Quantitative completion of the replica theory of the hard-sphere glass transition and amorphous close packing in finite dimension

Parisi and Zamponi (J. Chem. Phys. 123, 144501, 2005) computed, from first principles via the replica method applied to a replicated hard-sphere liquid with hypernetted-chain (HNC) input and a first-order small-cage expansion, the complexity, the Kauzmann (ideal glass transition) packing fraction, the ideal glass close-packing density, and the glass pair correlation function in three dimensions, and they explicitly listed what the computation lacked: a controlled liquid input beyond HNC, cage-expansion corrections beyond first order, mixtures, the stability endpoint of the complexity ($f_{\max}$), the relation between the ideal thermodynamic packing densities and the densities reachable by finite-time algorithms, and the neglected subleading corrections to the glass correlation functions. Later work by the same research program settled parts of this list: binary mixtures were treated at the mean-field level, and the exact solution in the limit of infinite spatial dimension removed the HNC and truncation issues in that limit and identified the Gardner transition as the instability of glassy states. What remains open is the finite-dimensional completion: a quantitatively controlled, systematically improvable theory in $d=3$, validated against simulation, for the transition densities, the complexity and its stability boundary, and the glass pair correlation, together with a characterization of the densities that finite-time local algorithms can actually reach and their scaling with protocol and system size.

Background

Hard spheres — identical particles whose only interaction is excluded volume — are the reference model for the glass transition in condensed-matter physics. The state of a hard-sphere system is characterized by the packing fraction $\varphi = N\pi D^3/(6V)$, with $D$ the sphere diameter, $N$ the number of spheres and $V$ the volume. A monodisperse hard-sphere liquid freezes into a crystal at $\varphi_f \simeq 0.494$; if crystallization is avoided, the metastable liquid can be compressed further, its relaxation time grows extremely fast, and around $\varphi \simeq 0.56$ it falls out of equilibrium on numerical and experimental timescales. Whether, deeper in the metastable region, a genuine thermodynamic ('ideal') glass transition exists — a Kauzmann density $\varphi_K$ at which the configurational entropy, i.e. the logarithm of the number of long-lived amorphous metastable states (also called the complexity $\Sigma$), vanishes — is a central open question of liquid-state physics. A related empirical landmark is random close packing (RCP), the density $\varphi_{\mathrm{RCP}} \simeq 0.64$ at which disordered sphere packings jam.

Parisi and Zamponi (2005) gave this question a first-principles quantitative treatment using the replica method, a technique originating in spin-glass theory: $m$ identical copies ('replicas') of the liquid are coupled so that all copies must sit in the same metastable state; in the glass phase each particle is confined to a small 'cage' of radius $\sqrt{A}$ centered on a common 'molecular' center, and the replicated free energy is expanded to first order in the cage size (the 'small-cage expansion'). The leading correction, of order $\sqrt{A}$, arises from the discontinuity of the liquid pair correlation function $g(r)$ at contact $r=D$. The equilibrium liquid enters through its free energy, for which they used the hypernetted-chain (HNC) approximation — a standard approximate closure of the exact liquid-state integral equations for the pair correlation function, known to be quantitatively poor at high density — cross-checked against the phenomenological Carnahan–Starling equation of state (an accurate fitting formula for the hard-sphere liquid pressure). The mean-field glass solution is the one-step replica symmetry breaking (1RSB) scheme, in which the equilibrium measure splits into exponentially many amorphous states enumerated by the complexity $\Sigma$. The outcome in $d=3$ was: $\Sigma \sim 1$; $\varphi_K = 0.582$ with HNC input ($0.617$ with Carnahan–Starling input); an ideal close-packing density $\varphi_c = 0.640$ (HNC) to $0.683$ (Carnahan–Starling) at which the glass pressure and the contact value of the glass pair correlation $\widetilde{g}(r)$ diverge — the thermodynamic close packing of the equilibrium glass, conceptually distinct from, though numerically close to, the protocol-dependent empirical RCP of the previous paragraph — and a mean contact number $z = 6 = 2d$ at $\varphi_c$, matching the isostaticity of jammed packings (marginally stable contact networks require $z = 2d$ by Maxwell counting). The authors stressed that these results do not prove the existence of a glass transition, and enumerated what the computation still lacked.

Later work by the same research program defines the present status. The 2010 review by Parisi and Zamponi systematized the theory: it introduced the 'glass close packing' density $\varphi_{\mathrm{GCP}} \simeq 0.68$ in $d=3$ — the jamming density of the equilibrium (ideal) glass, distinct from the protocol-dependent RCP $\simeq 0.64$ — showed that differently prepared glasses jam anywhere in an interval $[\varphi_{\mathrm{th}}, \varphi_{\mathrm{GCP}}]$, where $\varphi_{\mathrm{th}} \simeq 0.64$ is the density at which the pressure of the least dense, fastest-quenched glass states diverges, computed the full glass pair correlation function and the contact-force distribution in $d=3$ via a Percus–Yevick treatment of the effective liquid (Percus–Yevick being a second standard integral-equation closure), and attempted to go beyond first order in the cage expansion by resummation, with explicitly unsatisfactory results. Biazzo, Caltagirone, Parisi and Zamponi (2009) extended the theory to binary mixtures. Kurchan, Parisi, Urbani and Zamponi, with Charbonneau (2012–2014), then obtained the exact solution in the limit of infinite spatial dimension $d\to\infty$: no closure such as HNC is needed there, the cage expansion is exact, the thermodynamics of the Gaussian cage ansatz is exact once arbitrary replica symmetry breaking is allowed, and the stability of glassy states is settled — beyond a Gardner transition (named after E. Gardner's spin-glass work) each amorphous state splits into a hierarchy of marginally stable sub-basins described by full replica symmetry breaking (full RSB), the broken-symmetry pattern familiar from mean-field spin glasses. Numerical simulations of hard spheres in $d=3$–$12$ support this scenario. Mangeat and Zamponi (2016) constructed a systematic expansion around the exact $d\to\infty$ solution usable in finite dimension, but its outputs remain sensitive to the quality of the liquid input. The program's reviews by Charbonneau, Kurchan, Parisi, Urbani and Zamponi (2017) and by Berthier, Biroli, Charbonneau, Corwin, Franz and Zamponi (2019) state that quantitatively controlled finite-dimensional corrections are still 'an open area of research' and 'a work in progress', and that the densities reached by finite-time protocols are not yet theoretically understood.

Problem Statement

Complete the first-principles replica theory of the ideal glass transition and amorphous close packing of hard spheres into a quantitatively controlled theory in finite dimension, $d=3$ at minimum. The question is the one acknowledged by Parisi and Zamponi (2005) as the limitations of their computation, restricted to the core that is still open after the later developments recorded in the Background: the binary-mixture extension and the exact $d\to\infty$ solution already exist and need not be redone.

The open core comprises the following jointly posed parts:

(1) A controlled approximation scheme for the replicated-liquid free energy in finite $d$, with an explicit small parameter and error statement, replacing the uncontrolled hypernetted-chain input and its ad hoc phenomenological substitutes.

(2) The resulting quantitative predictions for the Kauzmann density $\varphi_K$, the ideal glass close-packing density $\varphi_{\mathrm{GCP}}$ (equivalently $\varphi_c$), and the complexity $\Sigma(\varphi)$ in $d=3$, including the effect of the corrections beyond the leading $O(\sqrt{A})$ cage term.

(3) The glass pair correlation function $\widetilde{g}(r)$, including the subleading corrections of order $\sqrt{A}$ and $A$ in the molecular correlation functions, compared quantitatively with numerical radial-distribution data of dense amorphous packings of monodisperse hard spheres in $d=3$, over the packing-fraction window $0.58 \le \varphi \le 0.64$ and prepared by the fixed protocol of Charbonneau, Ikeda, Parisi and Zamponi, Phys. Rev. Lett. 107, 185702 (2011) — event-driven molecular dynamics with diameters grown at a constant rate (Lubachevsky–Stillinger compression) at large system size, crystallization being suppressed by the fast compression; the reference data are defined by this protocol and are either those of that study and the cited simulation literature or a simulation generated by the solver on exactly this system and protocol.

(4) The stability range of the metastable states in finite $d$: the endpoint $f_{\max}$ of the complexity curve at high free-energy density $f$ and the Gardner instability line bounding the stable one-step-replica-symmetry-breaking states, i.e. the structure of the complexity at high free energies.

(5) The algorithmic-accessibility question: which packing densities can actually be reached by finite-time local algorithms in large systems, how the reachable density scales with the algorithmic protocol and with system size, and how it relates to the ideal thermodynamic $\varphi_{\mathrm{GCP}}$.

Items (1)–(4) form the thermodynamic-structural core; item (5) is the dynamical-protocol component; they jointly support one objective, the quantitative first-principles account of the hard-sphere glass transition and amorphous close packing in the physical dimension. Each of items (1)–(5) is an independently resolvable sub-question, and together they form the jointly posed set (1)–(5). The theory must be quantitatively controlled: it must reproduce the exact $d\to\infty$ results in the appropriate limit, and in $d=3$ its intrinsic errors — those obtained under refinement of the scheme's own expansion order, not agreement with any prior estimate — are of order $|\Delta\varphi| \sim 0.005$ in the Kauzmann density $\varphi_K$ and the glass close-packing density $\varphi_{\mathrm{GCP}}$, of order $|\Delta\Sigma| \sim 0.02$ in the complexity curve $\Sigma(\varphi)$ over the window $0.58 \le \varphi \le 0.64$ (an absolute bound, well defined where $\Sigma$ vanishes near $\varphi_K$), and a root-mean-square relative deviation of order $5\%$ for the glass pair correlation $\widetilde{g}(r)$ against the item-(3) reference data — thereby sharpening or refuting the 2005 estimates $\varphi_K = 0.582$ (HNC) / $0.617$ (Carnahan–Starling) and $\varphi_c = 0.640$ / $0.683$, the 2010 review's glass-close-packing value $\varphi_{\mathrm{GCP}} \simeq 0.68$, and the protocol-dependent random-close-packing value $\simeq 0.64$. The scheme supplied under item (1) yields items (2)–(4); the $f_{\max}$ endpoint and the Gardner line of item (4) are functions of $\varphi$ that reduce to the exact $d\to\infty$ results in the appropriate limit; and the reachable densities of item (5) are to be compared, over a sweep of compression rates (at least three rates spanning a decade) and of system sizes (at least three sizes, so as to resolve the finite-size trend), with measured jamming densities to within a packing-fraction accuracy of order $|\Delta\varphi| \sim 0.005$.

References

1. Giorgio Parisi and Francesco Zamponi, 'The ideal glass transition of hard spheres', The Journal of Chemical Physics 123, 144501 (2005). DOI: 10.1063/1.2041507, https://doi.org/10.1063/1.2041507 (arXiv: cond-mat/0506445, https://arxiv.org/abs/cond-mat/0506445)
2. Indaco Biazzo, Francesco Caltagirone, Giorgio Parisi, Francesco Zamponi, 'Theory of amorphous packings of binary mixtures of hard spheres', Physical Review Letters 102, 195701 (2009). DOI: 10.1103/PhysRevLett.102.195701, https://doi.org/10.1103/PhysRevLett.102.195701 (arXiv: 0903.5099, https://arxiv.org/abs/0903.5099)
3. Giorgio Parisi and Francesco Zamponi, 'Mean field theory of hard sphere glasses and jamming', Reviews of Modern Physics 82, 789 (2010). DOI: 10.1103/RevModPhys.82.789, https://doi.org/10.1103/RevModPhys.82.789 (arXiv: 0802.2180, https://arxiv.org/abs/0802.2180)
4. Patrick Charbonneau, Atsushi Ikeda, Giorgio Parisi, Francesco Zamponi, 'Glass transition and random close packing above three dimensions', Physical Review Letters 107, 185702 (2011). DOI: 10.1103/PhysRevLett.107.185702, https://doi.org/10.1103/PhysRevLett.107.185702 (arXiv: 1107.4666, https://arxiv.org/abs/1107.4666)
5. Jorge Kurchan, Giorgio Parisi, Francesco Zamponi, 'Exact theory of dense amorphous hard spheres in high dimension. I. The free energy', Journal of Statistical Mechanics (2012) P10012. DOI: 10.1088/1742-5468/2012/10/P10012, https://doi.org/10.1088/1742-5468/2012/10/P10012 (arXiv: 1208.0421, https://arxiv.org/abs/1208.0421)
6. Jorge Kurchan, Giorgio Parisi, Pierfrancesco Urbani, Francesco Zamponi, 'Exact theory of dense amorphous hard spheres in high dimension. II. The high density regime and the Gardner transition', The Journal of Physical Chemistry B 117, 12979 (2013). DOI: 10.1021/jp402235d, https://doi.org/10.1021/jp402235d (arXiv: 1303.1028, https://arxiv.org/abs/1303.1028)
7. Patrick Charbonneau, Jorge Kurchan, Giorgio Parisi, Pierfrancesco Urbani, Francesco Zamponi, 'Exact theory of dense amorphous hard spheres in high dimension. III. The full replica symmetry breaking solution', Journal of Statistical Mechanics (2014) P10009. DOI: 10.1088/1742-5468/2014/10/P10009, https://doi.org/10.1088/1742-5468/2014/10/P10009 (arXiv: 1310.2549, https://arxiv.org/abs/1310.2549; preprint title: 'Exact theory of dense amorphous hard spheres in high dimension. III. The full RSB solution')
8. Matthieu Mangeat, Francesco Zamponi, 'Quantitative approximation schemes for glasses', Physical Review E 93, 012609 (2016). DOI: 10.1103/PhysRevE.93.012609, https://doi.org/10.1103/PhysRevE.93.012609 (arXiv: 1510.03808, https://arxiv.org/abs/1510.03808)
9. Patrick Charbonneau, Jorge Kurchan, Giorgio Parisi, Pierfrancesco Urbani, Francesco Zamponi, 'Glass and jamming transitions: From exact results to finite-dimensional descriptions', Annual Review of Condensed Matter Physics 8, 265 (2017). DOI: 10.1146/annurev-conmatphys-031016-025334, https://doi.org/10.1146/annurev-conmatphys-031016-025334 (arXiv: 1605.03008, https://arxiv.org/abs/1605.03008)
10. Ada Altieri, Francesco Zamponi, 'Mean field stability map of hard-sphere glasses', Physical Review E 100, 032140 (2019). DOI: 10.1103/PhysRevE.100.032140, https://doi.org/10.1103/PhysRevE.100.032140 (arXiv: 1907.05513, https://arxiv.org/abs/1907.05513)
11. Harukuni Ikeda, Francesco Zamponi, 'Effect of particle exchange on the glass transition of binary hard spheres', Journal of Statistical Mechanics (2019) 054001. DOI: 10.1088/1742-5468/ab0c1a, https://doi.org/10.1088/1742-5468/ab0c1a (arXiv: 1812.08780, https://arxiv.org/abs/1812.08780)
12. Ludovic Berthier, Giulio Biroli, Patrick Charbonneau, Eric I. Corwin, Silvio Franz, Francesco Zamponi, 'Gardner physics in amorphous solids and beyond', The Journal of Chemical Physics 151, 010901 (2019). DOI: 10.1063/1.5097175, https://doi.org/10.1063/1.5097175 (arXiv: 1902.10494, https://arxiv.org/abs/1902.10494; preprint title: 'Perspective: Gardner Physics in Amorphous Solids and Beyond')
13. Sam Wilken, Rodrigo E. Guerra, Dov Levine, Paul M. Chaikin, 'Random Close Packing as a Dynamical Phase Transition', Physical Review Letters 127, 038002 (2021). DOI: 10.1103/PhysRevLett.127.038002, https://doi.org/10.1103/PhysRevLett.127.038002
14. Patrick Charbonneau, Peter K. Morse, 'Memory formation in jammed hard spheres', Physical Review Letters 126, 088001 (2021). DOI: 10.1103/PhysRevLett.126.088001, https://doi.org/10.1103/PhysRevLett.126.088001 (arXiv: 2009.00492, https://arxiv.org/abs/2009.00492)
15. Giampaolo Folena, Patrick Charbonneau, Peter K. Morse, Rafael Díaz Hernández Rojas, Federico Ricci-Tersenghi, 'The geometry of jamming algorithms in the random Lorentz gas', Proceedings of the National Academy of Sciences 122, e2422096122 (2025). DOI: 10.1073/pnas.2422096122, https://doi.org/10.1073/pnas.2422096122 (arXiv: 2410.05784, https://arxiv.org/abs/2410.05784)
16. Harukuni Ikeda, Francesco Zamponi, 'High-dimensional theory of the glass transition revisited: hopping and local defects' (2026). arXiv: 2607.26793, https://arxiv.org/abs/2607.26793