Knowledge-driven design of fluorinated ether electrolytes via a multi-model approach

doi:10.21203/rs.3.rs-4364688/v1

Download PDF

Article

Knowledge-driven design of fluorinated ether electrolytes via a multi-model approach

https://doi.org/10.21203/rs.3.rs-4364688/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Fluorinated ether solvents (FLS) can enhance the cycle life of Li-S batteries by mitigating the polysulfide shuttle effect. However, developing fluorinated electrolytes with reduced polysulfide solubility and uncompromised transport properties is underexplored. We integrate high-throughput density functional theory, molecular simulations, machine learning, and experimental analyses to explore ~1,000 FLS to be used as co-solvent with 1,3-dioxolane. Only 14 FLS in our library have been previously reported in Li-S literature. Through a rigorous screening process, we identify and test two new solvents which demonstrate reduced polysulfide solubility. One solvent exhibits electrochemical performance on par with the widely used 1,1,2,2-tetrafluoro-3-(1,1,2,2-tetrafluoroethoxy)propane (TTE) solvent, yet with superior electrolyte viscosity and ionic conductivity. Interpretable machine learning models indicate fluorination degree, steric effects on ether oxygen, and fluorine proximity to ether oxygen are crucial in dictating oxidation reactions and polysulfide solubility. This work not only introduces new promising co-solvents for Li-S batteries but also provides a framework for knowledge-driven electrolyte design.

Physical sciences/Materials science/Materials for energy and catalysis/Batteries

Physical sciences/Materials science/Theory and computation/Computational methods

The advancement of rechargeable batteries with high energy density, employing abundant resources, emerges as a pivotal factor in steering the transportation sector away from fossil fuels and fostering the widespread adoption of electric vehicles^1,2. Lithium-sulfur (Li-S) batteries are considered one of the most promising battery chemistries due to the Earth abundance, high theoretical capacity (1,675 mA h g^-1 ), and energy density (2,500 kW kg^-1) of sulfur cathode³. However, mass commercialization of Li-S batteries is hindered by the notorious “shuttle effect” originating from the initial reduction of sulfur, S₈, to highly soluble intermediate polysulfides (\({\text{S}}_{\text{n}}^{\text{2-}}\), n ≥ 4) in conventional ether electrolytes with ∼ 1 M Li salt^4,5. The polysulfide species migrate to the anode through electrolyte and react with lithium metal anode, leading to insoluble precipitants, e.g., lithium sulfide (Li₂S), that passivate the anode surface⁶. Consequences of this phenomenon include loss of active sulfur, high self-discharge rate, and a decrease in the battery capacity over time.

Over decades, considerable efforts have been invested in overcoming the shuttle effect by trapping the polysulfides in the pores of carbon materials^7–9 or by using polymer coatings^10–12. Although these strategies may be effective at retaining polysulfides to some extent, their functionality is ultimately limited by a huge structure change of cathode caused by large volume change of sulfur (80%) during discharge³. Two dimensional functionalized materials, such as graphene oxides, have been employed as cathode hosts, leveraging the binding affinity between functional groups and polysulfides to prevent polysulfide dissolution^13,14. Optimizing both liquid and solid electrolytes is also highly effective at enhancing the cyclability of a wide range of S/Li₂S cathodes^15–17.

Polysulfide solubility is an intrinsic property of the electrolyte system, primarily determined by the solvent and salt concentration. Meanwhile, liquid electrolytes are prevalent in Li-S batteries due to their high ionic conductivity and stability against Li metal anode. Redesign of organic solvents that sparingly solvate polysulfides is key to achieving long-lasting and efficient operation of Li-S cells at the strict requirement of low electrolyte/sulfur ratio (< 1 mL/g)^2,18. Recently, FLS have showed great promise in reducing polysulfide shuttling and limiting self-discharge in Li-S cells^19–25. Since 2010, FLS have become a popular avenue to develop functional electrolytes not only for Li-S batteries, but also for lithium-ion^26,27, lithium/selenium-sulfur²⁸, lithium-air²⁹, and sodium-ion batteries^30,31. FLS offer several advantages, including their low melting point³², high electrolyte anodic stability²⁶, and ability to form localized concentrated regions. These feature enable the utilization of high salt concentrations without sacrificing the electrolyte viscosity³³, the formation of stable solid electrolyte interphase (SEI) layers during cycling²¹, and the suppression of polysulfide shuttle effect³⁴.

Despite the extensive effort that has been made on the design of FLS-based electrolytes, there is still a vast design space that has not been explored. Many of the experimental studies are limited in the number of solvents investigated due to complicated synthesis of FLS or simply the availability of commercial FLS, which leads to selection bias, and overlook the electrolyte transport properties. Additionally, there is a lack of large-scale computational studies on FLS, which are essential to identify new promising candidates and uncover structure-property relations from systematic datasets. While some computational screening approaches have been employed for other types of electrolyte components, these studies have mainly focused on a few properties of isolated molecules or clusters^35–37 such as redox potentials and viscosities, rather than accounting for a broader range of properties or predicting their ensemble behavior.

In this work, we employed a multi-model approach by integrating high-throughput density functional theory (DFT), molecular dynamics (MD) simulations, COSMO-RS calculations, machine learning (ML) techniques, and experimental characterization to identify 18 new FLS with better stability and dynamical properties than the state-of-the-art fluorinated solvent. We experimentally tested two commercially available solvents out of the 18 screened FLS and confirmed their potential in reducing polysulfide shuttle effect while maintaining excellent electrolyte transport properties. Only 14 solvents of our initial library have previously been reported in the Li-S literature, highlighting the potential for new discoveries through computational screening. We also determined key FLS features that most significantly influence electrolyte properties. These features can serve as general guidelines for future FLS selection.

Figure 1a shows the hierarchical down-selection screening approach we employed to identify new FLS for Li-S batteries. We began with a search of the solvents reported in Li-S literature and identified 14 FLS compounds (Supplementary Table S1). We performed a structure similarity search using these compounds as inputs to query the PubChem³⁸ database and retrieved 1,250 fluorinated ether compounds. We filtered these compounds for neutral, non-aromatic, and C-O-H-F chemical systems, which reduced the size of the initial library to 922 compounds. At this stage, we performed path-based fingerprinting using OpenBabel³⁹ and used the resulting fingerprints to calculate the structural similarity between each pair of molecules via the Tanimoto coefficient. The similarity scores were used to quantify the structural diversity of the library. A similarity score of zero indicates dissimilar structures, while a value of one indicates identical molecules. The cumulative distribution function (CDF) in Fig. 1b shows that only 10% of the pairs have a similarity score above 0.75. Given that the library is restricted to fluorinated ether molecules, this level of diversity is considered sufficient. Supplementary Fig. 1 shows the distribution of similarity scores between the compounds in the library and each of the 14 FLS reported in the literature. Most of the distributions are unimodal and indicate an intermediate degree of similarity with previously reported solvents. This is desirable because a high similarity score implies that the solvents closely resemble those previously tested, while a low similarity score suggests that the library has deviated from the ones reported in the literature with promising properties⁴⁰.

In the next screening stage, we computed the solubility of lithium bis(trifluoromethane sulfonel) imide (LiTFSI) and lithium bis(fluorosulfonyl) imide (LiFSI) in each of the FLS using COSMO-RS calculations. LiTFSI and LiFSI were selected for this study due to their widespread application in Li-S batteries¹⁷. An ideal FLS should be capable of suppressing polysulfide solubility and dissolving sufficient amount of Li salt (~ 1 M) to facilitate the fast Li⁺ ion transport between the electrode while maintaining a low viscosity to enhance the wetting ability of the electrolyte to cell components such as electrode and separators. Out of the 922 solvents, only 141 compounds met the criterion of having Li salt solubility ≥ 1 M. All the FLS exhibited lower salt solubilities than conventional dioxolane (DOL) and 1,2-dimethoxyethane (DME) solvents. This result is expected since the solvating power of FLS molecules is restrained by the inductive effect imposed by the electron-withdrawing fluoroalkyl groups⁴¹. The solvation ability of FLS is not, however, necessarily inversely proportional to their fluorination degree, as will be explained later.

Safety considerations were accounted for by estimating the boiling point (BP) and flash point (FP) of the pure molecules using COSMO-RS calculations. BP and FP of electrolyte are critical parameters to ensure safe battery operation and reduce the potential harm to the batteries in the event of thermal runaway or ignition. Out of the 141 compounds, 86 met the criteria of BP ≥ 47 ℃ and FP ≥ -23 ℃. These parameters were selected to align with the acceptable and optimal temperature ranges for Li-ion batteries, which are − 20 ℃ to 60 ℃ and 15 ℃ to 35 ℃, respectively⁴². The BP threshold of 47 ℃ corresponds to the upper limit set by the majority of battery thermal management systems⁴³. Supplementary Table S2 shows the predicted and measured BP for DME and common FLS. The mean absolute deviation (MAD, 16 ℃) in the order of ~ 20% of the BP window and the Pearson’s R value of 0.93 indicate a sufficient level of correlation.

Ionization potential (IP) and electron affinity (EA) were computed from DFT calculations and used as proxies for the solvent stability towards initial oxidation and reduction, respectively. These properties were then used to determine the electrochemical window of the solvent, within which the reversible reactions of the anode and cathode are facilitated without the continuous decomposition of the electrolyte components. Considering that the working voltage of sulfur cathode is between 1.5 and 3.0 V vs. Li/Li⁺ in Li-S batteries^44,45, solvents with an electrochemical window greater than or equal to 2.7 V were down-selected to ensure electrochemical stability. Only two solvents from the previous step did not meet this criterion. Additional DFT calculations were performed to compute the relative binding energy (BE) of FLS with both LiTFSI and lithium polysulfide (LiPS), modeled as Li₂S₄. BE of DME with LiTFSI and Li₂S₄, -24.7 and − 22.3 kcal/mol, respectively, were used as reference zero. BE is calculated at the FLS ether oxygen-Li⁺ site for both FLS-LiTFSI and FLS-LiPS interactions. A desirable property of a potential FLS is higher selective solvating power than DME (second quadrant of Fig. 2a), i.e., the solvent is characterized with similar or higher BE with the weakly coordinating salt than DME but less binding affinity towards LiPS. As shown in Fig. 1a, 18 solvents out of 84 compounds selected from the previous step met this criterion. We note that DFT calculations were performed for the entire library of 922 compounds, extending beyond the subset identified through screening at that stage (Refer to Supplementary Fig. 2 for the distribution of computed properties of FLS in comparison with DOL, DME, and fluorinated ether solvents reported in Li-S literature). Figure 2b shows that fluorination degree has more significant positive impact on the solvent IP than its BE with LiTFSI. Here, fluorination degree refers to the percentage of hydrogen atoms substituted by fluorine atoms in a molecule. While many solvents have the same fluorination degree, their ionization potentials and salt binding affinities can differ as shown in Fig. 2b.

To understand these correlations, we employed an extreme gradient boosting (XGBoost) regression model that allows identifying attributes of high impact on FLS properties relevant to Li-S electrolytes. We featurized the FLS molecules based on their structural properties, such as shape index, steric hinderance, Verloop’s Sterimol parameters⁴⁶, radius, proximity of fluorine to oxygen, fluorination degree, among others. We also incorporated DFT-extracted properties like dipole moment, polarizability, and electrostatic partial charges (minimum and maximum absolute partial charges in the molecule). Refer to Supplementary Table S3 for detailed feature descriptions. We then used a Shapley Additive exPlanations (SHAP) framework⁴⁷ to interpret predictions made by the XGBoost model. This technique utilizes Shapley values, a game theory-derived metric that assesses the contribution by different players. In this case, the players are the derived FLS features, and the game outcome is the model predictions of IP and LiTFSI BE. Figure 2c presents a summary plot that demonstrates the effect of the top 10 most essential features on the solvent IP, ordered according to their importance. A similar analysis for EA and BE is shown in Supplementary Fig. 3. The IP is primarily governed by the fluorination degree in a positive relation, as discussed earlier. The proximity of fluorine to the ether oxygen atom is the second most critical factor, exhibiting an overall negative impact on IP. Steric effect also significantly influences IP, with increased steric hindrance on the oxygen atom leading to a higher IP.

SHAP analysis for EA of FLS molecules (Supplementary Fig. 3a) indicates that fluorination degree is also the most impactful feature on reduction. However, steric effects, unlike oxidation, are negligible under reduction, similar to previously reported results for phenothiazines⁴⁸. This behavior can be rationalized by the fact that the oxygen atom contributes considerably to the highest occupied molecular orbital but not much to the lowest unoccupied molecular orbital (Supplementary Fig. 4). Supplementary Fig. 3b shows that in general, a wider separation between fluorine and ether oxygens, a lower fluorination degree, and a higher FLS polarizability are correlated with an enhanced binding affinity of the solvent to LiTFSI. Supplementary Fig. 5 illustrates two molecules, both with a fluorination degree of 50%, yet with notably different IPs of 7.35 eV and 5.84 eV, and PS binding affinities of -16.5 kcal/mol and − 22.3 kcal/mol, respectively. A detailed examination reveals differing steric effects and fluorine-to-oxygen distances in these molecules: the left molecule shows a greater steric effect (2.41) and a closer fluorine atom, approximately 2.25 Å to the ether oxygen, in contrast to the corresponding values of 2.28 and 2.76 Å in the second molecule. These variations corroborate the SHAP analysis previously discussed, serving as an example of the discussed trends. Consequently, controlling factors like the fluorination degree, fluorine proximity to oxygen, and steric effects is crucial for manipulating IPs and binding affinities. Based on these observations, for a specific fluorination degree and to attain higher IPs and reduced PS solubility, we suggest adjusting the FLS structure to increase steric hindrance around the ether oxygen and position fluorine atoms closer to it.

For a thorough evaluation, we created two additional models, one that integrated 202 general RDKit⁴⁹ descriptors and another with 1,323 Mordred⁵⁰ descriptors, each combined with the custom FLS features described above. Supplementary Fig. 6 displays the SHAP analysis for IP, highlighting the top 10 features predicted from the decision tree model with a maximum tree depth of three to avoid overfitting. Upon incorporating RDKit and Mordred descriptors, the fluorination degree, proximity of fluorine to oxygen, and steric effect persist as the three dominant factors influencing IP. Among the top ten features, eight to nine were custom calculated FLS features. This finding emphasizes the significant advantage of developing features tailored to the specific nuances of the chemical system under study, as opposed to exclusive dependence on generic molecular descriptors. Moreover, the custom calculated FLS features offer an additional benefit in that they can be more readily fine-tuned experimentally, which is a flexibility often not afforded by many of the standardized RDKit or Mordred descriptors.

Supplementary Table S4 presents the structures of the 18 solvents selected from the last screening step. For each of these solvents, we performed MD simulations to gain insight into solvation and transport behavior. MD systems were composed of 1 M LiTFSI, 0.3 M LiNO₃ in DOL/FLS (1/1; v/v) with and without the presence of 0.25 M Li₂S₂, 0.25 M Li₂S₄, and 0.25 M Li₂S₈, each run separately. FLS0 was included in these simulations despite not being screened, due to its commercial availability and its similar properties to the commonly used TTE solvent. We calculated radial distribution functions (RDF; Supplementary Fig. 7–11) and probability of forming a polysulfide of cluster sizes 1–6 in the first and second solvation shell of each PS specie (Fig. 3). Cluster analysis results indicate a lower solubility of Li₂S₂ compared to longer chain PS (Li₂S₄ and Li₂S₈) in all electrolytes including the state-of-the-art DOL/DME, in accordance with previously reported experimental results⁵¹. The introduction of FLS leads to a lower solubility of Li₂S₄ and Li₂S₈ (more clustering or aggregation), compared to the system containing DME, as indicated by lighter red and purple hues for PS size one, and darker tones for larger PS sizes (See sample MD snapshots in Supplementary Fig. 12–14). Even solvents with a fluorination degree less than 20% (FLS2, FLS4, FLS6, FLS10, FLS12, FLS14, FLS16, FLS17, and FLS18; see Supplementary Table S4) led to substantial PS aggregation, particularly in Li₂S₈ systems. This is an interesting result indicating the effectiveness of these solvents in reducing the solubility of long chain PS despite their low fluorination degree. Recent studies have increasingly shown that highly fluorinated solvents may not always be preferable, with several groups highlighting the need to explore partially fluorinated electrolytes^52–54. We note that all screened FLS in this work, except FLS0, exhibit a lower fluorination degree than TTE.

The RDF analysis of \({\text{S}}_{\text{x}}^{2-}\)–solvent for various PS species (Supplementary Fig. 8) shows significantly weaker PS interactions with TTE and FLS than with DME. Such dramatic reduction in PS-solvent interaction in fluorinated systems explains the previously discussed long-range PS structures and consequently, the lower PS solubility in these systems. Additionally, Li-FLS interactions are weaker than the Li-DME counterpart (Supplementary Fig. 9), leading to more contact ion pairs in the FLS electrolytes, evidenced by the pronounced Li⁺-TFSI^- RDF peaks at around 2 Å (Supplementary Fig. 10). Additionally, the less coordinating DOL solvent compared to DME, has more opportunity to interact with Li⁺ in these systems than in the DOL/DME electrolyte (Supplementary Fig. 11). However, improved Li⁺ solvation occurs in electrolytes containing FLS2, FLS12, and FLS14-FLS18 (peak at 1.9 A in Li⁺-solvent RDF; Supplementary Fig. 9), suggesting the potential of these solvents to reduce PS solubility while retaining a degree of coordinating ability. Despite both having a fluorination degree of around 17%, FLS4 and FLS12 exhibit distinct behaviors in terms of Li₂S₈ aggregation, with FLS4 leading to higher aggregation levels (Fig. 3). This difference can be attributed to the greater steric hindrance around the ether oxygen atom in FLS4 coupled with the closer proximity of fluorine atoms to oxygen compared to FLS12. These observations are in alignment with the SHAP trends previously discussed, emphasizing the influence of fluorine atom placement on FLS solvating power. Finally, we obtained the distribution of diffusion coefficients of Li⁺ and the co-solvent (DME, TTE, or FLS) within each of the electrolyte systems (Supplementary Fig. 15). Compared to the FLS-containing systems, much more significant overlap occurs between the diffusion coefficients of the cation and the co-solvent in the DOL/DME system, indicating more correlated motion compared to the FLS systems. This dynamical behavior reinforces the less coordinating capability of these solvents compared to DME.

One of the critical targets of this work is to improve the ionic conductivity and address transport challenges associated with fluorinated solvents, while retaining their effectiveness in mitigating the PS shuttle effect. To this end, we selected FLS0 and FLS18 for further evaluation due to their commercial availability. Figure 4a shows the ionic conductivity and viscosity of electrolytes containing DME, TTE, FLS0, and FLS18, used as cosolvents with DOL. Bars represent results from MD simulations, whereas stars indicate experimental data collected at 25 ℃. All four electrolytes containing the new FLS exhibit lower viscosities and superior ionic conductivities compared to their TTE-based counterpart, while maintaining viscosities (< 2 cP) on par with the state-of-the-art DOL/DME system. Notably, the two equivolume ternary mixtures of DOL/DME/FLS0 and DOL/DME/FLS18 achieve ionic conductivities of 8.6 and 9.9 mS/cm, respectively. These values are comparable to state-of-the-art DOL/DME system and surpass the ionic conductivities of similar ternary mixtures containing fluorinated solvents^20,55,56 (note good agreement of MD predictions with measured viscosity, density, and ionic conductivity as function of temperature in Supplementary Fig. 16). Even binary mixtures of DOL/FLS show conductivities 1.5–2.3 times higher than that of DOL/TTE in the presence of 0.1 M LiNO₃, which can be attributed to the formation of less compact cation-anion clusters within these systems.

Encouraged by their potential in reducing PS solubility and their improved dynamical properties over the widely used TTE, we performed Li || S/C full battery tests of the following four electrolytes: 1 M LiTFSI, 0.3 M LiNO₃ in DOL/DME, 1 M LiTFSI, 0.3 M LiNO₃ in DOL/TTE, 1 M LiTFSI, 0.3 M LiNO₃ in DOL/DME/FLS0, and 1 M LiTFSI, 0.1 M LiNO₃ in DOL/DME/FLS18, each with composite sulfur/carbon cathode at ~ 1 mg cm^− 2 S loading and a glass fiber paper separator, with constant current cycling at C/10 rate. Here, it is important to note that we encountered solubility issues with LiNO₃ in the fluorinated electrolytes. Specifically, a maximum concentration of 0.1 M LiNO₃ could be dissolved in systems containing DOL/FLS0, DOL/FLS18, and DOL/DME/FLS18. In contrast, the DOL/DME/FLS0 system could dissolve up to 0.3 M LiNO₃. This observation underlines a limitation in our computational screening, which initially focused only on the solubility of the Li salt and not the additive. Consequently, future developments in designing these fluorinated solvents should also consider the additives solubility.

Figure 4 shows representative charge and discharge curves along with the discharge capacity and Coulombic efficiency (CE) as a function of the cycle number. The capacity fade is alleviated upon the use of FLS0 and TTE relative to that of the base DOL/DME electrolyte, consistent with prior results using TTE as a cosolvent⁵⁷. The degree of capacity retention between DOL/TTE and DOL/DME/FLS0 is comparable yielding similar discharge capacities at 200 cycles. Overall, the cell with DOL/DME/FLS0 attains the highest capacity after 200 cycles and the least capacity fading (Fig. 4b) followed by DOL/TTE. The cell with DOL/DME/FLS18 resulted in visible capacity degradation to ~ 9% within 200 cycles, likely attributed to LiNO₃ solubility issues and the strong reaction of FLS18 with Li metal in the presence of the additive, thus complicating interphase formation. At cycle 200th, cells with DOL/DME/FLS0 and DOL/TTE exhibit discharge capacities of 243 and 240 mAh/g, respectively, whereas the DOL/DME and DOL/DME/FLS18 systems resulted in capacities of only 55 and 86 mAh/g, respectively. Li-S cells using TTE and FLS presented more stable cycling than that containing DOL/DME, indicating a reduction in the shuttling effect due to the decrease of LiPS solubility.

Figure 4b shows the discharge and charge profiles for the same Li-S cells at cycle 10. The voltage of the first plateau corresponding to the solid S conversion to LiPSs is extracted at 15 mAh/g, and the second plateau corresponding to the LiPSs conversion to Li₂S is at 500 mAh/g. For the system containing DOL/DME/FLS0, the first plateau occurs at a lower voltage compared to the DOL/DME cell, while the second plateau occurs at a higher voltage (Supplementary Fig. 17). This pattern indicates that decreasing LiPS solubility suppresses the formation of LiPSs, thus reducing the shuttling effect. The system with DOL/DME/FLS18 showed high overpotential and low capacity utilization likely attributed to the solubility and reactivity issues mentioned earlier. Overall, the cell with DOL/DME/FLS0 exhibited the best performance with the highest discharge capacity of 785 mAh/g, compared to 708 mAh/g for the DOL/TTE system.

The lower CE of the DOL/DME/FLS0 cell compared to that of DOL/DME (Fig. 4d) indicates reduced self-discharge behavior due to the polysulfide shuttling effect. The dissolution of active S into the electrolyte as well as redox shuttling of soluble polysulfide chains during charging results in Li-S batteries often having a CE > 100%, where CE = Charge Capacity/Discharge Capacity⁵⁸. This overcharge effect may be even further magnified at a slow charge/discharge rate such as C/10 used herein. Improved CEs through screened FLS are observed in the voltage profiles in Fig. 4c, which show less additional capacity in the charge profile. In Cycle 10, DOL/DME, DOL/TTE, DOL/DME/FLS0, and DOL/DME/FLS18 show CEs of 104.6, 103.1, 100.9, and 102.6%, respectively. Figure 4d further shows that the DOL/DME/FLS0 and DOL/DME/FLS18 electrolytes screened solvents can maintain reasonable CEs and mitigate the shuttling effect over a long duration of 200 cycles.

In summary, we demonstrate a systematic high-throughput screening approach to identify new FLS exhibiting a promising balance between transport properties and PS solubility, outperforming the widely used fluorinated solvent, TTE. Interestingly, we find that even solvents with a low fluorination degree can potentially mitigate the shuttle effect in Li-S batteries. Such a feature is increasingly desirable for its potential to enhance battery performance while maintaining environmental considerations. However, our findings also suggest a need for further optimization of the electrolyte composition to address solubility issues associated with other electrolyte additives such as LiNO₃. We also highlight on three highly tunable properties of FLS—fluorination degree, oxygen steric effect, and proximity of fluorine atoms to ether oxygen—as critical features for refining solvent design. The SHAP analysis results emphasize the significance of developing tailored features that cater to the specific nuances of the chemical system under study, rather than depending on generic molecular descriptors. Although our evaluation was limited to only two solvents due to their commercial availability, we have revealed an additional 17 solvents with promising properties for future experimental investigation. This work informs and inspires viable paths forward for the future design and optimization of next-generation Li-S electrolyte solutions.

Density functional theory calculations

Binding energy. Binding energy calculations were performed using the binding energy workflow implemented in MISPR⁵⁹. Gaussian 16 Revision C.01⁶⁰ was used in the backend with the B3LYP/6–31 + G* level of theory in the gas phase for two sets of calculations: (1) between FLS and LiTFSI and (2) between FLS and Li₂S₄. The D3 version of Grimme’s dispersion⁶¹ was used to account for long-distance van der Waals interactions. The input structures (i.e., co-solvent and ligand) were first optimized individually and subjected to a vibrational frequency calculation to confirm a true minimum has been reached. The optimized structures were then automatically connected at a binding site specified as input to the workflow. The ligand binding site was chosen to be Li while the solvent binding site was the ether oxygen. Another sequence of optimization and frequency calculations was executed for the complex structure at the same level of theory. Finally, an analysis task was performed in which the binding energy was calculated as \({E}_{complex}-({E}_{solvent}+{E}_{ligand})\), where \({E}_{complex}\) is the energy of the molecular complex and \({E}_{solvent}\) and \({E}_{ligand}\) are the energies of the solvent and ligand, respectively.

Redox potentials. IP and EA calculations were performed using the redox potential workflow implemented in MISPR. Gaussian 16 Revision C.01 was used in the backend with the ωB97XD/Def2TZVPPD level of theory using the full thermodynamic cycle (explained in more detail in the MISPR manuscript) with the SMD solvation model and acetone as a solvation medium. Each FLS was subjected to an optimization and frequency step both in vacuum and in solution at each charge state for the first electron transfer reactions for the isolated molecules. Predicted redox potentials were converted to the Li⁺/Li scale by subtracting 1.4 V.

Electrostatic partial charges. For each FLS, we extracted the electrostatic partial charges (ESP) by running the ESP workflow implemented in MISPR. The optimized FLS structures were fed to the workflow and an ESP step at the B3LYP/6–31 + G* level of theory in Gaussian 16 Revision C.01⁶⁰ was performed.

Molecular dynamics simulations

MD simulations for electrolytes consisting of 1 M LiTFSI, 0.3 M LiNO₃, with and without 0.25 M Li₂S_x (x = 2, 4, and 8) in DOL/co-solvent (1/1, v/v; co-solvent: DME, TTE, or FLS; FLS corresponds to one of the 19 solvents presented in Supplementary Table S4) were performed using the MD workflow in MISPR with the default recipe in version 0.0.4. Additional set of simulations were performed for electrolytes consisting of 1 M LiTFSI, 0.1 M LiNO₃ in DOL/FLS0, 1 M LiTFSI, 0.1 M LiNO₃ in DOL/FLS18, 1 M LiTFSI, 0.3 M LiNO₃ in DOL/DME/FLS0, and 1 M LiTFSI, 0.1 M LiNO₃ in DOL/DME/FLS18. Within the MD workflow in MISPR, the initial configuration of each system was built using the PACKMOL⁶² package by randomly placing the structures in a 6×6×6 nm³ cubic box periodic in XYZ direction. The OPLS-2005⁶³ force field of Schrödinger⁶⁴ was employed for the solvents, while the Lopes-Pádua^65,66 force field parameters were used for TFSI⁻ and \({\text{N}\text{O}}_{3}^{-}\). Li⁺ parameters were taken from Dang et al.⁶⁷ while \({S}_{x}^{2-}\) bonded parameters were adopted from the work of Rajput et al.⁵¹ DRIEDING⁶⁸ force field was used for Lennard-Jones (LJ) parameters of S atoms of \({S}_{x}^{2-}\) with the corresponding partial charges from the work of Park et al.⁶⁹ Electronic polarizability effects were accounted for using the electronic continuum correction (ECC) method^70,71, in which the partial charges (\({q}_{i}\)) of all ionic species (Li⁺, TFSI⁻, \({NO}_{3}^{-}\)and \({S}_{x}^{2-}\)) were replaced by rescaled charges using a scaling factor of ~ 0.7. MD simulations were performed using the LAMMPS code version 3Mar2020. LJ interactions were truncated at 1.2 nm and the particle-particle particle-mesh (PPPM)⁷² method was used for long-range electrostatic interactions. Specifically, the systems were first minimized in two steps first using steepest descent with convergence criterion of 1,000 kcal/mol Å followed by conjugated-gradient energy minimization scheme with convergence criterion of 10 kcal/mol Å. The systems were then equilibrated to maintain the temperature of 298.15 K and pressure of 1 atm in the isothermal-isobaric ensemble (NPT) using the Nosé/Hoover temperature thermostat and pressure barostat with a time constant of 0.001 ps for 2 ns. The systems were then melted at 400 K for 2 ns and quenched from 400 to 298.15 K in three steps for 2.25 ns. Finally, production runs of 5 ns were obtained in the canonical ensemble (NVT) using a timestep of 0.001 ps at 298.15 K. Snapshots, sampled every 50 ps, were used to derive the structural properties (RDF and PS cluster analysis) and diffusion coefficients for electrolyte systems containing PSs. For the viscosity calculations, 40 additional independent production runs per system were initialized from the equilibrated NVT configurations for 1–5 ns with pressure tensors output every 0.004 ps. For conductivity calculations, NVT simulations were run for 1 ns and at least three independent replicates per system were performed with atomic velocities dumped every 0.02 ps. Viscosity and conductivity calculations were performed for systems not containing PSs. The analysis was performed using MDPropTools version 0.0.4.

Machine learning

An XGBoost model was employed to establish correlations between FLS features and each of the following properties: IP, EA, and binding energy of FLS with LiTFSI salt. The xgboost⁷³ library was used for this purpose and the shap⁴⁷ package was employed for the feature importance analysis.

Electrochemical characterization

Sulfur/carbon composite was prepared by combining 2 g nanosulfur (SkySpring Nanomaterials, 99.99%, <55 nm) with 1 g graphene nanopowder multilayer flakes (Graphene Supermarket, AO 4) and ball-milling for 30 min at 20 Hz. The composite was then combined with Super P conductive carbon (Timcal) and PVDF at weight ratios of 80:5:15, respectively. The resulting cathode contains 53.3 wt.% sulfur in the total electrode composite. The mixture was combined with approximately 3:1 weight ratio of dry materials to n-methylpyrrolidone solvent and homogenized in a Thinky planetary mixer for 30 min. The resulting slurry was blade-coated onto carbon-coated aluminum foil and dried under vacuum at 45℃ for 12 h. Electrodes were punched into 12 mm cathode disks. Lithium anodes were prepared in an argon glovebox (VAC) with H2O and O2 kept below 0.1 and 0.5 ppm, respectively, by rolling lithium rods (Aldrich, 99.9%) in a polyethylene bag and punched into 14mmdisks. For the electrolytes, DOL (anhydrous 99.8%), DME (anhydrous 99.5%), and LiTFSI (anhydrous 99.99%) were purchased from Aldrich, LiNO₃ additive (anhydrous 99%) was purchased from Alfa Aesar, and FLS0 and FLS18 were purchased from Synquest Laboratories. Cells were assembled with CR2032 coin casings (MTI) and electrolyte-flooded glass fiber paper (GFP, Whatman) separator under argon atmosphere. Electrochemical cycling occurred on MTI BST8-3 Battery Analyzer.

Dynamical measurements

Viscosity and density of the electrolyte solution were measured using the Anton Paar DMA 4500 and Lovis 2000 ME densimeter and viscometer. The system was purged with N₂ gas for 15 minutes before measurement to remove moisture and oxygen from the system. Four milliliters of each electrolyte were loaded into an airtight syringe (vacuum-dried at 60°C for 4 hours) in an Ar-filled glovebox (with oxygen and moisture concentrations less than 1 ppm). The same was then quickly injected into the densimeter and viscometer and remained closed until the measurement was completed. Viscosity and density were measured between 5°C and 40°C with 5°C intervals. Ionic conductivity of the electrolyte was measured with a fully integrated multichannel conductivity spectrometer (Bio-Logic MCS10) between 5°C and 40°C with 5°C intervals.

Data availability

The code and dataset associated with this work can be found in the GitHub repository, https://github.com/rashatwi/ebatt.

Acknowledgements

The authors acknowledge financial support provided by LG Energy Solutions. R.A. acknowledges support received through a Junior Research Award from the Institute for Advanced Computational Science (IACS) at Stony Brook University. This research used resources of the Seawulf cluster provided by IACS. D.T.N., V.J., and M.S. acknowledge the support of the Joint Center for Energy Storage Research (JCESR), an Energy Innovation Hub funded by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences (BES).

Author contributions

R.A. conceived the idea and designed the project. N.N.R. supervised the project. R.A. ran the COSMO-RS, DFT, and MD simulations. R.A. and A.B. performed the SHAP analysis. D.A.G. performed the electrochemical characterization. D.T.N. and M.S. performed the density, viscosity, and ionic conductivity measurements. V.M., V.G.P., and N.N.R. provided important insights during the project design and supervision. R.A. wrote the paper, with contributions from all authors.

Competing interests

The authors declare no competing financial interest.

Chu S, Majumdar A (2012) Opportunities and challenges for a sustainable energy future. nature 488, 294–303
Cheng L et al (2016) Sparingly solvating electrolytes for high energy density lithium–sulfur batteries. ACS energy Lett 1:503–509
Liang X et al (2015) A highly efficient polysulfide mediator for lithium–sulfur batteries. Nat Commun 6:1–8
Manthiram A, Fu Y, Chung S-H, Zu C, Su Y-S (2014) Rechargeable lithium–sulfur batteries. Chem Rev 114:11751–11787
Mikhaylik YV, Akridge JR (2004) Polysulfide shuttle study in the Li/S battery system. J Electrochem Soc 151:A1969
Bruce PG, Freunberger SA, Hardwick LJ, Tarascon J-M (2012) Li–O2 and Li–S batteries with high energy storage. Nat Mater 11:19–29
Ji X, Lee KT, Nazar L (2009) F. A highly ordered nanostructured carbon–sulphur cathode for lithium–sulphur batteries. Nat Mater 8:500–506
Schuster J et al (2012) Spherical ordered mesoporous carbon nanoparticles with high porosity for lithium–sulfur batteries. Angew Chem 124:3651–3655
He G et al (2013) Tailoring porosity in carbon nanospheres for lithium–sulfur battery cathodes. ACS Nano 7:10920–10930
Yang Y et al (2011) Improving the performance of lithium–sulfur batteries by conductive polymer coating. ACS Nano 5:9187–9193
Zhou W, Yu Y, Chen H, DiSalvo FJ, Abruña H (2013) c. D. Yolk–shell structure of polyaniline-coated sulfur for lithium–sulfur batteries. J Am Chem Soc 135:16736–16743
Fu Y, Manthiram A (2012) Orthorhombic bipyramidal sulfur coated with polypyrrole nanolayers as a cathode material for lithium–sulfur batteries. J Phys Chem C 116:8910–8915
Ji L et al (2011) Graphene oxide as a sulfur immobilizer in high performance lithium/sulfur cells. J Am Chem Soc 133:18522–18525
Song M-K, Zhang Y, Cairns EJ (2013) A long-life, high-rate lithium/sulfur cell: a multifaceted approach to enhancing cell performance. Nano Lett 13:5891–5899
Gupta A, Bhargav A, Manthiram A (2019) Highly solvating electrolytes for lithium–sulfur batteries. Adv energy Mater 9:1803096
Umeshbabu E, Zheng B, Yang Y (2019) Recent progress in all-solid-state Lithium – Sulfur batteries using high Li-ion conductive solid electrolytes. Electrochem Energy Reviews 2:199–230
Li Y, Guo S (2021) Material design and structure optimization for rechargeable lithium-sulfur batteries. Matter 4:1142–1188
Liu Y et al (2021) Electrolyte solutions design for lithium-sulfur batteries. Joule 5:2323–2364
Shin W et al (2020) Fluorinated co-solvent promises Li-S batteries under lean-electrolyte conditions. Mater Today 40:63–71
Gu S et al (2016) Suppressing the dissolution of polysulfides with cosolvent fluorinated diether towards high-performance lithium sulfur batteries. Phys Chem Chem Phys 18:29293–29299
Azimi N, Weng W, Takoudis C, Zhang Z (2013) Improved performance of lithium–sulfur battery with fluorinated electrolyte. Electrochem Commun 37:96–99
Gordin ML et al (2014) Bis (2, 2, 2-trifluoroethyl) ether as an electrolyte co-solvent for mitigating self-discharge in lithium–sulfur batteries. ACS applied materials & interfaces 6, 8006–8010
Zheng J et al (2019) High-fluorinated electrolytes for Li–S batteries. Adv energy Mater 9:1803774
Zhao C et al (2022) Pushing Lithium–Sulfur Batteries towards Practical Working Conditions through a Cathode–Electrolyte Synergy. Angew Chem Int Ed 61:e202203466
Drvarič Talian S et al (2017) Fluorinated ether based electrolyte for high-energy lithium–sulfur batteries: Li + solvation role behind reduced polysulfide solubility. Chem Mater 29:10037–10044
Zhang Z et al (2013) Fluorinated electrolytes for 5 V lithium-ion battery chemistry. Energy Environ Sci 6:1806–1810
Doi T, Shimizu Y, Hashinokuchi M, Inaba M (2017) Dilution of highly concentrated LiBF4/propylene carbonate electrolyte solution with fluoroalkyl ethers for 5-V LiNi0. 5Mn1. 5O4 positive electrodes. J Electrochem Soc 164:A6412
Xu GL et al (2019) Solid-State Lithium/Selenium–Sulfur Chemistry Enabled via a Robust Solid‐Electrolyte Interphase. Adv Energy Mater 9:1802235
Zhao Q et al (2018) Binary mixtures of highly concentrated tetraglyme and hydrofluoroether as a stable and nonflammable electrolyte for Li–O2 batteries. ACS Appl Mater Interfaces 10:26312–26319
Zheng J et al (2018) Extremely stable sodium metal batteries enabled by localized high-concentration electrolytes. ACS Energy Lett 3:315–321
Terada S et al (2018) Glyme–Sodium Bis (fluorosulfonyl) amide Complex Electrolytes for Sodium Ion Batteries. J Phys Chem C 122:16589–16599
Weng W, Pol VG, Amine K (2013) Ultrasound assisted design of sulfur/carbon cathodes with partially fluorinated ether electrolytes for highly efficient Li/S batteries. Adv Mater 25:1608–1615
Yu L et al (2018) A localized high-concentration electrolyte with optimized solvents and lithium difluoro (oxalate) borate additive for stable lithium metal batteries. ACS Energy Lett 3:2059–2067
Meisner QJ et al (2019) Lithium–sulfur battery with partially fluorinated ether electrolytes: Interplay between capacity, coulombic efficiency and Li anode protection. J Power Sources 438:226939
Park MS, Park I, Kang Y-S, Im D, Doo S-G (2016) A search map for organic additives and solvents applicable in high-voltage rechargeable batteries. Phys Chem Chem Phys 18:26807–26815
Husch T, Korth M (2015) Charting the known chemical space for non-aqueous lithium–air battery electrolyte solvents. Phys Chem Chem Phys 17:22596–22603
Borodin O, Olguin M, Spear CE, Leiter KW, Knap J (2015) Towards high throughput screening of electrochemical stability of battery electrolytes. Nanotechnology 26:354003
Kim S et al (2021) PubChem in 2021: new data content and improved web interfaces. Nucleic Acids Res 49:D1388–D1395
O'Boyle NM et al (2011) Open Babel: An open chemical toolbox. J Cheminform 3:33
Willett P, Barnard JM, Downs GM (1998) Chemical similarity searching. J Chem Inf Comput Sci 38:983–996
Su CC, He M, Amine R, Amine K (2019) A Selection Rule for Hydrofluoroether Electrolyte Cosolvent: Establishing a Linear Free-Energy Relationship in Lithium–Sulfur Batteries. Angew Chem 131:10701–10705
Ma S et al (2018) Temperature effect and thermal impact in lithium-ion batteries: A review. Progress Nat Science: Mater Int 28:653–666
Zhou H, Dai C, Liu Y, Fu X, Du Y (2020) Experimental investigation of battery thermal management and safety with heat pipe and immersion phase change liquid. J Power Sources 473:228545
Yan B et al (2020) Design, synthesis, and application of metal sulfides for Li–S batteries: progress and prospects. J Mater Chem A 8:17848–17882
Su Y-S, Fu Y, Cochell T, Manthiram A (2013) A strategic approach to recharging lithium-sulphur batteries for long cycle life. Nat Commun 4:2985
Brethome AV, Fletcher SP, Paton RS (2019) Conformational effects on physical-organic descriptors: the case of sterimol steric parameters. ACS Catal 9:2313–2323
Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 30
Elliott CF, Fraser KE, Odom SA, Risko C (2021) Steric Manipulation as a Mechanism for Tuning the Reduction and Oxidation Potentials of Phenothiazines. J Phys Chem A 125:272–278
Landrum G, RDKit (2022) Open-Source Cheminformatics
Moriwaki H, Tian Y-S, Kawashita N, Takagi T (2018) Mordred: a molecular descriptor calculator. J Cheminform 10:1–14
Rajput NN et al (2017) Elucidating the solvation structure and dynamics of lithium polysulfides resulting from competitive salt and solvent interactions. Chem Mater 29:3375–3379
Tan J, Li X, Shen J (2023) Building Better Lithium-Sulfur Batteries—A Reassessment of the Working Mechanism. Batteries Supercaps 6:e202200457
Yu Z et al (2022) Tuning Fluorination of Linear Carbonate for Lithium-Ion Batteries. J Electrochem Soc 169:040555. 10.1149/1945-7111/ac67f5
Lin Y et al (2024) Impact of Fluorination Degree of Ether-Based Electrolyte Solvent on Li-metal Battery Performance. J Mater Chem A
Lu H et al (2015) Application of partially fluorinated ether for improving performance of lithium/sulfur batteries. J Electrochem Soc 162:A1460
Yue Z, Dunya H, Aryal S, Segre CU, Mandal B (2018) Synthesis and electrochemical properties of partially fluorinated ether solvents for lithiumsulfur battery electrolytes. J Power Sources 401:271–277
Cuisinier M et al (2014) Unique behaviour of nonsolvents for polysulphides in lithium–sulphur batteries. Energy Environ Sci 7:2697–2705
Sun M et al (2018) Assessment on the self-discharge behavior of lithium–sulfur batteries with LiNO3-possessing electrolytes. ACS Appl Mater Interfaces 10:35175–35183
Atwi R, Bliss M, Makeev M, Rajput NN (2022) MISPR: an open-source package for high-throughput multiscale molecular simulations. Sci Rep 12:15760. 10.1038/s41598-022-20009-w
Gaussian (2016) 16 Rev. C.01Wallingford, CT
Grimme S, Antony J, Ehrlich S, Krieg H (2010) A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu. J Chem Phys 132:154104
Martínez L, Andrade R, Birgin EG, Martínez JM (2009) PACKMOL: a package for building initial configurations for molecular dynamics simulations. J Comput Chem 30:2157–2164
Banks JL et al (2005) Integrated modeling program, applied chemical theory (IMPACT). J Comput Chem 26:1752–1780
Schrödinger LLC (2021) (New York, NY
Canongia Lopes JN, Pádua AA (2004) Molecular force field for ionic liquids composed of triflate or bistriflylimide anions. J Phys Chem B 108:16893–16898
Canongia Lopes JN, Deschamps J, Pádua AA (2004) Modeling ionic liquids using a systematic all-atom force field. J Phys Chem B 108:2038–2047
Dang LX (1992) Development of nonadditive intermolecular potentials using molecular dynamics: Solvation of Li + and F – ions in polarizable water. J Chem Phys 96:6970–6977
Mayo SL, Olafson BD, Goddard W (1990) A. DREIDING: a generic force field for molecular simulations. J Phys Chem 94:8897–8909
Park C et al (2019) Structural and transport properties of Li/S battery electrolytes: role of the polysulfide species. J Phys Chem C 123:10167–10177
Leontyev I, Vener M, Rostov I, Basilevsky M, Newton MD (2003) Continuum level treatment of electronic polarization in the framework of molecular simulations of solvation effects. J Chem Phys 119:8024–8037
Leontyev I, Stuchebrukhov A (2009) Electronic continuum model for molecular dynamics simulations. J Chem Phys 130, 02B609
Hockney RW, Eastwood JW (1988) Computer simulation using particles. crc
Chen T, Guestrin C in Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining. 785–794

There is NO Competing Interest.

supplementaryinformation.docx
Additional information Supplementary information Detailed description of DFT calculations, MD simulation setups and analyses, feature extraction, ML model, and experimental methods (PDF).

Download PDF

Version 1

posted

You are reading this latest preprint version

Knowledge-driven design of fluorinated ether electrolytes via a multi-model approach

Status:

Version 1

Abstract

Figures

Introduction

Results and Discussion

Methods

Experimental methods

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1