WorksPhysics and Astronomy
Contact-number hyperuniformity at glass close packing in infinite dimensions
Abstract
We investigate contact-number fluctuations of harmonic spheres near glass close packing (GCP), the amorphous jamming endpoint reached by compressing an ideal glass. Using replica theory, we show that the Gardner-unstable region shrinks toward jamming as the spatial dimension increases, allowing GCP to be approached along a stable one-step replica symmetry breaking branch in the high-dimensional limit. By coupling an external field to the local contact number, we derive the variance of the total contact number and the wave-number-dependent structure factor directly from the microscopic interactions. At GCP, the structure factor vanishes quadratically at small scaled wave numbers, demonstrating contact-number hyperuniformity. Upon weak compression above GCP, the structure factor develops a low-wave-number plateau whose height is proportional to the excess coordination. These results provide an analytical description of contact-number hyperuniformity and its breakdown upon compression.
Introduction
The jamming transition is characterized by the emergence of mechanical rigidity upon compression of a particle assembly [1–3]. Systems of frictionless repulsive spheres have been extensively studied as a simple model of this transition [4]. Numerical and theoretical studies have established isostaticity at the jamming transition and revealed scaling laws for elastic moduli, vibrational properties, and the distributions of contact forces and interparticle gaps [5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] [4].
In addition to these scaling laws, numerical studies have revealed an anomalous suppression of contact-number fluctuations on large length scales near jamming [17, 18]. This behavior is known as contact-number hyperuniformity. In general, hyperuniformity is characterized by a variance of integrated fluctuations that grows more slowly than the volume of an observation window [19, 20]. For contact-number fluctuations at jamming, the variance was found to scale with the surface area of the window rather than its volume [17]. This suppression is also reflected in the contact-number structure factor , which vanishes with power law as . The original numerical study reported an approximate scaling [17, 21, 22]. More recently, Wang et al. found in jammed harmonic-sphere packings with a tuned particle-size distribution [23].
To explain the origin of contact-number hyperuniformity, Shang et al. proposed a minimal network model based on global isostaticity and local mechanical stability constraints [24]. Numerical simulations of this model yielded contact-number hyperuniformity with an exponent . However, to our knowledge, the contact-number structure factor near jamming has not yet been derived analytically from a microscopic model of interacting particles.
Replica theory provides a framework for describing glassy states starting from microscopic particle interactions [25, 26]. Early calculations for harmonic spheres used a one-step replica symmetry breaking (1RSB) approximation to study glassy and jamming transitions [27]. In the infinite-dimensional limit, the replicated free-energy functional can be derived exactly [28]. As approaches jamming, the 1RSB solution can become unstable at a Gardner transition [29, 30, 31, 32, 33, 34, 35]. Beyond the Gardner transition, a hierarchy of glassy subbasins emerges, requiring a full replica symmetry breaking (fullRSB) description. The fullRSB solution predicts the nontrivial critical scaling of the force and gap distributions near jamming [9, 10, 13, 14]. However, calculating spatial contact-number correlations within the fullRSB framework remains technically challenging.
To make analytical progress, we focus on glass close packing (GCP), the amorphous jamming endpoint of the ideal-glass branch [26]. The reduced packing fraction at GCP, , grows as , and the cage size becomes small [26, 30]. The Gardner boundary moves toward jamming as the reduced density increases [9]. For harmonic spheres, we show that the width of the Gardner-unstable region, expressed in terms of the relative excess coordination, decreases as . In the limit , GCP can thus be approached along a stable 1RSB branch.
Within this 1RSB description, we introduce an external field conjugate to the local contact number. The response to a uniform field yields the mean contact number and the variance of the total contact number. At GCP, the mean reaches the isostatic value and the variance per particle vanishes. Using a spatially varying field, we derive an explicit expression for the contact-number structure factor . At GCP, it vanishes quadratically in the scaled wave number , yielding contact-number hyperuniformity. Upon weak compression above GCP, a finite plateau develops at small , with a height proportional to the excess coordination.
The remainder of this paper is organized as follows. In Sec. II, we introduce the model and explain how to compute the mean contact number and its fluctuations. In Sec. III, we derive the replicated free energy in the high-dimensional and zero-temperature limits. In Sec. IV, we analyze the response to a uniform field and locate the Gardner stability boundary. In Sec. V, we extend the calculation to a spatially varying field and derive the contact-number structure factor. We conclude in Sec. VI.
Setup
Model
We consider harmonic spheres of diameter in dimensions [4]. The total potential energy is
where is the position of particle and the pair potential is
Here, sets the energy scale and is the Heaviside step function.
Contact-number observables
The contact number of particle is
We define the contact field as
Its Fourier transform is
The contact-number structure factor is
where . An alternative definition uses the Fourier transformation of , where is the mean contact number [17], leading to
The two definitions agree to leading order in the high-dimensional limit considered here because density fluctuations contribute only at subleading order (see Appendix A).
Partition function
To compute the mean contact number and its correlations, we introduce a partition function with an external field conjugate to ,
where is the inverse temperature. A functional derivative with respect to the field gives
For a uniform field , the mean contact number per particle is
Using the same Fourier convention for as for , Parseval’s identity gives
Differentiating with respect to the Fourier components of the field then gives
Replica method
Replicated free energy
Following Monasson [25], we consider replicas constrained to occupy the same metastable state. The replicated free energy is defined by
where is the replicated partition function. After analytic continuation in , the equilibrium liquid corresponds to , whereas the ideal-glass branch has . On the glass branch considered here, as the temperature goes to zero.
When the replicas occupy the same metastable state, the copies of particle remain localized near one another and can be treated as a single molecule [26]. We denote its coordinates by and write for integration over all replica coordinates of molecule . The replicated partition function on this glass branch is then written as
where
Here, is the contact number of particle in replica . The source term can be written as a sum over pairs,
The Boltzmann weight thus factorizes over molecular pairs, giving
where the molecular Mayer function is defined by
Following the standard procedure of Morita and Hiroike [36, 37], we obtain the replicated free energy as a functional of the molecular density ,
The higher-order terms are subleading in the high-dimensional limit [14, 26]. The first term is the ideal-gas contribution , and the second is the leading interaction contribution .
Gaussian ansatz
Tomita and Yoshino developed a replicated liquid theory in which the density and the glass order parameter vary along a single spatial direction [38]. Here we allow the density and cage variances to depend on the full molecular-center coordinate . We use a Gaussian ansatz for the molecular density,
where is a normalized Gaussian with variance per Cartesian component. The local density of molecular centers, , is normalized to the total number of molecules,
Since the source can differ between replicas, we initially allow the cage variance to depend on the replica index . In the zero-temperature limit, however, the saddle-point equations admit a solution with the same rescaled cage variance for all replicas at each , as we will see below.
High-dimensional limit
Substituting the Gaussian ansatz for the density outside the logarithm gives
The logarithm contains a separate integral over the molecular center,
At large , the cage variances are of order [26]. At fixed , the outer Gaussian factors therefore localize the replica coordinates around the molecular center . The Gaussian product inside the logarithm is appreciable when is close to these replica coordinates, so the dominant contribution to the inner integral comes from near . We thus approximate and by and , respectively, in the inner integral,
We perform the Gaussian integral over and average the resulting logarithm over the replica coordinates at fixed . To express the result, we introduce the dimensionless variances through
The ideal contribution then reads [26]
Substituting the Gaussian ansatz into the interaction contribution gives
where
When the sources vary slowly on the cage scale, we evaluate and at the corresponding molecular centers and . We define the scaled gap and inverse temperature as
At fixed in the high-dimensional limit, we obtain [26, 34]
Zero-temperature limit
We take the joint limit and at fixed ratio, writing
For fixed , the saddle point method leads to
where
The sources enter the free-energy only through their sum . The free-energy is thus invariant under permutations of the replica labels on the cage fields . We seek a saddle that preserves this symmetry,
With this choice, the stationarity conditions for the individual replicas reduce to a single equation for . At this saddle, taking in the ideal term gives
and the interaction term becomes
Homogeneous case
Free energy
We first consider a spatially uniform source, . The homogeneous saddle has and . From Eq. (35), the ideal contribution per particle is
Let be the volume of a -dimensional ball of radius . Then , where . At the GCP, [26]. Using Stirling’s formula for , we obtain
After angular integration, the radial measure in Eq. (36) becomes in the high-dimensional limit. With , the interaction contribution per particle is
where
For small , this function has the expansion
Cage size
The cage size is calculated by minimizing the free-energy
The cage size is determined by the saddle-point condition:
which gives
We introduce the dimensionless ratio
For weak compression along the ideal-glass branch in the high-dimensional limit, increases linearly with compression above GCP and measures the relative excess coordination, as we will show in the following subsections. Using in Eq. (44), we find
Thus, at fixed and , the cage size decreases as , which validates the small-cage expansion at large $d.
Complexity
The zero-complexity condition selects the ideal-glass branch. Together with the saddle-point equation, it determines and hence as functions of , as we will see below. At zero source, the complexity per particle is
where is the replicated entropy per particle at finite [25, 26]. Writing and taking the zero-temperature limit gives [27]
To reach the jamming endpoint, we first take at fixed . This suppresses particle overlaps and reduces the model to hard spheres. We then take along the hard-sphere glass branch, reaching its infinite-pressure jamming endpoint [26]. This order of limits corresponds to , for which . In the density regime , the leading terms are
Setting gives the GCP density,
consistent with Ref. [26].
To determine how changes upon compression above GCP, we expand the complexity at small . The stationarity of with respect to gives
where the last equality follows from Eq. (44) at zero source. With , we obtain
At , differentiation with respect to density gives
Expanding about and imposing , we find
for weak compression above GCP.
Gardner transition
At zero source, we locate the Gardner transition of the 1RSB solution using the condition derived by Biroli and Urbani [34],
where
We set and . In the limit at fixed , we obtain, for ,
Substituting these limits into Eq. (55) gives
Using the zero-source saddle-point condition (44), we eliminate to obtain
For small , the ratio in this equation has the expansion
Using , we find that the Gardner boundary occurs at
where is the cage size at the transition. Near GCP, , so the width of the unstable region is
The Gardner boundary thus approaches jamming as increases. At fixed near GCP, the 1RSB solution is stable against the Gardner instability in the limit . The jamming endpoint is marginal in this limit [30].

Figure 1. Schematic stability diagram of the 1RSB solution for harmonic spheres near GCP. The parameter measures the relative excess coordination to leading order at large . (a) At large but finite , the 1RSB solution is unstable for , with . (b) As , the Gardner boundary approaches the marginal jamming endpoint , and the 1RSB solution is stable at fixed near GCP.
Mean contact number and its fluctuations
The uniform source couples to the sum of the particle contact numbers in replica :
where
We obtain the mean contact number per particle by differentiating the free-energy with respect to . Since is stationary with respect to , only the explicit source dependence contributes to the first derivative,
Using Eq. (44), we can also write
The leading small-cage result is
Thus, near GCP in the high-dimensional limit, measures the relative excess coordination
At , the limiting mean contact number reaches the isostatic value . Combining Eq. (67) with the zero-complexity condition gives a linear increase upon weak compression in the high-dimensional 1RSB regime,
In finite-dimensional simulations of harmonic spheres, the excess coordination grows as near jamming [4]. The difference between these scaling laws reflects the replica structure governing the approach to jamming. At GCP, where , the Gardner-unstable region shrinks as , allowing jamming to be approached along a stable 1RSB branch. At finite , a glass can undergo a Gardner transition before reaching jamming. This can also occur in the limit when the jamming density remains finite [34, 35]. In either case, full replica symmetry breaking governs the asymptotic scaling near jamming [9, 10].
Further derivative with respect to gives the contact-number susceptibility,
Using the leading small-cage expression and differentiating at fixed and , we obtain
Taking the logarithm of the small-cage form of Eq. (44) and differentiating with respect to gives
Substituting this result into Eq. (71) yields
Figure 2 shows this result. As jamming is approached, , where denotes the excess coordination. The 1RSB calculation thus gives

Figure 2. Contact-number susceptibility as a function of . The parameter measures the relative excess coordination to leading order at large . The solid line shows the leading 1RSB result and the dashed line shows its small- asymptote, .
Numerical simulations in three dimensions suggest – [18, 39, 40]. Whether extending our calculation to the fullRSB regime yields an exponent closer to these estimates remains an open question.
Inhomogeneous case
In this section, we extend the calculation to a spatially varying source to determine how the contact-number fluctuations depend on wave number.
Free energy
We apply a spatially varying source with zero spatial average and expand Eqs. (35) and (36) about the homogeneous saddle at zero source. We write the density and cage fields as
For a source at nonzero wave number, the linear responses and have zero spatial average. Near GCP, and, at fixed , is of the same order. The Gaussian in the interaction term is therefore confined to a narrow shell on the inside of the contact surface . To leading order in the shell width [41],
The square-root factor is the integrated weight of the Gaussian over , and converts the delta function in to one in the radial distance.
We use for spatial averages. For functions of two positions, we define averages over the surface and interior of the contact sphere,
where is the surface area of a sphere of radius . With , , and , the ideal and interaction contributions are, up to additive constants,
Here is the mean contact number of the homogeneous saddle to leading order.
For a modulation at fixed nonzero scaled wave number , the excluded-volume term gives a density stiffness of order near GCP. The couplings of the density mode to the cage modulation and the contact-number source are both of order . The density saddle point therefore gives
so that we can set when computing the leading-order contact-number response. Setting and expanding to second order in the cage modulation and the source , we obtain
With the Fourier convention introduced above, the surface average of two fields and takes the form
where
is the Fourier transform of the normalized surface measure on the contact sphere. In the limit , it converges to
The quadratic contribution to is then
with
Figure 3. Contact-number structure factor as a function of the scaled wave number for several values of . The parameter measures the relative excess coordination to leading order at large . The small- plateau decreases as approaches zero. At GCP (), the structure factor vanishes quadratically at small .
Contact-number structure factor
Stationarity with respect to gives the cage response,
Substituting this response into the quadratic free energy gives
where denotes the value at the saddle. Since each replica source enters through , differentiating twice with respect to the source in one replica yields the connected correlation function
In a periodic volume , the Dirac delta is replaced by times the corresponding Kronecker delta. Dividing the covariance at by therefore gives
For fixed in the high-dimensional limit, using Eq. (84), we get
Figure 3 shows the resulting structure factor as a function of . The plateau at small decreases as jamming is approached. Taking in the limiting spectrum gives
At GCP, where , the plateau vanishes and the leading small- behavior is
The quadratic decay establishes contact-number hyperuniformity at GCP in the high-dimensional limit. The exponent 2 is larger than the value of about 1.5 reported by Hexner et al. [17]. Whether full replica symmetry breaking changes the wave-number exponent in the marginal regime remains an open question.
Summary and discussion
We derived the replicated free-energy functional of harmonic spheres near glass close packing (GCP) in the high-dimensional limit within the 1RSB ansatz. By coupling a source to the contact field, we obtained both the uniform susceptibility and the wave-number-dependent structure factor. For a uniform source, the mean contact number reaches the isostatic value at GCP and increases linearly with the excess packing fraction upon compression. The variance of the total contact number per particle, , vanishes at the transition point as with . For a spatially varying source, we obtained the contact-number structure factor. Taking the high-dimensional limit at fixed scaled wave number gives
At GCP, and the contact number exhibits hyperuniformity with . Above GCP, , and thus, hyperuniformity is not observed.
Our results and deviate from those of three-dimensional simulations, which reported [17] and – [18, 39, 40]. The difference may be related to the stability of the glass on the approach to jamming. At high-dimensional GCP, the width of the Gardner-unstable region scales as . In the limit , this region shrinks to the jamming point, which can then be approached along a stable 1RSB branch and is marginal at its endpoint. A direction for future work is to extend the current calculation to the fullRSB regime and determine whether the hierarchy of glass subbasins yields values of and closer to the numerical results in finite dimension.
Wang et al. found in jammed harmonic-sphere packings with a tuned particle-size distribution [23], in agreement with our 1RSB prediction. They also found enhanced vibrational stability in glasses prepared by particle resizing well above jamming, where low-frequency quasilocalized modes were strongly suppressed. This enhanced stability raises the possibility that these glasses approach jamming along a stable 1RSB branch, which could explain the agreement in . Further work is needed to elucidate this point.
Acknowledgments
This project has received JSPS KAKENHI Grant Numbers 23K13031, and 25H01401.
AI disclosure. I used ChatGPT–6.1 (OpenAI) to assist with drafting and editing the manuscript, and checking intermediate steps in the derivations. I checked the derivations against my calculations, and reviewed and revised the AI-assisted text. I take full responsibility for the content of the manuscript.
Data availability
This study is based on analytical calculations. All quantitative plots can be reproduced from the expressions given in the article.
Comparison of the two contact-number structure factors
We compare the two contact-number structure factors and introduced in Sec. II. Throughout this appendix, we take at fixed nonzero scaled wave number .
The density stiffness derived in § 2.4 is of order . Near GCP, where , its inverse gives
The same order is obtained by evaluating the high-dimensional hard-sphere liquid result, , at the GCP density scale [42].
For , translational invariance gives . The first term in Eq. (7) is therefore
where the scaling follows from the calculation in the main text. Since , the density term is
The Cauchy–Schwarz inequality bounds the mixed term as
Thus, at fixed , the density-dependent terms are subleading, and and agree to leading order as .
References
- [5]M. Wyart, L. E. Silbert, S. R. Nagel, and T. A. Witten, Phys. Rev. E 72, 051306 (2005).
- [6]C. P. Goodrich, A. J. Liu, and S. R. Nagel, Phys. Rev. Lett. 109, 095704 (2012).
- [7]E. DeGiuli, A. Laversanne-Finot, G. Düring, E. Lerner, and M. Wyart, Soft Matter 10, 5628 (2014).
- [8]E. DeGiuli, E. Lerner, C. Brito, and M. Wyart, Proceedings of the National Academy of Sciences 111, 17054–17059 (2014).
- [9]P. Charbonneau, J. Kurchan, G. Parisi, P. Urbani, and F. Zamponi, Nature Communications 5 (2014), 10.1038/ncomms4725.
- [10]P. Charbonneau, J. Kurchan, G. Parisi, P. Urbani, and F. Zamponi, Journal of Statistical Mechanics: Theory and Experiment 2014, P10009 (2014).
- [11]P. Charbonneau, E. I. Corwin, G. Parisi, and F. Zamponi, Phys. Rev. Lett. 114, 125504 (2015).
- [12]L. Yan, E. DeGiuli, and M. Wyart, EPL (Europhysics Letters) 114, 26003 (2016).
- [13]P. Charbonneau, J. Kurchan, G. Parisi, P. Urbani, and F. Zamponi, Annual Review of Condensed Matter Physics 8, 265–288 (2017).
- [14]G. Parisi, P. Urbani, and F. Zamponi, Theory of Simple Glasses: Exact Solutions in Infinite Dimensions (Cambridge University Press, 2019).
- [15]P. Charbonneau, E. I. Corwin, R. C. Dennis, R. Díaz Hernández Rojas, H. Ikeda, G. Parisi, and F. Ricci-Tersenghi, Phys. Rev. E 104, 014102 (2021).
- [16]H. Ikeda and M. Shimada, Phys. Rev. E 106, 024904 (2022).
- [17]D. Hexner, A. J. Liu, and S. R. Nagel, Phys. Rev. Lett. 121, 115501 (2018).
- [18]D. Hexner, P. Urbani, and F. Zamponi, Phys. Rev. Lett. 123, 068003 (2019).
- [19]S. Torquato and F. H. Stillinger, Phys. Rev. E 68, 041113 (2003).
- [20]S. Torquato, Physics Reports 745, 1–95 (2018).
- [21]S. Franz, A. Sclocchi, and P. Urbani, SciPost Physics 9 (2020), 10.21468/scipostphys.9.1.012.
- [22]J. Zhang, J. Si, N. Xu, and H. Tong, Phys. Rev. Lett. 137, 078201 (2026).
- [23]Y. Wang, Z. Qian, H. Tong, and H. Tanaka, Nature Communications 16 (2025), 10.1038/s41467-025-56283-1.
- [24]J. Shang, Y. Wang, D. Pan, Y. Jin, and J. Zhang, Proceedings of the National Academy of Sciences 123 (2026), 10.1073/pnas.2517241123.
- [25]R. Monasson, Phys. Rev. Lett. 75, 2847 (1995).
- [26]G. Parisi and F. Zamponi, Rev. Mod. Phys. 82, 789 (2010).
- [27]L. Berthier, H. Jacquin, and F. Zamponi, Phys. Rev. E 84, 051103 (2011).
- [28]J. Kurchan, G. Parisi, and F. Zamponi, Journal of Statistical Mechanics: Theory and Experiment 2012, P10012 (2012).
- [29]E. Gardner, Nuclear Physics B 257, 747–765 (1985).
- [30]J. Kurchan, G. Parisi, P. Urbani, and F. Zamponi, The Journal of Physical Chemistry B 117, 12979–12994 (2013).
- [31]P. Charbonneau, Y. Jin, G. Parisi, C. Rainone, B. Seoane, and F. Zamponi, Phys. Rev. E 92, 012316 (2015).
- [32]L. Berthier, P. Charbonneau, Y. Jin, G. Parisi, B. Seoane, and F. Zamponi, Proceedings of the National Academy of Sciences 113, 8397–8401 (2016).
- [33]A. Seguin and O. Dauchot, Phys. Rev. Lett. 117, 228001 (2016).
- [34]G. Biroli and P. Urbani, SciPost Physics 4 (2018), 10.21468/scipostphys.4.4.020.
- [35]C. Scalliet, L. Berthier, and F. Zamponi, Phys. Rev. E 99, 012107 (2019).
- [36]T. Morita and K. Hiroike, Progress of Theoretical Physics 23, 1003–1027 (1960).
- [37]J.-P. Hansen and I. R. McDonald, Theory of simple liquids: with applications to soft matter (Academic press, 2013).
- [38]Y. Tomita and H. Yoshino, The Journal of Chemical Physics 163 (2025), 10.1063/5.0300245.
- [39]J. A. Giannini, E. Lerner, F. Zamponi, and M. L. Manning, The Journal of Chemical Physics 160 (2024), 10.1063/5.0176713.
- [40]K. Shiraishi and H. Mizuno, Soft Matter 22, 2017–2025 (2026).
- [41]M. Mangeat and F. Zamponi, Phys. Rev. E 93, 012609 (2016).
- [42]B. Schmid and R. Schilling, Phys. Rev. E 81, 041502 (2010).
