huang, J., Knowles, A., Yau, H.-T.: Edge rigidity and universality of random regular graphs of intermediate degree. preprint, arXiv:1910.10121 (2019)
Bourgade, P.: Extreme gaps between eigenvalues of Wigner matrices. preprint, arXiv:1812.10376 (2018)
Bourgade, P., Erdős, L., Yau, H.-T., Yin, J.: Fixed energy universality for generalized Wigner matrices. Commun. Pure Appl. Math. 69(10), 1815–1881 (2016)
Bourgade, P., Erdős, L., Yau, H.-T.: Bulk universality of general \(\beta \)-ensembles with non-convex potential. J. Math. Phys. 53(9), 095221 (2012). 19
Bourgade, P., Erdős, L., Yau, H.-T.: Edge universality of beta ensembles. Commun. Math. Phys. 332(1), 261–353 (2014)
de Monvel, A.B., Khorunzhy, A.: Asymptotic distribution of smoothed eigenvalue density. I. Gaussian random matrices. Random Oper. Stoch. Equ. 7(1), 1–22 (1999)
de Monvel, A.B., Khorunzhy, A.: Asymptotic distribution of smoothed eigenvalue density. II. Wigner random matrices. Random Oper. Stoch. Equ. 7(2), 149–168 (1999)
Chan, T.: The Wigner semi-circle law and eigenvalues of matrix-valued diffusions. Probab. Theory Relat. Fields 93(2), 249–272 (1992)
Cipolloni, G., Erdős, L., Krüger, T., Schröder, D.: Cusp universality for random matrices, II: the real symmetric case. Pure Appl. Anal. 7, 615–707 (2019)
Claeys, T., Its, A., Krasovsky, I.: Higher-order analogues of the Tracy–Widom distribution and the Painlevé II hierarchy. Commun. Pure Appl. Math. 63(3), 362–412 (2010)
Claeys, T., Neuschel, T., Venker, M.: Critical behavior of non-intersection Brownian motions. preprint, arXiv:1912.02142 (2019)
Duits, M., Johansson, K.: On mesoscopic equilibrium for linear statistics in Dyson’s Brownian motion. Mem. Am. Math. Soc. 255(1222), v+118 (2018)
Dyson, F.J.: A Brownian-motion model for the eigenvalues of a random matrix. J. Math. Phys. 3, 1191–1198 (1962)
Erdős, L., Krüger, T., Schröder, D.: Random matrices with slow correlation decay. Forum Math. Sigma 7(e8), 89 (2019)
Erdős, L., Krüger, T., Schröder, D.: Cusp universality for random matrices, I: local law and the complex Hermitian case. preprint, arXiv:1809.03971 (2018)
Erdős, L., Schnelli, K.: Universality for random matrix flows with time-dependent density. Ann. Inst. Henri Poincaré Probab. Stat. 53(4), 1606–1656 (2017)
Erdős, L., Yau, H.-T.: A dynamical approach to random matrix theory, volume 28 of Courant Lecture Notes in Mathematics. Courant Institute of Mathematical Sciences, New York; American Mathematical Society, Providence, RI (2017)
Erdős, L., Knowles, A., Yau, H.-T., Yin, J.: Spectral statistics of Erdős-Rényi graphs I: local semicircle law. Ann. Probab. 41(3B), 2279–2375 (2013)
Erdős, L., Péché, S., Ramírez, J.A., Schlein, B., Yau, H.-T.: Bulk universality for Wigner matrices. Commun. Pure Appl. Math. 63(7), 895–925 (2010)
Erdős, L., Ramírez, J., Schlein, B., Tao, T., Van, V., Yau, H.-T.: Bulk universality for Wigner Hermitian matrices with subexponential decay. Math. Res. Lett. 17(4), 667–674 (2010)
Erdős, L., Ramírez, J.A., Schlein, B., Yau, H.-T.: Universality of sine-kernel for Wigner matrices with a small Gaussian perturbation. Electron. J. Probab. 15(18), 526–603 (2010)
Erdős, L., Schlein, B., Yau, H.-T.: Local semicircle law and complete delocalization for Wigner random matrices. Commun. Math. Phys. 287(2), 641–655 (2009)
Erdős, L., Schlein, B., Yau, H.-T.: Semicircle law on short scales and delocalization of eigenvectors for Wigner random matrices. Ann. Probab. 37(3), 815–852 (2009)
Erdős, L., Schlein, B., Yau, H.-T.: Universality of random matrices and local relaxation flow. Invent. Math. 185(1), 75–119 (2011)
Erdős, L., Yau, H.-T.: Gap universality of generalized Wigner and \(\beta \)-ensembles. J. Eur. Math. Soc. JEMS 17(8), 1927–2036 (2015)
Erdős, L., Yau, H.-T., Yin, J.: Bulk universality for generalized Wigner matrices. Probab. Theory Relat. Fields 154(1–2), 341–407 (2012)
Erdős, L., Yau, H.-T., Yin, J.: Rigidity of eigenvalues of generalized Wigner matrices. Adv. Math. 229(3), 1435–1515 (2012)
Guionnet, A., Huang, J.: Rigidity and edge universality of discrete \(\beta \)-ensembles. Commun. Pure Appl. Math. 72(9), 1875–1982 (2019)
He, Y.: Mesoscopic linear statistics of Wigner matrices of mixed symmetry class. J. Stat. Phys. 175, 932–959 (2019)
He, Y., Knowles, A.: Mesoscopic eigenvalue statistics of Wigner matrices. Ann. Appl. Probab. 27(3), 1510–1550 (2017)
He, Y., Knowles, A.: Mesoscopic eigenvalue density correlations of Wigner matrices. Probab. Theory Relat. Fields (2019)
Huang, J., Landon, B.: Rigidity and a mesoscopic central limit theorem for Dyson Brownian motion for general \(\beta \) and potentials. Probab. Theory Relat. Fields 175(1–2), 209–253 (2019)
Huang, J., Landon, B., Yau, H.-T.: Transition from Tracy-Widom to Gaussian fluctuations of extremal eigenvalues of sparse Erdős-Rényi graphs. preprint, arXiv:1712.03936 (2017)
Landon, B., Sosoe, P.: Applications of mesoscopic CLTs in random matrix theory. preprint, arXiv:1811.05915 (2018)
Landon, B., Sosoe, P., Yau, H.-T.: Fixed energy universality of Dyson Brownian motion. Adv. Math. 346, 1137–1332 (2019)
Landon, B., Yau, H.-T.: Convergence of local statistics of Dyson Brownian motion. Commun. Math. Phys. 355(3), 949–1000 (2017)
Landon, B., Yau, H.-T.: Edge statistics of Dyson Brownian motion. preprint, arXiv:1712.03881 (2017)
Lee, J.O.: Edge universality for deformed Wigner matrices. Rev. Math. Phys. 27(8), 1550018 (2015). 94
Lee, J.O., Schnelli, K., Stetler, B., Yau, H.-T.: Bulk universality for deformed Wigner matrices. Ann. Probab. 44(3), 2349–2425 (2016)
Li, S., Li, X.-D., Xie, Y.-X.: Generalized Dyson Brownian motion, McKean-Vlasov equation and eigenvalues of random matrices. preprint, arXiv:1303.1240 (2013)
Li, S., Li, X.-D., Xie, Y.-X.: On the law of large numbers for the empirical measure process of generalized Dyson Brownian motion. preprint, arXiv:1407.7234 (2015)
Lodhia, A., Simm, N.J.: Mesoscopic linear statistics of Wigner matrices. preprint, arXiv: 1503.03533 (2015)
Ramírez, J.A., Rider, B., Virág, B.: Beta ensembles, stochastic Airy spectrum, and a diffusion. J. Am. Math. Soc. 24(4), 919–944 (2011)
Revuz, D., Yor, M.: Continuous martingales and Brownian motion, volume 293 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], 3rd edn. Springer, Berlin (1999)
Rogers, L., Shi, Z.: Interacting Brownian particles and the Wigner law. Probab. Theory Relat. Fields 95(4), 555–570 (1993)
Sosoe, P., Wong, P.: Regularity conditions in the CLT for linear eigenvalue statistics of Wigner matrices. Adv. Math. 249, 37–87 (2013)
Tao, T., Van, V.: Random matrices: universality of local eigenvalue statistics up to the edge. Commun. Math. Phys. 298(2), 549–572 (2010)
Tao, T., Van, V.: Random matrices: universality of local eigenvalue statistics. Acta Math. 206(1), 127–204 (2011)
Tao, T., Van, V.: Random matrices: sharp concentration of eigenvalues. Random Matrices Theory Appl. 2(3), 1350007 (2013). 31
Wigner, E.P.: Characteristic vectors of bordered matrices with infinite dimensions. Ann. Math. 2(62), 548–564 (1955)
Wigner, E.P.: Characteristic vectors of bordered matrices with infinite dimensions. II. Ann. Math. 2(65), 203–207 (1957)
Acknowledgements
We want to thank the anonymous reviewers for their comments and suggestions, which significantly improved the readability of this manuscript.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
A Proof of Proposition 3.4
The proof of Proposition 3.4 is based on performing a power series expansion of the analytic functions \(A_0\) and \(B_0\) in a neighborhood around \(E_0\) and solving the McKean–Vlasov equation (2.2) term by term. \(A_0(z)\) and \(B_0(z)\) have power series representations
such that
The following proposition states that the Stieltjes transform of \(\hat{\mu }_t\) has the form \(A_t+\sqrt{B_t}\), and \(A_t, B_t\) have power series representations in a neighborhood of \(E_0\). Proposition 3.4 is a natural consequence of the proceeding proposition.
Proposition A.1
We assume (3.3) holds. We fix a sufficiently small \(T>0\) and let \(L= 1/T\). Then for \(t\in [0,T]\), the solution of (2.2) is given by
with \(A_t(z)\) and \(B_t(z)\) analytic functions in a small neighborhood of \(E_0\). In this neighborhood, \(A_t(z)\) and \(B_t(z)\) have power series representations,
where the coefficients satisfy
Moreover, in a small neighborhood of \(E_0\), \(B_t(z)\) has a unique simple root at \(z=E_t\), which satisfies the differential equation
Proof
We suppose that the solution of (2.2) is given by (A.1), and \(A_t(z)\), \(B_t(z)\) have power series representations given by (A.2). We plug (A.1) into (2.2) to get
where
are analytic in a neighborhood of \(E_0\). One should note here that our goal in this section is not to show the existence of solutions to the original McKean–Vlasov equation, but only that said solution is analytic in a small neighborhood of \(E_0\). Thus, the fact that we use \(\hat{\mu }_t\) in our expression of \(R_t(z)\) is not an issue.
For (A.5) to hold, it is sufficient that
We solve (A.7) using the power series representations. Let
Then, (A.7) is equivalent to the following infinite system of ordinary differential equations. For \(i=0,1,2,\ldots \),
Although (A.8) is not Lipschitz, we can still solve this system by the Picard iteration. Let
and define recursively
As we have noted before, the existence of the measure \(\hat{\mu }_t\) is not in question and, thus, we do not need to perform an iteration of the \(r_i(t)\) terms in n.
We take large constants \(C>0\), \(L>0\) and a small constant \(T=1/L\). We first prove by induction that uniformly for \(t\in [0,T]\),
Since \(R_t(z)\) is analytic in a neighborhood of \(E_0\), we can take \(C_0, M>0\) large enough such that
We assume that (A.10) holds for n. Using (A.9), we have
provided \(C>4C_0\) and \(L>24eCM\). Similarly, for \(b_i^{(n+1)}(t)\),
provided \(C>2C_0\) and \(L>36eCM\). This finishes the proof of claim (A.10).
In the following, we prove that \(d_i^{(n)}(t), b_i^{(n)}(t)\) converge uniformly as n goes to infinity. This follows from
We prove (A.12) by induction. We assume that (A.12) holds for n. By using (A.9), the difference \(|d_i^{(n+1)}(t)-d_i^{(n)}(t)|\) is bounded by
provided that \(L\geqslant 24eCM\). Similarly, the difference \(|b_i^{(n+1)}(t)-b_i^{(n)}(t)|\) is bounded by
provided \(L\geqslant 72eCM\).
We denote for \(i=0,1,2,\ldots \),
They satisfy the system of differential equations (A.9), and (A.3) holds.
Using the same argument we used for (A.11) gives us
And, thus, for \(T>0\) small enough, times \(0\leqslant t\leqslant T\), and z on a small circle centered at \(E_0\),
Thus, by Rouché’s theorem, \(B_t(z)\) has a unique simple root at \(z=E_t\) inside a small neighborhood of \(E_0\). Moreover, from our construction, \(\bar{B}_t(z)=B_t(\bar{z})\). Thus, \(E_t\) is real. By taking the derivative of \(B_t(E_t)=0\) with respect to t and using (A.7), we get
This finishes the proof of Proposition A.1. \(\square \)
Proof of Proposition 3.4
It follows from Proposition A.1 that \(\hat{\mu }_t\) has square root behavior. In addition, \(\hat{m}_t(E_t) = A_t(E_t) +\sqrt{B_t(E_t)}\) is uniformly bounded for \(0\leqslant t\leqslant T\). The claim that \(E_t\) is Lipschitz in time simply follows from integrating (A.4).
Next, we prove that \({\mathfrak {C}}_t\) is Lipschitz in time. We notice that \({\mathfrak {C}}_t^2=\partial _z B_t(E_t)\). To prove \(|{\mathfrak {C}}_t-{\mathfrak {C}}_s|={{\,\mathrm{O}\,}}(t-s)\), it suffices to prove this for \(s=0\). Using the series expansion, as in the notation of (A.2), we see that
Using the differential equation (A.8), we see that \(b_1(t) -b_1(0)={{\,\mathrm{O}\,}}(t)\). Earlier, we showed that \((E_t - E_0)\) is of \({{\,\mathrm{O}\,}}(t)\). The infinite sum converges provided we take T sufficiently small. Therefore, it follows that \({\mathfrak {C}}_t^2 - {\mathfrak {C}}_0^2 ={{\,\mathrm{O}\,}}(t)\). Since \({\mathfrak {C}}_0\) is bounded away from 0, this would imply that \({\mathfrak {C}}_t -{\mathfrak {C}}_0 = {{\,\mathrm{O}\,}}(t)\) as desired. \(\square \)
B Rigidity bounds in the Proof of the Mesoscopic Central Limit Theorem
We start this section by collecting some lemmas that analyze the movement of characteristic flows that terminate in a radius \(\eta \) (\(N^{-2/3} \ll \eta \ll 1\)) neighborhood of the edge \(E_t\). Knowing the precise values of \(\kappa _s\) and \(\eta _s\) along a characteristic of the form \(z_s = E_s + \kappa _s + \mathrm{i}\eta _s\) will allow us to apply many of the improved rigidity bounds from Theorem 4.8 and the following remarks.
Lemma B.1
Suppose V satisfies Assumption 2.1 and let \({\mathfrak {d}}\) be a small parameter. Consider a characteristic \(z_t=E_t+\kappa _t+\mathrm{i}\eta _t\) as in (2.4) with terminal time \(t=o(1)\). If \(z_t\) satisfies
then for any \(0\leqslant s\leqslant t\), we have:
Proof
From the equation determining the motion of the characteristics (2.4), we know that each characteristic moves at a rate \({{\,\mathrm{O}\,}}(1)\) in time. Thus, if \(\eta _t \leqslant N^{-20{\mathfrak {d}}}\), we have \(\eta _s \leqslant N^{-20{\mathfrak {d}}}+C(t-s)\). Thanks to the estimates (2.7), if \(\eta _t \geqslant N^{-2/3+ 20{\mathfrak {d}}}\), we have \(\eta _s \geqslant N^{-2/3+ 20{\mathfrak {d}}}(1-C(t-s))\).
In the following, we estimate the ratio \( \kappa _s/\eta _s\). Let \(z_t - E_t =\kappa _t+\mathrm{i}\eta _t=(R_t+\mathrm{i}a_t)^2\), and \(z_s-E_s= \kappa _s+\mathrm{i}\eta _s=(R_s+\mathrm{i}a_s)^2\), where \(R_t, a_t, R_s\), and \(a_s>0\). By our assumption on \(z_t\), we have
and it follows that \(N^{-{\mathfrak {d}}}/C\leqslant \frac{R_t}{a_t}\leqslant C\). Thanks to (2.7) and the assumption of square root behavior, we have \(a_t\asymp \mathrm{Im}[\hat{m}_t(z_t)]\asymp \mathrm{Im}[\hat{m}_s(z_s)]\asymp a_s\). Moreover, using (2.7) again, we have \(2a_sR_s=\eta _s\asymp (\eta _t+(t-s)a_t)\asymp (a_s R_t+(t-s)a_s)\). We conclude that \(R_s\asymp R_t+(t-s)\), and
It follows, by rearranging, that
This finishes the proof of Lemma B.1. \(\square \)
Lemma B.2
Suppose V satisfies Assumption 2.1. Consider a characteristic \(z_t=E_t+\kappa _t+\mathrm{i}\eta _t\) as in (2.4) with terminal time \(t={{\,\mathrm{o}\,}}(1)\). If its terminal point \(z_t = E_t +\kappa _t + \mathrm{i}\eta _t\) satisfies \(-\eta _t \leqslant \kappa _t \leqslant C \eta _t \), then we have, for \(t-q \geqslant \sqrt{\eta _t}\), that \(\sqrt{|\kappa _q| +\eta _q} \asymp (t-q)\), and, for \(t-q \leqslant \sqrt{\eta _t}\), it holds that \(\sqrt{|\kappa _q|+ \eta _q} \asymp \sqrt{\eta _t}\). We always have \(\eta _q \asymp \eta _t + (t-q) \sqrt{\eta _t}\).
If we consider a characteristic whose terminal point instead satisfies \(-N^{{\mathfrak {d}}} \eta _t \leqslant \kappa _t \leqslant -\eta _t \), we instead find that \(\sqrt{\kappa _q + \eta _q} \asymp (t-q)\) for \(t-q \geqslant \sqrt{|\kappa _t| + \eta _t}\). For \(t-q \leqslant \sqrt{|\kappa _t| +\eta _t}\), we have \(\sqrt{|\kappa _q| + \eta _q} \asymp \sqrt{|\kappa _t| + \eta _t}\). We always have \(\eta _q \asymp \eta _t +(t-q) \sqrt{|\kappa _t| +\eta _t}\).
Proof
Consider a characteristic whose terminal point \(z_t\) satisfies \(-\eta _t\leqslant \kappa _t \leqslant C \eta _t\). By square root behavior, this will imply \(\mathrm{Im}[\hat{m}_t(z_t)] \asymp \sqrt{\eta _t}\). Finally, by the final line of (2.7), we have that \(\eta _q \asymp \eta _t + (t-q) \sqrt{\eta _t}\). We see that, for \(t-q \leqslant \sqrt{\eta _t}\), we have that \(\eta _q \asymp \eta _t\), while, for \((t-q) \geqslant \sqrt{\eta _t}\), we have that \(\eta _q \asymp (t-q) \sqrt{\eta _t}\).
We additionally have the relation \(\mathrm{Im}[\hat{m}_q(z_q)] \asymp \mathrm{Im}[\hat{m}_t(z_t)]\) by the second line of (2.7). This will certainly imply that \(z_q\) cannot move far inside the spectrum, and, instead, we must have
This will imply
Thus, when \(t-q \geqslant \sqrt{\eta _t}\), we have that \(\sqrt{|\kappa _q|}\asymp (t-q)\).
Now let us consider, instead, the case that \(- N^{{\mathfrak {d}}} \eta _t \leqslant \kappa _t \leqslant - \eta _t\). We will have the relation \(\eta _q \asymp \eta _t + (t-q) \sqrt{|\kappa _t|}\). For \(t-q \geqslant \eta _t/\sqrt{\kappa _t}\), this will be comparable to \((t-q)\sqrt{|\kappa _t|}\). To maintain \(\mathrm{Im}[m_q(z_q)]\asymp \mathrm{Im}[m_t(z_t)]\), when \(t-q \leqslant \sqrt{\kappa _t}\), would necessitate \(|\kappa _q| + \eta _q \asymp |\kappa _t| + \eta _t\). When \((t-q) \geqslant \sqrt{\kappa _t}\), we would instead have the relation
This would then imply that \(\sqrt{\kappa _q + \eta _q}\asymp (t-q)\). \(\square \)
The improved rigidity bounds in Theorem 4.8 allow us to get sharp estimates on the derivatives of the Stieltjes transform as in the following lemma.
Lemma B.3
Recall the times s and t from Eq. (5.1) as well as the event \({\mathcal {S}}\) from Sect. 5. Under the assumptions of Theorem 5.1, consider a characteristic \(z_t\) such that \(\eta _t\asymp \eta \) and \(-N^{{\mathfrak {d}}}\eta \leqslant \kappa _t \lesssim \eta \). On \({\mathcal {S}}\), we have the following estimate uniformly for times q satisfying \(s\leqslant q\leqslant t\).
where \(\tilde{{\mathfrak {d}}}\) is any parameter strictly greater than \({\mathfrak {d}}\).
Proof
Since both \(m_q\) and \(\hat{m}_q\) are analytic on the upper half-plane, by Cauchy’s integral formula,
where \({\mathcal {C}}\) is a small contour in the upper half-plane centered at \(z_q\) with radius \(\eta _q/2\). We take the absolute value of both sides to get
If we have that \(\kappa _q \geqslant \eta _q \gg N^{-2/3}\), we will be in a region in which we can apply the improved error estimates (4.18) to argue that \(|m_q(w) - \hat{m}_q(w)| \leqslant \frac{M(\log N)^2}{N (\kappa _q + \eta _q)}\). Otherwise, we have, instead, that \(\eta _q \geqslant \kappa _q \geqslant - N^{{\mathfrak {d}}} \eta _q\) by applying Lemma B.1. At this point, we can apply the error estimate from (4.13) to conclude that \(|m_q(w) -\hat{m}_q(w)| \leqslant \frac{N^{{\mathfrak {d}}} (\log N) M}{N(|\kappa _q|+\eta _q)}\).
We can regard the M and \(\log N\) factors as unimportant relative to the \(N^{{\mathfrak {d}}}\) factor in the numerator for sufficiently large N. We can additionally bound \(|w-z_q|\) in the denominator from below by \(\eta _q\). Applying these considerations to the contour integral (B.3), we derive the bound
where \(\tilde{{\mathfrak {d}}}\) is any parameter greater than \({\mathfrak {d}}\). \(\square \)
We will often have to perform the time integral of the error term coming from our rigidity estimates. This lemma explicitly performs the time integrals that frequently appear in our analysis.
Lemma B.4
Recall the times s and t from Eq. (5.1). Under the assumptions of Theorem 5.1, consider a characteristic \(z_\tau = E_\tau + \kappa _\tau + \mathrm{i}\eta _\tau \) that, at time t, satisfies the conditions \(\eta _t \asymp \eta \) and \(-N^{{\mathfrak {d}}}\eta \lesssim \kappa _t \lesssim \eta \). We have
where the implicit constant is universal.
Proof
We can perform explicit calculations thanks to (2.7)
The claim follows from our assumption of square root behavior. We chose the time difference \(t -s\) large enough so that the quantity \(\eta _t + (t-s) \mathrm{Im}[\hat{m}_t(z_t)]\) is much larger than \(\eta _t\). Finally, we used the fact that, at \(z_t\), we have \(\mathrm{Im}[\hat{m}_t(z_t)] \leqslant C \sqrt{\eta _t}\) for some large constant C. \(\square \)
The following lemma computes the main contribution of the quadratic variance terms from Proposition 5.3.
Lemma B.5
We recall \({\mathfrak {C}}_q\) from (3.5) and the times s and t from Eq. (5.1). Under the assumptions of Theorem 5.1 and for any time q satisfying \(s\leqslant q\leqslant t\), we have the following integral evaluation,
where the characteristics \(z_t=E_t+\kappa _t+\mathrm{i}\eta _t,z_t'=E_t+\kappa _t'+\mathrm{i}\eta _t'\) satisfy \(\eta _t, \eta _t'\asymp \eta \), \(-N^{{\mathfrak {d}}}\eta \leqslant \kappa _t,\kappa _t'\leqslant \eta \).
Proof
We recall from Proposition 3.4 that the measure \(\hat{\mu }_q\) is supported on \((-\infty , E_q]\), and has a density in a small neighborhood of \(E_q\) of the form
Now, we try to explicitly identify the main term found in (B.4). We have
The second term of the above equation can clearly be bounded by \({{\,\mathrm{O}\,}}((\log N)^4)\) if one takes into account the fact that \(z_q\) and \(z_q'\) are far away from \(E_q-(\log N)^{-1}\). This can easily be seen to be an \({{\,\mathrm{O}\,}}(N^{-{\mathfrak {d}}^2})\) factor of the main term in (B.4).
To estimate the other quantity, we do a Taylor expansion of \(\hat{\rho }_q(x)\) around the point 0 as \(|\hat{\rho }_q(x) -{\mathfrak {C}}_q\sqrt{x}| = {{\,\mathrm{O}\,}}(x^{3/2})\). Without loss of generality, we will assume that \(|\kappa _q| \geqslant |\kappa _q'|\). To illustrate the computation, we will only consider the case that \(-\kappa _q \geqslant \eta _q\), and \(-\kappa _q' \geqslant \eta _q'\). Similar techniques can be used in all other cases and are generally simpler. The above computation can be divided into two cases; the first case is when \(|\kappa _q - \kappa _q'| \geqslant |\kappa _q|/2\). We will perform a decomposition of the integral as follows:
We need to show that the above quantity will be less than \(N^{-{\mathfrak {d}}^2} |\kappa _q|^{-2} |\kappa _q'|^{-1/2}\). This is implied by the two inequalities \(\eta _q \leqslant N^{-20{\mathfrak {d}}}\) and \(|\kappa _q| \leqslant N^{{\mathfrak {d}}} \eta _q \) from Lemma B.1.
Next, we consider the case in which \(|\kappa _q - \kappa _q'| \leqslant |\kappa _q|/2\). We can divide the integral as
To show that the above error is smaller than \(N^{-{\mathfrak {d}}^2} |\kappa _q|^{-2} |\kappa _q'|^{-1/2}\), we would only need to show that \(|\kappa _q|^{5} \leqslant N^{-{\mathfrak {d}}^2} \eta _q^2 \eta _q'^2\). Applying \(\eta _q \asymp \eta _q'\) and the fact that \(|\kappa _q| \leqslant N^{{\mathfrak {d}}} \eta _q \), we see that we would only need to prove that \(\eta _q \leqslant N^{- {\mathfrak {d}}^2 - 5 {\mathfrak {d}}}\). This is due to the fact that \(\eta _q \leqslant N^{-20{\mathfrak {d}}}\).
We now only need to compute the following integral
where the square roots come from the branch with positive real part, and the error term \((\log N)^{5/2}\) can be absorbed into the \(N^{-{\mathfrak {d}}^2}\) factor of the main term. This finishes the proof of Lemma B.5. \(\square \)
We now compute the time integrals of the main terms in Proposition 5.3 weighted by the factors \({\mathcal {I}}_t ({\mathcal {I}}_s)^{-1}\). By using the equation of the characteristics, we can integrate this by a simple change of variables formula.
Lemma B.6
We recall \({\mathfrak {C}}_q\) from (3.5) and times s and t from Eq. (5.1). Under the assumptions of Theorem 5.1, take two characteristics \(z_t=E_t+\kappa _t+\mathrm{i}\eta _t\) and \(z_t'=E_t+\kappa _t'+\mathrm{i}\eta _t'\) such that \(\eta _t, \eta _t'\asymp \eta \), \(-N^{{\mathfrak {d}}}\eta \lesssim \kappa _t,\kappa _t'\lesssim \eta \). We have that
Proof
Using the results of Lemma B.5, we have
Let \(A= \sqrt{\kappa _s + \mathrm{i}\eta _s}\) , \(A' = \sqrt{\kappa _s' + \mathrm{i}\eta _s'}\) and \(B_q = \pi /2 \int _{q}^{t} {\mathfrak {C}}_\tau \mathrm{d}\tau \). By using our characteristic estimates from (5.27), we would like to replace \(\sqrt{\kappa _q + \mathrm{i}\eta _q}\) with the sum \(\sqrt{\kappa _t + \mathrm{i}\eta _t} + \pi /2 \int _q^t {\mathfrak {C}}_\tau \mathrm{d}\tau \) up to some small error. For times \((t-q) \ll (|\kappa _t| + \eta _t)^{1/4}\), this replacement is valid up to a multiplicative factor of \((1+ {{\,\mathrm{O}\,}}(N^{-{\mathfrak {d}}^2}))\).
We will change the integration variable from \(q \rightarrow B_q\); the Jacobian of this transform will be \(\pi /2 {\mathfrak {C}}_q\). Upon this transformation, we see that we would have to integrate \(\int _{B_s=0}^{B_t} \frac{1}{(A+A' -2 x)^3} \mathrm{d}x\). This can be evaluated explicitly and will give us the last line of (B.7) up to a multiplicative factor of the form \(1+{{\,\mathrm{O}\,}}(N^{-{\mathfrak {d}}^2})\), due to the fact that we have to invert the substitution \(\sqrt{\kappa _t + \mathrm{i}\eta _t} + \pi /2 \int _q^t {\mathfrak {C}}_\tau \mathrm{d}\tau \rightarrow \sqrt{\kappa _q + \mathrm{i}\eta _q}\). We remark here that if we have that \(t-s\gg \sqrt{\eta }\), the last line of (B.7) can be subsumed as an \({{\,\mathrm{O}\,}}(N^{- {\mathfrak {d}}^2})\) error of the second term. \(\square \)
The following lemma collects the last computation appearing in Sect. 5. It follows from the arguments of Lemma B.5 and Lemma B.6. We will only mention the result here for reference.
Lemma B.7
We recall \({\mathfrak {C}}_q\) from (3.5) and times s and t from Eq. (5.1). Under the assumptions of Theorem 5.1, consider a characteristic \(z_t=E_t+\kappa _t+\mathrm{i}\eta _t\) such that \( \eta _t\asymp \eta \) and \(-N^{{\mathfrak {d}}}\eta \lesssim \kappa _t\lesssim \eta \). We have
and
Rights and permissions
About this article
Cite this article
Adhikari, A., Huang, J. Dyson Brownian motion for general \(\beta \) and potential at the edge. Probab. Theory Relat. Fields 178, 893–950 (2020). https://doi.org/10.1007/s00440-020-00992-9
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00440-020-00992-9
Mathematics Subject Classification
- 60H15
- 60F05
- 60B20