Abstract
Finding the global minimum of a cost function given by the sum of a quadratic and a linear form in N real variables over (N−1)-dimensional sphere is one of the simplest, yet paradigmatic problems in Optimization Theory known as the “trust region subproblem” or “constraint least square problem”. When both terms in the cost function are random this amounts to studying the ground state energy of the simplest spherical spin glass in a random magnetic field. We first identify and study two distinct large-N scaling regimes in which the linear term (magnetic field) leads to a gradual topology trivialization, i.e. reduction in the total number \(\mathcal{N}_{tot}\) of critical (stationary) points in the cost function landscape. In the first regime \(\mathcal{N}_{tot}\) remains of the order N and the cost function (energy) has generically two almost degenerate minima with the Tracy-Widom (TW) statistics. In the second regime the number of critical points is of the order of unity with a finite probability for a single minimum. In that case the mean total number of extrema (minima and maxima) of the cost function is given by the Laplace transform of the TW density, and the distribution of the global minimum energy is expected to take a universal scaling form generalizing the TW law. Though the full form of that distribution is not yet known to us, one of its far tails can be inferred from the large deviation theory for the global minimum. In the rest of the paper we show how to use the replica method to obtain the probability density of the minimum energy in the large-deviation approximation by finding both the rate function and the leading pre-exponential factor.
Similar content being viewed by others
Notes
Note that the value of C(N,ν′,ν) given in Eq. (5.2) of [20] misses the factor \(\frac{\nu'}{\sqrt{2\nu''}}\).
Note that though the treatment of [20] was formally restricted to covariances of the form ν(y)=y 2+⋯, one can check that inclusion of the linear term in the expansion does not invalidate their formalism.
Note that s ∗ vanishes at the typical energy and becomes negative for e typ <e<e c , i.e. that region is controlled by negative number of replica.
The actual situation in the vicinity of \(\mathcal{E}_{c}\) may appear to be even more complicated, see a note about the announced recent rigorous analysis of the problem by Dembo and Zeitouni in the Conclusion section.
For σ=0 the same calculation can be easily extended to any T and the corresponding tail of the free energy distribution f (coming from the large deviation regime) is found to be \(\sim e^{- \frac{2}{3} (1-T)^{3} (-2 y)^{3/2}}\) for T<1 with f=f typ+Jy, cf. [26]. It would be interesting to investigate how the small deviation distribution of f relates to the Tracy-Widom in the whole phase T<1.
We use that \(\prod_{k=0}^{n-1} \varGamma((n-k)/2) = G(\frac{N+1}{2}) G(1+ \frac {N}{2})/(G(\frac{N-n+1}{2}) G(1+ \frac{N-n}{2}))\) in terms of the Barnes function G(x).
Of course, as noted above the exponent term is correct, i.e. \(\mathcal{L}_{0}(\mathcal{E}) = \psi _{+}(s)\) there.
References
Conn, A.R., Gould, N.I.M., Toint, Ph.L.: Trust Region Methods. SIAM, Philadelphia (2000)
Tisseur, F., Meerberger, K.: The quadratic eigenvalue problem. SIAM Rev. 41(2), 235–286 (2001)
Gander, W., Golub, G.H., von Matt, U.: A constrained eigenvalue problem. Linear Algebra Appl. 114–115, 815–839 (1989)
Hager, W.W.: Minimizing a quadratic over a sphere. SIAM J. Optim. 12(1), 188–208 (2001)
Sorensen, D.C.: Minimization of a large-scale quadratic function subject to a spherical constraint. SIAM J. Optim. 7(1), 141–161 (1997)
Forsythe, G.E., Golub, G.H.: On the stationary values of a second degree polynomial on the unit sphere. J. Soc. Ind. Appl. Math. 13, 1050–1068 (1965)
Kosterlitz, J.M., Thouless, D.J., Jones, R.C.: Spherical model of a spin glass. Phys. Rev. Lett. 36, 1217–1220 (1976)
De Dominisis, C., Giardina, I.: Random Fields and Spin Glasses. Cambridge Univ. Press, Cambridge (2006). 213 pp
Tracy, C., Widom, H.: On orthogonal and symplectic matrix ensembles. Commun. Math. Phys. 177(3), 727–754 (1996)
Ben Arous, G., Dembo, A., Guionnet, A.: Aging of spherical spin glasses. Probab. Theory Relat. Fields 120, 1–67 (2001)
Dean, D.S., Majumdar, S.N.: Large deviations of extreme eigenvalues of random matrices. Phys. Rev. Lett. 97, 160201 (2006). [4 pages]
Majumdar, S.N., Vergassola, M.: Large deviations of the maximum eigenvalue for Wishart and Gaussian random matrices. Phys. Rev. Lett. 102, 060601 (2009). [4 pages]
Borot, G., Eynard, B., Majumdar, S.N., Nadal, C.: Large deviations of the maximal eigenvalue of random matrices. J. Stat. Mech. Theory Exp. 11, 11024 (2011)
Forrester, P.J.: Spectral density asymptotics for Gaussian and Laguerre β-ensembles in the exponentially small region. J. Phys. A 45(7), 075206 (2012). 145201 [15pp]
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 (2004). [4 pages]; Erratum ibid 92 (2004), Issue 16, 240601 [1p]
Bray, A.J., Dean, D.: The statistics of critical points of Gaussian fields on large-dimensional spaces. Phys. Rev. Lett. 98(15), 150201 (2007). [4p]
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)
Fyodorov, Y.V., Nadal, C.: Critical behavior of the number of minima of a random landscape at the glass transition point and the Tracy-Widom distribution. Phys. Rev. Lett. 109(16), 167203 (2012). [5 pages]
Auffinger, A., Ben Arous, G., Cerny, C.: Random matrices and complexity of spin glasses. Commun. Pure Appl. Math. 66(2), 165–201 (2013)
Auffinger, A., Ben Arous, G.: Complexity of random smooth functions of many variables. e-preprint. arXiv:1110.5872
Cugliandolo, L.F., Dean, D.S.: Full dynamical solution for a spherical spin-glass model. J. Phys. A 28, 4213–4234 (1995)
Zippold, W., Kuhn, R., Horner, H.: Nonequilibrium dynamics of a simple spherical spin models. Eur. Phys. J. B 13, 531–537 (2000)
Cugliandolo, L.F., Dean, D.S.: On the dynamics of a spherical spin-glass in a magnetic field. J. Phys. A, Math. Gen. 28, L453–L459 (1995)
Fyodorov, Y.V.: High-dimensional random fields and random matrix theory. e-preprint. arXiv:1307.2379
Parisi, G., Rizzo, T.: Large deviations in the free energy of mean-field spin glasses. Phys. Rev. Lett. 101, 117205 (2008). [4pp]
Parisi, G., Rizzo, T.: Phase diagram and large deviations in the free energy of mean-field spin glasses. Phys. Rev. B 79, 134205 (2009). [12pp]
Parizi, G., Rizzo, T.: Universality and deviations in disordered systems. Phys. Rev. B 81, 094201 (2010). [6 pp]
Crisanti, A., Sommers, H.-J.: The spherical p-spin interaction spin glass model: the statics. Z. Phys. B 87(3), 341–354 (1992)
de Almeida, J.R.L., Thouless, D.J.: Stability of the Sherrington-Kirkpatrick solution of a spin glass model. J. Phys. A 11(5), 983–990 (1978)
Fyodorov, Y.V., Sommers, H.-J.: Classical particle in a box with random potential: exploiting rotational symmetry of replicated Hamiltonian. Nucl. Phys. B 764(3), 128–167 (2007)
Fyodorov, Y.V., Bouchaud, J.P.: Statistical mechanics of a single particle in a multiscale random potential: Parisi landscapes in finite-dimensional Euclidean spaces. J. Phys. A, Math. Theor. 41, 324009 (2008). [25pp]
Calabrese, P., Le Doussal, P., Rosso, A.: Free-energy distribution of the directed polymer at high temperature. Europhys. Lett. 90, 20002 (2010)
Dotsenko, V.: Bethe ansatz derivation of the Tracy-Widom distribution for one-dimensional directed polymers. Europhys. Lett. 90, 20003 (2010). Replica Bethe ansatz derivation of the Tracy–Widom distribution of the free energy fluctuations in one-dimensional directed polymers. J. Stat. Mech. P07010 (2010)
Dotsenko, V., Klumov, B.: Bethe ansatz solution for one-dimensional directed polymers in random media. J. Stat. Mech., P03022 (2010)
Calabrese, P., Le Doussal, P.: Exact solution for the Kardar-Parisi-Zhang equation with flat initial conditions. Phys. Rev. Lett. 106, 250603 (2011). The KPZ equation with flat initial condition and the directed polymer with one free end. J. Stat. Mech. P06001 (2012)
Peché, S.: The largest eigenvalue of small rank perturbations of Hermitian random matrices. Probab. Theory Relat. Fields 134, 127–173 (2006)
Féral, D., Peché, S.: The largest eigenvalue of rank one deformation of large Wigner matrices. Comm. Math. Phys. 272 (2007)
Benaych-Georges, F., Guionnet, A., Maida, M.: Fluctuations of the extreme eigenvalues of finite rank deformations of random matrices. Electron. J. Probab. 16(60), 1621–1662 (2011)
Bassler, K.E., Forrester, P.J., Frankel, N.E.: Edge effects in some perturbations of the Gaussian unitary ensemble. J. Math. Phys. 51, 123305 (2010)
Benaych-Georges, F., Guionnet, A., Maida, M.: Large deviations of the extreme eigenvalues of random deformations of matrices. Probab. Theory Relat. Fields 154(3–4), 703–751 (2012)
Fyodorov, Y.V.: Level curvature distribution: from bulk to the soft edge of random Hermitian matrices. Acta Phys. Pol. A 120(6), 100–113 (2012). arXiv:1108.0950
von Oppen, F.: Exact distribution of eigenvalues curvatures for time-reversal invariant chaotic systems. Phys. Rev. E 51(3), 2647–2650 (1995)
Fyodorov, Y.V., Sommers, H.-J.: Universality of “level curvature” distribution for large random matrices: systematic analytical approaches. Z. Phys. B 99(1), 123–135 (1995)
Mehta, M.L.: Random Matrices, 3rd edn. Pure and Applied Mathematics (Amsterdam), vol. 142, Elsevier/Academic Press, Amsterdam (2004). xviii+688 pp
Mehta, D., Stariolo, D.A., Kastner, M.: Energy landscape of the finite-size mean-field 3-spin spherical model. Phys. Rev. E 87, 052143 (2013). [9 pages]
Forrester, P.J.: The spectrum edge of random matrix ensembles. Nucl. Phys. B 402, 709–728 (1993)
Parisi, G.: Constraint optimization and statistical mechanics. In: Proc. Int. Sch. Physics ENRICO FERMI, vol. 155, pp. 205–228 (2004). arXiv:cs/0312011
Mézard, M., Parisi, G.: On the solution of the random link matching problems. J. Phys. 48, 1451–1459 (1987)
Parisi, G., Ratieville, M.: On the finite size corrections to some random matching problems. Eur. Phys. J. B 29(3), 457–468 (2002)
Le Doussal, P., Wiese, K.J.: First-principle derivation of static avalanche-size distribution. Phys. Rev. E 85(6), 061102 (2012). [29pp]
Acknowledgements
We are grateful to Antonio Auffinger for a useful communication related to the content of the paper [20], to Jean-Philippe Bouchaud and Satya Majumdar for lively discussions of results and encouraging interest in the subject, to Peter Forrester and Mark Mezard for bringing a few relevant references to our attention, and to Ofer Zeitouni for informing us on his forthcoming rigorous large-deviation analysis of the problem. YF was supported by EPSRC grant EP/J002763/1 “Insights into Disordered Landscapes via Random Matrix Theory and Statistical Mechanics”. PLD was supported by ANR Grant No. 09-BLAN-0097-01/2.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Fyodorov, Y.V., Le Doussal, P. Topology Trivialization and Large Deviations for the Minimum in the Simplest Random Optimization. J Stat Phys 154, 466–490 (2014). https://doi.org/10.1007/s10955-013-0838-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10955-013-0838-1