Abstract
We investigate the energy landscape of the mixed even p-spin model with Ising spin configurations. We show that for any given energy level between zero and the maximal energy, with overwhelming probability there exist exponentially many distinct spin configurations such that their energies stay near this energy level. Furthermore, their magnetizations and overlaps are concentrated around some fixed constants. In particular, at the level of maximal energy, we prove that the Hamiltonian exhibits exponentially many orthogonal peaks. This improves the results of Chatterjee (Disorder chaos and multiple valleys in spin glasses, 2009) and Ding et al. (Ann Probab 43(6):3468–3493, 2015), where the former established a logarithmic size of the number of the orthogonal peaks, while the latter proved a polynomial size. Our second main result obtains disorder chaos at zero temperature and at any external field. As a byproduct, this implies that the fluctuation of the maximal energy is superconcentrated when the external field vanishes and obeys a Gaussian limit law when the external field is present.
Similar content being viewed by others
References
Amit, D.J., Gutfreund, H., Sompolinsky, H.: Spin-glass models of neural networks. Phys. Rev. A 32, 1007–1018 (1985)
Auffinger, A., Ben Arous, G.: Complexity of random smooth functions on the high-dimensional sphere. Ann. Probab. 41(6), 4214–4247 (2013)
Auffinger, A., Ben Arous, G., Černý, J.: Random matrices and complexity of spin glasses. Commun. Pure Appl. Math. 66(2), 165–201 (2013)
Auffinger, A., Chen, W.-K.: The Parisi formula has a unique minimizer. Commun. Math. Phys. 335(3), 1429–1444 (2015)
Auffinger, A., Chen, W.-K.: Parisi formula for the ground state energy in the mixed p-spin model. ArXiv e-prints, June (2016)
Auffinger, A., Chen, W.-K.: The Legendre structure of the Parisi formula. Commun. Math. Phys. 348(3), 751–770 (2016)
Ben Arous, G., Gayrard, V., Kuptsov, A.: A new REM conjecture. In: In and out of equilibrium. 2, volume 60 of Progr. Probab., pp. 59–96. Birkhäuser, Basel (2008)
Bhamidi, S., Dey, P.S., Nobel, A.B.: Energy Landscape for large average submatrix detection problems in Gaussian random matrices. ArXiv e-prints, November (2012)
Boucheron, S., Lugosi, G., Massart, P.: Concentration inequalities. Oxford University Press, Oxford. A nonasymptotic theory of independence, With a foreword by Michel Ledoux (2013)
Bovier, A., Klimovsky, A.: Fluctuations of the partition function in the generalized random energy model with external field. J. Math. Phys. 49(12), 125202, 27 (2008)
Bovier, A., Kurkova, I.: Derrida’s generalised random energy models. I. Models with finitely many hierarchies. Ann. Inst. H. Poincaré Probab. Stat. 40(4), 439–480 (2004)
Bovier, A., Kurkova, I.: Derrida’s generalized random energy models. II. Models with continuous hierarchies. Ann. Inst. H. Poincaré Probab. Stat. 40(4), 481–495 (2004)
Bovier, A., Kurkova, I.: Energy statistics in disordered systems: the local REM conjecture and beyond. Acta Phys. Pol. B 36(9), 2621–2634 (2005)
Bovier, A., Kurkova, I.: Local energy statistics in disordered systems: a proof of the local REM conjecture. Commun. Math. Phys. 263(2), 513–533 (2006)
Bovier, A., Kurkova, I.: A tomography of the GREM: beyond the REM conjecture. Commun. Math. Phys. 263(2), 535–552 (2006)
Bovier, A., Kurkova, I.: Local energy statistics in spin glasses. J. Stat. Phys. 126(4–5), 933–949 (2007)
Bruckner, A.: Differentiation of real functions, volume 5 of CRM Monograph Series, 2nd edn. American Mathematical Society, Providence, RI (1994)
Charbonneau, P., Kurchan, J., Parisi, G., Urbani, P., Zamponi, F.: Fractal free energy landscapes in structural glasses. Nature Commun. 5 (2014). doi:10.1038/ncomms4725
Chatterjee, S.: Chaos, concentration, and multiple valleys. ArXiv e-prints, October (2008)
Chatterjee, S.: Disorder chaos and multiple valleys in spin glasses. ArXiv e-prints, July (2009)
Chatterjee, S.: Superconcentration and related topics. Springer Monographs in Mathematics. Springer, Cham (2014)
Chen, W.-K.: Disorder chaos in the Sherrington–Kirkpatrick model with external field. Ann. Probab. 41(5), 3345–3391 (2013)
Chen, W.-K.: Chaos in the mixed even-spin models. Commun. Math. Phys. 328(3), 867–901 (2014)
Chen, W.-K.: Variational representations for the Parisi functional and the two-dimensional Guerra–Talagrand bound. ArXiv e-prints, January (2015)
Chen, W.-K., Hsieh, H.-W., Hwang, C.-R., Sheu, Y.-C.: Disorder chaos in the spherical mean-field model. J. Stat. Phys. 160(2), 417–429 (2015)
Chen, W.-K., Sen, A.: Parisi formula, disorder chaos and fluctuation for the ground state energy in the spherical mixed \(p\)-spin models. ArXiv e-prints, December (2015)
Choromanska, A., Henaff, M., Mathieu, M., Ben Arous, G., LeCun, Y.: The loss surfaces of multilayer networks. In: Lebanon, G., Vishwanathan, S.V.N. (eds.) Proceedings of the eighteenth international conference on artificial intelligence and statistics, AISTATS 2015, San Diego, California, USA, May 9–12, 2015, volume 38 of JMLR workshop and conference proceedings. JMLR.org (2015)
Dembo, A., Montanari, A., Sen, S.: Extremal cuts of sparse random graphs. ArXiv e-prints, March (2015)
Ding, J., Eldan, R., Zhai, A.: On multiple peaks and moderate deviations for the supremum of a Gaussian field. Ann. Probab. 43(6), 3468–3493 (2015)
Fyodorov, Y.V.: Complexity of random energy landscapes, glass transition, and absolute value of the spectral determinant of random matrices. Phys. Rev. Lett. 92(24):240601, 4, (2004)
Fyodorov, Y.V., Williams, I.: Replica symmetry breaking condition exposed by random matrix calculation of landscape complexity. J. Stat. Phys. 129(5–6), 1081–1116 (2007)
Gamarnik, D., Li, Q.: Finding a Large Submatrix of a Gaussian Random Matrix. ArXiv e-prints, February (2016)
Jagannath, A., Tobasco, I.: A dynamic programming approach to the Parisi functional. Proc. Am. Math. Soc. 144(7), 3135–3150 (2016)
Jagannath, A., Tobasco, I.: Low temperature asymptotics of spherical mean field spin glasses. ArXiv e-prints, February (2016)
Karatzas, I., Shreve, S.E.: Brownian motion and stochastic calculus, volume 113 of graduate texts in mathematics, 2nd edn. Springer, New York (1991)
Kurchan, J., Parisi, G., Virasoro, M.A.: Barriers and metastable states as saddle points in the replica approach. J. Phys. I 3(8), 1819–1838 (1993)
Lee, Y.C., Doolen, G., Chen, H.H., Sun, G.Z., Maxwell, T., Lee, H.Y., Giles, C.L.: Machine learning using a higher order correlation network. Phys. D Nonlinear Phenom. 22(1), 276–306 (1986)
Mézard, M., Montanari, A.: Information, physics, and computation. Oxford graduate texts. Oxford University Press, Oxford (2009). Autre tirage: 2010, 2012
Mézard, M., Parisi, G., Virasoro, M.A.: Spin glass theory and beyond, volume of 9 World Scientific Lecture Notes in Physics. World Scientific Publishing Co., Inc., Teaneck (1987)
Panchenko, D.: On differentiability of the Parisi formula. Electron. Commun. Probab. 13, 241–247 (2008)
Panchenko, D.: The Sherrington–Kirkpatrick Model. Springer Monographs in Mathematics. Springer, New York (2013)
Panchenko, D.: The Parisi formula for mixed \(p\)-spin models. Ann. Probab. 42(3), 946–958 (2014)
Peretto, P., Niez, J.J.: Long term memory storage capacity of multiconnected neural networks. Biol. Cybern. 54(1), 53–63 (1986)
Rizzo, T.: Chaos in mean-field spin-glass models. In: Spin glasses: statics and dynamics, volume 62 of Progr. Probab. pp. 143–157. Birkhäuser Verlag, Basel (2009)
Subag, E.: The complexity of spherical \(p\)-spin models—a second moment approach. ArXiv e-prints, April (2015)
Subag, E.: The geometry of the Gibbs measure of pure spherical spin glasses. ArXiv e-prints, April (2016)
Subag, E., Zeitouni, O.: The extremal process of critical points of the pure \(p\)-spin spherical spin glass model. ArXiv e-prints, September (2015)
Talagrand, M.: The Parisi formula. Ann. Math. (2) 163(1), 221–263 (2006)
Tyrrell Rockafellar, R.: Convex analysis. Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ. Reprint of the 1970 original, Princeton Paperbacks (1997)
Acknowledgements
The authors are indebted to N. Krylov and M. Safonov for illuminating discussions on the regularity properties of the Parisi PDE. They thank S. Chatterjee, D. Panchenko, and the anonymous referees for a number of suggestions regarding the presentation of the paper. The research of W.-K. C. is partially supported by NSF Grant DMS-16-42207 and Hong Kong Research Grants Council GRF-14-302515. The research of M. H. and G. L. is partially supported by NSF Grant DMS-14-18386.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
This appendix is devoted to establishing Proposition 2. We prove a priori estimate first:
Lemma 10
Let \(0<r_0<r_1<r_2<\infty .\) Suppose that \(\kappa _1,\kappa _2\in L^\infty ([r_0 ,r_2]\times \mathbb {R})\) and \(g\in L^\infty (\mathbb {R})\) with \(\Vert \kappa _i\Vert _\infty \le C_i\) for \(i=1,2\) and \(\Vert g\Vert _\infty \le C_0\). Assume that u is the classical solution to
with initial condition \(u(r_0 ,x)=g(x).\) If
is continuous on \([r_0 ,{r_2} ]\), then there exists a nonnegative continuous function F on \([0,\infty )^3\) depending only on \(r_0 ,r_2\) such that
Proof
From the Duhamel principle,
where for any \(f\in L^\infty (\mathbb {R})\),
A direct computation leads to
Note that
from which
A similar computation also yields that
As a result,
where
From the Gronwall inequality,
where
and the last inequality was obtained by using \(\int _{r_0 }^r1/\sqrt{(r-w)(w-r_0 )}dw =\pi .\) This finishes our proof. \(\square \)
With the help of Lemma 10, we obtain some controls on the spacial derivatives of \(\Phi _\gamma .\)
Lemma 11
Let \(s_1\in (0,1)\). For any \(\gamma \in \mathcal {U}_d,\)
and for \(k\ge 2\)
where \(F_k\) is nonnegative continuous on \([0,\infty )\) depending on \(s_1\) only and independent of \(\gamma .\)
Proof
Assume that \(\gamma \in \mathcal {U}_d.\) One can explicitly solve \(\Phi _\gamma \) in the classical sense by performing the Cole-Hopf transformation. In fact, if \(\gamma =\sum _{i=0}^{m}a_i1_{[q_{i},q_{i+1})}\) for some sequences
then for any \(1\le i\le m,\)
for all \((s,x)\in [q_i,q_{i+1})\times \mathbb {R},\) where z is a standard normal random variable. Using the initial condition \(\Phi _\gamma (1,x)=|x|\) and an iteration argument, it can be easily checked that for any \(k\ge 0\), \(\partial _x^k\Phi \in C([0,1)\times \mathbb {R})\) and for any \(k\ge 0\) and \(s_0\in (0,1)\), \(\partial _s\partial _x^k\Phi (s,\cdot ) \in C(\mathbb {R})\). Furthermore, it can be verified that (62) holds, and for any \(s_0\in (0,1)\), there exists a constant C such that
This inequality implies that \(s\mapsto \Vert \partial _x^k\Phi _\gamma (s,\cdot )\Vert _\infty \) is a continuous function for \(s\in [0,1).\) Now define \(\zeta (s)=(\xi '(1)-\xi '(s))/2\). Fix \(0<s_1<s_0<1\). Set \(r_0=\zeta (s_0)\), \(r_1=\zeta (s_1)\), and \(r_2=\zeta (0)\). Evidently, the function \(u(r,x):=\partial _{x}\Phi _\gamma (\zeta ^{-1}(r),x)\) satisfies
for \((r,x)\in [r_0,r_2]\times \mathbb {R}\) with initial condition \(\partial _xu(r_0,x)=\partial _x\Psi _\gamma (\zeta ^{-1}(r_0),x),\) where
Since \(\Vert \kappa _1\Vert _\infty \le \gamma (\zeta ^{-1}(r_0))<\infty \), we can apply Lemma 10 to get that
where F is a nonnegative continuous function on \([0,\infty )^3\) depending only on \(s_1.\) Letting \(F_1(y)=F(1,y,0)\) gives (63) with \(k=1.\) For \(k\ge 2,\) note that
where K(s, x) is the sum of products of spatial derivatives of \(\Phi _\gamma \) of order at most k. One may use a similar argument as the case \(k=1\) together with an induction procedure to obtain the proof of (63) for all \(k\ge 2\). \(\square \)
Following a similar definition of the weak solution for the original Parisi PDE in Jagannath and Tobasco [33], we define the weak solution of the Parisi PDE (5) as follows.
Definition 1
(Weak solution) Let \(\gamma \in \mathcal {U}.\) Let \(\Phi \) be a continuous function on \([0,1]\times \mathbb {R}\) with essentially bounded weak derivative \(\partial _x\Phi .\) We say that \(\Phi \) is a weak solution to
on \([0,1)\times \mathbb {R}\) with \(\Phi (1,x)=|x|\) if
for all smooth \(\phi \) on \((0,1]\times \mathbb {R}\) with compact support.
Proof of Proposition 2
Let us pick any \((\gamma _n)_n\subset \mathcal {U}_d\) with weak limit \(\gamma .\) From the estimates in Lemma 11, one readily sees that for any \(k\ge 0,\) the sequence \((\partial _x^k\Phi _{\gamma _n})_{n\ge 1}\) is equicontinuous and uniformly bounded on \([0,s_1)\times \mathbb {R}\) for all \(s_1\in (0,1)\). Thus, from the Arzela–Ascoli theorem combined with a diagonal process, on any \([0,s_1]\times [-M,M],\) \((\partial _x^k\Phi _{\gamma _n})_{n\ge 1}\) converges uniformly on any compact subset of \([0,1)\times \mathbb {R}\) for any \(k\ge 0\). Here, without loss of generality, we use the same sequence \((\gamma _n)_{n\ge 1}\) instead of adapting a subsequence for notational convenience. The above discussion makes the following well-defined,
for \((s,x)\in [0,1]\times \mathbb {R}.\) Note that for \(s\in [0,1)\), \(\Phi _{\gamma }(s,x)\) is differentiable in \(x\in \mathbb {R}\) and satisfies \(\Vert \partial _x\Phi _\gamma (s,\cdot )\Vert _\infty <\infty \). Since \(\Phi _{\gamma _n}\) satisfies
for \((s,x)\in [0,1)\times \mathbb {R}\) with boundary condition \(\Phi _{\gamma _n}(1,x)=|x|,\) passing to the limit gives
for all smooth functions \(\phi \) on \((0,1]\times \mathbb {R}\) with compact support. Therefore, \(\Phi _\gamma \) exists in the weak sense and the fact that it satisfies the Lipschitz property (7) follows by applying (9) and noting that |x| has Lipschitz constant 1. Note that this Lipschitz property also implies that the definition of \(\Phi _\gamma \) is independent of the choice of the sequence \((\gamma _n)_{n\ge 1}\). To see the uniqueness of \(\Phi _\gamma \), it can be obtained via a fixed point argument identical to [33, Lemma 13]. These together give (i). The above discussion and Lemma 11 imply (ii). Since one can also pick this sequence \((\gamma _n)\) from \(\mathcal {U}_c\) and perform a similar procedure as above, we also obtain (iii). \(\square \)
Rights and permissions
About this article
Cite this article
Chen, WK., Handschy, M. & Lerman, G. On the energy landscape of the mixed even p-spin model. Probab. Theory Relat. Fields 171, 53–95 (2018). https://doi.org/10.1007/s00440-017-0773-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00440-017-0773-1