Skip to main content

Edge statistics for a class of repulsive particle systems

Abstract

We study a class of interacting particle systems on \(\mathbb {R}\) which was recently investigated by Götze and Venker (Ann Probab 42(6):2207–2242, 2014. doi:10.1214/13-AOP844). These ensembles generalize eigenvalue ensembles of Hermitian random matrices by allowing different interactions between particles. Although these ensembles are not known to be determinantal one can use the stochastic linearization method of Götze and Venker (Ann Probab 42(6):2207–2242, 2014. doi:10.1214/13-AOP844) to represent them as averages of determinantal ones. Our results describe the transition between universal behavior in the regime of the Tracy–Widom law and non-universal behavior for large deviations of the rightmost particle. Moreover, a detailed analysis of the transition that occurs in the regime of moderate deviations, is provided. We also compare our results with the corresponding ones obtained recently for determinantal ensembles (Eichelsbacher et al. in Symmetry Integr Geom Methods Appl 12:18, 2016. doi:10.3842/SIGMA.2016.093; Schüler in Moderate, large and superlarge deviations for extremal eigenvalues of unitarily invariant ensembles. Ph.D. thesis, University of Bayreuth, 2015). In particular, we discuss how the averaging effects the leading order behavior in the regime of large deviations. In the analysis of the averaging procedure we use detailed asymptotic information on the behavior of Christoffel–Darboux kernels that is uniform for perturbative families of weights. Such results have been provided by K. Schubert, K. Schüler and the authors in Kriecherbauer et al. (Markov Process Relat Fields 21(3, part 2):639–694, 2015).

This is a preview of subscription content, access via your institution.

References

  1. Abramowitz, M., Stegun, I.A.: Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables. National Bureau of Standards Applied Mathematics Series, For Sale by the Superintendent of Documents, vol. 55. U.S. Government Printing Office, Washington (1964)

    MATH  Google Scholar 

  2. Adler, R.J., Taylor, J.E.: Random Fields and Geometry. Springer Monographs in Mathematics. Springer, New York (2007)

    MATH  Google Scholar 

  3. Akemann, G., Baik, J., Di Francesco, P. (eds.): The Oxford Handbook of Random Matrix Theory. Oxford University Press, Oxford (2011)

    MATH  Google Scholar 

  4. Anderson, G.W., Guionnet, A., Zeitouni, O.: An Introduction to Random Matrices, Cambridge Studies in Advanced Mathematics, vol. 118. Cambridge University Press, Cambridge (2010)

    MATH  Google Scholar 

  5. Baik, J., Buckingham, R., DiFranco, J.: Asymptotics of Tracy–Widom distributions and the total integral of a Painlevé II function. Commun. Math. Phys. 280(2), 463–497 (2008). doi:10.1007/s00220-008-0433-5

    Article  MATH  Google Scholar 

  6. Baik, J., Deift, P., McLaughlin, K.T.R., Miller, P., Zhou, X.: Optimal tail estimates for directed last passage site percolation with geometric random variables. Adv. Theor. Math. Phys. 5(6), 1207–1250 (2001)

    Article  MathSciNet  MATH  Google Scholar 

  7. Bekerman, F., Figalli, A., Guionnet, A.: Transport maps for \(\beta \)-matrix models and universality. Commun. Math. Phys. 338(2), 589–619 (2015). doi:10.1007/s00220-015-2384-y

    Article  MathSciNet  MATH  Google Scholar 

  8. Borot, G., Eynard, B., Orantin, N.: Abstract loop equations, topological recursion and new applications. Commun. Number Theory Phys. 9(1), 51–187 (2015)

    Article  MathSciNet  MATH  Google Scholar 

  9. Borot, G., Guionnet, A., Kozlowski, K.K.: Large-\(N\) asymptotic expansion for mean field models with Coulomb gas interaction. Int. Math. Res. Not. 2015(20), 10451–10524 (2015). doi:10.1093/imrn/rnu260

    Article  MathSciNet  MATH  Google Scholar 

  10. Bourgade, P., Erdös, L., Yau, H.T.: Edge universality of beta ensembles. Commun. Math. Phys. 332(1), 261–353 (2014). doi:10.1007/s00220-014-2120-z

    Article  MathSciNet  MATH  Google Scholar 

  11. Boutet de Monvel, A., Pastur, L., Shcherbina, M.: On the statistical mechanics approach in the random matrix theory: integrated density of states. J. Statist. Phys. 79(3–4), 585–611 (1995). doi:10.1007/BF02184872

  12. Chafaï, D., Gozlan, N., Zitt, P.A.: First-order global asymptotics for confined particles with singular pair repulsion. Ann. Appl. Probab. 24(6), 2371–2413 (2014). doi:10.1214/13-AAP980

    Article  MathSciNet  MATH  Google Scholar 

  13. Chafaï, D., Péché, S.: A note on the second order universality at the edge of Coulomb gases on the plane. J. Stat. Phys. 156(2), 368–383 (2014). doi:10.1007/s10955-014-1007-x

    Article  MathSciNet  MATH  Google Scholar 

  14. Cramér, H.: Sur un nouveau théorème-limite de la théorie des probabilités. Actual. sci. industr (Confér. internat. Sci. math. Univ. Genève. Théorie des probabilités. III: Les sommes et les fonctions de variables aléatoires), vol. 736, pp. 5–23 (1938)

  15. Deift, P.: Orthogonal Polynomials and Random Matrices: A Riemann–Hilbert Approach, Courant Lecture Notes in Mathematics, vol. 3. American Mathematical Society, New York University, Courant Institute of Mathematical Sciences, New York (1999)

  16. Deift, P., Gioev, D.: Random Matrix Theory: Invariant Ensembles and Universality, Courant Lecture Notes in Mathematics, vol. 18. American Mathematical Society, Courant Institute of Mathematical Sciences, New York (2009)

  17. Eichelsbacher, P., Kriecherbauer, T., Schüler, K.: Precise deviations results for the maxima of some determinantal point processes: the upper tail. SIGMA Symmetry Integr. Geom. Methods Appl. 12, 18 (2016). doi:10.3842/SIGMA.2016.093

    MathSciNet  MATH  Google Scholar 

  18. Erdos, L.: Random matrices, log-gases and Hölder regularity. In: Proceedings of ICM 2014, Seoul, vol. III, pp. 213–236 (2015)

  19. Figalli, A., Guionnet, A.: Universality in several-matrix models via approximate transport maps. Acta Math. 217, 81 (2016). doi:10.1007/s11511-016-0142-4

  20. Freud, G.: Orthogonale Polynome. Birkhäuser Verlag, Basel-Stuttgart (1969). Lehrbücher und Monographien aus dem Gebiete der Exakten Wissenschaften, Mathematische Reihe, Band 33

  21. Götze, F., Venker, M.: Local universality of repulsive particle systems and random matrices. Ann. Probab. 42(6), 2207–2242 (2014). doi:10.1214/13-AOP844

    Article  MathSciNet  MATH  Google Scholar 

  22. Johansson, K.: On fluctuations of eigenvalues of random Hermitian matrices. Duke Math. J. 91(1), 151–204 (1998). doi:10.1215/S0012-7094-98-09108-6

    Article  MathSciNet  MATH  Google Scholar 

  23. Johansson, K.: Shape fluctuations and random matrices. Commun. Math. Phys. 209(2), 437–476 (2000). doi:10.1007/s002200050027

    Article  MathSciNet  MATH  Google Scholar 

  24. Kriecherbauer, T., Schubert, K., Schüler, K., Venker, M.: Global asymptotics for the Christoffel–Darboux kernel of random matrix theory. Markov Process. Relat. Fields 21(3, part 2), 639–694 (2015)

    MathSciNet  MATH  Google Scholar 

  25. Kriecherbauer, T., Shcherbina, M.: Fluctuations of eigenvalues of matrix models and their applications (2010). arXiv:1003.6121v1 [math-ph]

  26. Krishnapur, M., Rider, B., Virág, B.: Universality of the stochastic airy operator. Commun. Pure Appl. Math. 69(1), 145–199 (2016). doi:10.1002/cpa.21573

    Article  MathSciNet  MATH  Google Scholar 

  27. Ledoux, M., Rider, B.: Small deviations for beta ensembles. Electron. J. Probab. 15(41), 1319–1343 (2010). doi:10.1214/EJP.v15-798

    Article  MathSciNet  MATH  Google Scholar 

  28. Löwe, M., Merkl, F.: Moderate deviations for longest increasing subsequences: the upper tail. Commun. Pure Appl. Math. 54(12), 1488–1520 (2001). doi:10.1002/cpa.10010

    Article  MathSciNet  MATH  Google Scholar 

  29. Löwe, M., Merkl, F., Rolles, S.: Moderate deviations for longest increasing subsequences: the lower tail. J. Theor. Probab. 15(4), 1031–1047 (2002). doi:10.1023/A:1020649006254

    Article  MathSciNet  MATH  Google Scholar 

  30. Lukacs, E., Szász, O.: On analytic characteristic functions. Pac. J. Math. 2, 615–625 (1952)

    Article  MathSciNet  MATH  Google Scholar 

  31. Pastur, L., Shcherbina, M.: Eigenvalue Distribution of Large Random Matrices, Mathematical Surveys and Monographs, vol. 171. American Mathematical Society, Providence (2011). doi:10.1090/surv/171

    MATH  Google Scholar 

  32. Ramírez, J.A., Rider, B., Virág, B.: Beta ensembles, stochastic Airy spectrum, and a diffusion. J. Am. Math. Soc. 24(4), 919–944 (2011). doi:10.1090/S0894-0347-2011-00703-0

    Article  MathSciNet  MATH  Google Scholar 

  33. Saff, E.B., Totik, V.: Logarithmic potentials with external fields, Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences], vol. 316. Springer, Berlin (1997) (Appendix B by Thomas Bloom)

  34. Schubert, K., Venker, M.: Empirical spacings of unfolded eigenvalues. Electron. J. Probab. 20, 37 (2015)

    MathSciNet  MATH  Google Scholar 

  35. Schüler, K.: Moderate, large and superlarge deviations for extremal eigenvalues of unitarily invariant ensembles. Ph.D. thesis, University of Bayreuth (2015)

  36. Shcherbina, M.: Orthogonal and symplectic matrix models: universality and other properties. Commun. Math. Phys. 307(3), 761–790 (2011). doi:10.1007/s00220-011-1351-5

    Article  MathSciNet  MATH  Google Scholar 

  37. Tao, T., Vu, V.: Random matrices: the universality phenomenon for Wigner ensembles. In: Modern Aspects of Random Matrix Theory, Proceedings of Symposia in Applied Mathematics, Am. Math. Soc., vol. 72, pp. 121–172. Providence (2014)

  38. Titchmarsh, E.C.: The Theory of Functions, 2nd edn. Oxford University Press, Oxford (1939)

    MATH  Google Scholar 

  39. Totik, V.: Asymptotics for Christoffel functions with varying weights. Adv. Appl. Math. 25(4), 322–351 (2000). doi:10.1006/aama.2000.0698

    Article  MathSciNet  MATH  Google Scholar 

  40. Tracy, C.A., Widom, H.: Level-spacing distributions and the airy kernel. Commun. Math. Phys. 159(1), 151–174 (1994)

    Article  MathSciNet  MATH  Google Scholar 

  41. Tracy, C.A., Widom, H.: Correlation functions, cluster functions, and spacing distributions for random matrices. J. Stat. Phys. 92(5–6), 809–835 (1998). doi:10.1023/A:1023084324803

    Article  MathSciNet  MATH  Google Scholar 

  42. Venker, M.: Particle systems with repulsion exponent \(\beta \) and random matrices. Electron. Commun. Probab. 18(83), 1–12 (2013). doi:10.1214/ECP.v18-2864

    MathSciNet  MATH  Google Scholar 

Download references

Acknowledgements

Funding was provided by Deutsche Forschungsgemeinschaft (DE) (CRC 701).

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Martin Venker.

Additional information

Martin Venker has been supported by the CRC 701 at Bielefeld University.

Appendix: Asymptotics for determinantal ensembles

Appendix: Asymptotics for determinantal ensembles

In this appendix we extract those results from [24] on the asymptotics of the determinantal ensembles \(P_{N,V,f;L}:=P_{N,V-f/N;L}\), defined on \([-L,L]^N\) that are relevant in our context.

We begin by recalling the notion of the equilibrium measure with respect to a general convex external field \(\tilde{V}\). By this we mean the unique Borel probability measure \(\mu _{\tilde{V}}\) which minimizes the energy functional

$$\begin{aligned} \mu \mapsto \int \int \log \left|t-s\right|^{-1}d\mu (t)d\mu (s)+\int \tilde{V}(t)d\mu (t). \end{aligned}$$

A general reference on equilibrium measures with external fields is [33]. It is known that under mild assumptions such a unique minimizer exists. In our case \(\tilde{V}\) will be convex and the density of \(\mu _{\tilde{V}}\) can be described as follows (see e.g. [24, Sections 1 and 2]): There exist real numbers \(a_{\tilde{V}} < b_{\tilde{V}}\) that are uniquely determined by the two equations for a and b,

$$\begin{aligned} \int _a^b \frac{\tilde{V}'(t)}{\sqrt{(b-t)(t-a)}} dt = 0 \ , \quad \int _a^b \frac{t \tilde{V}'(t)}{\sqrt{(b-t)(t-a)}} dt = 2 \pi . \end{aligned}$$

We denote by \(\lambda _{\tilde{V}}\) the linear rescaling that maps the interval \([-1, 1]\) onto \([a_{\tilde{V}}, b_{\tilde{V}}]\),

$$\begin{aligned} \lambda _{\tilde{V}}(t):=\frac{b_{\tilde{V}}-a_{\tilde{V}}}{2}t+\frac{a_{\tilde{V}}+b_{\tilde{V}}}{2} \ , \end{aligned}$$
(74)

and introduce a function \(G_{\tilde{V}}\) on \(\mathbb {R}\) by

$$\begin{aligned} G_{\tilde{V}}(t):=\frac{1}{\pi }\int _{-1}^1\int _0^1\frac{(\tilde{V}\circ \lambda _{\tilde{V}})''(t+u(s-t))}{\sqrt{1-s^2}}duds \ \end{aligned}$$
(75)

that inherits real analyticity and positivity from \(\tilde{V}''\). The density of the equilibrium measure is then given by (see [24, (2.1), (1.11–1.14)])

$$\begin{aligned} d\mu _{\tilde{V}}(t)=\frac{2}{(b_{\tilde{V}}-a_{\tilde{V}})^2\pi }\sqrt{(t-a_{\tilde{V}})(b_{\tilde{V}}-t)}G_{\tilde{V}}(\lambda _{\tilde{V} }^{-1}(t))\mathbbm {1}_{[a_{\tilde{V}},b_{\tilde{V}}]}(t) dt. \end{aligned}$$

The linearization technique of [21] requires to consider ensembles \(P_{N,V-f/N; L}\) that arise from restricting the \(P_{N,V-f/N}\) to \([-L,L]^N\) and renormalizing them as probability measures. Note that the choice of L throughout the paper ensures that the support \([a_{V-f/N}, b_{V-f/N}]\) of \(\mu _{V-f/N}\) is contained in the interior of \([-L,L]\). Moreover, the functions f are always defined on some fixed domain D in the complex plane that contains \([-L,L]\). They belong to the real Banach space \(X_D:=\{f:\, D\, \longrightarrow \, \mathbb {C}\,:\,f\) analytic and bounded, \(f(D\cap \mathbb {R})\subset \mathbb {R}\}\), equipped with the norm \(\Vert f\Vert :=\Vert f\Vert _{D}:=\sup _{z\in D}\left|f(z)\right|\). By [24, Lemma 2.4], the maps \(\tilde{V}\mapsto a_{\tilde{V}}\), \(\tilde{V}\mapsto b_{\tilde{V}}\) that relate the external fields to the endpoints of the support of their equilibrium measures are \(C^1\) with bounded derivatives on a neighborhood of V in \(X_D\). Thus, for sufficiently large values of N,

$$\begin{aligned} a_{V,f}-a_V,b_{V,f}-b_V=\mathcal {O}\left( \frac{\Vert f\Vert }{N}\right) , \end{aligned}$$
(76)

where the subscript \(_{V,f}\) is short for \(_{V-f/N}\). We can now formulate a first result on kernels \(K_{N,V,f;L}\) associated to determinantal ensembles of the form \(P_{N,V,f;L}\) that has essentially been proved in [24].

Proposition 2

(cf. [24, Theorem 1.8]) Let \(V:\, [-L,L]\, \longrightarrow \, \mathbb {R}\) be real-analytic with \(\inf _{t\in [-L,L]}V''(t)>0\). Assume that the support \([a_V,b_V]\) of the equilibrium measure \(\mu _V\) is contained in \((-L,L)\) and recall the notations introduced in this appendix. Moreover, define

$$\begin{aligned}&\gamma _V:=\frac{(2G_V(1))^{2/3}}{b_V-a_V} \, , \, \\&\widehat{K}_{N,V,f; L} (s, t) := \frac{1}{N^{2/3}\gamma _V} K_{N,V,f; L} \left( b_V+ \frac{s}{N^{2/3}\gamma _V}, b_V+ \frac{t}{N^{2/3}\gamma _V} \right) , \end{aligned}$$

where s, \(t \in \mathbb {R}\). Then we have for fixed \(0<\kappa <1/3\), \(0< \varepsilon < \min (2/15, 2/3 -2 \kappa )\), \(q \in \mathbb {R}\),

$$\begin{aligned} \widehat{K}_{N,V,f; L} (s, t) = {\left\{ \begin{array}{ll} K_\text {Ai}(s,t) + \mathcal {O}\left( N^{\kappa - 1/3}\right) \, , &{}\textit{if } q\le s, t \le 2,\\ K_\text {Ai}(s,t) \left( 1 + \mathcal {O}\left( N^{\kappa +\varepsilon /2 - 1/3} \right) \right) \, , &{}\textit{if } 1\le s, t \le N^{\varepsilon },\\ K_\text {Ai}(s,t) + \text {Ai}'(t) \mathcal {O}\left( N^{\kappa - 1/3} \right) \, , &{}\textit{if } q\le s\le 1; 2\le t \le N^{\varepsilon } \end{array}\right. } \end{aligned}$$

with the \(\mathcal {O}\) terms being uniform in N, in \(f\in X_D\) with \(\Vert f\Vert _{D}\le N^\kappa \) and in st within the respective regions.

Proof

The result in [24, Theorem 1.8] provides leading order information on

$$\begin{aligned} \frac{1}{N^{2/3}\gamma _{V,f}}K_{N,V,f; L}\left( b_{V,f}+\frac{s}{N^{2/3}\gamma _{V,f}},b_{V,f}+\frac{t}{ N^ {2/3 } \gamma _{V,f}} \right) \end{aligned}$$

together with error bounds that have the uniformity needed to prove Proposition 2. We note in passing that (76) ensures that, uniformly for \(f\in X_D\) with \(\Vert f\Vert _{D}\le N^\kappa \), \([a_{V,f},b_{V,f}]\subset (-L,L)\) for N large enough.

The reader should be aware of some differences in notation, e.g. the quantities Q, V, \(\gamma _V^+\), \(\mathbb {A}{\text {i}}\) in [24] correspond to V, \(V - f/N\), \(\gamma _{V,f}(b_{V,f}-a_{V,f})/2\), \(K_\text {Ai}\) in the present paper.

Our first task is to replace \(b_{V,f}\) by \(b_V\) and \(\gamma _{V,f}\) by \(\gamma _V\). Using

$$\begin{aligned}&[(V-f/N)\circ \lambda _{V,f}]'' - [V\circ \lambda _{V}]'' \\&\quad = \left[ (\lambda _{V,f}')^2 V''\circ \lambda _{V,f} - (\lambda _{V}')^2 V''\circ \lambda _{V}\right] -(\lambda _{V,f}')^2 (f''/N)\circ \lambda _{V,f}, \end{aligned}$$

relations (74)–(76), the thrice differentiability of V on \([-L, L]\), the analyticity of f in D, and the positivity of \(G_V\), one obtains

$$\begin{aligned} G_{V,f}(t)=G_V(t)(1+\mathcal {O}(\Vert f\Vert /N)) \end{aligned}$$
(77)

uniformly for \(f\in X_D\) with \(\Vert f\Vert _{D}\le N^\kappa \), \(0<\kappa <1\) and also uniformly in \(t\in \mathbb {R}\) with both \(\lambda _V(t)\), \(\lambda _{V,f}(t) \in [-L,L]\). This yields in particular \(\gamma _{V,f}=\gamma _V(1+\mathcal {O}(\Vert f\Vert /N))\). In order to be able to apply the results of [24] we relate to any \(q \le t \le N^{\varepsilon }\) a number \(\widehat{t}=\widehat{t}(N, V, f)\) by

$$\begin{aligned} b_V+\frac{t}{N^{2/3}\gamma _V}=b_{V,f}+\frac{\widehat{t}}{N^{2/3}\gamma _{V,f}} \end{aligned}$$

and in the same way we relate \(\widehat{s}\) to s. For all relevant values of f, t, and s we obtain the uniform estimates

$$\begin{aligned} \widehat{t}&=t+t\mathcal {O}(\Vert f\Vert /N)+N^{2/3}\gamma _{V,f}(b_V-b_{V,f})=t+\mathcal {O}(N^{\kappa -1/3})\, ,\nonumber \\ \widehat{s}&= s +\mathcal {O}(N^{\kappa -1/3}). \end{aligned}$$
(78)

Next we discuss the different regions in the (st) plane one by one.

(a) \(q \le s, t \le 2\): Application of Theorem 1.8. in [24] yields

$$\begin{aligned} \widehat{K}_{N,V,f; L} (s, t) = \frac{\gamma _{V,f}}{\gamma _V} \left( K_\text {Ai}(\widehat{s},\widehat{t}) + \mathcal {O} (N^{- 2/3}) \right) \end{aligned}$$

and the claim follows from (78) and from \(\gamma _{V, f}/\gamma _V = 1 + \mathcal {O}(N^{\kappa -1})\). Observe that we need a slightly modified version of the result in [24], because one only has that \(\widehat{s}\), \(\widehat{t}\) lie in some small neighborhood of [q, 2]. However, the proof in [24] shows that the statement also holds true in such an slightly enlarged neighborhood. This remark applies to the next two cases as well.

(b) \(1 \le s, t \le N^{\varepsilon }\): Theorem 1.8 of [24] together with \(\varepsilon < 2/15\) show

$$\begin{aligned} \widehat{K}_{N,V,f; L} (s, t) = K_\text {Ai}(\widehat{s},\widehat{t}) \left( 1 + \mathcal {O} (N^{-1/3}) \right) . \end{aligned}$$

The asymptotic formula for the Airy kernel (see e.g. [24, (4.23)]) implies the existence of some \(C > 1\) with

$$\begin{aligned} \frac{1}{C} \le K_\text {Ai}(s, t) (st)^{1/4} (\sqrt{s} + \sqrt{t}) \exp \left[ \textstyle {\frac{2}{3}}(s^{3/2} + t^{3/2})\right] \le C \end{aligned}$$
(79)

for all s, \(t \ge 1\). Using e.g. [24, Proposition 4.2] together with the asymptotics of the Airy function and its derivative (see [1, 10.4.59, 10.4.61]) one obtains a similar formula for the partial derivatives of the Airy kernel leading up to

$$\begin{aligned} |\nabla K_\text {Ai}(s, t)| = K_\text {Ai}(s, t) \mathcal {O}\left( \sqrt{s} + \sqrt{t}\right) = K_\text {Ai}(s, t) \mathcal {O}\left( N^{\varepsilon / 2}\right) . \end{aligned}$$

Since \(\varepsilon < 2/3 -2\kappa \) relation (79) implies for all \(s^*\), \(t^*\) with distance \(\mathcal {O}(N^{\kappa - 1/3})\) from s, resp. t

$$\begin{aligned} K_\text {Ai}(s^*, t^*) = K_\text {Ai}(s, t) \mathcal {O}(1)\, , \end{aligned}$$

yielding the claim.

(c) \(q \le s \le 1; 2 \le t \le N^{\varepsilon }\): This time application of [24] gives for \(\varepsilon < 2/15\)

$$\begin{aligned} \widehat{K}_{N,V,f; L} (s, t) = \left( K_\text {Ai}(\widehat{s},\widehat{t}) + \text {Ai}'(\widehat{t}) \mathcal {O} (N^{-1/3}) \right) \left( 1 +\mathcal {O} (N^{\kappa -1})\right) . \end{aligned}$$

The standard techniques used to derive the asymptotics of the Airy kernel in [24] (see in particular Proposition 4.2 there) yield the following estimates for the values of s and t considered in the present case:

$$\begin{aligned} K_\text {Ai}(s, t) = \text {Ai}'(t) \mathcal {O}(t^{-1})\, , \quad |\nabla K_\text {Ai}(s, t) | = \text {Ai}'(t) \mathcal {O}(t^{-1/2})\, , \quad \text {Ai}'(\widehat{t}) = \text {Ai}'(t) \mathcal {O}(1)\, , \end{aligned}$$

where we have used \(\varepsilon < 2/3 -2\kappa \) to derive the last relation. This completes the proof.\(\square \)

In the regime of moderate and large deviations Proposition 2 does not provide the leading order description, which is the content of the following proposition. As it turns out, it suffices to consider the Christoffel–Darboux kernel on the diagonal. In order to simplify the resulting formulas we now make use of the assumed evenness of Q and h that implies the evenness of V (cf. Remark 1 b)) and hence \(a_V=-b_V\).

Proposition 3

(cf. [24, Theorem 1.5]) Let V satisfy the assumptions of Proposition 2 and assume in addition \(a_V=-b_V\). In addition to the notations introduced in this appendix, recall the definition of the function \(\eta _V\) in (15). There exist constants \(c_0 > 0\) and \(C_0 \ge 1\) only depending on V such that for

  1. a)

    \(b_V + C_0N^{-2/3}< t < L\):

    $$\begin{aligned}&K_{N,V,f;L}(t,t)\\&\quad =\frac{b_V e^{-N\eta _V(t/b_V) + \mathcal {O}\left( \sqrt{t-b_V} \Vert f\Vert \right) }}{4\pi (t^2-b_V^2)} \left[ 1 + \mathcal {O} \left( \frac{\Vert f\Vert }{N\left|t-b_{V}\right|} \right) +\mathcal {O} \left( \frac{1}{N\left|t-b_{V}\right|^{3/2}} \right) \right] . \end{aligned}$$
  2. b)

    \(b_V + C_0N^{-2/3}< t < L - 1\):

    $$\begin{aligned}&\int _t^L K_{N,V,f;L}(y,y) dy\\&\quad =\frac{b_V^2 e^{-N\eta _V(t/b_V) + \mathcal {O}\left( \sqrt{t-b_V} \Vert f\Vert \right) }}{4\pi N (t^2-b_V^2) \eta _V'(t/b_V)} \left[ 1 + \mathcal {O} \left( \frac{\Vert f\Vert }{N\left|t-b_{V}\right|} \right) +\mathcal {O} \left( \frac{1}{N\left|t-b_{V}\right|^{3/2}} \right) \right] . \end{aligned}$$

All \(\mathcal {O}\) terms appearing in statements a) or b) are uniform in N, in \(t\in (b_V + C_0N^{-2/3}, L)\) resp. in \(t\in (b_V + C_0N^{-2/3}, L-1)\), and in \(f\in X_D\) with \(\Vert f\Vert \equiv \Vert f\Vert _{D}\le c_0N(t-b_V)\).

Proof

From the arguments given in the beginning of the proof of Proposition 2 it follows that we can apply the result [24, Theorem 1.5 (ii)] to the ensembles \(P_{N,V-f/N; L}\) provided that \(\Vert f\Vert /N\) is sufficiently small which we can always achieve by choosing the constant \(c_0\) in the statement of Proposition 3 small enough. Hence there exists a positive number \(\tilde{C}\) such that for all \(x>1+\tilde{C}N^{-2/3}\) with \(\lambda _{V,f}(x) \le L\)

$$\begin{aligned}&K_{N,V,f; L}(\lambda _{V,f}(x),\lambda _{V,f}(x))=\frac{1}{2\pi (b_{V,f}-a_{V,f})}e^{-N\eta _{V,f}(x)}\nonumber \\&\quad \times \frac{1}{x^2-1}\left[ 1+\mathcal {O}\left( \frac{1}{N\left|x - 1\right|^{3/2}}\right) \right] \end{aligned}$$

holds with the required uniformity. Setting \(x=\lambda _{V,f}^{-1}(t)\) we immediately arrive at

$$\begin{aligned}&K_{N,V,f; L}(t,t)=\frac{1}{2\pi } e^{-N\eta _{V,f}\left( \lambda _{V,f}^{-1}(t)\right) } \frac{b_{V,f}-a_{V,f}}{4(t-b_{V,f})(t-a_{V,f})}\nonumber \\&\quad \times \left[ 1 +\mathcal {O}\left( \frac{1}{N\left|t-b_{V,f}\right|^{3/2}}\right) \right] . \end{aligned}$$
(80)

Using again the smallness of \(\Vert f\Vert /[N(t-b_V)]\), we learn from (76) that for real exponents \(\alpha \) one has

$$\begin{aligned} (t-b_{V, f})^{\alpha } = (t-b_V)^{\alpha }\left[ 1 + \mathcal {O}_\alpha \left( \frac{\Vert f\Vert }{N\left|t-b_{V}\right|} \right) \right] \end{aligned}$$
(81)

and consequently

$$\begin{aligned}&\textstyle \frac{b_{V,f}-a_{V,f}}{4(t-b_{V,f})(t-a_{V,f})} \left[ 1 +\mathcal {O}\left( \frac{1}{N\left|t-b_{V,f}\right|^{3/2}}\right) \right] \nonumber \\&\quad = \frac{b_V}{2(t^2-b_V^2)} \left[ 1 + \mathcal {O}\left( \frac{\Vert f\Vert }{N\left|t-b_{V}\right|} \right) +\mathcal {O}\left( \frac{1}{N\left|t-b_{V}\right|^{3/2}}\right) \right] . \end{aligned}$$
(82)

We now turn to the exponential term. Applying (77) to the definition of \(\eta _V\) in (15), we see \(\eta _{V,f}=\eta _V(1+\mathcal {O}(\Vert f\Vert /N))\) uniformly on the domain of interest. Thus

$$\begin{aligned} \eta _{V,f}\left( \lambda _{V,f}^{-1}(t)\right) -\eta _{V}\left( \lambda _{V,f}^{-1}(t)\right) = \eta _{V}\left( \lambda _{V,f}^{-1}(t)\right) \mathcal {O}(\Vert f\Vert /N) . \end{aligned}$$

Moreover, it is straightforward to see that \(\lambda _V^{-1}(t) = t/b_V\), \(\lambda _V^{-1}(t) -\lambda _{V,f}^{-1}(t) = \mathcal {O}(\Vert f\Vert /N)\), \(\eta _V (x) = \mathcal {O}(\left|x-1\right|^{3/2})\), and \(\eta _V' (x) = \mathcal {O}(\left|x-1\right|^{1/2})\). Using in addition (81) we obtain

$$\begin{aligned} \eta _{V,f}\left( \lambda _{V,f}^{-1}(t)\right) -\eta _{V}\left( \lambda _{V,f}^{-1}(t)\right)&= \mathcal {O}\left( \left|t-b_{V}\right|^{3/2} \Vert f\Vert /N\right) , \\ \eta _{V}\left( \lambda _{V,f}^{-1}(t)\right) - \eta _{V}\left( t/b_V\right)&= \mathcal {O}\left( \left|t-b_{V}\right|^{1/2} \Vert f\Vert /N\right) , \end{aligned}$$

that leads to

$$\begin{aligned} e^{-N\eta _{V,f}\left( \lambda _{V,f}^{-1}(t)\right) } = e^{-N\eta _V(t/b_V) + \mathcal {O}\left( \sqrt{t-b_V} \Vert f\Vert \right) }. \end{aligned}$$
(83)

Combining (80) with (82) and (83) completes the proof of statement a).

The essential relation for establishing the second claim is

$$\begin{aligned} \int _t^L \frac{e^{-N\eta (y)}}{(y-b)(y-a)} dy = \frac{e^{-N\eta (t)}}{N(t-b)(t-a)\eta '(t)} \left[ 1 +\mathcal {O}\left( \frac{1}{N\left|t-b\right|^{3/2}}\right) \right] \end{aligned}$$
(84)

with an \(\mathcal {O}\) term that is uniform in N, t, and f in their respective domains and where \(a = a_{V,f}\), \(b = b_{V,f}\) and \(\eta = \eta _{V,f} \circ \lambda _{V,f}^{-1}\). We first sketch the derivation of (84) following [35, Lemma 4.8] (cf. [17]). Since \(\eta \) is strictly increasing the substitution \(u(y) := \eta (y)-\eta (t)\) is invertible and we obtain

$$\begin{aligned}&\int _t^L \frac{e^{-N\eta (y)}}{(y-b)(y-a)} dy \nonumber \\&\quad = e^{-N\eta (t)} \int _0^{\eta (L) - \eta (t)} e^{-N u}k(u)du, \ \ \ k(u) := \frac{1}{(y(u)-b)(y(u)-a)\eta '(y(u))}.\nonumber \end{aligned}$$

Observe that \(k(u) = k(0) + u k'(\tilde{u})\) for some \(\tilde{u} \in (0, u)\). In addition, straightforward estimates yield

$$\begin{aligned} \frac{k'(\tilde{u})}{k(\tilde{u})}&= \mathcal {O}\left( \frac{1}{\left|y(\tilde{u})-b\right|^{3/2}}\right) \quad \text {and by monotonicity of }k\text { and } y:\\ \ k'(\tilde{u})&= k(0) \mathcal {O}\left( \frac{1}{\left|t-b\right|^{3/2}}\right) . \end{aligned}$$

We arrive at

$$\begin{aligned} \int _t^L \frac{e^{-N\eta (y)}}{(y-b)(y-a)} dy = e^{-N\eta (t)} k(0) \left[ \frac{1}{N} \left( 1 + \mathcal {O}(e^{-dN})\right) + \frac{1}{N^2} \mathcal {O}\left( \frac{1}{\left|t-b\right|^{3/2}}\right) \right] \nonumber \end{aligned}$$

with \(d = \eta (L) - \eta (L-1) > 0\). Evaluating k(0) provides (84).

Note that (84) implies in particular that \(\int _t^L K_{N,V,f;L}(y,y) dy\) has exactly the same representation as \(K_{N,V,f;L}(t,t)\) in (80) except for additional factors N and \((\eta _{V,f} \circ \lambda _{V,f}^{-1})'(t)\) in the denominator. Thus the arguments in the proof of statement a) below (80) together with

$$\begin{aligned} (\eta _{V,f} \circ \lambda _{V,f}^{-1})'(t)&= \left( \frac{2}{b_{V,f} - a_{V,f}}\right) ^2 \sqrt{(t-b_{V,f})(t-a_{V,f})} G_{V,f}(\lambda _{V,f}^{-1}(t)) \nonumber \\&=\frac{1}{b_V} \eta _V'(t/b_V)\left[ 1 + \mathcal {O} \left( \frac{\Vert f\Vert }{N\left|t-b_{V}\right|} \right) \right] \end{aligned}$$
(85)

prove statement b).\(\square \)

Rights and permissions

Reprints and Permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Kriecherbauer, T., Venker, M. Edge statistics for a class of repulsive particle systems. Probab. Theory Relat. Fields 170, 617–655 (2018). https://doi.org/10.1007/s00440-017-0765-1

Download citation

  • Received:

  • Revised:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00440-017-0765-1

Keywords

  • Universality
  • Airy kernel
  • Tracy–Widom distribution
  • Random matrices
  • Moderate deviations
  • Large deviations

Mathematics Subject Classification

  • 60B20
  • 82C22