Open analytic problems for the instantaneous normal mode spectrum of low-density liquids: closed-form tail coefficient, extension to $d>1$, and localization edges

The instantaneous normal mode (INM) spectrum -- the density of eigenvalues of the Hessian of a liquid's Hamiltonian averaged over the equilibrium Boltzmann distribution -- is a standard microscopic descriptor of short-time liquid dynamics and of the unstable-mode sector implicated in mode-coupling-theory accounts of the glass transition. Cavagna, Giardina, and Parisi (Phys. Rev. Lett. 83, 108 (1999)) computed this spectrum exactly within a low-density, pair-factorized approximation for a one-dimensional soft-sphere liquid, mapping the Hessian onto a diluted (sparse) random-Laplacian ensemble and solving a closed self-consistent equation that goes beyond the Gaussian ansatz of earlier analytic treatments. Their paper and companion review leave a coherent set of analytic problems open: (1) the exponential high-eigenvalue tail $D(\lambda)\sim e^{-\alpha\lambda}$ was established numerically, with $\alpha$ increasing with the scaled inverse temperature, but no closed-form $\alpha(\beta_s,m)$ was obtained; (2) the computation is restricted to $d=1$, where the Hessian is positive definite, while the physically relevant $d>1$ case has negative eigenvalues whose extended-mode fraction is tied to mode-coupling scenarios; and (3) the localization edges separating extended and localized modes were found only numerically via inverse participation ratios -- alongside scope caveats recorded by the authors (the restriction of the low-density scaling argument to homogeneous potentials, and the delicate $\epsilon\to 0$ limit of the bimodal comparison ensemble) that remain unsettled. Later work in the adjacent Euclidean-random-matrix and sparse-random-matrix lines (notably Ciliberti et al., Phys. Rev. B 71, 153104 (2005)) extended the program to three dimensions only as numerically solved random-graph equations, leaving the analytic core -- the tail coefficient, analytic mobility edges, and a controlled $d>1$ computation -- open.

Background

Short-time liquid dynamics and the INM framework. At times shorter than the structural relaxation time, a liquid behaves 'solid-like': particles vibrate about quasi-equilibrium positions, with occasional jumps between local minima of the potential energy surface. The instantaneous normal mode (INM) approach to liquid-state dynamics (reviewed by Keyes) exploits this by diagonalizing the Hessian matrix $A$ of the Hamiltonian $H=\sum_{k>l}V(r_{kl})$ -- the matrix of second derivatives of the potential energy with respect to the particle coordinates -- at instantaneous equilibrium configurations. Its eigenvalues $\lambda$ define the INM spectrum $D(\lambda)$, the density of eigenvalues normalized per particle; its eigenvectors are the instantaneous normal modes; and $\omega=\sqrt{\lambda}$ plays the role of a vibrational frequency, so that a negative $\lambda$ corresponds to an imaginary frequency, i.e. a locally unstable direction of the energy landscape. Because $A$ is sampled over the Boltzmann distribution, $D(\lambda)$ is an equilibrium property. The fraction and spatial character of unstable modes have been tied to diffusion and to the mode-coupling theory (MCT) dynamical transition: Bembenek and Laird argued that this transition, which marks the crossover from non-Arrhenius to Arrhenius viscosity behavior in supercooled liquids, occurs when the fraction of negative delocalized modes drops to zero. An extended mode involves an $O(N)$ number of particles; a localized mode involves only a few. The boundary eigenvalues between extended and localized regions are called localization edges or mobility edges.

Prior analytic treatments and their limitation. Wu and Loring, and later Wan and Stratt (both characterized in the source paper), computed the INM spectrum analytically by liquid-theory methods, but both introduced a Gaussian ansatz for the auxiliary field distribution $\Omega(\phi)$ that enters the resolvent computation (the resolvent $G(\lambda)=(\lambda-A)^{-1}$ determines the spectrum through $D(\lambda)=-\lim_{\epsilon\to0}\mathrm{Im}\,\mathrm{Tr}\,G(\lambda+i\epsilon)/\pi N$). A Gaussian $\Omega$ enforces a quadratic effective action and therefore cannot represent the non-quadratic field statistics expected to control nontrivial spectral features such as localization.

The source result. Cavagna, Giardina, and Parisi considered $N$ soft spheres, $V(r)=1/r^m$, in one spatial dimension at low density, where three-particle correlations are negligible. The off-diagonal Hessian elements $J_{kl}=V''(r_{kl})$ then become independent random variables, and the row-sum-zero condition imposed by translational invariance makes $A$ a random Laplacian built on a diluted (sparse) random matrix: most entries are zero, and each row carries only $O(1)$ non-zero entries with a continuous distribution $p(J)$ induced by the pair correlation function $g^{(2)}(r)=e^{-\beta V(r)}$ (the probability of finding a particle at separation $r$ from another, relative to an ideal gas, taken at the first order of the virial expansion, the low-density expansion in the density; soft-sphere homogeneity makes density and temperature scale, so at fixed unit density the scaled inverse temperature $\beta_s=\beta\,[m(m+1)]^{1/(m+2)}$ is the single control parameter). The element distribution is heavy-tailed at the origin, $p(J)\sim e^{-\beta_s J^b}\,J^{-(1+c)}$ with $b=m/(m+2)$ and $c=1/(m+2)$, and is regularized by an infrared cutoff (the potential is set to zero beyond a distance $\bar r$, i.e. a minimum element size $\eta=V(\bar r)$). Using replicas (the disordered-systems technique of averaging over the disorder by introducing fictitious copies of the system) and a cavity-like single-site recursion for the resolvent (a self-consistency equation built by integrating out all sites but one, in the spirit of the Cizeau-Bouchaud treatment of Levy matrices -- random matrices with independent, power-law-distributed entries -- and of the diluted random matrices of Bray and Rodgers), the authors derived a closed nonlinear self-consistent integral equation for the function $g(x)$ that measures the deviation of $\Omega$ from a Gaussian; they proved $g(x)\sim -x^{2c}$ at large argument, so $\Omega$ is definitely non-Gaussian, and solved the equation numerically by fixed-point iteration, obtaining $D(\lambda)$ exactly (without further approximation) within the diluted ensemble, including the localized sector. They validated the result against direct diagonalization of sampled finite-$N$ matrices, computed also the spectrum of the classic Bray-Rodgers bimodal ensemble ($p(J)=\delta(J)+(p/N)[\delta(J-1/p)-\delta(J)]$ with connectivity $p$), improving on the approximate single-defect solution of Biroli and Monasson, and studied the inverse participation ratio (IPR) $Y(\lambda_\alpha)=\sum_i w_{\alpha i}^2$, where $w_{\alpha i}=|\langle\lambda_\alpha|i\rangle|^2$ is the weight of site $i$ in eigenmode $\alpha$ ($Y=O(1)$ for localized modes, $O(1/N)$ for extended ones), finding two localization edges with localized tails (IPR tending to $1/2$ for strongly bound pairs at large $\lambda$, and to $1$ for nearly decoupled single particles as $\lambda\to0$).

What was left open. The authors explicitly left the following for future work: the closed-form dependence of the tail coefficient $\alpha$ in $D(\lambda)\sim e^{-\alpha\lambda}$ for $\lambda\gg1$ on the microscopic parameters ($\beta_s$ and $m$; numerically $\alpha$ increases monotonically with $\beta_s$); the extension to $d>1$, where each pair coupling becomes a tensor and the Hessian is no longer positive definite, so negative eigenvalues appear -- the sector tied to the MCT-type scenarios above; and a detailed analytic localization study (mobility edges, critical states, finite-size scaling) beyond the numerical IPR results. As scope caveats, the source also records the restriction of the low-density scaling argument to homogeneous potentials and the delicate $\epsilon\to0$ (resolvent-regularization) limit of the bimodal ensemble.

Problem Statement

Extend the analytic computation of the instantaneous normal mode (INM) spectrum of low-density liquids, as established for $d=1$ by Cavagna-Giardina-Parisi, to full analytic control of the same diluted Hessian ensembles (defined by the source paper's pair-factorized distribution of the Hessian elements). The common objective is derivation-based understanding of the spectrum; the requested, jointly supporting results, each an analytic statement about these ensembles obtained by derivation, are:

(1) Tail coefficient. Derive the dependence of the coefficient $\alpha$ of the exponential high-eigenvalue tail, $D(\lambda)\sim e^{-\alpha\lambda}$ for $\lambda\gg1$, on the microscopic parameters -- the scaled inverse temperature $\beta_s$ and the soft-sphere exponent $m$ (numerically, $\alpha$ increases monotonically with $\beta_s$) -- in closed form, or as an equivalent explicit characterization.

(2) Higher dimensions. Extend the computation to $d>1$, where each pair coupling becomes a $d\times d$ tensor combining $V''$ and $V'/r$ terms, the Hessian is no longer positive definite, and negative eigenvalues appear: formulate the corresponding self-consistent equations, obtain the spectrum including the negative-eigenvalue sector, and give the parameter dependence of the fraction of negative extended (delocalized) modes that mode-coupling-type scenarios tie to the dynamical transition.

(3) Localization. Determine analytically the localization (mobility) edges of the spectrum -- the two boundaries found numerically in the inverse-participation-ratio analysis -- and characterize the edge eigenstates through their inverse-participation-ratio tails, their localization lengths, and their multifractal dimensions, together with the finite-size scaling of these localization quantities.

References

1. Andrea Cavagna, Irene Giardina, Giorgio Parisi, 'Analytic Computation of the Instantaneous Normal Modes Spectrum in Low-Density Liquids', Phys. Rev. Lett. 83, 108-111 (1999). DOI: 10.1103/PhysRevLett.83.108; arXiv:cond-mat/9903155. URL: https://doi.org/10.1103/PhysRevLett.83.108 (source paper)
2. Andrea Cavagna, Irene Giardina, Giorgio Parisi, 'A new approach for the analytic computation of the instantaneous normal modes spectrum', J. Phys.: Condens. Matter 12, 6295-6304 (2000). DOI: 10.1088/0953-8984/12/29/301; arXiv:cond-mat/9910289. URL: https://doi.org/10.1088/0953-8984/12/29/301 (companion review)
3. Ten-Ming Wu, Roger F. Loring, 'Phonons in liquids: A random walk approach', J. Chem. Phys. 97, 8568-8575 (1992). DOI: 10.1063/1.463375. URL: https://doi.org/10.1063/1.463375 (prior analytic INM computation, characterized in the source paper as using a Gaussian ansatz)
4. Yi Wan, Richard M. Stratt, 'Liquid theory for the instantaneous normal modes of a liquid', J. Chem. Phys. 100, 5123-5138 (1994). DOI: 10.1063/1.467178. URL: https://doi.org/10.1063/1.467178 (prior analytic INM computation, characterized in the source paper as using a Gaussian ansatz)
5. T. Keyes, 'Instantaneous Normal Mode Approach to Liquid State Dynamics', J. Phys. Chem. A 101, 2921 (1997). DOI: 10.1021/jp963706h. URL: https://doi.org/10.1021/jp963706h (review of the INM approach)
6. Scott D. Bembenek, Brian B. Laird, 'The role of localization in glasses and supercooled liquids', J. Chem. Phys. 104, 5199 (1996). DOI: 10.1063/1.471147. URL: https://doi.org/10.1063/1.471147 (negative delocalized modes and the MCT transition scenario)
7. A. J. Bray, G. J. Rodgers, 'Diffusion in a sparsely connected space: A model for glassy relaxation', Phys. Rev. B 38, 11461 (1988). DOI: 10.1103/PhysRevB.38.11461. URL: https://doi.org/10.1103/PhysRevB.38.11461 (bimodal diluted random Laplacian ensemble)
8. G. J. Rodgers, A. J. Bray, 'Density of states of a sparse random matrix', Phys. Rev. B 37, 3557 (1988). DOI: 10.1103/PhysRevB.37.3557. URL: https://doi.org/10.1103/PhysRevB.37.3557 (early sparse random matrix density of states)
9. P. Cizeau, J. P. Bouchaud, 'Theory of Lévy matrices', Phys. Rev. E 50, 1810 (1994). DOI: 10.1103/PhysRevE.50.1810. URL: https://doi.org/10.1103/PhysRevE.50.1810 (recursive resolvent method for heavy-tailed random matrices)
10. G. Biroli, R. Monasson, 'A single defect approximation for localized states on random lattices', J. Phys. A: Math. Gen. 32, L255-L261 (1999). DOI: 10.1088/0305-4470/32/24/101; arXiv:cond-mat/9902032. URL: https://doi.org/10.1088/0305-4470/32/24/101 (approximate bimodal-ensemble solution improved on by the source paper)
11. M. Mézard, G. Parisi, A. Zee, 'Spectra of Euclidean random matrices', Nucl. Phys. B 559, 689-701 (1999). DOI: 10.1016/S0550-3213(99)00428-9; arXiv:cond-mat/9906135. URL: https://doi.org/10.1016/S0550-3213(99)00428-9 (Euclidean random matrix framework)
12. A. Zee, Ian Affleck, 'Hopping between random locations: spectrum and instanton', J. Phys.: Condens. Matter 12, 8863-8873 (2000). DOI: 10.1088/0953-8984/12/41/312; arXiv:cond-mat/0006342. URL: https://doi.org/10.1088/0953-8984/12/41/312 (instanton tail computation for a non-translationally-invariant Euclidean random matrix class)
13. S. Ciliberti, T. S. Grigera, V. Martin-Mayor, G. Parisi, P. Verrocchio, 'Anderson localization in Euclidean random matrices', Phys. Rev. B 71, 153104 (2005). DOI: 10.1103/PhysRevB.71.153104; arXiv:cond-mat/0403122. URL: https://doi.org/10.1103/PhysRevB.71.153104 (main continuation: 3D INM density of states and numerical mobility edge via population dynamics)
14. Stefano Ciliberti, Tomas S. Grigera, 'Localization threshold of instantaneous normal modes from level-spacing statistics', Phys. Rev. E 70, 061502 (2004). DOI: 10.1103/PhysRevE.70.061502; arXiv:cond-mat/0404469. URL: https://doi.org/10.1103/PhysRevE.70.061502 (numerical mobility-edge determination)
15. Giorgio Parisi, 'Euclidean random matrices: solved and open problems', arXiv:cond-mat/0512004 (2005); also published as a book chapter in 'Applications of Random Matrices in Physics', ed. E. Brezin (Springer, NATO Science Series II: Mathematics, Physics and Chemistry), pp. 219-260. DOI: 10.1007/1-4020-4531-X_7. URL: https://arxiv.org/abs/cond-mat/0512004 (status review)
16. Reimer Kühn, 'Spectra of sparse random matrices', J. Phys. A: Math. Theor. 41, 295002 (2008). DOI: 10.1088/1751-8113/41/29/295002; arXiv:0803.2886. URL: https://doi.org/10.1088/1751-8113/41/29/295002 (sparse random matrix spectra and localization transitions)
17. Fernanda P. C. Benetti, Giorgio Parisi, Francesca Pietracaprina, Gabriele Sicuro, 'Mean-field model for the density of states of jammed soft spheres', Phys. Rev. E 97, 062157 (2018). DOI: 10.1103/PhysRevE.97.062157; arXiv:1804.02705. URL: https://doi.org/10.1103/PhysRevE.97.062157 (related random-graph Hessian program at jamming)
18. Stefano Mossa, Taras Bryk, Giancarlo Ruocco, Walter Schirmacher, 'Heterogeneous-elasticity theory of instantaneous normal modes in liquids', Sci. Rep. 13, 21442 (2023). DOI: 10.1038/s41598-023-46248-z. URL: https://doi.org/10.1038/s41598-023-46248-z (alternative continuum approach to 3D INM spectra)
19. Jeferson D. da Silva, Diego Tapias, Peter Sollich, Fernando L. Metz, 'Spectral properties, localization transition and multifractal eigenvectors of the Laplacian on heterogeneous networks', SciPost Phys. 18, 047 (2025). DOI: 10.21468/SciPostPhys.18.2.047; arXiv:2408.13322. URL: https://doi.org/10.21468/SciPostPhys.18.2.047 (modern sparse-Laplacian localization study)
20. Pietro Valigi, Joseph W. Baron, Izaak Neri, Giulio Biroli, Chiara Cammarota, 'Eigenvalue spectral tails and localization properties of asymmetric networks', arXiv:2507.20225 (2025). URL: https://arxiv.org/abs/2507.20225 (recent analytic Lifshitz-tail theory for sparse asymmetric ensembles)
21. Ban-Chiech Huang, Cheng-Hung Chang, 'Localization-delocalization transition of the instantaneous normal modes of liquid water', Phys. Rev. E 88, 042116 (2013). DOI: 10.1103/PhysRevE.88.042116. URL: https://doi.org/10.1103/PhysRevE.88.042116 (numerical localization study of INM in a real liquid)
22. B. J. Huang, Ten-Ming Wu, 'Multifractality of instantaneous normal modes at mobility edges', Phys. Rev. E 82, 051133 (2010). DOI: 10.1103/PhysRevE.82.051133. URL: https://doi.org/10.1103/PhysRevE.82.051133 (numerical multifractality study of INM at mobility edges)