betaFeedback

WorksPhysics and Astronomy

Colloidal gel morphology and phase behavior differs based on the electrostatic interaction model in concentrated electrolytes

Abstract

Predicting how nanoparticles interact in concentrated electrolytes remains a central challenge in nanoscience, with implications spanning energy storage, catalysis, and soft materials design. While Derjaguin-Landau-Verwey-Overbeek (DLVO) theory provides a widely used framework for modeling colloidal interactions, its electrostatic component relies on electric double-layer (EDL) theories whose validity in ionic liquids and concentrated electrolytes is consistently debated. Experiments have revealed significant deviations from classical screening behavior, including underscreening and oscillatory charge ordering, yet the consequences of these nonclassical effects for collective nanoparticle behavior remain unresolved. Here, we demonstrate that the extended range and oscillatory nature of electrostatic screening in concentrated electrolytes predicts different phase behavior and morphology of colloidal ionogels, yielding distinct thermodynamic signatures, percolation thresholds, and amorphous network structures. Using molecular dynamics simulations with DLVO-type pair interactions parameterized by competing EDL models, we systematically compare the consequences of different screening theories on gel formation and configurations.

Introduction

Understanding nanoparticle interactions in concentrated electrolytes is a fundamental challenge in nanoscience, with direct significance for technologies ranging from electrochemical energy storage[1, 2] to nanoparticle-enabled catalysis[3] and soft functional materials.[4] Ionic liquids, in particular, provide a highly tunable and strongly correlated environment, enabling the formation of complex materials such as colloidal ionogels.[5, 6] For decades, Derjaguin–Landau–Verwey–Overbeek (DLVO) theory has served as the standard framework for modeling colloidal interactions, combining van der Waals attraction and electrostatic repulsion into an effective pair potential.[7] This coarse-grained description is essential for bridging the large length and time scale disparity between nanoparticles and solvent molecules, which prohibits atomistic resolution.

DLVO theory relies critically on the so-called “screening” of nanoparticle surface charges by ions in the surrounding electrolyte. Classical models of the electric double layer (EDL), notably the Gouy-Chapman-Stern model, employ mean-field approximations to describe an exponential decay of electrostatic interactions over the Debye screening length.

However, extensive experimental evidence has shown that this picture breaks down in concentrated electrolytes and ionic liquids. Surface force measurements revealed that the characteristic length of the exponential decay of the electrostatic interaction, λS\lambda_S, becomes larger than the Debye length λD\lambda_D at high ion concentration,[8, 9, 10, 11] leading to longer-ranged electrostatic surface interactions. These deviations are attributed to ion-ion correlations, the ion size and shape, and molecular ordering effects that are necessarily neglected by mean-field approximations.[12, 13, 14] Others have demonstrated that high ion concentrations can give rise not only to longer-ranged interactions, but also to oscillatory charge-density profiles associated with ion layering and self-assembly near charged surfaces.[12, 15, 16, 17]

Despite significant theoretical and experimental progress, the microscopic origins and macroscopic consequences of these nonclassical EDL behaviors are not fully understood. Direct experimental probes, such as scanning probe techniques,[18] and X-ray-based methods[19] can resolve ion structuring near interfaces but are typically limited to only a few molecular layers and struggle to capture the fluctuating diffuse region. At the same time, atomistically resolving ion dynamics at nanoparticle surfaces remains prohibitively expensive for simulations of collective colloidal behavior. As a result, it is still unclear to what extent competing EDL theories influence emergent, mesoscale properties of colloidal materials.

alizability of this work.

Figure 1. Differing theories of ion order at charged surfaces. The colored shade represents net charge distribution: red for positive and blue for negative. The electrostatic potential follows similar decay as well. Debye-Hückel theory treats ions as point charges without explicit ion-ion correlation and shows the shortest smooth decay. We can extend the simple decay to compensate for the overpopulation near the surface in Debye-Hückel theory and account for underscreening observed in experiments. Oscillatory decay arises when we include the effect of finite size of the ions. The discrete ions in the figure represent this nature that distinguishes this model from the other two.

This gap itself represents an opportunity: colloidal particles are experimentally-resolvable objects whose collective behavior can be characterized via optical microscopy. If the differences in ion ordering near nanoparticle surfaces lead to meaningful differences in colloidal interactions, then macroscopic structural and thermodynamic behavior can serve as an indirect indicator of interfacial electrostatics.

In this work, we establish this connection by systematically investigating how competing models of electrostatic screening influence colloidal assembly in ionic liquids. Using molecular dynamics simulations of a model system with DLVO-type pair interactions parameterized by 1) classical Debye–Hückel decay, 2) longer-ranged monotonic decay (henceforth called “underscreened”), or 3) oscillatory decay consistent with ion layering, we quantify differences in phase behavior, local structure, and mesoscale morphology. We show that changes in the form of electrostatic interactions lead to pronounced divergences in gel topology and thermodynamic signatures, even when local structural descriptors remain nearly indistinguishable. These results demonstrate that macroscopic colloidal behavior provides a sensitive probe of interfacial electrostatics and reveal that EDL models with different features lead to different predictions for nanoparticle assembly in concentrated electrolytes. We find that several other colloidal ionogels share similar differences in pair interactions arising from different electrostatic interaction models, implying the potential generalizability of this work.

Results and Discussion

The following results are based on equilibrium NVT molecular dynamics simulations of three different electrostatic double layer theories, as incorporated into the DLVO potential (Figure 2) and performed over several simulation replicates. For our study, we assume bare silica nanoparticles (r=100r = 100 nm) submerged in an ionic liquid, 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl)imide ([EMIM][TFSI]), as a technologically-relevant[20] model system. The relevant properties are listed in Table I, which result in a screening length of roughly 0.05 nm (DH) and 5 nm (underscreened). Details on simulation setup and protocols can be found in the Methods section. Longer screening length under the underscreened model results in the emergence of a local energetic maximum at around d=5d = 5 nm and the following secondary minimum (Figure 2).

Figure 2. The DLVO pair potentials (UDLVOU_{\mathrm{DLVO}}, solid line) of our model colloidal system, resulting from the different electrostatic potentials (UelectrostaticU_{\mathrm{electrostatic}}, dashed line). The extension of electrostatic screening length from 0.05 nm in Debye-Hückel to 5 nm in Underscreened leads to the emergence of an energetic maximum around d=5d = 5 nm. Additional inclusion of oscillatory screening results in several energy maxima and minima. (b) and (c) show parts of (a) in detail. The equations and parameters used to derive the pair potentials are in Methods and Table I.

The trends discussed in the following sections have the potential to be generalized in colloidal systems with diverse ionic liquids, even though we performed simulations on a model system. As context, we collected bulk properties of 34 ionic liquids that had enough information to construct the DLVO-type pair potentials using the same equations we used for [EMIM][TFSI]. All the ionic liquids except for two—2-hydroxyethylammonium formate and tris(2-hydroxyethyl)methylammonium methylsulfate—demonstrate similar qualitative features as our model system, showing the existence of a maximum and a secondary minimum in their pair potentials (Figure 3). In Figure 3, UUS∗/UDH∗U^*_{US}/U^*_{DH} represents the significance of the long-range electrostatic component in the pair potential; higher significance leads to a lower value, considering UUS∗/UDH∗→1U^*_{US}/U^*_{DH} \to1 when the magnitude of electrostatic interaction is negligible compared to that of the van der Waals interaction. λ∗/λD\lambda^*/\lambda_D represents the extent of nonclassical deviation from the Debye-Hückel theory; larger deviation leads to a higher value. Most ionic liquids have similar or smaller UUS∗/UDH∗U^*_{US}/U^*_{DH} compared to [EMIM][TFSI]. In addition, λ∗/λD\lambda^*/\lambda_D values are mostly similar to or larger than [EMIM][TFSI] and even many with smaller λ∗/λD\lambda^*/\lambda_D still show secondary minimum in their pair potentials. This implies that the different electrostatic interaction models can lead to different predictions in the phase and structural behavior of colloidal ionogels other than our model system.

Figure 3. The features of pair potential of 34 ionic liquids under the Debye-Hückel theory or the underscreened (US) model, assuming constant surface potential (ϕs=−40\phi_s = -40 mV) condition. λD\lambda_D is the Debye length and λ∗\lambda^* is the decay length under the US model. The color represents the existence of a secondary minimum, hence a local maximum, of each pair potential. U∗U^* is the depth of the primary energy minimum, calculated by the difference between the primary minimum and either maximum (if energy maximum exists) or zero (if it does not exist). The point size represents the magnitude of dielectric constants. [EMIM][TFSI] is marked as green.

Energetics and phase behavior

To analyze how different electrostatic interaction theories impact collective colloidal phases, we investigated phase behavior through the analyses of the potential energies and the heat capacity. The potential energy per particle (Figure 4, solid lines) shows an increase across particle fractions, indicating a decrease in the average coordination number as branching occurs in particulate gels. This is confirmed by multiplying the average coordination number and the value of the potential minimum (dashed lines), which shows that these two are the main factors affecting the potential energy.

Figure 4. Potential energy per particle for different UelectrostaticsU_{\mathrm{electrostatics}} along the particle fraction. The solid lines are the potential energy values from molecular dynamics simulations. The dashed lines represent the multiplication of primary energy minimum in Figure 2 and the normalized mean coordination number of the particles.

The isochoric heat capacity (cVc_V, Figure 5) provide a key understanding of phase boundaries and evolution. Since the simulations are conducted at a constant temperature, the observed discontinuity in cVc_V directly indicate the discontinuity in the second derivative of the intensive Helmholtz free energy, implying the phase transition. In the Debye-Hückel case, this discontinuity occurs between ϕ=8.5%\phi= 8.5\% and 9.8%9.8\%, coinciding with the complete transition to a 3D-percolated network. In the longer-range decay (underscreening), we also observe a discontinuity in cVc_V, albeit at a much lower particle fraction (ϕ=4.6%−5.9%\phi= 4.6\% - 5.9\%). In contrast to the other models, the oscillatory model shows no distinct cVc_V discontinuity.

Figure 5. Intensive volumetric heat capacity (cVc_V) for different UelectrostaticsU_{\mathrm{electrostatics}} along the particle volume fraction. For Debye-Hückel and underscreened models, the discontinuity in cVc_V is observed at the volume fraction between two separated lines. Each line is a least-square fit based on nine data points (three volume fraction values) it passes.

After full percolation, we see a weaker signature of

the phase transition between gel and glassy states, as signified by the leveling off of cVc_V across all theories at ϕ=12.5%−17.5%\phi= 12.5\% - 17.5\%. Considering cVc_V serves as a proxy for the system’s available degrees of freedom, this still indicates that the systems develop into their more arrested, glassy structures where increase particle concentration restricts configurational freedom.

Local configurations

To quantify structural changes across different theories, we utilized two local descriptors: coordination number and Smooth Overlap of Atomic Positions (SOAP).[21] Coordination number is a simple and intuitive way to describe the local configuration, but includes little information about the packing or distribution of neighboring particles. SOAP accounts for this nuance by placing a Gaussian distribution on each particle center and evaluating the resulting local particle density. SOAP has proven to be effective for distinguishing local environments for self-assembled materials,[22] making it well suited for characterizing local structure of our systems.

At low particle fractions, all three gels show flatter distributions of coordination numbers, corresponding to non-dense, variably-coordinated systems. As particle fraction increases, all three theories demonstrate a more well-defined peak near 6-8 neighbors, with a small difference in the average coordination number between the theories (with underscreening corresponding to approximately 0.1-0.4 more neighbors/particle). Part of this trend can be interpreted as an effect of the deeper primary minimum of the interaction potential (Figure 2), based on the same trend observed in a simulation study of a non-dense colloidal system under the Morse potential of different interaction strengths.[23] As the particle fraction increases, the coordination numbers of differing theories converge. To more exhaustively quantify local coordination, we employ SOAP descriptors, which provide a mathematical signature to the coordination environments, taking into account radial and angular components to the neighboring particles. Despite the sophistication of these descriptors and our employment of a mapping technology, PCovC, specifically intentioned to separate distributions in latent space,[24] these local descriptors show the same qualitative similarities between theories. This is demonstrated in Fig. S1 (showing the PCA) and Figure 7, where the PCA shows full overlap of these statistical distributions, albeit with some small shifts for the underscreened model.

Figure 7. Distribution of local environments represented by SOAP and mapped by PCovC, based on (a) 1st (rcut=201r_{\mathrm{cut}} = 201 nm) and (b) 2nd (rcut=401r_{\mathrm{cut}} = 401 nm) neighbor shell analysis across different theories. PCovC provides a latent-space visualization of the data that best separates the distributions corresponding to each theory. In other words, this constitutes the “viewing angle” on the high-dimensional data that best separates the configurations resulting from each theory. The Debye-Hückel and the oscillatory potential show almost same distributions. The underscreened potential exhibits some differences along PCov1 axis, but mostly overlaps with the other two.

Global topology

Finally, we use the Euler Characteristic (EC), a feature vector that encodes the topology of the particle point cloud, as a metric to compare the resultant gel network morphologies. EC has been shown to be a powerful and information-dense description of several systems.[25, 26, 27] Importantly, it has been shown to be closely correlated with the mechanical and rheological properties of colloidal gels such as elastic and viscous moduli.[28] We randomly extract 10 snapshots after the equilibration period of each simulation and calculate their EC vectors. Three simulations with different random initial configurations were run for every (electrostatic model, ϕ\phi) pair, resulting in 30 EC vectors for each particle fraction and theory.

Notably, each snapshot corresponds to one EC vector, unlike the local descriptors that are assigned to a particle. The EC encodes the particle configuration as an alternating sum of β0\beta_0, the number of clusters, β1\beta_1, the number of two-dimensional holes, and β2\beta_2, the number of void spaces across a scan of neighborhood distances rr (called the “filtration value” in topological data analysis). For example, a small void space in the middle of a tetrahedron of touching particles is captured in β2\beta_2 at a short rr, versus a large void resulting from the gel network is captured in β2\beta_2 at a longer rr.

Contrary to the local descriptors, the EC vectors clearly demonstrate key similarities and divergences between the morphologies stabilized by each electrostatic theory, even when projected along principal components, even though PCA is not natively intended to separate these distributions. In Figure 8, we show the first three principal components (PCs) of EC vectors for three electrostatic theories, where Debye-Hückel theory is in blue, underscreening theory is in red, and the oscillatory theory in yellow/green. The darker the color of the point, the higher the particle fraction and the error bars represent the standard deviation across data points for a given particle fraction and electrostatic theory. The first principal component clearly aligns with the particle fraction of each simulation, smoothly increasing from sol phases on the right to gel and glass phases on the left of Figure 8 (a). Our next two PCs, by definition, contain orthogonal information to the first PC and show clear divergences between theories – particularly between the Debye-Hückel and the oscillatory theories and the theory with the decay length of Lee et al. 11``.

Figure 8. PCA of the Euler Characteristic vectors for different electrostatic theories across particle fractions (ϕ\phi). PC1 aligns with ϕ\phi, while PC2 and PC3 capture theory-specific variations perpendicular to ϕ\phi. All theories overlap at low ϕ\phi but diverge at higher fractions. The underscreened potential differs significantly from the others. The Debye-Hückel and the oscillatory potentials remain relatively closer, still showing distinct differences especially at moderate ϕ\phi (5.9 − 9.8% in (a), 9.8 − 12.4% in (b)).

The differences in the second and third PCs among the theories become starker with a higher particle fraction, corresponding to the formation of longer-ranged structures. This trend disagrees with what the local descriptors show, where the differences decrease or remain similar along the particle fraction axis (Figure 6). This disagreement demonstrates that the structures of dense colloidal systems cannot be fully understood solely with these local descriptors that capture structural information up to the second neighbor shell. In other words, the local particle environment may not change, yet the global morphology, which is captured by EC, changes greatly between the different theories.

Figure 6. (a)–(c) Distribution of coordination numbers for different UelectrostaticsU_{\mathrm{electrostatics}} along the particle fraction. The darker colors represent higher particle fraction in each plot. The median coordination number for each particle fraction is presented as a dot with the same color on each line.

Surprisingly, the morphology derived from the oscillatory theory more closely aligns with that derived from Debye-Hückel theory. The same trend was observed in terms of structure factor S(q)S(q) (Fig. S7). This is of note, as the oscillatory potential retains higher electrostatic repulsions and attractions at longer length scales, while the Debye-Hückel theory gives an electrostatic potential and force that become negligible quickly as we move from the particle surface. Still, their structural topologies show meaningful differences in PC2 at ϕ=5.9−9.8%\phi= 5.9 - 9.8\% and PC3 at ϕ=9.8−12.4%\phi= 9.8 - 12.4\%. These ϕ\phi regimes overlap with the lack of distinct discontinuity in cVc_V under the oscillatory theory at ϕ=8.5−9.8%\phi= 8.5 - 9.8\% (Figure 5), implying that the DH and the oscillatory theory lead to different phase and structural transitions along the increase in ϕ\phi.

These results, namely the difference in sol-gel phase transition and three-dimensional percolation, suggest that extended screening leads to more open network gels of significantly different topology than phases suggested by other theories. To confirm this, we identified the pore sizes of each system across particle fraction. Across all particle fractions, the extended electrostatic interaction model consistently yields larger maximum pore sizes than the Debye-Hückel theory (Figure 9). The maximum pore sizes predicted by the oscillatory model fall between these two cases, approaching the Debye-Hückel results at high particle fractions. This trend corresponds to a systematic broadening of the pore size distribution toward larger volumes, rather than a few statistical outliers (Fig. S4). These findings codify that the extended range of electrostatic repulsion not only enables the formation of colloidal gels at lower particle fractions but also increases their openness relative to the shorter-ranged Debye-Hückel screening. This effect is presumably due to long-range repulsion between non-neighboring particles.

Figure 9. Maximum pore diameters of different electrostatic interaction models and particle volume fractions. The pore sizes were evaluated at 10 frames after equilibration timestep for each simulation, thus 30 distributions for each point plotted. Each point represents the mean and error bar represents the standard deviation of maximum pore diameters of corresponding distributions. The inset shows the relative sizes, the maximum pore diameters divided by the average value of those under the Debye-Hückel theory.

Conclusions

In this work, we have demonstrated that the specific form of the electrostatic interaction terms within DLVO-type potentials plays a decisive role in determining the phase behavior and structural morphology of colloidal systems in ionic liquids and concentrated electrolytes. By comparing three representative screening models: classical Debye-Hückel theory, an underscreened single-decay model, and an oscillatory decay model, we have shown that differences in screening length and functional form lead to distinct phase boundaries, percolation behavior, and gel morphologies.

We have demonstrated these differences through a variety of analysis and characterization techniques. First, we showed distinct differences in phase behavior across different electrostatic interaction models. The extended monotonic screening model demonstrated the discontinuity of cVc_V at a lower particle fraction (around 5%) compared to the Debye-Hückel theory (around 10%), while the oscillatory screening model showed no distinct discontinuity. This implies different phase behavior for all the three models studied. Local structural descriptors such as coordination number and SOAP vectors revealed only small differences between electrostatic models. Conversely, global analyses through persistent homology, captured through the Euler characteristic, demonstrated strong divergences in mesoscale topology that become pronounced at higher particle fractions. This is particularly interesting, as topological descriptors have been directly tied to the rheological properties of gel systems,[29] inferring that dynamical simulations of colloidal gels would reveal experimentally-measurable changes in gel rheology dependent on electrostatic models. Interestingly, oscillatory electrostatic interactions lead to gels topologically closer to those suggested by Debye-Hückel-type screening, yet, produce structural features distinct from the other two models. Finally, pore size analysis indicated that the increased screening length broadens the pore size distribution toward larger volumes. Hence, this work demonstrates that the formation of a highly open gel network can serve as a strong indicator of long-range interaction forces.

These findings also have important implications for computational simulations of colloidal materials. Implicit-solvent interaction models are widely used to enable simulations of large-scale systems at a tractable computational cost, particularly in the context of materials discovery where broad parameter spaces require high-throughput exploration. Our results suggest that the electrostatic assumptions embedded in these models strongly influence phase stability and structure, and the choice of screening length can artificially bias the results of simulations of implicit solvents representing charged electrolytes and solutions. In this regard, further studies on phase diagrams and gel formation mechanisms, spanning a wider range of temperature and particle fraction, would be valuable and provide more insights on the impact of long-range interactions onto colloidal behavior.[30]

More broadly, this sensitivity of collective colloidal properties to the underlying electrostatic model suggests that macroscopic observables, such as phase transitions, connectivity, and gel morphology, can serve as experimentally accessible fingerprints of interfacial ion ordering. Rather than relying on hard-to-obtain microscopic measurements of interfacial structure, future studies could combine experimental techniques such as scattering and confocal microscopy with simulations. They could provide a promising pathway for assessing and refining electric double-layer theories in complex ionic environments using experimentally accessible measurements.

Methods

Implementation of the DLVO potential

To model interactions, we use functionalities and interaction models provided by LAMMPS.[31] The van der Waals attraction term is

UvdW(d)=−A6[2r24rd+d2+2r2(2r+d)2+ln⁡(4rd+d2(2r+d)2)]U_{\mathrm{vdW}}(d)=-\frac{A}{6}\left[\frac{2r^{2}}{4rd+d^{2}}+\frac{2r^{2}}{(2r+d)^{2}}+\ln\left(\frac{4rd+d^{2}}{(2r+d)^{2}}\right)\right]

where rr is the radius of the nanoparticle, dd is the closest separation distance between the particle surfaces, and AA is the Hamaker constant, here assumed to be 1.583×10−211.583 \times10^{-21} J for silica nanoparticles, derived from Lifshitz theory.[32]

A=34kBT(ϵS−ϵMϵS+ϵM)2+3hνe162(nS2−nM2)2(nS2+nM2)3/2,A=\frac{3}{4}k_{\mathrm{B}}T\left(\frac{\epsilon_{\mathrm{S}}-\epsilon_{\mathrm{M}}}{\epsilon_{\mathrm{S}}+\epsilon_{\mathrm{M}}}\right)^{2}+\frac{3h\nu_{\mathrm{e}}}{16\sqrt{2}}\frac{\left(n_{\mathrm{S}}^{2}-n_{\mathrm{M}}^{2}\right)^{2}}{\left(n_{\mathrm{S}}^{2}+n_{\mathrm{M}}^{2}\right)^{3/2}},

where ϵM\epsilon_{\mathrm{M}} and ϵS\epsilon_{\mathrm{S}} are the dielectric constant of the solvent and particle, respectively, nn is the respective refractive index, kBk_{\mathrm{B}} is the Boltzmann constant, and TT is the temperature of the system. We introduce an additional steric term[33] to avoid particle overlap:

Equation not extracted
(1)
Usteric=ACC37800σ6d+2r[d2−10dr+30r2d7+d2+18dr+86r2(d+4r)7−2d2+4dr−26r2(d+2r)7](2)U_{\mathrm{steric}}=\frac{A_{\mathrm{CC}}}{37800}\frac{\sigma^6}{d+2r}\left[\frac{d^2-10dr+30r^2}{d^7}+\frac{d^2+18dr+86r^2}{(d+4r)^7}-2\frac{d^2+4dr-26r^2}{(d+2r)^7}\right] \tag*{(2)}

where σ\sigma is the characteristic length scale for silica, here assumed to be 0.295 nm.[34] For the electrostatic term, we use a tabulated potential derived from Sader et al.[35] which yields the following:

Uelectrostatic(d)=εMε0(kBTe0)2Y2(d)(r22r+d)ln⁡(1+e−d/λS)(3)U_{\mathrm{electrostatic}}(d)=\varepsilon_M\varepsilon_0\left(\frac{k_{\mathrm{B}}T}{e_0}\right)^2Y^2(d)\left(\frac{r^2}{2r+d}\right)\ln\left(1+e^{-d/\lambda_S}\right) \tag*{(3)}

where

Y(d,λS)=4ed/2λStanh⁡−1[e−d/2λStanh⁡(ys4)].Y(d,\lambda_S)=4e^{d/2\lambda_S}\tanh^{-1}\left[e^{-d/2\lambda_S}\tanh\left(\frac{y_s}{4}\right)\right].

This potential term is largely governed by the “screening length”, λS\lambda_S, the characteristic decay length of the electrostatic repulsion of the nanoparticles. For the λS\lambda_S, we evaluated the effect of different theories prevailing on these particles. Traditionally, the Debye-Hückel theory estimates the screening length as

λD=(εMε0kBT∑jnjqj2)1/2.\lambda_D=\left(\frac{\varepsilon_M\varepsilon_0 k_{\mathrm{B}}T}{\sum_j n_j q_j^2}\right)^{1/2}.

where {j}\{j\} includes all ionic species, njn_j is the number density of ion jj, qjq_j is the charge of ion jj, and ε0\varepsilon_0 is the vacuum permittivity. This is the simplest representation of the screening length, assuming no ion-ion correlations, and thus assumes screening to directly follow the Debye length. However, Debye-Hückel theory is a poor approximation when there exist many ion-ion correlations,[14] as is common in highly ionic systems. As a result, λD=0.06\lambda_D=0.06 nm for pure [EMIM][TFSI], which is smaller than the dimension of the constituent molecules.

The failure of Debye-Hückel theory is reflected as underscreening, a positive deviation of screening length from Debye length at high ion concentration. Here we incorporate this effect by setting an extended screening length λ∗\lambda^* following an experimental scaling relationship,[11]

λ∗λD∝(aλD)3/2.\frac{\lambda^*}{\lambda_D}\propto\left(\frac{a}{\lambda_D}\right)^{3/2}.

where aa is the size of the constituent ions. aa is calculated as v1/3v^{1/3} where vv is half of the volume of an ion pair. If we set λ∗/λD=(a/λD)3/2\lambda^*/\lambda_D=(a/\lambda_D)^{3/2}, λ∗=4.56\lambda^*=4.56 nm for [EMIM][TFSI], which is on the same order of experimental value of 4.4–8.3 nm.[9, 10, 36] This value also strongly corresponds with the Bjerrum length, which represents the length scale of electrostatic correlations between two charges in a system. The Bjerrum length of [EMIM][TFSI] calculated from the experimentally measured dielectric constant (Table I) is 4.6 nm. In this

we use a tabulated potential derived from Sader et al.[35] which yields the following:

context, this relationship was used to estimate λ∗\lambda^* and derive pair potential under the underscreened model for other ionic liquids in Figure 3.

Lastly, we consider the oscillatory nature of ion ordering at the nanoparticle surface suggested by Kjellander and Mitchell[17, 37] on top of longer-range screening. In this approach, the effective force between nanoparticles is given as a function of a complex-conjugate inverse screening length κ±κ′i\kappa\pm\kappa'i:

f(d)=Cexp⁡(−κd)2r+dcos⁡(κ′d+α)f(d)=C\frac{\exp(-\kappa d)}{2r+d}\cos(\kappa'd+\alpha)

where CC is a coefficient in units of energy, and α\alpha is a phase of charge oscillation. We construct f(d)f(d) to act as an oscillatory analog to the potential defined by λ∗\lambda^*. Namely, we assume parameters that make the non-oscillatory part of f(d)f(d) to follow Uelectrostatic(d)U_{\mathrm{electrostatic}}(d), which corresponds to κ−1=λ∗\kappa^{-1}=\lambda^*, κ′=2π/1.5a\kappa'=2\pi/1.5a, α=π/2\alpha=\pi/2.

This value of κ′\kappa' was determined by numerically solving the equations from Kjellander[37] to derive the complex decay length of simple spherical ions of the same size:[37]

[κ′κD]2=eκ′a1+ακ′\left[\frac{\kappa'}{\kappa_D}\right]^2=\frac{e^{\kappa'a}}{1+\alpha\kappa'}

where κD=λD−1\kappa_D=\lambda_D^{-1}. This force was then tabulated in place of UelectrostaticU_{\mathrm{electrostatic}} for LAMMPS simulation. This serves to represent the resultant effects of introducing oscillations into slowly-decaying DLVO interactions.

Simulation protocol

We initialize systems ranging from 1,024–10,462 particles in random configurations for concentrations ranging between 2–20% volume fraction of nanoparticles at room temperature (298.15K) employing a Nosé-Hoover thermostat. The well-dispersed random configurations were generated such that the surface distance between two particles is greater than 10 nm. The arrested systems, such as nonequilibrium colloidal gels, depend on the kinetic path chosen in the simulations such as an equilibration protocol. We aim to imitate the evolution of the system after ultrasonication, which is used to disperse colloidal particles in experimental studies,[42, 43] by simulating these systems in the canonical (NVT) ensemble with periodic boundary conditions. To ensure no finite size effects, we queried system sizes L=4−16L = 4 - 16 µm to test for energy convergence, where energies converged at L=6L = 6 µm (Fig. S5). The cutoff radius for the pair potential was set to 260 nm, corresponding to d=60d = 60 nm, to include the long-range nature of the electrostatic component. The magnitude of UDLVOU_{DLVO} was less than 0.01kBT0.01k_BT for all models, and the differences in UDLVOU_{DLVO} between the models were negligible at the cutoff; in the order of 10−6−10−5kBT10^{-6} - 10^{-5}k_BT.

PropertyValueDefinition
εS\varepsilon_S3.8Dielectric constant of particle (amorphous silica)[38]
εM\varepsilon_M12.3Dielectric constant of medium ([EMIM][TFSI])[39]
nSn_S1.475Refractive index of particle[40]
nMn_M1.422Refractive index of medium[41]
νe\nu_e3.00×10153.00 \times10^{15}Main electronic absorption frequency (Hz)[32]
σ\sigma0.295Dimension of constituent atoms of particle (nm)[34]
ψs\psi_s-0.04Surface potential (V)[42]
AA1.58Hamaker constant (10−2110^{-21} J)
λD\lambda_D0.06Debye length (nm)
λ∗\lambda^*4.6Rescaled screening length (nm)
κ\kappa0.22Real part of the complex conjugate inverse screening length (nm−1^{-1})
κ′\kappa'7.0Imaginary part of the complex conjugate inverse screening length (nm−1^{-1})

Table 1. Properties of the system materials, and the resulting screening lengths calculated for different models. Main electronic absorption frequency is assumed to be the same for both materials. The dimension of constituent atoms of silica is obtained by averaging the characteristic length of Lennard-Jones potential for Si and O in silica.

All simulations were run via LAMMPS software, with the time of equilibration determined by potential energy convergence, employing the algorithms of Chodera [44]. This resulted in simulations on the order of 20 µs–5 ms with a timestep size of 0.1 ns. We ran three separate replicas of each configuration at each volume fraction. Using the LAMMPS framework, we defined a unique trial as one that had a different random seed to initialize velocities and positions. This way, we ensured that a different particle configuration alone did not fundamentally change the final structure. With these completed trials, we extracted the thermodynamic statistics of the system, such as particle configuration and potential energy.

As a note, we do not implement Langevin dynamics; the random fluctuation force is significantly smaller than what is required to disrupt the steep pair potential well in Figure 2. Furthermore, the high viscosity of [EMIM][TFSI] (32.42 mPa · s,[45]) compared to other organic solvents would lead to prohibitively long equilibration times[46] and thus incorporating Langevin dynamics would incur significant computational overhead without much anticipated impact on equilibrium results.

We conducted our simulation on two high-performance computing systems: Anvil[47] at Purdue University through an allocation supported by NSF ACCESS,[48] and a high-performance computing cluster at the Center for High Throughput Computing[49] in the University of Wisconsin-Madison.

Thermodynamic and structural definitions of phase phenomena

The rigorous study of phase behavior in colloidal, disordered materials, is a challenge for molecular simulation and statistical theory, largely because of their dynamical behavior and amorphous morphology. There is considerable literature on characterizing phase phenomena in glassy or crystallizing systems,[50, 51, 52, 53, 54, 55, 56, 57] which, while relevant to many problems, does not extend naturally to the problem of gelating systems studied via implicit solvent models.

We investigate the phase behavior of our systems via response functions; this approach follows the classical Ehrenfest approach of determining phase boundaries based on divergences in free energy. Many thermodynamic derivatives, such as pressure and isothermal compressibility, are non-trivial or ill-founded in implicit solvent simulations.[58] More straightforward is the isochoric heat capacity CVC_V, which is related to the second derivative of Helmholtz free energy and can be calculated from the fluctuations of internal energy of the system:

CV=−T(∂2A∂T2)V,N=⟨E2⟩−⟨E⟩2kBT2C_V = -T\left(\frac{\partial^2 A}{\partial T^2}\right)_{V,N} = \frac{\langle E^2\rangle- \langle E\rangle^2}{k_B T^2}

For our simulations, we compute ⟨E2⟩\langle E^2\rangle and ⟨E⟩\langle E\rangle as moving averages with a window of 10,000 data points after the convergence of potential energies over simulation timeframes. We do this running average to avoid the problems of internal energy drift that can occur with slowly-equilibrating systems, as the fluctuations often converge well-before this timeframe (Fig. S6). CVC_V was calculated as an average of the corresponding values calculated from all the windows. Intensive volumetric heat capacity cVc_V is a number-averaged value of the extensive value: cV=CV/Nc_V = C_V/N.

Local descriptors of state

To quantify changes in structure across phase transitions, a classic approach is through structural or bond-order analysis; for non-crystallizing systems, however, these “order parameters” fall short of quantifying “disordered” structures.[59]

For our local structure characterization, we employ two local descriptors: coordination number (CN) and Smooth Overlap of Atomic Positions (SOAP).[21] While the former (CN) is a more standard approach, the latter (SOAP) assumes each particle represented as a density field and creates a structural signature by expanding over spherical harmonics and radial bases. Conceptually, this results in a vector which contains a sensitive description of the three-body correlation functions for any given particle. The cutoff radius (rcutr_{cut}) is set to 201 nm for both CN and SOAP to include the first neighbor shell of interactions. In order to analyze the structure of the second neighbor shell, the difference between SOAP vectors with rcut=401r_{cut} = 401 nm and rcut=201r_{cut} = 201 nm is evaluated. The maximum number of radial basis function (nn) and angular degree of spherical harmonics (ll) are set to 8 and 6, respectively. These two parameters are resolution parameters of SOAP; that is, increasing nn and ll allows finer decomposition of particle density field, but it does not make a meaningful difference in the results.

High-dimensional SOAP descriptors were analyzed by reducing their dimensionality with Principal Component Analysis (PCA) and Principal Covariates Classification (PCovC).[24] PCA maps high-dimensional data in a way that best preserves overall variance; PCovC is a hybrid mapping technique that additionally incorporates class information, herein defined as our choice of EDL theory. The importance ratio of preserving either variance or classes is determined by a mixing parameter, taken as 0.5 when both inputs (here the SOAP vectors) and labels (here the EDL theory) are properly scaled. In essence, PCovC shows how separable two distributions in a high-dimensional space are, and is here used to show that the distribution of nearest- and next-nearest-neighbors under different theories are near-identical in SOAP space. PCA and PCovC are computed via scikit-learn[60] and scikit-matter[61] Python packages, respectively. Due to the large amount of data, we use farthest point sampling[61, 62] to reduce the data to train the model on while maintaining data diversity. scikit-matter[61] was used to apply farthest point sampling to select 200 particles in each trajectory. The PCovC model is trained on the sub-selected data and then applied to the entire dataset of SOAP vectors.

Global descriptors of state

While local descriptors provide excellent insight into particle-level structures, mesoscale network structures are crucial to explain the properties of colloidal systems.[63] In this sense, topological data analysis (TDA) provides a powerful way of analyzing morphology in various systems by quantifying the shape of complex data with computational topology.[28, 29, 64, 65, 66]

To apply TDA, we first construct geometric objects such as points, edges, and faces from a set of particle coordinates. A popular TDA method for this purpose is persistent homology, which defines the connectivity of the points by introducing a filtration value.[67] A simple example of filtration value is the radius of spheres centered at particle coordinates, where two data points are considered to be connected if the two corresponding spheres touch each other. The connectivity of the system evolves as we increase the filtration value, and persistent homology captures this evolution of topology by sweeping through a wide range of filtration values.

From this evolving network, we extract topological summaries. Among these, Euler characteristic (EC) is particularly useful because it is closely correlated to the mechanical properties of colloidal gels.[28] Furthermore, the EC serves as a robust descriptor because it is stable against noise, such as minor thermal fluctuations.[67] The EC is an alternating sum of Betti numbers βi\beta_i:

EC=β0−β1+β2\mathrm{EC} = \beta_0 - \beta_1 + \beta_2

where β0\beta_0, β1\beta_1, and β2\beta_2 represent the number of isolated clusters, two-dimensional holes, and three-dimensional voids, respectively. An EC value is defined for each filtration value, so each configuration yields a high-dimensional EC vector for a list of filtration values. Principal Component Analysis (PCA) is used to project these EC vectors into lower-dimensional space, allowing us to examine the topological changes of our colloidal systems.

In addition to EC, pore size distributions were determined by converting particle configurations into a voxelized representation using ASE[68] and freud.[69] To capture complete pore distribution in periodic boundary condition, the particle coordinates were periodically replicated into a 2×2×22 \times2 \times2 supercell. Only the pores centered in the central 1×1×11 \times1 \times1 simulation box were considered. The structure was then discretized using cubic voxels of 20 nm side length. Each voxel was classified as ‘filled’ if its center is contained in one or more particles and as ‘void’ otherwise, yielding a binary field. The void space was then analyzed using image processing tools from SciPy[70] and scikit-image.[71] A Euclidean distance transform of SciPy was applied to calculate the distance to the nearest filled voxel from each void voxel, which represents the radius of the largest sphere that can be inscribed without overlapping the filled voxels. We then identified local maxima of the distance field which correspond to the pore centers using scikit-image. The resulting set of diameters was used to evaluate the pore size distribution.

The static structure factor S(q)S(q) was computed from the molecular dynamics trajectories using the Debye scattering method implemented in the freud[69] package. Ten configurations were sampled from the equilibrated portion of each trajectory by selecting one frame from each of ten equal temporal intervals. Particle coordinates, normalized by the particle diameter 2r2r, were used to calculate S(q)S(q) in the range (0.5×box length/2r)−1≤2rq≤1(0.5 \times\text{box length} / 2r)^{-1} \le2rq \le1.

Acknowledgments

Support for this research was primarily provided by the University of Wisconsin–Madison Office of the Vice Chancellor for Research via the Wisconsin Alumni Research Foundation, the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE-1747503, and the National Science Foundation Division of Materials Research under Grant No. DMR-2309000. We also acknowledge support from the Michael and Virginia Conway Professorship. The funders played no role in study design, data collection, analysis and interpretation of data, or the writing of this manuscript.

Research was sponsored by the Army Research Laboratory and was accomplished under Grant Number W911NF-26-1-A040. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Army Research Laboratory or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation herein.

This work used Anvil at Purdue University through allocation CHM250049 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by U.S. National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296.

References

References

  1. [1]Li, Y.; Fedkiw, P. S. Effect of gel electrolytes containing silica nanoparticles on aluminum corrosion. Electrochimica Acta 2007, 52, 2471–2477.DOI
  2. [2]Chen, N.; Zhang, H.; Li, L.; Chen, R.; Guo, S. Ionogel Electrolytes for High-Performance Lithium Batteries: A Review. Advanced Energy Materials 2018, 8, 1702675, eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/aenm.201702675.DOI
  3. [3]Yan, C.-C.; Li, W.; Liu, Z.; Zheng, S.; Hu, Y.; Zhou, Y.; Guo, J.; Ou, X.; Li, Q.; Yu, J.; Li, L.; Yang, M.; Liu, Q.; Yan, F. Ionogels: Preparation, Properties and Applications. Advanced Functional Materials 2024, 34, 2314408.DOI
  4. [4]Fan, X.; Liu, S.; Jia, Z.; Justin Koh, J.; Chuan Yeo, J. C.; Wang, C.-G.; Erdeanna Surat’man, N.; Jun Loh, X.; Bideau, J. L.; He, C.; Li, Z.; Loh, T.-P. Ionogels: recent advances in design, material properties and emerging biomedical applications. Chemical Society Reviews 2023, 52, 2497–2527.DOI
  5. [5]Ueno, K.; Watanabe, M. From Colloidal Stability in Ionic Liquids to Advanced Soft Materials Using Unique Media. Langmuir 2011, 27, 9105–9115.DOI
  6. [6]Le Bideau, J.; Viau, L.; Vioux, A. Ionogels, ionic liquid based hybrid materials. Chemical Society Reviews 2011, 40, 907–925.DOI
  7. [7]Israelachvili, J. N. Intermolecular and surface forces, 3rd ed.; Academic Press: Burlington, MA, 2011; OCLC: ocn656213150.
  8. [8]Gebbie, M. A.; Valtiner, M.; Banquy, X.; Fox, E. T.; Henderson, W. A.; Israelachvili, J. N. Ionic liquids behave as dilute electrolyte solutions. Proceedings of the National Academy of Sciences 2013, 110, 9674–9679.DOI
  9. [9]Smith, A. M.; Lee, A. A.; Perkin, S. The Electrostatic Screening Length in Concentrated Electrolytes Increases with Concentration. The Journal of Physical Chemistry Letters 2016, 7, 2157–2163.DOI
  10. [10]Gebbie, M. A.; Smith, A. M.; Dobbs, H. A.; Lee, A. A.; Warr, G. G.; Banquy, X.; Valtiner, M.; Rutland, M. W.; Israelachvili, J. N.; Perkin, S.; Atkin, R. Long range electrostatic forces in ionic liquids. Chemical Communications 2017, 53, 1214–1224.DOI
  11. [11]Lee, A. A.; Perez-Martinez, C. S.; Smith, A. M.; Perkin, S. Scaling Analysis of the Screening Length in Concentrated Electrolytes. Physical Review Letters 2017, 119, 026002.DOI
  12. [12]Perkin, S. Ionic liquids in confined geometries. Physical Chemistry Chemical Physics 2012, 14, 5052.DOI
  13. [13]Rotenberg, B.; Bernard, O.; Hansen, J.-P. Underscreening in ionic liquids: a first principles analysis. Journal of Physics: Condensed Matter 2018, 30, 054005.DOI
  14. [14]Elliott, G. R.; Gregory, K. P.; Robertson, H.; Craig, V. S. J.; Webber, G. B.; Wanless, E. J.; Page, A. J. The known-unknowns of anomalous underscreening in concentrated electrolytes. Chemical Physics Letters 2024, 843, 141190.DOI
  15. [15]Kirkwood, J. G. Statistical Mechanics of Liquid Solutions. Chemical Reviews 1936, 19, 275–307.DOI
  16. [16]Kirkwood, J. G.; Poirier, J. C. The Statistical Mechanical Basis of the Debye–Hückel Theory of Strong Electrolytes. The Journal of Physical Chemistry 1954, 58, 591–596.DOI
  17. [17]Kjellander, R.; Mitchell, D. J. Dressed ion theory for electric double layer structure and interactions; an exact analysis. Molecular Physics 1997, 91, 173–188, eprint: https://doi.org/10.1080/00268979717490.
  18. [18]Schott, C. M.; Schneider, P. M.; Song, K.-T.; Yu, H.; Götz, R.; Haimerl, F.; Gubanova, E.; Zhou, J.; Schmidt, T. O.; Zhang, Q.; Alexandrov, V.; Bandarenka, A. S. How to Assess and Predict Electrical Double Layer Properties. Implications for Electrocatalysis. Chemical Reviews 2024, 124, 12391–12462, PMID: 39527623.DOI
  19. [19]Wang, P.; Li, Y.; Wang, L.; Klos, J.; Peng, Z.; Kim, N.; Bluhm, H.; Gaskell, K.; Liu, P.; Lee, S. B.; Eichhorn, B. W.; Wang, Y. Probing the electrical double layer by operando X-ray photoelectron spectroscopy through a graphene-carbon nanotube composite window. EcoMat 2020, 2, e12023.DOI
  20. [20]Chen, X.; Put, B.; Sagara, A.; Gandrud, K.; Murata, M.; Steele, J. A.; Yabe, H.; Hantschel, T.; Roeffaers, M.; Tomiyama, M.; Arase, H.; Kaneko, Y.; Shimada, M.; Mees, M.; Vereecken, P. M. Silica gel solid nanocomposite electrolytes with interfacial conductivity promotion exceeding the Li-ion conductivity of the ionic liquid electrolyte filler. Science Advances 2020, 6, eaav3400.DOI
  21. [21]Bartók, A. P.; Kondor, R.; Csányi, G. On representing chemical environments. Physical Review B 2013, 87, 184115.DOI
  22. [22]Gardin, A.; Perego, C.; Doni, G.; Pavan, G. M. Classifying soft self-assembled materials via unsupervised machine learning of defects. Communications Chemistry 2022, 5, 82.DOI
  23. [23]Mangal, D.; Vera, G. S.; Aime, S.; Jamali, S. Small variations in particle-level interactions lead to large structural heterogeneities in colloidal gels. Soft Matter 2024, 20, 4692–4698.DOI
  24. [24]Jorgensen, C. A.; Lin, A. Y.; Vasavada, R.; Cercosnsky, R. K. Interpretable visualizations of data spaces for classification problems. Machine Learning: Science and Technology 2026, 7, 025008.
  25. [25]Smith, A.; Zavala, V. M. The Euler characteristic: A general topological descriptor for complex data. Computers & Chemical Engineering 2021, 154, 107463.arxiv.org/abs/2103.03144
  26. [26]Pranav, P.; van de Weygaert, R.; Vegter, G.; Jones, B. J. T.; Adler, R. J.; Feldbrugge, J.; Park, C.; Buchert, T.; Kerber, M. Topology and geometry of Gaussian random fields I: on Betti numbers, Euler characteristic, and Minkowski functionals. Monthly Notices of the Royal Astronomical Society 2019, 485, 4167–4208.DOI
  27. [27]Amézquita, E. J.; Quigley, M. Y.; Ophelders, T.; Landis, J. B.; Koenig, D.; Munch, E.; Chitwood, D. H. Measuring hidden phenotype: quantifying the shape of barley seeds using the Euler characteristic transform. in silico Plants 2022, 4, diab033.
  28. [28]Smith, A. D.; Donley, G. J.; Del Gado, E.; Zavala, V. M. Topological Data Analysis for Particulate Gels. ACS Nano 2024, 18, 28622–28635.arxiv.org/abs/2404.02991
  29. [29]Smith, A.; Runde, S.; Chew, A. K.; Kelkar, A. S.; Maheshwari, U.; Van Lehn, R. C.; Zavala, V. M. Topological Analysis of Molecular Dynamics Simulations using the Euler Characteristic. Journal of Chemical Theory and Computation 2023, 19, 1553–1567.DOI
  30. [30]Indouolak, K.; Kanduč, M.; Li, L.; Frenkel, D.; Dobnikar, J.; Del Gado, E. The crucial effect of early-stage gelation on the mechanical properties of cement hydrates. Nature Communications 2016, 7, 12106.
  31. [31]Thompson, A. P.; Aktulga, H. M.; Berger, R.; Bolintineanu, D. S.; Brown, W. M.; Crozier, P. S.; in ’t Veld, P. J.; Kohlmeyer, A.; Moore, S. G.; Nguyen, T. D.; Shan, R.; Stevens, M. J.; Tranchida, J.; Trott, C.; Plimpton, S. J. LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Computer Physics Communications 2022, 271, 108171.DOI
  32. [32]Israelachvili, J. N. In Intermolecular and Surface Forces (Third Edition), third edition ed.; Israelachvili, J. N., Ed.; Academic Press: San Diego, 2011; pp 253–289.
  33. [33]Everaers, R.; Ejtehadi, M. R. Interaction potentials for soft and hard ellipsoids. Physical Review E 2003, 67, 041710.DOI
  34. [34]Anagnostopoulos, A.; Alexiadis, A.; Ding, Y. Simplified force field for molecular dynamics simulations of amorphous SiO2 for solar applications. International Journal of Thermal Sciences 2021, 160, 106647.DOI
  35. [35]Sader, J. E.; Cairnie, S. L.; Chan, D. Y. C. Accurate Analytic Formulas for the Double-Layer Interaction between Spheres. Journal of Colloid and Interface Science 1995, 171, 46–54.DOI
  36. [36]Groves, T. S.; Perkin, S. Wave mechanics in an ionic liquid mixture. Faraday Discussions 2024, 253, 193–211.arxiv.org/abs/2402.17480
  37. [37]Kjellander, R. A multiple decay-length extension of the Debye–Hückel theory: to achieve high accuracy also for concentrated solutions and explain under-screening in dilute symmetric electrolytes. Physical Chemistry Chemical Physics 2020, 22, 23952–23985.DOI
  38. [38]Arpe, H. Ullmann’s Encyclopedia of Industrial Chemistry, Fibers, 5. Synthetic Inorganic to Formaldehyde; Ullmann’s Encyclopedia of Industrial Chemistry; Wiley, 1988.
  39. [39]Krossing, I.; Slattery, J. M.; Daguenet, C.; Dyson, P. J.; Oleinikova, A.; Weingartner, H. Why Are Ionic Liquids Liquid? A Simple Explanation Based on Lattice and Solvation Energies. Journal of the American Chemical Society 2006, 128, 13427–13434.pubmed.ncbi.nlm.nih.gov/17031955
  40. [40]Khlebtsov, B. N.; Khanadeev, V. A.; Khlebtsov, N. G. Determination of the Size, Concentration, and Refractive Index of Silica Nanoparticles from Turbidity Spectra. Langmuir 2008, 24, 8964–8970.DOI
  41. [41]Rahman Beigi, A. A.; Abdouss, M.; Yousefi, M.; Pourmor-tazavi, S. M.; Vahid, A. Investigation on physical and electrochemical properties of three imidazolium based ionic liquids (1-hexyl-3-methylimidazolium tetrafluoroborate, 1-ethyl-3-methylimidazolium bis(trifluoromethylsulfonyl) imide and 1-butyl-3-methylimidazolium methylsulfate). Journal of Molecular Liquids 2013, 177, 361–368.
  42. [42]Ueno, K.; Inaba, A.; Kondoh, M.; Watanabe, M. Colloidal Stability of Bare and Polymer-Grafted Silica Nanoparticles in Ionic Liquids. Langmuir 2008, 24, 5253–5259.DOI
  43. [43]Bertsch, P.; Fischer, P. Interfacial Rheology of Charged Anisotropic Cellulose Nanocrystals at the Air–Water Interface. Langmuir 2019, 35, 7937–7943.DOI
  44. [44]Chodera, J. D. A Simple Method for Automated Equilibration Detection in Molecular Simulations. Journal of Chemical Theory and Computation 2016, 12, 1799–1805.
  45. [45]Liu, Q.; Ma, L.; Wang, S.; Ni, Z.; Fu, X.; Wang, J.; Zheng, Q. Study on the properties of density, viscosity, excess molar volume, and viscosity deviation of [C2mim][NTf2], [C2mmim][NTf2], [C4mim][NTf2], and [C4mmim][NTf2] with PC binary mixtures. Journal of Molecular Liquids 2021, 325, 114573.
  46. [46]Hijazi, M.; Wilkins, D. M.; Ceriotti, M. Fast-forward Langevin dynamics with momentum flips. The Journal of Chemical Physics 2018, 148, 184109.DOI
  47. [47]Song, X. C.; Smith, P.; Kalyanam, R.; Zhu, X.; Adams, E.; Colby, K.; Finnegan, P.; Gough, E.; Hillery, E.; Irvine, R.; Maji, A.; St. John, J. Anvil - System Architecture and Experiences from Deployment and Early User Operations. Practice and Experience in Advanced Research Computing 2022: Revolutionary: Computing, Connections, You. New York, NY, USA, 2022.DOI
  48. [48]Boerner, T. J.; Deems, S.; Furlani, T. R.; Knuth, S. L.; Towns, J. ACCESS: Advancing Innovation: NSF’s Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support. Practice and Experience in Advanced Research Computing 2023: Computing for the Common Good. New York, NY, USA, 2023; p 173–176.
  49. [49]Center for High Throughput Computing. 2006; https://chtc.cs.wisc.edu/.
  50. [50]Steinhardt, P. J.; Nelson, D. R.; Ronchetti, M. Bond-orientational order in liquids and glasses. Physical Review B 1983, 28, 784–805.
  51. [51]Mickel, W.; Kapfer, S. C.; Schröder-Turk, G. E.; Mecke, K. Shortcomings of the bond orientational order parameters for the analysis of disordered particulate matter. The Journal of Chemical Physics 2013, 138, 044501.DOI
  52. [52]Spellings, M.; Glotzer, S. C. Machine learning for crystal identification and discovery. AIChE Journal 2018, 64, 2198–2206.arxiv.org/abs/1710.09861
  53. [53]Martirosyan, M. M.; Spellings, M.; Pan, H.; Dshemuchadse, J. Local Structural Features Elucidate Crystalization of Complex Structures. ACS Nano 2024, 18, 14989–15002.
  54. [54]Coslovich, D.; Jack, R. L.; Paret, J. Dimensionality reduction of local structure in glassy binary mixtures. The Journal of Chemical Physics 2022, 157, 204503.DOI
  55. [55]Byu-Mai, S.-C.; Huse, D. A. Stacking entropy of hard-sphere crystals. Physical Review E 1999, 59, 4396–4401.
  56. [56]Middleton, T. F.; Hernández-Rojas, J.; Mortenson, P. N.; Wales, D. J. Crystals of binary Lennard-Jones solids. Physical Review B 2001, 64, 184201.DOI
  57. [57]Fan, X.; Xu, D.; Zhang, J.; Hu, H.; Tan, P.; Xu, N.; Tanaka, H.; Tong, H. Ideal non-crystals as a distinct form of ordered states without symmetry breaking. Nature Materials 2026, 25, 1020–1027.DOI
  58. [58]Roux, B.; Simonson, T. Implicit solvent models. Biophysical Chemistry 1999, 78, 1–20.
  59. [59]Yang, Y.; Zhou, J.; Zhu, F.; Yuan, Y.; Chang, D. J.; Kim, D. S.; Pham, M.; Rana, A.; Tian, X.; Yao, Y.; Osher, S. J.; Schmid, A. K.; Hu, L.; Ercius, P.; Miao, J. Determining the three-dimensional atomic structure of an amorphous solid. Nature 2021, 592, 60–64.DOI
  60. [60]Pedregosa, F. et al. scikit-learn: Machine Learning in Python. Journal of Machine Learning Research 2011, 12, 2825–2830.arxiv.org/abs/1201.0490
  61. [61]Goscinski, A.; Jørgensen, C. A.; Principe, V. P.; Fraux, G.; Klivinek, S.; Helfrecht, B. A.; Vasavada, R.; Loche, P.; Ceriotti, M.; Cersonsky, R. K. scikit-matter : A Suite of Generalisable Machine Learning Methods Born out of Chemistry and Materials Science. Open Research Europe 2025, 3, 81.
  62. [62]Cersonsky, R. K.; Helfrecht, B. A.; Engel, E. A.; Klivinek, S.; Eriotti, M. Improving sample and feature selection with principal covariates regression. Machine Learning: Science and Technology 2021, 2, 035038.arxiv.org/abs/2012.12253
  63. [63]Nabizadeh, M.; Nasirian, F.; Li, X.; Saraswat, Y.; Waheibi, R.; Hsiao, L. C.; Bi, D.; Ravandi, B.; Jamali, S. Network physics of attractive colloidal gels: Resilience, rigidity, and phase diagram. Proceedings of the National Academy of Sciences 2024, 121, e2316394121.DOI
  64. [64]Smith, A. D.; Dłotko, P.; Zavala, V. M. Topological data analysis: Concepts, computation, and applications in chemical engineering. Computers & Chemical Engineering 2021, 146, 107202.arxiv.org/abs/2006.03173
  65. [65]Dabaghian, Y.; Mémoli, F.; Frank, L.; Carlsson, G. A Topological Paradigm for Hippocampal Spatial Map Formation Using Persistent Homology. PLOS Computational Biology 2012, 8, e1002581.DOI
  66. [66]Townsend, J.; Micucci, C. P.; Hymel, J. H.; Maroulas, V.; Vogtiazis, K. D. Representation of molecular structures with persistent homology for machine learning applications in chemistry. Nature Communications 2020, 11, 3230.DOI
  67. [67]Otter, N.; Porter, M. A.; Tillmann, U.; Grindrod, P.; Harrington, H. A. A roadmap for the computation of persistent homology. EPJ Data Science 2017, 6, 17.
  68. [68]Hjorth Larsen, A. et al. The atomic simulation environment—a Python library for working with atoms. Journal of Physics: Condensed Matter 2017, 29, 273002.DOI
  69. [69]Ramasubramani, V.; Dice, B. D.; Harper, E. S.; Spellings, M. P.; Anderson, J. A.; Glotzer, S. C. freud: A Software Suite for High Throughput Analysis of Particle Simulation Data. Computer Physics Communications 2020, 254, 107275.
  70. [70]Virtanen, P. et al. SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nature Methods 2020, 17, 261–272.
  71. [71]van der Walt, S.; Schönberger, J. L.; Nunez-Iglesias, J.; Boulogne, F.; Warner, J. D.; Yager, N.; Gouillart, E.; Yu, T.; the scikit-image contributors scikit-image: image processing in Python. PeerJ 2014, 2, e453.

Paper details

vlnc:W_k5ywfw53
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents