Abstract
We derive the existence of infinite level GREM-like K-processes by taking the limit of a sequence of finite level versions of such processes as the number of levels diverges. The main step in the derivation is obtaining the convergence of the sequence of underlying finite level clock processes. This is accomplished by perturbing these processes so as to turn them into martingales, and resorting to martingale convergence to obtain convergence for the perturbed clock processes; nontriviality of the limit requires a specific choice of parameters of the original process; we conclude the step by showing that the perturbation washes away in the limit. The perturbation is done by inserting suitable factors into the expression of the clocks, as well as rescaling the resulting expression suitably; the existence of such factors is itself established through martingale convergence.
Similar content being viewed by others
Notes
This is one of the phenomenological models alluded to above.
As pointed out in earlier papers, \(\infty \)’s have also a physical meaning: they represent the high energy configurations lumped together.
Under the resulting dynamics, the levels for which \(\Xi _k=\) id would however not be at low temperature, and their corresponding coordinates would be equal \(\infty \) at all times, the remaining coordinates undergoing an ordinary GK-process.
References
Barlow, M.T., Černý, J.: Convergence to fractional kinetics for random walks associated with unbounded conductances. Probab. Theory Relat. Fields 149, 639–673 (2011)
Barreto-Souza, W., Fontes, L.R.G.: Long-range trap models on \({\mathbb{Z}}\) and quasistable processes. J. Theor. Probab. 28, 1500–1519 (2015)
Ben Arous, G., Černý, J.: Scaling limit for trap models on \({{\mathbb{Z}}}^d\). Ann. Probab. 35, 2356–2384 (2007)
Ben Arous, G., Bovier, A., Černý, J.: Universality of the REM for dynamics of mean-field spin glasses. Commun. Math. Phys. 282, 663–695 (2008)
Ben Arous, G., Černý, J., Mountford, T.: Aging in two-dimensional Bouchaud’s model. Probab. Theory Relat. Fields 134, 1–43 (2006)
Ben Arous, G., Bovier, A., Gayrard, V.: Glauber dynamics of the random energy model. II. Aging below the critical temperature. Commun. Math. Phys. 236(1), 1–54 (2003)
Bezerra, S.C., Fontes, L.R.G., Gava, R.J., Gayrard, V., Mathieu, P.: Scaling limits and aging for asymmetric trap models on the complete graph and K processes. ALEA Lat. Am. J. Probab. Math. Stat. 9, 303–321 (2012)
Bouchaud, J.-P., Dean, D.S.: Aging on Parisi’s tree. J. Phys. I France 5, 265–286 (1995)
Bovier, A., Gayrard, V.: Convergence of clock processes in random environments and ageing in the p-spin SK model. Ann. Probab. 41, 817–847 (2013)
Bovier, A., Kurkova, I.: Derrida’s Generalised Random Energy models I. Models with finitely many hierarchies. Ann. I. H. Poincaré - PR 40, 439–480 (2004)
Bovier, A., Kurkova, I.: Derrida’s Generalised Random Energy models I. Models with continuous hierarchies. Ann. I. H. Poincaré - PR 40, 481–495 (2004)
Bovier, A., Kurkova, I.: Much ado about Derrida’s GREM. In: Spin Glasses, Lecture Notes in Math., vol. 1900, pp. 81–115. Springer (2007)
Černý, J., Wassmer, T.: Aging of the metropolis dynamics on the random energy model. Probab. Theory Relat. Fields 167, 253–303 (2017)
Cortines, A., Gold, J., Louidor, O.: Dynamical freezing in a spin glass system with logarithmic correlations. Electron. J. Probab. 23, 59 (2018)
Derrida, B.: A generalization of the random energy model which includes correlations between energies. J. Phys. Lett. 46, L401–L407 (1985)
Derrida, B.: Random-energy model: limit of a family of disordered models. Phys. Rev. Lett. 45, 79–82 (1980)
Enriquez, N., Sabot, C., Zindy, O.: Aging and quenched localization for one-dimensional random walks in random environment in the sub-ballistic regime. Bull. Soc. Math. France 137, 423–452 (2009)
Ethier, S.N., Kurtz, T.G.: Markov Processes. Characterization and Convergence. Wiley, New York (1986)
Fontes, L.R.G., Isopi, M., Newman, C.M.: Random walks with strongly inhomogeneous rates and singular diffusions: convergence, localization and aging in one dimension. Ann. Probab. 30, 579–604 (2002)
Fontes, L..R..G., Isopi, M., Newman, C..M., Stein, D.: Aging in 1\({}{D}\) discrete spin models and equivalent systems. Phys. Rev. Lett. 87(11), 110201-1-110201–4 (2002)
Fontes, L.R.G., Gava, R.J., Gayrard, V.: The K-process on a tree as a scaling limit of the GREM-like trap model. Ann. Appl. Probab. 24, 857–897 (2014)
Fontes, L.R.G., Gayrard, V.: Asymptotic behavior and aging of a low temperature cascading 2-GREM dynamics at extreme time scales. Electron. J. Probab. 24, 142 (2019)
Fontes, L.R.G., Lima, P.H.S.: Convergence of symmetric trap modelsin the hypercube. In: Sidoravičius, V. (ed.) New Trends in Mathematical Physics, pp. 285–297. Springer, Dordrecht (2009)
Fontes, L.R.G., Mathieu, P.: K-processes, scaling limit and aging for the trap model on the complete graph. Ann. Probab. 36, 1322–1358 (2008)
Fontes, L.R.G., Mathieu, P.: On the dynamics of trap models in \({{\mathbb{Z}}}^d\). Proc. Lond. Math. Soc. 108, 1562–1592 (2014)
Fontes, L.R.G., Peixoto, G.R.C.: Elementary results on K processes with weights. Markov Process. Relat. Fields 19, 343–370 (2013)
Gava, R.J.: Scaling limits of trap models on a tree. Ph.D. thesis; University of São Paulo (in portuguese) (2011)
Gayrard, V.: Aging in metropolis dynamics of the REM: a proof. Probab. Theory Relat. Fields 174, 501–551 (2019)
Gayrard, V., Gün, O.: Aging in the GREM-like trap model. Markov Process. Relat. Fields 22, 165–202 (2016)
Gayrard, V., Švejda, A.: Convergence of clock processes on infinite graphs and aging in Bouchaud’s asymmetric trap model on \({{\mathbb{Z}}}^d\). ALEA Lat. Am. J. Probab. Math. Stat. 11, 781–822 (2014)
Jara, M., Landim, C., Teixeira, A.: Quenched scaling limits of trap models. Ann. Probab. 39, 176–223 (2011)
Jara, M., Landim, C., Teixeira, A.: Universality of trap models in the ergodic time scale. Ann. Probab. 42, 2497–2557 (2014)
Kingman, J.F.C.: Poisson Processes. Clarendon, Oxford (1993)
Laue, G.: Remarks on the relation between fractional moments and fractional derivatives of characteristic functions. J. Appl. Probab. 17, 456–466 (1980)
Mathieu, P., Mourrat, J.-C.: Aging of asymmetric dynamics on the random energy model. Probab. Theory Relat. Fields 161, 351–427 (2015)
Ruelle, D.: A mathematical reformulation of Derrida’s REM and GREM. Commun. Math. Phys. 108, 225–239 (1987)
Samorodnitsky, G., Taqqu, M.: Stable Non-Gaussian Random Processes: Stochastic Models with Infinite Variance. Chapman and Hall, New York (1994)
Sasaki, M., Nemoto, K.: Analysis on aging in the generalized random energy model. J. Phys. Soc. Jpn. 69, 3045–3050 (2000)
Acknowledgements
The authors would like to thank NUMEC-USP for the hospitality shown to the second author during the development of this work, and also two anonymous referees, whose thoughtful comments and criticism much helped improve the presentation.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Aernout van Enter.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
L. R. Fontes: Partially supported by CNPq grant 307884/2019-8, and FAPESP grant 2017/10555-0. G. R. C. Peixoto: Supported by CNPq fellowship 140177/2012-4 and a CAPES institutional fellowship.
Appendices
Martingality Inducing Factors—Proofs
Proof of Proposition 2.1
Without loss of generality, let us assume \(k = 0\), and define the following random variables. For \(n > 0\) and \(\lambda > 0\), let
Let us show that these random variables form a martingale with respect to the filtration \(({\mathcal {D}}_n)\), where \({\mathcal {D}}_n\) is the \(\sigma \)-algebra generated by \(\{{\tilde{\gamma }}_j(x|_j): x|_j \in {\mathbb {N}}^j, \,j \le n \}\). Let us at this point note that it readily follows from Proposition B.2 below that
is a Poisson point process, with intensity measure \({\tilde{\mu }}(dt) := \frac{c^*_k}{t^{1+\alpha _k}}\,dt, t > 0\).
We may thus use Campbell’s Theorem (see e.g. [33], Section 3.2) in the following:
Since \((Z_n(1))\) is a positive martingale, it converges a.s.; thus, its exponent must converge as well.
The second claim can be obtained by using (A.2) to explicitly compute:
since \(\alpha _n\rightarrow 1\) as \(n\rightarrow \infty \), where the latter equality follows from Campbell’s Theorem, once we note that \(\{[{\tilde{\gamma }}_1(x|_1)]^{\alpha _2},\,x|_1\in {\mathbb {N}}\}\) is a Poisson point process with intensity measure \(\mu ^*_1(dt)=(c^*_1/\alpha _2)/t^{1+\alpha _1/\alpha _2}\,dt\).
To establish the final claim, let us fix a realization such that (2.2) holds for every \(x|_k \in {\mathbb {N}}^k\), \(k\ge 0\), and fix an arbitrary \(\epsilon > 0\), then:
Let us treat these two terms separately, and show that their respective limits exist in probability. For (A.4), since the outermost sum is finite and \(\alpha _n \rightarrow 1\) as \(n \rightarrow \infty \), we have that
Let \({\mathcal {D}}'_k={\mathcal {D}}'_k(x|_k)\) be the \(\sigma \)-algebra generated by
\(\{{\tilde{\gamma }}_{k+1}(x|_{k+1}) : x_{k+1} \in {\mathbb {N}}\}\). Using (A.3), we can compute the Laplace transform of (A.5) as:
We used Campbell’s Theorem in the last equality. With this we can conclude that the expression in (A.5) converges in probability to 0 as \(\epsilon \rightarrow 0\).
Finally, note that we have shown that (A.4) \(+\) (A.5) converges in probability to both \(W_{k}(x|_k)\) and \(\sum _{x_{k+1}} {\tilde{\gamma }}_{k+1}(x|_{k+1}) W_{k+1}(x|_{k+1})\) as \(\epsilon \rightarrow 0\), which implies that these latter two quantities are almost surely equal. \(\square \)
Proof of Corollary 3.1
Let \(A_n := \{ x|_n \in {\mathbb {N}}^n: \hat{\gamma }_n(x|_n) > 1 \}\) and \(m_n := \max \{ \hat{\gamma }_n(x|_n): x|_n \in {\mathbb {N}}^n\}\). Using Proposition 2.1, we get that almost surely
Since \(W_0(\emptyset ) < \infty \), we have that \(\limsup _n |A_n| < \infty \) and \(\limsup _n m_n < \infty \) a.s. Thus,
From the previous remark, the limsup of the first term of this last sum is bounded a.s. by \(W^2_0(\emptyset )\), while the \(\limsup \) of the second term is bounded by \(W_0(\emptyset )\). \(\square \)
Proof of Lemma 3.2
Recall from Proposition 2.1 that \(W_k(x|_k)\) has a stable distribution, which may be denoted, in the notation of [37, Sections 1.1. and 1.2], as follows: \(W_k(x|_k)\sim S_{\alpha _{k+1}}(\sigma ,0,1)\), with \(\sigma ^{\alpha _{k+1}}= \cos \left( \frac{\pi \alpha _{k+1} }{2}\right) \). We may thus write the characteristic function of \(W_k(x|_k) - 1\) as follows:
Theorem 2.2 from [34] states that
Note that \( \frac{\alpha _k}{\cos \left( \frac{\pi \alpha _k}{2}\right) \Gamma (1-\alpha _k)}\) converges to \(\frac{2}{\pi }\) as \(k\rightarrow \infty \). So we are left with showing that the real part of the above integral vanishes as \(k\rightarrow \infty \). Fixing an arbitrary \(\epsilon > 0\), we write
To control (A.8), note that the integrand converges to zero as \(k \rightarrow \infty \) and can be bounded by \(2/u^{1+\alpha _k} \le 2/u^{3/2}\) for large enough k. Therefore, by dominated convergence, (A.8) vanishes as \(k \rightarrow \infty \) for any \(\epsilon > 0\).
We control (A.7), using \(1-e^{-x} \le x\), as follows. It is bounded above by
Since \(\frac{1-\alpha _{k+1}}{1-\alpha _k} \xrightarrow {k \rightarrow \infty } 0\) by hypothesis and \(\cos (\pi x /2)/(1-x) \xrightarrow {x \rightarrow 1} \frac{\pi }{2}\), we can rewrite the leftmost term from (A.9) as
To control the integral in (A.9), let us first remark that there exists an \(\epsilon _0 > 0\) such that if \(|x| < \epsilon _0\), then \(-\log \cos x < x^2\). Thus, for small enough \(\epsilon > 0\) we can write
\(\square \)
Auxiliary Results; Proof of Lemma 2.2
We collect here a few elementary results used in the main part of the text, and provide a proof of Lemma 2.2.
Lemma B.1
Recall (1.4,1.4b). Almost surely \(X_{k-1}(\sigma _i^{k, x}) \in {\mathbb {N}}^{k-1}\), \(k,x\ge 1\).
Proof
This is obvious for \(k=1\); let then \(k\ge 2\). We first note that for \(1\le j < k\), \(X_{k-1, j}(s) = \infty \) only if s belongs to the range of \(\theta _j^{k-1}\), and this set has null Lebesgue measure, since \(\theta _j^{k-1}\) increases only by jumps (at the points of the Poisson processes \( N^{j,x}\), \(x \in {\mathbb {N}}\)). Combining the latter point with that fact that the Poisson processes \( N^{k,x}\), \(x \in {\mathbb {N}}\), are independent from \(\theta _j^{k-1}\), we readily find that the probability that \(\sigma _i^{k, x}\) belongs to the range of \(\theta _j^{k-1}\) for some i, x vanishes, and the result follows. \(\square \)
Proposition B.2
Let \(\{\xi _i: i \in {\mathbb {N}}\}\) be the marks of a Poisson point process on \({\mathbb {R}}^+\) with intensity measure \(\mu (d t) = \frac{c}{t^{1+\alpha }}\,dt\), \(t >0\), with some \(\alpha \in (0, 1)\) and \(c > 0\). Let \(\{ R_i : i \in {\mathbb {N}}\}\) be i.i.d. positive random variables, independent from the Poisson process, such that \({\mathbb {E}}(R_1^\alpha ) <\infty \). Then \(\{ R_i\xi _i : i \in {\mathbb {N}}\}\) is also a Poisson point process, with intensity measure \({\mathbb {E}}(R_1^\alpha ) \mu \).
Proof
Define \(S := \{ (\xi _i, R_i): i \in {\mathbb {N}}\}\) and note that S is the set of the marks of a Poisson point process on the first quadrant of \({\mathbb {R}}^2\) with intensity measure \(\pi = \mu \times \nu \), where \(\nu \) is the probability measure of \(R_1\).
Note that \(T(x, y) = xy\) is a continuous transformation with no accumulation point except 0, so \(\{ T(s): s \in S\} = \{R_i\xi _i : i \in {\mathbb {N}}\}\) is a Poisson point process (see eg. [33], Mapping Theorem, Section 2.3); its intensity measure is readily determined to be \({\mathbb {E}}(R_1^\alpha ) \mu \). \(\square \)
Proposition B.3
Let R be a positive random variable with Laplace transform \(\phi (\lambda ) := {\mathbb {E}}(e^{-\lambda R}) = e^{-c \lambda ^\alpha }\), \(\lambda >0\), for some \(c > 0\), \(\alpha \in (0, 1)\). Then for \(0< \beta < \alpha \):
Proof
By Fubini and the fact that \(\phi ^\prime (\lambda ) = - {\mathbb {E}}(R e^{-\lambda R})\), we have that
It follows that \( {\mathbb {E}}(R^\beta ) = - \frac{1}{\Gamma (1-\beta )} \int _0^\infty \phi ^\prime (\lambda ) \lambda ^{-\beta } d \lambda , \) and differentiating \(\phi (\lambda ) = e^{-c \lambda ^\alpha }\) and substituting in the latter expression yields the result. \(\square \)
We now justify a claim made at the beginning.
Corollary B.4
Claim (1.3) holds true.
Proof
By the definition of \(\gamma _\cdot (\cdot )\), we have that \({\mathcal {R}}_k(x|_{k-1}):=\sum _{x_k\in {\mathbb {N}}}\gamma _k(x|_{k-1},x_k)\) is an \(\alpha _k\)-stable random variable for every \(x|_{k-1}\in {\mathbb {N}}^{k-1}\). It follows from Proposition B.2 that for every \(x|_{k-2}\in {\mathbb {N}}^{k-2}\), \(\left\{ \gamma _k(x|_{k-2},x_{k-1})\sum _{x_k\in {\mathbb {N}}}\gamma _k(x|_{k-2},x_{k-1},x_k),\,x_{k-1}\in {\mathbb {N}}\right\} \) is a Poisson point process with intensity measure const \(\times \mu _{k-1}\), with const \(={\mathbb {E}}({\mathcal {R}}_k(1,\ldots ,1)^{\alpha _{k-1}})\) finite by Proposition B.3. The result follows by iteration. \(\square \)
Proof of Lemma 2.2
(2.11) is established in the proof of Lemma 4.6 in [21]. We present an explicit argument for \({{\theta }}_1^n\); the case of \({\hat{\theta }}_1^n\) is similar. We proceed by induction on n. The case \(n = 1\) is a direct consequence of the independence and stationarity of increments of Poisson point processes, as well as of the fact that the variables used to construct \(\theta _1^1 = {\Xi }_1\) are independent.
Assuming \(n \ge 2\), note that for \(t, s > 0\):
Fix \(0 = t_0< t_1< \cdots < t_k\) and let us look at the joint law of \((\theta _1^n(t_i) - \theta _1^{n}(t_{i-1}))_{i = 1, \cdots , k}\).
\(\theta _1^n(t_i) - \theta _1^{n}(t_{i-1})\), for varying values of i, depends on disjoint intervals of the Poisson processes \(N^{n, x}\), each one with length \(\theta _1^{n-1}(t_{i}) - \theta _1^{n-1}(t_{i-1})\). Because of the induction hypothesis these lengths are independent and each one has the same law as \(\theta _1^{n-1}(t_i - t_{i-1})\).
Note that, as follows from Lemma B.1, we have that for fixed \(t > 0\), \(X_{n-1}(\theta _1^{n-1}(t)) = (\infty , \cdots , \infty )\) a.s. The instant \(\theta _1^{n-1}(t)\) is thus a renewal time for \(X_{n-1}\), and, by construction, the law of \(X_{n-1}\) at such a renewal is the same as at time 0.
\(\theta _1^n(t_i) - \theta _1^n(t_{i-1})\) also depends on the values of \(X_{n-1}(r)\) for \(r \in (\theta _1^{n-1}(t_{i-1}), \theta _1^{n-1}(t_i))\). Both ends of this interval are renewals, so what happens to \(X_{n-1}\) inside such an interval is independent of what happens on other such intervals (namely, \((\theta _1^{n-1}(t_{j-1}),\theta _1^{n-1}(t_j))\), \(j\ne i\)).
Therefore we have that \(\theta _1^n(t_i) - \theta _1^n(t_{i-1})\), \(i\ge 1\), are independent, and the law of each \(\theta _1^n(t_i) - \theta _1^n(t_{i-1})\) depends only on \(t_i - t_{i-1}\). That is, \(\theta _1^n\) has independent and stationary increments. \(\square \)
Rights and permissions
About this article
Cite this article
Fontes, L.R., Peixoto, G.R.C. Infinite Level GREM-Like K-Processes Existence and Convergence. J Stat Phys 182, 50 (2021). https://doi.org/10.1007/s10955-021-02713-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10955-021-02713-5
Keywords
- Spin glass dynamics
- Mean field spin glasses
- Generalized Random Energy Model
- Trap models
- Infinite volume
- K-processes
- Infinitely many levels