Competitive Salt Precipitation/Dissolution During Free‐Water Reduction in Water‐in‐Salt Electrolyte

Abstract Water‐in‐salt electrolytes based on highly concentrated bis(trifluoromethyl)sulfonimide (TFSI) promise aqueous electrolytes with stabilities nearing 3 V. However, especially with an electrode approaching the cathodic (reductive) stability, cycling stability is insufficient. While stability critically relies on a solid electrolyte interphase (SEI), the mechanism behind the cathodic stability limit remains unclear. Now, two distinct reduction potentials are revealed for the chemical environments of free and bound water and that both contribute to SEI formation. Free water is reduced about 1 V above bound water in a hydrogen evolution reaction (HER) and is responsible for SEI formation via reactive intermediates of the HER; concurrent LiTFSI precipitation/dissolution establishes a dynamic interface. The free‐water population emerges, therefore, as the handle to extend the cathodic limit of aqueous electrolytes and the battery cycling stability.


Electrolyte preparation
The electrolyte solutions were prepared by dissolving salt in the corresponding quantity of water. Due to the hygroscopic character of the salt, in order to minimize calculation errors, the weighing was carried out under argon atmosphere. These solutions were prepared by molality (moles number of salt in 1 kg of water).

Electrode fabrication
Pica based electrodes: AC, CB and PTFE were mixed in a 75:15:10 weight ratio. AC was used as active material, CB as electronic conductor additive and PTFE as binder. After adding acetone, the solution was stirred and heated at 55°C until solvent evaporation. The resulting paste was kneaded and spread until resulting in a homogeneous film with a thickness of about 150 µm. Carbone nanofiber based electrode: a quantity of carbon nanofibers were dispersed in ethanol during 1 hour by stirring and sonication. The suspension was then filtered and the filtrate was pressed several times to obtain a self-standing layer. Chevrel phase Mo6S8 based electrode: the electrodes composition was prepared according to K. Xu et al., Mo6S8, CB and PTFE were mixed in a 80:10:10 weight ratio. The mixture was pressed in a thin film and punched out in 6 mm diameter electrodes.

Electrode preparation for surface analysis
The electrodes were first polarized in 12 m electrolyte to various reducing potentials for 15 min to form a surface film, and the pristine electrode was immersed overnight in 12m solution. The electrodes were then recovered and rinsed intensively with water to remove excess electrolyte or precipitated salt, then dried at 80 °C for 24 h. Electrodes were polarized to 1.2 V and 0.4 V vs. Li/Li + , corresponding to the reduction of free water and bound water, respectively.

Characterization
All the electrochemical analysis were recorded using VMP-300 Multi Potentiostat -Bio-Logic instrument. Scanning electron spectroscopy (SEM) and energy dispersive spectroscopy (EDS) analysis were recorded with an FEI scanning electron microscope. The X-ray photoelectron spectroscopy (XPS) analyzes are performed with the Thermo Electron ESCALAB 250 instrument. The excitation source is the monochromatic source, line Al Kα (1486.6 eV). The analyzed surface has a diameter of 400 μm. The photoelectron spectra are calibrated as binding energy with respect to the energy of the C = C component of the C1s carbon at 284.4 eV. X-ray diffraction recorded on PANalytical X'Pert in Bragg-Brentano configuration with CuKα radiation.

Linear polarization characterization
Rotating disk electrode: linear polarization was recorded in a three electrodes cell on a glassy carbon working electrode at different LiTFSI·H2O concentrations, platinum was used as counter electrode and Ag/AgCl as a reference. The scan rate was 1mV/s and before all measurements. Before all measurements, the working electrode was cleaned by mechanical polishing and the solutions were degassed for 10 min under nitrogen. The ohmic drop is compensated. Same experiments were performed using an internal reference, 2mM of ferrocene methanol was introduced in different salt concentration. Carbon nanofiber: linear polarization was recorded in a three electrodes Swagelok cell at 5mV/s scan rate. Freestanding carbon nanofiber as working electrode, platinum disk as a counter electrode and Ag/AgCl disk as a reference. Chevrel phase Mo6S8 based electrode: linear polarization was recorded in a three electrodes Swagelok cell at 5mV/s scan rate. Freestanding Mo6S8 based electrodes as working electrode, LiCoO2 based electrode as a counter electrode and Ag/AgCl disk as a reference.

Differential electrochemical mass spectrometry (DEMS)
The gases evolved upon polarization, were studied using differential electrochemical mass spectrometry coupled to linear polarization. In cell DEMS we used completely delithiated LFP as a counter electrode, activated carbon (PICA) based material as the working electrode and the partially delithiated LFP as a reference. We chose activated carbon due to the high specific area, which increases the reactivity of the electrode with the electrolyte. Several counter electrodes were used in order to avoid its polarization out of the stable potential window. We did the measurements with two different concentrations of LiTFSI.H2O electrolyte: 0.3 molal and 20 molal corresponding to salt-in-water and water-in-salt solutions respectively.

MD simulations
We use molecular dynamics to simulate two cells consisting of 3.5 and 20 m LiTFSI, respectively, placed between two graphite electrodes. The first simulation box (3.5 m LiTFSI) contains 116 LiTFSI ion pairs and 1839 water molecules as the electrolyte, the second one (20 m LiTFSI) contains 255 LiTFSI ion pairs and 707 water molecules as the electrolyte, and both of them contains 2496 carbon atoms as the two graphite electrodes. We exploit the atomistic model with modified OPLS force filed for ions and the SPC/E model for water, [1,2] which has been validated as in our previous work. [3] The electrodes consist of three fixed graphene layers on each side with in-plane dimensions of Lx = 32.25 Å and Ly = 34.37 Å. The length of the cell in z-direction was set to Lx = 90 and 96 Å (for 3.5 m and 20 m, respectively) to match the experimental density of the electrolyte in the bulk region of the simulation cells. A vacuum region was added in the z-direction, and the Yeh−Berkowitz condition for slab correction was used to mimic 2D periodic boundary condition. [4] The simulations were conducted in the NVT ensemble with a time step of 1 fs at room temperature (298.15 K) by using LAMMPS code. [5] The systems were first equilibrated during 80 ns at zero constant charge on each electrode and followed by 12 ns equilibration at zero constant potential. During the constant potential runs, the carbon atom charges on the electrode were allowed to fluctuate, which ensured an adequate description of the surface polarization by ions and water during simulations. The simulation data were finally collected in 8 and 25 ns production runs (for 3.5 and 20 m, respectively) at constant potential with configurations saved every 1 ps for each simulation.

Cathodic stability
With the used rotating disk electrode, the diffusion layer is in a stationary state as shown in Fig. S5. Its electrochemical response generally gives a current plateau or sharply rising current for low/high reactant concentration, respectively. The latter is seen for up to 7 m in Fig. 1a and Fig S1. At higher salt / lower free water concentrations, no plateau is seen, but a bell shaped peak. This is explicable by either the reduction of a finite amount of water in the diffusion layer and/or electrode blocking as a result of the water reduction. This resembles a thin-layer type behaviour, in which the electro-active molecules are confined in a finite volume. [6,7] The first reduction wave may therefore be associated to the reduction of the population of water which is a minority in the diffusion layer, i.e. the free water molecules, as shown by the simulation results in Fig.1c. Further, it means that this population is the less replenished from the bulk electrolyte reservoir the higher the salt concentration. Figure S1. Electrochemical stability of LiTFSI-H2O water-in-salt electrolytes. Linear polarization on a glassy carbon rotating disk electrode with different salt concentrations (molality). Platinum was used as counter electrode and Ag/AgCl as reference electrode. Curves were measured at 1000 min -1 , a scan rate of 1 mV·s -1 , and 25°C. a) and b) HER polarization curves versus Fc/Fc + andAg/AgCl, respectively.

Anodic stability
Linear polarization experiments performed in WIS with increasing concentrations on a rotating glassy carbon electrode on oxidation as shown in Fig. S1. During positive polarization, the oxidation potential for water continuously shifts to increasing potentials with increasing concentration in agreement with previous work. [8] This increase in the stability of water was recently explained by molecular dynamics (MD) simulations that showed that the strong adsorption of TFSI at the positive electrode tends to push lithium ions and water molecules away from the electrode surface [8] . Our MD simulations for electrolyte concentrations of 3.5 and 20 m at carbon electrodes are in agreement with these findings (see the density profiles on Fig. S2a). Although for the more dilute 3.5 m electrolyte the density of TFSI is not sufficient to fully displace water, in other words to "dry" the first adsorbed layer, it is indeed the case in the WIS regime at 20 m concentration, as illustrated on the representative snapshots shown on Fig. S2b. Hence, high salt concentrations result in an additional thermodynamic barrier for water to reach the positive electrode surface and to get oxi dized.

Water reduction at Mo6S8 electrodes
The results in Fig. 1 explain the reductive (cathodic) stability only on the basis of water reduction; direct TFSI reduction to form a passivating layer as suggested in previous reports on Chevrel phase Mo6S8 electrodes [8,9] appears not to be involved. To check whether the electrode material has an influence on the mechanism, we performed polarization measurements with porous electrodes made of the same Mo6S8. The results are compared to the ones on a porous carbon nanofiber (CNF) electrode in Fig. S6. The carbon electrode with 12 m solution shows analogously to Fig. 1a, a peaking higher voltage process and further reduction at lower voltage. The slight shift of the potentials is related to the different scan rate applied and to the porosity of the electrode surface. The Mo6S8 electrode shows a plateau with an onset at ~2.4 V with the 3.5, 12, and 20 m electrolytes, which was assigned to the reduction of Mo6S8 [8,10] . As with graphite electrodes, there is a reduction onset at ~1.8 V. The current density of the second reduction decreases with the electrolyte concentration from 3.5 to 20 m, suggesting a kinetic effect of the reaction, which we finally attribute to water reduction. The same two reduction waves confirm that water reduction at two distinct potentials at carbon and Mo6S8 electrodes is governed in the same way by the chemical environment of water molecules.

Precipitation/dissolution mechanism at the interface
The laminar flow at a rotating disk electrode conveys a steady stream of material from the bulk solution to the electrode surface. The rotating structure acts as a pump, pulling the solution upward and then throwing it outward reaching a steady state rather quickly as demonstrated in Fig. S5a. [11,12] According to the diffusion layer model, the electrolyte can be divided into two zones (Fig. S5b): 1. A first region close to the surface of the electrode with thickness δ, where it is assumed that there is a totally stagnant layer and thereby diffusion is the only mode of mass transport. 2. A second zone outside the first region where a strong convection occurs, and all species concentrations are inchanged during the redox mechanism. [11,12] The thickness of the stationary diffusion layer at a rotating disc electrode is illustrated in Fig. S5a  where ν is the kinematic viscosity, D the diffusion coefficient and ω the angular speed of the electrode. In the first zone where there is a concentration gradient, the slope of the curve can be determined by using these following coordinates: concentration at δ distance ( , δ) and the average concentration at δ/2 ( , δ/2), figure S5b: The concentration change due to water reduction in the diffusion layer can be calculated by using Faraday's equation corresponding to the number of moles in the volume when a charge Q has been consumed Here Q is the charge, n is the number of electrons exchanged during the reaction, F is the Faraday constant and S is the surface of the electrode. The average concentration is therefore the remaining water concentration.
Thus, the slope can be expressed as a function of the charge, the diffusion layer and the concentration in the bulk: The concentration in the first zone ( ( ), at all value of x) can be expressed as function of the slope and the concentration at x=0: For any point in the diffusion layer with 0< x ≤ δ, the linear equation is valided. Hence, at x = δ the concentration gradient equation reads: Therefore, the concentration at the interface (x = 0) is given as follow: The concentration of LiTFSI varies linearly with the water concentration as shown in the Fig. S10. The linear equation was used to extrapolate the salt concentration at the interface. Note that, the determined concentration are apparent values because we have included concentration higher than the solubility limit. The results express instead the solubility of the salt at the interface. A concentration exceeding the solubility means that the interface is continuously renewed with water enabling the solubility of LiTFSI salt therefore the high concentration values.

Interface formation at Mo6S8
Mo6S8 electrodes were polarized to 2.5 V and 1.4 V vs. Li + /Li, corresponding to the Mo6S8 reduction and water reduction potential, respectively. The electrodes were first polarized in 12 m electrolyte to various reducing potentials for 15 min to form a surface film and the pristine electrode was immersed overnight in 12m solution.
The electrodes recovered and rinsed intensively with water to remove excess electrolyte or precipitated salt, then dried at 80 °C for 24 h.
In contrast to the CNF electrodes in Fig. S12, polarized Mo6S8 electrodes appear in the SEM virtually identical to the pristine electrode (Fig. S15). Nevertheless, EDS shows a drastically increased oxygen content at 2.5 V with the other atom ratios unchanged. At 1.4 V, the O fraction decreases slightly whilst leaving the other atom ratios unchanged (Fig. S16). Figure S17 shows the C1s, O1s, F1s, S2p, N1s, and Li1s spectra of the Mo6S8 electrodes. The electrode polarized to 2.5 V shows peaks very similar to those of the pristine electrode without any deposit, except for C1s spectra showing the presence of CO3 and CO2 peaks, which can be explained by the adsorption of CO2 during cell assembly. Therefore, polarization to the potential, where Mo6S8 is reduced does not result in a surface layer via electrolyte reduction. In contrast, largely the same decomposition products as on CNFs are observed at 1.4 V, with an additional peak characteristic for Li2CO3. The strong intensity of this latter peak may be assigned to the oxidation of the carbon counter electrode. This assertion contrasts with the reduction of trace CO2 as suggested by Suo et al. [9] Our interpretation is supported by the absence of Li2CO3 at the CNF electrode, where the counter electrode was a platinum disk, excluding reductive processes at the negative electrode in the absence of CO2 as the Li2CO3 source. To confirm that the positive electrode is the origin of CO2, we polarized a symmetric cell with CNF electrodes as both cathode and anode. XRD of the negative electrode confirms the presence of Li2CO3 (Fig. S18) and thus carbon from the positive electrode to be the Li2CO3 source at the Mo6S8. Overall, carbon and Mo6S8 electrodes are covered with the same surface species when polarized to potentials that drive water reduction.