The use of binary mixtures of 1-butyl-1-methylpyrrolidinium bis{(trifluoromethyl)sulfonyl}imide and aliphatic nitrile solvents as electrolyte for supercapacitors

The development of high voltage electrolytes is one of the key aspects for increasing both energy and powerdensityofelectrochemicaldoublelayercapacitors(EDLCs).Theusageofblendsofionicliquidsand organic solvents has been considered as a feasible strategy since these electrolytes combine high usable voltages and good transport properties at the same time. In this work, the ionic liquid 1-butyl-1-methylpyrrolidinium bis{(tri ﬂ uoromethyl)sulfonyl}imide ([Pyrr 14 ][TFSI]) was mixed with two nitrile- based organic solvents, namely butyronitrile and adiponitrile, and the resultingblends were investigated regardingtheirusageinelectrochemicaldoublelayercapacitors.Bothblendshaveahighelectrochemical stability, which was con ﬁ rmed by prolonged ﬂ oat tests at 3.2V, as well as, good transport properties. In fact,thebutyronitrileblendreachesaconductivityof17.14mS (cid:1) cm (cid:3) 1 andaviscosityof2.46mPa (cid:1) sat20 (cid:4) C, which is better than the state-of-the-art electrolyte (1mol (cid:1) dm (cid:3) 3 of tetraethylammonium tetra ﬂ uoroborate in propylene carbonate). ã 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


. Introduction
In the field of electrochemical storage devices electrochemical double layer capacitors (EDLCs), which are also known as supercapacitors, are typically the desired technology when confronted with applications where fast delivery and/or uptake of energy is needed [1][2][3][4]. The energy in EDLCs is stored in a double layer, which is formed during charging by separating the positively and negatively charged ions in the electrolyte and adsorbing them at the electrodes' surface with the opposite charge. This charging process is much faster than the redox reactions taking place in battery systems leading to charging and discharging times in the order of seconds, an excellent cycle life (>500,000 cycles) and high power outputs (10 kWÁkg -1 ) [1][2][3][4].
Typically, activated carbons (ACs) are nowadays used as active materials, while electrolytes usually consist of organic solvents like propylene carbonate (PC) or acetonitrile (ACN) and tetraalkylammonium-salts like tetraethylammonium tetrafluoroborate ([Et 4 N][BF 4 ]) [5][6][7][8]. EDLCs using this setup of organic electrolyte and AC-based electrodes deliver an energy in the order of 5 WhÁkg À1 and a maximum usable voltage of 2.7 V-2.8 V [3,8,9]. Since the energy stored, E, in EDLCs is defined by the expression, E = 0.5ÁCV 2 , where C and V represent the capacitance and the operative voltage of the device, respectively, it is evident that an increase in the EDLCs' usable voltage will have a great impact on the EDLCs' supplied energy, which in turn would allow their usage in wider range of applications and, therefore, increase the market size of EDLCs [7].
The limiting factor of the maximum usable voltage in EDLCs is the electrochemical stability of the electrolyte [2]. One group of electrolytic materials that offers a wider range of electrochemical stability compared to the commonly used organic solvents based electrolytes, is the group of ionic liquids (ILs) [2]. ILs offer a unique set of properties which make them interesting for many energy applications, i.e. lithium and sodium ion batteries [10][11][12], EDLCs [2,13,14], dye sensitized solar cells [15][16][17], solar thermal collectors [18,19], actuators [20,21], fuel cells [22,23], thermoelectric cells [24,25] and water splitting [26,27], as well as other applications like CO 2 capture [28], material synthesis [29][30][31] or pharmaceuticals [31]. The interest in ILs for electrochemical applications primarily stems from the mentioned high electrochemical stability, which can be as high as 6 V [32]. In EDLCs, the high electrochemical stability of ILs translates into a maximum usable voltage for the final device as high as 3.2 V-3.5 V, affording an increase in energy density when compared to organic solvent based electrolytes [13]. Additionally, ILs are in general nonflammable and their vapor pressure is negligible, which has the potential to create safer devices compared to the usage of electrolytes based on volatile solvents like ACN. However, there are also some disadvantages regarding the usage of ILs: They are typically more cost-intensive than organic solvents and their typically high viscosity and low conductivity lead to an increased resistance of the device, which limits the power output of EDLCs [33]. However, high power EDLCs based on ILs with low relaxation time constants have been reported, showing that this limitation can be overcome [34]. Researching the processes taking place at the electrode-IL-interface appears to be a crucial step for tackling this issue [35][36][37][38].
Work done by Krause [45]. Besides improving both conductivity and viscosity at room temperature, also the low temperature behaviour was improved down to temperatures as low as À50 C, possibly enabling low temperature applications.
Ruiz [TFSI] with different carbonate and nitrile based solvents in order to assess the physical and electrochemical properties of these mixtures showing that transport properties of the IL may be drastically improved with the addition of solvent [41]. For example, the conductivity could be increased up to 45 mSÁcm À1 (measured at 25 C) with a blend of [Pyrr 14 ][TFSI] and ACN [41]. However, while electrochemical stability windows from 4.19 V up to 5.90 V were reported for the different blends, EDLC testing was only performed using a maximum voltage of 2.5 V; possibly not exploiting the electrochemical stability to its full capability.
In the following we report about the use of binary mixtures of nitrile-based solvents (butyronitile, adiponitrile) and an IL ([Pyrr 14 ][TFSI]) as possible electrolytes for EDLC application. Starting with the evaluation of the transport properties (conductivity and viscosity), all important aspects for their implementation as EDLC electrolyte are investigated, including the electrochemical stability window, the maximum operative voltage, as well as, full cell tests to test their electrochemical stability in both short and long term measurements.
The ionic conductivity of the electrolytes and the viscosity were measured as indicated elsewhere [51]. In short, a sealed glass conductivity cell with platinized Pt electrodes was filled with a 0.01 molÁdm À3 aqueous solution of KCl (VWR) of known conductivity to determine its cell constant. Afterwards, the conductivity for the electrolytes was determined in the range extending from -30 C to 80 C. The determination of the conductivity was done by performing impedance measurements using a Solartron model 1260 Impedance coupled with a potentiostat/galvanostat 273A PAR. The estimated maximum error of the conductivity values is 5%. Regarding the viscosity, an Anton Paar Modular Compact MCR 102 Rheometer (measurement uncertainty: AE0.25%) was used, measuring in the same temperature range of À30 C to 80 C (AE0.2 C) as the conductivity.
Composite AC-based electrodes were prepared as reported previously [52]. The prepared electrodes had a final composition of 90 % active material (activated carbon, DLC Super 30, Norit), 5 % conductive additive (carbon black, SuperC65, Imerys) and 5 % binder (sodiumcarboxymethylcellulose, CMC, Walocell CRT 2000, Dow Wolff Cellulosics) with an average mass loading of the active material of 4.6 mgÁcm À2 . Additionally, free standing electrodes with high mass loadings (30 mgÁcm À2 ) were prepared. For these electrodes the same AC and conductive additive as for the composite electrodes were used, but the binder was replaced by polytetrafluoroethylene (PTFE, Sigma Aldrich, 60% aqueous dispersion). The final composition was 85:10:5 (activated carbon: carbon black: binder) in this case. All components were dispersed in an excess amount of ethanol and the dispersion was stirred at elevated temperatures until the slurry became a highly viscous, dough-like mass. This mass was put onto a glass plate and rolled out several times before punching out disks with a diameter of 12 mm (area = 1.13 cm 2 ). The obtained electrodes were dried in an oven (ED 115, Binder) set to 80 C, before drying them for 24 h under vacuum at 180 C to remove any residual water.
For measuring the electrochemical properties of the electrolytes, Swagelok 1 -type cells were used, which were assembled in a glove-box under argon atmosphere, with oxygen and water contents below 1 ppm. As measuring devices, a Solartron model 1287A potentiostat and a VMP multichannel potentiostaticgalvanostatic system (Biologic Science Instruments), both connected to climatic chambers set to 20 C (KBF 115 Binder), were utilized.
The measurements for the electrochemical stability window (ESW) and the operative voltage were performed using a threeelectrode setup, in which a silver-wire was used as a quasireference electrode and a heavy, self-standing AC electrode was used as counter electrode. As working electrode, a platinum microelectrode (embedded in PEEK; active area = 0.79 mm 2 ) was employed for the determination of the ESW and an AC-based composite electrode was utilized for the determination of the operative voltage. In both cases, a Whatman GF/D glass microfiber filter (675 mm thickness, 13 mm diameter), drenched with 150 mL of electrolyte, was used as separator.
The ESW was determined by linear sweep voltammetry (LSV) at 1 mVÁs À1 using the Solartron controlled by Corrware 1 software.
Starting from the open circuit potential (OCP), the cell potential was scanned towards more positive (anodic limit) or more negative (cathodic) limits until a strong current response, indicating the decomposition of the electrolytes, was detected.
The operative voltages of devices using the considered electrolytes were identified by cyclic voltammetry applying a scan rate of 5 mVÁs À1 using the VMP workstation. The cell potential was scanned from OCP to a final potential of 0.8 V/-0.8 V vs. OCP, which was afterwards increased/decreased stepwise by 0.1 V, until the efficiency of the CV cycle dropped below 99 %.
With knowledge about the ESW and the operative voltage of the electrolytes, full cell testing was performed in a two-electrode setup using 100 mL of electrolyte by conducting cyclic voltammetry (CV) measurements and galvanostatic charge-discharge cycling. Due to the asymmetric voltage excursion of the electrolytes, the cells were balanced using electrodes with different electrode masses (m) based on the expression C + Ám + ÁDV + = C À Ám À ÁDV À using the values of specific capacitance (C) and voltage excursion (DV) revealed by the operative voltage test [53].
CV measurements were conducted with varying scan rates of 5, 20, 50, 100 and 200 mVÁs À1 . The specific capacitances reported for the CV measurements refer to the value, which was achieved at the half of the maximum voltage and were calculated using the total active mass of both electrodes.
Cycling experiments were performed by applying current densities of 0.5, 1, 2, 5 and 10 AÁg À1 for 1000 cycles each. The Coulombic efficiency (h), equivalent series resistance (ESR), specific capacitance of the total active material (C), average power (P) and average energy (E) were calculated as before [39,52].
In order to assess the stability of the investigated electrolytes, float tests were carried out, keeping the EDLCs charged at their maximum voltage for a prolonged time. In detail, the cells were kept at their maximum voltage for 20 h, followed by a short cycling test of 50 cycles and impedance measurements, in order to determine changes in specific capacitance and ESR (based on the 50th cycle). This sequence was repeated 25 times for a total of 500 h of float test. Cycling during this test was performed by applying a current density of 1 AÁg À1 with the maximum voltage matching the voltage during voltage hold. The impedance spectra were recorded with a 5 mV ac perturbation in the frequency region from 500 kHz to 10 mHz.

Results and discussion
Blends for an IL molar fraction (x[Pyrr 14 ][TFSI]) of 0.14 and 0.29 in BTN and ADN, respectively, have been identified as interesting electrolytes for EDLCs (detailed information about the chemicalphysical properties of blends with different ionic liquid/solvent ratios are available elsewhere) [54]. These particular compositions are further characterized in this work and referred to as the BTN-blend and the ADN-blend for the binary mixtures containing butyronitrile and adiponitrile, respectively.
The conductivity and viscosity were investigated in a temperature range of À30 C to 80 C (Fig. 1). For the ADN-based electrolyte, no conductivity or viscosity was determined for temperatures below 0 C, since the melting point of this blend is À12 C as revealed by DSC measurements [54]. For the BTN-based electrolyte no viscosity was determined for temperatures above 40 C, due to the high vapor pressure of BTN and, therefore, accelerated evaporation of the nitrile at higher temperatures. It is obvious that the BTN-based electrolyte has the better transport properties in comparison with the ADN based electrolyte, since the conductivity is considerably higher and the viscosity considerably lower over the whole temperature range. For example, at 20 C the ADN blend has a conductivity and viscosity of 5.27 mSÁcm À1 and 17.45 mPaÁs, whereas the BTN reaches a conductivity of 17.14 mSÁcm À1 and a viscosity of 2.46 mPaÁs. This observation is primarily the result of the lower viscosity of the neat BTN solvent (0.55 mPaÁs at 25 C) [55] relative to the neat ADN solvent (5.74 mPaÁs at 25 C) [47]. As seen in Fig. 1c and d, the temperature dependence of both conductivity and viscosity could be well described by the Vogel-Tammann-Fulcher (VTF) equations where s 0 and h 0 are the limiting conductivity and viscosity, respectively, and B s /B h and T 0,s /T 0,h are fitting parameters related to the pseudo-activation energy and glass transition temperature, respectively. The VTF equation is commonly used to describe the temperature-viscosity relationship in glassy-forming liquids, such as ILs, which typically do not follow Arrhenius-type dependence.
The VTF fitting parameters are reported in Table 1. For ADN, the T 0,s and T 0,h temperatures obtained by the conductivity and viscosity measurements are very close to each other (164 K and 172 K, respectively), whereas for BTN there is a higher difference (131 K and 163 K, respectively). Since in each case the T 0,s and T 0,h fitting parameters are expected to be the same, the values for ADN fit quite well, whereas for BTN, there is a difference of 32 K between T 0,s and T 0,h values, which may be related to the fact that the fit of the viscosity of BTN does not have a good coefficient of determination (R 2 = 0.99820). These fitting issues in turn are a result of the "open" experimental setup of the viscosity measurements (see experimental), which allows the BTN to slowly evaporate due to its high vapor pressure even for lower temperatures.  [58], suggesting that the [BF 4 ] À anion might be more stable than the [TFSI] À anion in ADN.
After the determination of the ESW, the operative voltages of the electrolytes were determined via conducting 3-electrode setup CV measurements. Fig. 3 presents the obtained CV curves as well as the obtained Coulombic efficiencies. The positive and negative potential limits were defined by the highest and lowest potentials  at which the efficiency of the charge-discharge remained higher than 99 %. With this condition, the potential limits for the BTN blend were 1.65 V vs. Ag on the positive side and À2.05 V vs. Ag on the negative side. For the ADN blend, the potential limits are slightly shifted to more positive potentials (1.85 V and À1.85 V vs. Ag, respectively). Thus, the maximum operative voltage for both blends adds up to 3.7 V, the same value reported for the neat IL [39]. Therefore, the investigated nitriles do not have a negative impact on the applicable voltage for full cell EDLCs in regards to electrolyte stability. Also, the same operative voltage has been reported for an electrolyte based on ADN and 0.7 molÁdm À3 [Et 4 N] [BF 4 ] [58]. After the operative voltages of the different electrolytes were obtained, full cells with the two blends were assembled. Due to the different voltage excursion on the positive and negative side of the BTN blend, the masses of the electrodes were balanced in order to avoid premature electrolyte decomposition on either electrode [53]. To further decrease the chance of electrolyte decomposition, cell voltages of 3.5 V and 3.2 V were applied for full cell testing. Fig. 4 shows the CV obtained for these full cells using a maximum cell voltage of 3.5 V (Fig. 4a and b) or 3.2 V (Fig. 4c and  d). No Faradaic reactions are observed in any of the investigated electrolytes. Regardless of the [Pyrr 14 ][TFSI]/nitrile blend and operating voltages employed both blends show an increase in distortion from the typical rectangular shape for the CVs for higher scan rates. At lower scan rates the difference in capacitance is negligible for the two blends, and 24 FÁg À1 in both BTN and ADN were observed at 20 mVÁs -1 , using an operative voltage of 3.5 V. For a maximum cell voltage of 3.2 V, the specific capacitance is slightly lower, displaying 23 FÁg À1 for both BTN and ADN. Due to the aforementioned differences in conductivity and viscosity, the ADN blend is only able to retain ca. 50 % of its initial specific capacitance upon increasing the scan rate from 5 to 200 mVÁs À1 whereas the BTN is able to retain ca. 80 %. In both cases the retention is independent of the maximum cell voltage employed.
To further investigate the performance of the electrolytes, galvanostatic cycling experiments were performed using current densities ranging from 0.5 AÁg À1 to 10 AÁg À1 (Fig. 5). Similar to the CV measurements, operative voltages of 3.5 V and 3.2 V were investigated. The corresponding voltage profiles for the lowest and highest current densities used are presented in Fig. 6. For all electrolytes the Coulombic efficiency is stable for all cycles and all applied current densities, reaching almost 100 %. The ESR reaches a value of $10 VÁcm 2 for BTN and $16 VÁcm 2 for ADN and remains stable upon cycling, independent of applied cell voltage. This is somewhat surprising given the differences in conductivity and viscosity for both blends; however, the higher resistance in the case of the ADN system was expected. At lower current densities the specific capacitance for both blends is almost identical, reaching ca. 21 FÁg À1 at a current density of 1 AÁg À1 , which is a comparable value to other AC based EDLCs using binary mixtures of organic solvents and ILs [46,61]. However, at higher current densities, the better transport properties in conductivity and viscosity for the BTN-based blend allow for a higher retention of the capacitance. For example, at 10 AÁg À1 the cells using this electrolyte are still able to deliver 18 FÁg À1 (82 % of the capacitance at 0.5 AÁg À1 ), whereas the cell using ADN as the electrolyte solvent no longer displays any capacitive behavior at this rate. Upon cycling, a stronger decrease in the specific capacitance can be observed for the higher cell voltage of 3.5 V, especially for the ADN blend. Initially, the specific capacitance was the same at 0.5 AÁg -1 (22 FÁg À1 ), but after 4000 cycles the specific capacitance for the 3.2 V cell is higher than for the 3.5 V cell (16 FÁg À1 and 14 FÁg À1 , respectively). The difference in capacitance loss for the different voltages is not too large (27 % at 3.2 V and 36 % at 3.5 V), it may, nevertheless, be an indication of long term stability problems.
The average energy and average power were calculated, according to reference 39, for the devices and plotted in a Ragone-like plot (Fig. 7). The mass used for the calculation refers to the sum of the active mass of both positive and negative electrodes. Also, to get a better insight into the responsiveness of the devices, the time constants t have been calculated, as indicated [ 1 0 _ T D $ D I F F ] by Taberna et al. [62], and values of 12 s for the BTN blend and 32 s for the ADN blend were found. At 1 AÁg À1 and a maximum cell voltage of 3.5 V the devices with the BTN blend deliver an average energy and power of 32.2 WhÁkg À1 and 1.7 kWÁkg À1 , respectively. With a maximum cell voltage of 3.2 V, these values decrease to 26.0 WhÁkg À1 and 1.5 kWÁkg À1 . Since the voltage of the cell has a strong impact on the energy and power delivered by EDLCs, a decrease in energy and power when going from 3.5 V to 3.2 V was expected. Similarly, the energy and power density for the ADN blend decreases from 23.3 WhÁkg À1 and 1.4 kWÁkg À1 , respectively (at 3.5 V) to 21.8 WhÁkg À1 and 1.3 kWÁkg À1 , respectively (at 3.2 V). For higher current densities the BTN blend clearly outperforms the ADN blend, due to its higher conductivity and lower viscosity. The values for average energy and power achieved here are very comparable to systems using similar combinations of electrode materials and other type of non-conventional electrolytes [46,61,63]. Particularly, the performance of the device containing the ADN blend appears comparable to data already reported for this type of solvent, while the performance of the device containing the BTN blend appears similar to results obtained for high voltage PC-based devices [39,56,58,61]. Finally, float tests were carried out as a stress test, keeping the cells charged at their respective maximum voltage for 200 h. In intervals of 20 h, short charge/discharge steps as well impedance measurements were performed to follow the development of capacitance and ESR. As seen in Fig. 8, the stability of the electrolytes is strongly dependent on the applied voltage. Independent of the [Pyrr 14 ][TFSI]/nitrile blend employed, a huge capacitance loss and ESR increase can be observed for a voltage of 3.5 V. In the corresponding Nyquist plots, semicircles can be observed in the high frequency region, which can be attributed to contact resistance between the current collector and the carbon coating. More importantly, as the number investigated intervals of 20 h increases, the size of these semicircles also increases. These observations highlight the degradation processes taking place at a maximum voltage of 3.5 V. In comparison, when using a maximum of 3.2 V, the stability of the electrolytes is greatly increased with only a small decrease of capacitance and increase of ESR. Furthermore, the specific capacitance decreases from 21 FÁg À1 to 15 FÁg À1 and 19 FÁg À1 for the BTN and ADN blends, respectively, whereas the ESR increases from 17 VÁcm 2 to 20 VÁcm 2 (ADN) and  visible holes for the same electrochemical conditions. Therefore, it is reasonable to assume that the solubility of the Al-[TFSI]complexes might be lower in ADN compared to BTN, leading to the higher stability of the ADN blend in the float test. The lower solubility may be related to the higher viscosity of ADN; however, more investigations are necessary to completely assess the different processes occurring in the BTN and ADN electrolyte blends.

Conclusion
In this manuscript we have reported about the use of blends of nitriles, namely BTN and ADN, and [Pyrr 14 ][TFSI] in EDLCs. Investigations included the determination of physical (conductivity, viscosity) and electrochemical (electrochemical stability, operative voltage, specific capacitance) properties.
Based on the results of the conductivity and viscosity measurements, the BTN blend has the superior transport properties with higher conductivity (17.14 mSÁcm À1 at 20 C) and lower viscosity (2.46 mPaÁs at 20 C) than the ADN blend. In fact, these values are slightly better than the commonly used organic solvent based electrolyte of 1 molÁdm À3 Et 4 NBF 4 in PC (13 mSÁcm À1 /2.6 mPaÁs at 25 C, respectively). By using a three-electrode configuration, the electrochemical investigations revealed a good electrochemical stability allowing a maximum operative voltage up to 3.7 V for both blends. However, while both blends show reasonable values for specific capacitance and ESR using a maximum voltage of 3.5 V, applying this voltage for prolonged time during float testing led to severe losses in specific capacitance and increases in ESR. By reducing the maximum cell voltage to 3.2 V, only small changes in specific capacitance and ESR were observed after 200 h of float testing. Cells with the ADN blend, for example, were able to still display 89 % of the initial capacitance. The strong degradation of the electrochemical performance at 3.5 V was attributed to the anodic dissolution of the aluminum current collector by the [TFSI] À -anion, however more in-depth studies are necessary to fully confirm this.
Due to their high stability with an operative voltage as high as 3.2 V and, in case of the BTN blend, similar transport properties to PC/[Et 4 N][BF 4 ], the investigated blends surely seem to be promising alternatives to conventional electrolytes. As a matter of fact, [ ( F i g . _ 7 ) T D $ F I G ] using these electrolytes it is possible to realize EDLCs which are able to deliver energy and power values among the highest reported so far for devices containing non-conventional electrolyte [13]. These results confirm that the use of blends of organic solvent and ionic liquids represent one of the most interesting and successful strategies for the realization of advanced EDLCs[ 1 1 _ T D $ D I F F ] .[ 6 _ T D $ D I F F ]