Experimental measurement and prediction of ionic liquid ionisation energies

Ionic liquid (IL) valence electronic structure provides key descriptors for understanding and predicting IL properties. The ionisation energies of 60 ILs are measured and the most readily ionised valence state of each IL (the highest occupied molecular orbital, HOMO) is identified using a combination of X-ray photoelectron spectroscopy (XPS) and synchrotron resonant XPS. A structurally diverse range of cations and anions were studied. The cation gave rise to the HOMO for nine of the 60 ILs presented here, meaning it is energetically more favourable to remove an electron from the cation than the anion. The influence of the cation on the anion electronic structure (and vice versa) were established; the electrostatic effects are well understood and demonstrated to be consistently predictable. We used this knowledge to make predictions of both ionisation energy and HOMO identity for a further 516 ILs, providing a very valuable dataset for benchmarking electronic structure calculations and enabling the development of models linking experimental valence electronic structure descriptors to other IL properties, e.g. electrochemical stability. Furthermore, we provide design rules for the prediction of the electronic structure of ILs.


Introduction
Ionisation energy, E i , is a key descriptor for chemical, photochemical and electrochemical reactivity, 1-5 especially any application that involves exchange of electrons, particularly formal donation of an electron (ionisation) or donation of electron density (partial ionisation). For ionic liquids (ILs), these potential applications include: electrochemical energy storage; gas capture/separation/storage; as solvents for catalysis and metal extraction/separation. [6][7][8][9][10][11] The identity of the most readily ionised valence state, often called the highest occupied molecular orbital (HOMO), 12 is also a reactivity descriptor, particularly for ILs given the HOMO could come from the anion or from the cation. Furthermore, given their importance, E i and the HOMO identity can be used for quantitative validation of calculations of ILs. 13 E i can be used to validate methods, e.g. choice of functional/basis set in density functional theory (DFT) can be benchmarked. 13 HOMO identity can be used to validate the ability of calculations to capture the solvation effects of ions in liquid phase. However, for ILs there is limited experimental data on electronic structure, including E i and HOMO identity.
Most measurements of E i have been made using nonresonant X-ray photoelectron spectroscopy (XPS) and ultraviolet photoelectron spectroscopy (UPS). 14 Almost all E i values for ILs, E i (IL), have been measured on vaporised gas phase neutral ion pairs. [15][16][17][18][19][20][21][22][23][24][25][26] Whilst measuring E i is relatively facile in the gas phase, a major problem is that most ILs are very tricky to vaporise without significant thermal decomposition occurring/ dominating, meaning many IL ion pairs cannot be easily studied in the vapour phase; [27][28][29][30][31][32] furthermore, vapour phase ion pairs do not have the complete solvation environment of the liquid phase. A major hurdle for measuring reliable, reproducible, and comparable binding energies (E B ) 14 and E i for liquid phase ILs is dealing with sample charging during XPS measurements, which is not understood. 33,34 A widely-used, very robust method of charge referencing XP spectra for ILs is using E B (C alkyl 1s) = 285 eV for ILs with relatively long alkyl chains, usually -C 8 H 17 . 33,[35][36][37][38][39][40][41][42][43][44][45][46][47][48] E B values of valence states using this charge referencing method have been reported, e.g. E B (HOMO), E B (cation HOFO) and E B (anion HOFO), where the ion HOFOs are the highest occupied fragment orbital of each ion (one of which is the IL HOMO). 48 However, all of these E B values were effectively charge referenced to an apparent Fermi level for the alkyl chains, not the vacuum level. Reporting of experimental E i (IL) values (by definition, charge referenced to the vacuum level) for liquid phase ILs have been very limited, with little or no mention of charge referencing given; [49][50][51] these studies were published before IL sample charging was an acknowledged problem. Recently, E i (C C-C/C-H 1s) = 289.58 AE 0.14 eV was given as a reference to the vacuum level for C-C/C-H carbon adsorbed on conducting metal surfaces, [52][53][54][55][56] although this value has not been applied to IL XPS data to date.
A key challenge is to identify the valence states for ILs. Most importantly, which of the cation and the anion gives rise to the HOMO? For traditional salts such as NaCl, the anion is very clearly the HOMO, with the Na 2p cation HOFO valence state B26 eV larger E B than the Cl 3p anion HOFO. 57 However, for ILs E B of the cation HOFO and anion HOFO are far more similar. Furthermore, ILs have a relatively large number of valence electrons per the cation-anion molecular unit. Water and NaCl both have eight valence electrons and therefore the valence states are relatively easy to identify using XPS. [57][58][59][60] In contrast, common ILs can have between 50 and 300 valence electrons. 61 Therefore, ILs have many valence states at similar E B . Compounding this problem, the large range of ion solvation environments in the liquid phase is expected to give a significant range of E B for the nominally the same valence state, as demonstrated for Cl À ion solvated in water. 62 Consequently, valence XP spectra of ILs often have many overlapping contributions, making the separation of cation and anion contributions very difficult. The most common laboratory-based XPS apparatus employs Al Ka radiation at hn = 1486.6 eV, giving non-resonant XPS. Valence XPS data from hn = 1486.6 eV has been analysed using a visual fingerprint method and subtraction, 48 although this approach is difficult when using results measured on different apparatus. Furthermore, for XPS measured at hn = 1486.6 eV the most important contributors to cation-based valence states, C 2p and N 2p, have very low photoionisation cross-sections relative to many of the common anion-based valence states, e.g. Cl 3p, S 3p, making identification of cationic-based valence states very challenging in particular. Measuring photoelectron spectroscopy with a second hn, e.g. hn = 21.2 eV from He(I) giving non-resonant UPS, helps valence state identification due to variation in photoionisation cross-sections with varying hn. 48,[63][64][65][66][67][68] However, most valence state identification for ILs to date has relied on comparisons to calculations; this situation is less than ideal when trying to use experimental data to validate calculations. Valence state identification for ILs has mostly been limited to ILs comprised of [C n C 1 Im] + (1-alkyl-3-methylimidazolium), with a small selection of commonly studied anions, particularly cyano-based anions and [NTf 2 ] À (bis[(trifluoromethane)sulfonyl]imide). 48,[63][64][65][66][67][68][69][70][71][72][73] A recent development has been the use of resonant Auger electron spectroscopy (RAES, also known as resonant XPS, RXPS), which allows identification of valence states, particularly those states with strong p-bonding contributions, e.g. imidazolium ring, cyano-based anions. 48 Using an approach combining variable hn XPS and RXPS, key values for 37 ILs were determined: E B (HOMO), E B (cation HOFO) and E B (anion HOFO). 48 4 ] the cation (rather than the anion) has been identified as giving rise to the HOMO. 48,49,[74][75][76] The possibility, out of the potentially vast number of ILs, that an ideal IL exists for a particular application is an appealing prospect. The challenge of synthesising, characterising and testing a large number of potential ILs for an application is daunting and makes screening using predictions hugely advantageous. One important question for understanding and predicting IL properties is: how independent is the electronic structure of the cation from the anion and vice versa? Using XPS it has been demonstrated implicitly, i.e. by studying core state E B for elements located specifically in the cation, that the anion influenced the valence electronic structure of the cation (for the cations imidazolium, 35,37 pyridinium, 41,45 ammonium (linear 39 and cyclic 36,44 ) and phosphonium 39 3 ] À . 69 Gas phase E i (IL) have been compared to electrochemical stability for a number of ILs (the IL choice being limited to those ILs that can be vaporised). 25 Comparisons exist between liquid phase experimental XP/UP spectra for ILs and calculated data, but the structural range of ILs studied is limited. 24,49,[64][65][66][67][68][69][70][72][73][74][75][76]79 88 such comparisons should be made against experimental data measured on inert electrodes to minimise the importance of specific chemical reactivity with the electrode materials. 84 For quantitative structure-property relationships (QSPR), IL electronic structure descriptors such as E i have been used to understand and predict IL properties. 94 Importantly, the accuracy of these calculated electronic structure descriptors is not routinely validated against IL experimental electronic structure data, most likely due to a lack of available experimental data.
In this article, we investigate IL valence electronic structure using experimental methods, without the aid of calculations. Key ions studied are given in Fig. 1; all 60 ILs studied in this paper are given in ESI, † Table S1. All 60 ILs were liquid at room temperature, making XPS experiments relatively straightforward, as no heating was required for any IL studied here. Compared to the ILs studied in ref. 48, two new cation cores were studied here, [C n C 1 Pyrr] + and [C n Py] + . Furthermore, 20 new anions were studied, with a variety of properties/reasons to study; how the new anions were chosen is given in ESI, † Section S1. We have studied a total of 60 ILs using laboratorybased XPS; 37 ILs from ref. 48

IL synthesis
Details of IL synthesis are given in the ESI, † Section S1. Fig. 1 Key ions studied in this paper. A full list of ILs studied is given in the ESI, † Table S1.

Laboratory XPS
Laboratory-based XPS was carried out using four separate XP spectrometers for the 23 ILs studied here. In general, a drop of IL was placed directly onto a stainless steel sample plate (one IL was studied on a glass substrate). This sample was placed in a loadlock and the pressure reduced to 10 À7 mbar by pumping down for 46 hours. After attaining the required pressure, the IL was transferred to the analysis chamber. Etching (where necessary) was carried out using a 500 eV Ar + ion gun (B10 minutes per sample). Acquisition parameters were matched where possible to give comparable energy resolution; generally, a pass energy of 20 eV was used for core states and B40 eV for valence states.
(i) Non-resonant XPS of 16 ILs were recorded at University College London on a Thermo Scientific K-alpha monochromated Al Ka source (hn = 1486.6 eV) spectrometer. Charge compensation was achieved using a dual beam flood gun which applied both electrons and low energy Ar + ions to the sample.
(ii) Non-resonant XPS of 16 ILs were recorded at University College London on a Thermo Scientific Theta Probe monochromated Al Ka source (hn = 1486.6 eV) spectrometer. Charge compensation was achieved using a dual beam flood gun which applied both electrons and low energy Ar + ions to the sample.
(iii) Non-resonant XPS of four ILs were recorded at the University of Nottingham on a Kratos Axis Ultra equipped with a monochromatic Al Ka source (hn = 1486.6 eV). The core states were published already in ref. 36 and 41-43. Charge compensation was achieved using a flood gun which applied low energy Ar + ions to the sample.
(iv) Non-resonant XPS of one IL was recorded at Harwell XPS using a Kratos Axis Ultra DLD equipped with a monochromatic Al Ka source (hn = 1486.6 eV). The X-ray source was operated at 150 W (10 mA Â 15 kV). Charge compensation was achieved using a flood gun which applied low energy Ar + ions to the sample.

Synchrotron XPS and resonant XPS
Synchrotron XPS and resonant XPS were carried out using two separate beamlines, I09 and B07 at Diamond Light Source. In both cases a thin IL film was spread from less than 0.1 ml droplet on a tantalum sample holder so no drop could be observed by eye.
The soft synchrotron XPS for [C 8 C 1 Im][SnCl 3 ] was performed on the I09 beamline at Diamond Light Source (UK). 95 The XP spectra and RXP/RAE spectra were acquired using a VG Scienta EW4000 HAXPES analyser, which had an angular acceptance of AE301. The analyser was mounted with its lens axis approximately 901 away from the direction of the incident X-ray light in a horizontal plane; the analyser slits (and thus the angular acceptance direction) were also in the horizontal plane. Due to significant observable beam damage/sample charging (ESI, † Section S2), the flux of the synchrotron light was decreased by first defocussing the incident light B20 fold and by detuning the undulator (i.e. offsetting the undulator gap) away from the maximum intensity so as to detune the flux a further 100-fold. Prior to XPS measurements the sample was Ar + sputtered for 30 minutes at a voltage of 500 V.
The soft synchrotron XPS for 10 ILs was performed on the B07 beamline at Diamond Light Source (UK). 96 A thin film (less than 0.1 ml, essentially so no drop could be observed) of the IL sample was placed on a tantalum sample holder. For the T-cup apparatus, nine ILs were measured at the N 1s edge. Due to significant observable beam damage, the flux was reduced by using the 1200 l mm À1 grating (T-cup apparatus only). For the T-pot apparatus, one IL was measured at the C 1s edge. Due to significant observable beam damage, the sample was rastered continually perpendicular to the analyser entrance nozzle during X-ray irradiation (T-pot apparatus only); this rastering allowed a higher flux (400 l mm À1 grating) to be used than for the T-cup apparatus.
The RXPS/RAES data were acquired across the N 1s edge (hn B 402 eV) or the C 1s edge (hn B 285 eV); at each hn a RXP spectrum was acquired. Partial electron yield near edge X-ray absorption fine structure (NEXAFS) spectra for the N 1s and C 1s edges were recorded by summing the recorded RAE/RXPS intensity at each hn.

Analysing XP spectra
All non-resonant XP spectra were fitted using the CASAXPSt software. Fitting was carried out using a Shirley background and GL30 line shapes (70% Gaussian, 30% Lorentzian). The peak constraints used for core XP spectra are outlined ESI, † Section S3 and peak constraints used for valence XP spectra are outlined ESI, † Section S4. The purity of the ILs studied here is demonstrated in the ESI, † Section S7.

Charge referencing methods for XP spectra
All XP spectra for ILs were effectively charge referenced to C-C/C-H carbon for long alkyl chains. Two different values of E B (C alkyl 1s) were used.
(a) E B (C alkyl 1s) = 285.00 eV, which is equivalent to charge referencing to the Fermi level for long alkyl chains. This E B value is standard in the IL literature. 33,[35][36][37][38][39][40][41][42][43][44][45][46][47][48] (b) E i (C alkyl 1s) = 289.58 eV, [52][53][54][55][56] which is equivalent to charge referencing to the vacuum level for long alkyl chains. Adding the work function 97 for long alkyl chains would convert E B to E i . 14 The C-C/C-H carbon contribution to adventitious carbon has been found to match the vacuum level when setting E i (C alkyl 1s) = 289.58 AE 0.14 eV. [52][53][54][55][56] For our measurements, this value effectively means the work function was 289.58 À 285.00 = 4.58 eV. This value for the effective work function of alkyl carbon matches to expected work functions, which often range from 4 eV to 5 eV. 98 Therefore, to produce E i values referenced to the vacuum level from our E B values charge referenced to E B (C alkyl 1s) = 285.0 eV, we added 4.58 eV (Table 1). This charge referencing approach has not been used for ILs to date.
For the 60 ILs studied here, six different charge referencing methods were used to charge reference to E B (C alkyl 1s); all charge referencing was carried out after the measurements were completed.
(i) 36 ILs had a sufficiently long alkyl chain that a fitted component for E B (C alkyl 1s) for long alkyl chains was used for charge referencing all XP spectra. This approach to charge reference to E B (C alkyl 1s) has a very high confidence, with an error of less than AE0.1 eV.
(ii) 16 [C n C m Im][A] (where n r 4 and m = 1 or 0) ILs where [C 8 C 1 Im][A] IL with the same anion had already been studied, so E B (N cation 1s) or E B (element anion core) was used for charge referencing all XP spectra, effectively charge referenced to E B (C alkyl 1s) for long alkyl chains. This approach has a very high confidence, with an error of less than AE0.1 eV.
(iii) 1 IL, [C 4 C 1 Im][MeSO 4 ], E B (N cation 1s) for [C 4 C 1 Im][OcSO 4 ] was used to charge reference all XP spectra. As both anions are alkylsulfate, the same functional group was interacting with the countercations; hence, this approach to effectively charge reference to E B (C alkyl 1s) has a very high confidence, with an estimated error of less than AE0.1 eV.
(iv) 1 IL, [C 4 C 0 Im][HSO 4 ] with a protic cation, E B (C alkyl 1s) for [C 4 C 1 Im][HSO 4 ] was used to charge reference all XP spectra. As n = 4, based on data in ref. 33, this approach to effectively charge reference to E B (C alkyl 1s) has an estimated error of AE0.2 eV.
(v) 4 ILs with relatively short alkyl chains, where an IL with the same anion and a different cation with a long alkyl chain had already been studied (i.e. ILs from charge referencing method i) but the cation was new for XPS studies, so E B (element anion core) from [C 8 C 1 Im][A] was used for charge referencing all XP spectra. Based on data for [C][A] (where [C] + = cation) where the cation was varied (all with sufficiently long alkyl chains) and the anion kept constant, E B (element anion core) varied by a maximum of 0.4 eV. 78 Therefore, this approach to effectively charge reference to E B (C alkyl 1s) has an error of AE0.2 eV. Details of the charge referencing method applied to the synchrotron XP spectra are given in ESI, † Section S5.

Determining key valence electronic structure descriptors
The onset method used to determine E B (anion onset), E B (cation onset) and E B (IL onset) is explained in ref. 48. Threshold energies, E th (IL), were obtained by adding 4.58 eV to E i (IL onset), the IL onset energy charge referenced to the vacuum level); these E th (IL) values are compared to literature values.
The valence electronic structure descriptors charge referenced to the Fermi level are given in column 1 of Table 1 Table S8).
The valence electronic structure descriptors charge referenced to the vacuum level are given in column 3 of Table 1, and how they were determined in column 4. Values for valence electronic structure descriptors charge referenced to  (Table 3), E i (anion) and E i (cation) (ESI, † Tables S6 and S7 respectively) and E i (cation,pred.) (Table 4 and ESI, † Table S8). The valence electronic structure descriptors for which the reference level does not matter are given in column 5 of Table 1, and how they were determined in column 6. Values for valence electronic structure descriptors for which the reference level does not matter are given: DE B (ion HOFO), DE B (ion onset) and HOMO identity in Table 3 Fig. S9f, S12f, S13f, S14f, S15f). However, for 18 ILs studied here, features due to cationic valence states were not readily observed in non-resonant valence XP spectra recorded at hn = 1486.6 eV due to features from the anion valence states dominating (ESI, † Fig. S10-S32); the photoionisation crosssections of N 2p and C 2p atomic orbitals (AOs) are much lower than many of the anion-based AOs, e.g. Cl 3p. 99 For RXPS of the 15 [C n C 1 Im] + -based ILs reported here and in ref. 48 (Fig. 3a, c, 4 and ESI, † Fig. S44). hn E 402 eV corresponded to X-ray absorption from the N cation 1s core state to imidazolium ring p* unoccupied valence state(s) (Fig. 3b and d). 48,100 The feature at 3.5 eV o E B (N cation RXPS) o 7.5 eV was from participator Auger transitions involving valence states with good overlap with the N cation 1s core hole. Consequently, for [C n C 1 Im] + -based ILs, valence states at 3.5 eV o E B (N cation RXPS) o 7.5 eV had strong contributions from N cation in the imidazolium ring, i.e. from N cation 2p-based AOs. The anion charge ([A] À or [A] 2À ) did not have a strong effect on E B (N cation RXPS) (Fig. 4). The average E B (N cation 2p) for [C 8 C 1 Im][A] was estimated as E B B 5.7 eV (Fig. 4). Based on results presented in ESI, † Fig. S45 Fig. 2a and b) 48 investigating the effect of the counteranion on the cation contributions to valence XP spectra is very challenging. However, E B (N cation RXPS) potentially can be used to probe the effect of the counteranion on E B (cation HOFO). E B (N cation RXPS) showed some variation with respect to the anion identity (Fig. 4). There was a link between E B (N cation RXPS) (i.e. E B (N cation 2p)) and E B (N cation 1s); for [C 8 C 1 Im][NTf 2 ] both E B (N cation 2p) and E B (N cation 1s) were relatively large compared to E B (N cation 2p) and E B (N cation 1s) for [C 8 C 1 Im]Cl. This variation was not easy to discern given uncertainty that was principally from the subtraction process. These tentative observations suggest a linear correlation between E B (N cation 2p) and E B (N cation 1s) 101 For features at E B 4 12 eV (ESI, † Fig. S48), the dominant contributions were from spectator Auger transitions (i.e. not from participator Auger transitions). When charge referenced to E B (N cation 1s) (ESI, † Fig. S48), the subtracted N cation traces (which include peaks due to both participator and spectator Auger transitions) for [C n C 1 Im][A] where the anion was varied were the same (ESI, † Fig. S48)

][A], [C 8 C 1 Pip][A]) and [C 8 Py][A]
, the central group 15 N (or P) atom showed E B (N cation 1s) (or E B (P cation 2p 3/2 )) differences due to the counteranion when charge referenced to E B (C alkyl 1s) = 285.0 eV. 36,39,41,44 At present, it is not clear if this counteranion effect on the cation core state translates to any counteranion effect on the cation valence states, i.e. E B (cation HOFO). Given the lack of strong cation participator features for many of these ILs 48 and the significant impact of the alkyl chain length on E B (cation HOFO) for these ILs, observing any counteranion effect on the cation valence states appears very challenging.
3.2.3. Quantifying the effect of counterions E B and E i : summary. Countercations affect the anion electronic structure and counteranions affect the cation electronic structure. For the ILs studied here, these solvation (i.e. counterion) effects were not due to interactions between individual cation valence states and individual anion valence states, but can best be described as arising from electrostatic, non-specific interactions affecting anionic valence states relative to cationic valence states.

HOMO identification
The HOMO identity was judged mainly using DE B (ion HOFO) = E B (cation HOFO) À E B (anion HOFO) and DE B (ion onset) = E B (cation onset) À E B (anion onset), in combination with a visual assessment of the both resonant and non-resonant valence XP spectra (Table 1). For example, for [C 8 C 1 Im][SnCl 3 ] DE B (ion onset) = 1.6 AE 0.5 eV and DE B (ion HOFO) = 1.7 AE 0.6 eV; no peak due to resonant enhancement was observed at E B (anion HOFO) = 3.1 eV (i.e. only the same non-resonant XPS contribution can be observed at all hn values, Fig. 3a), demonstrating that the peak at lowest E B for [C 8 C 1 Im][SnCl 3 ] was from the [SnCl 3 ] À anion. Therefore, for [C 8 C 1 Im][SnCl 3 ] the HOMO was from the [SnCl 3 ] À anion (Table 3). These RXPS traces were produced by subtraction of resonant XP spectrum minus non-resonant XP spectrum using the procedure outlined in ref. 48. All electron spectra were charge referenced using methods outlined in Section 2.4. For the 60 ILs studied here and in ref. 48,39 ILs had the anion as the HOMO, 7 ILs had the cation as the HOMO, and for 14 ILs the HOMO was either the cation or the anion as it was too close to judge (Table 3 and ESI, † Table S5). Unambiguously, a significant number of ILs had the cation as the HOMO.

Predictions
The consistent E B shift of all valence states when varying the counterion demonstrates that IL valence electronic structure can be predicted, as the non-specific, electrostatic-based E B shift can be applied.
For  (Table 4 for select ILs, ESI, † Table S8 for all 36 ILs). However, the effect of the counteranion on E B (cation HOFO) for [P 6,6,6,14 ] + , [C n Py] + , ammonium or [S 2,2,n ] + has not been determined. Therefore, for predictions given in Fig. 6-8, no effects of the counteranion on E B (cation HOFO) were included, i.e. E B (cation HOFO) for each cation was kept constant whatever the identity of the anion, e.g. for all imidazolium-based ILs E B (cation HOFO) = 4.8 AE 0.4 eV was used for the predictions presented in Fig. 6-8.
Predictions of E i (IL), DE B (ion HOFO) and HOMO identity for 576 ILs are presented in Fig. 6-8 respectively; 60 ILs for which experimental data exists and 516 ILs for which experimental data has not been measured. For these predictions, the IL might not be liquid at room temperature, unlike the 60 ILs studied experimentally here. Furthermore, for some cation-anion combinations, the speciation of the metal complex may be affected by the cation identity.

E i (IL) predictions
The predicted E B (cation HOFO) and E B (cation HOFO) values were used to obtain E i (cation), E i (anion) and E i (IL) (Fig. 6). As with the experimentally determined values of E i (IL), the lowest value of E i (cation) and E i (anion) for each IL represents E i (IL).

HOMO identity predictions
DE B (ion HOFO) = E B (cation HOFO) À E B (anion HOFO) = E i (cation) À E i (anion) was calculated for 576 ILs to produce Fig. 7. Positive DE B (ion HOFO) values (red in Fig. 7) represent the anion as the HOMO, whereas negative DE B (ion HOFO) values (blue in Fig. 7) represent the cation as the HOMO; the ILs represented by near white have the cation/anion as the HOMO. The decision over which category (HOMO = anion, cation/anion or cation) each IL was placed into to produce Fig. 8 was based mainly on the predicted DE B (ion HOFO) value for that IL, although the experimental data was also taken into account for e.g. the alkylsulfate-based ILs. For most ILs, the choice was easy, but for a few ILs the judgement was trickier. This area is expanded upon in the discussion section. Overall, of the 576 ILs, 431 were predicted to have an anion HOMO, 59 were predicted to have a cation/anion HOMO, and 86 were predicted to have a cation HOMO.

Electrostatic effects of counterions on E B and E i
The electronic structure of the cation was not independent from the identity of the counteranion, and vice versa; nonspecific, electrostatic interactions dominated and specific, directional ion-ion interactions were not important. In comparison, for the NaI dissolved in water, solvation effects on the water caused changes to some valence states of the water but not to other valence states of the water, i.e. the solvation effects were due to specific, directional ion-water interactions between individual water valence states and iodide anion valence states. 102 All values were recorded to two decimal places, but the values are reported here to one decimal place; hence, the subtracted values do not appear to match the original values for some ILs.

Design rules for tuning E i and HOMO identity
Given the structural diversity of 36 anions and 16 cations studied here, gathering the anions and cations into groups is very challenging. All anions with E i (anion) larger than Cl À (i.e. E i (anion) 4 8.1 eV) are defined as superhalogen anions, demonstrating the relative stability of IL anions in general with respect to ionisation. 81 4 ] À was formed from cyano ligand AOs only. Furthermore, the [SnCl 3 ] À anion also contains a formal lone pair. 79 For [SCN] À , the central atom is carbon, but for this design rule the key atom is sulfur, which contains two formal lone pairs in the most favoured Lewis structure. One outlier to this design rule was [NTf 2 ] À , which has two formal lone pairs (or one lone pair, depending on the resonance structure drawn 104 ) on the central N atom but gave a relatively large E i (anion) value. This design rule has also been observed for anions with F and O ligands. 105 A number of the anions with the smaller E i (anion) values with a formal lone pair on the key atom also contained the soft and polarisable elements S and I; I À , [I 3 ] À , [SCN] À .
A design rule for producing (closed shell) anions with relatively large E i (anion) values is the presence of fluorine, e.g. A design rule for producing relatively small E i (cation) values is either: (a) aromatic cations with readily ionised p systems (e.g. imidazolium and pyridinium) or (b) very long alkyl chains (e.g. [P 6,6,6,14 ] + ). Conversely, a design rule for producing relatively large E i (cation) values is a non-aromatic cation with

Ionisation: competition between cation and anion
For two thirds of the anions studied here and in ref. 48, the anion was comfortably the HOMO. An electron was more readily removed from the anion than the cation, i.e. easier to remove an electron from the already negatively charged anion to form a neutral species, [A] À -A , rather than remove an electron from the positively charged cation to form a dicationic ion, B10% of the ILs studied here gave the cation as the HOMO.
Using the design rules laid out in Section 5.2, a combination of a cation with p-bonding/long alkyl chains and fluorinated anion/anion with no lone pair on central atom will likely give the cation as the HOMO, e.g. [C 8

Comparisons of E i with other data sources
From UPS measurements of a microscopically thick but macroscopically thin film of [C 8 C 1 Im][BF 4 ] produced using physical vapour deposition, the work function was 4.2 eV and the first peak came at approximately 5.5 eV from visually judging the UP spectra. 107 Therefore, E i (IL) using this approach was E i (IL) = 4.2 eV + 5.5 eV = 9.7 eV. 107 4 ] from our work, strongly suggesting a problem with the charge referencing for the data in ref. 49.
Using gas phase UPS of neutral ion pairs, E th (ion pair) values were measured by two groups, Leone and co-workers 15-18 and Kuusik and co-workers. [19][20][21][22][23][24][25] Most of these E th (ion pair) values were for a combination of imidazolium cations and an imide anion, e.g. [NTf 2 ] À ; E th (ion pair) B 8.5 eV for these ILs, which matches well to our E th (IL,pred.) = 8.5 eV. Kuusik (Table 4). E i (IL) was always larger than these absorption energies (and transfer energies) from absorption spectroscopy (and resonant X-ray emission spectroscopy), demonstrating that the transitions in both absorption and resonant X-ray emission spectroscopies were to bound states; in the case of [C n C 1 Im][A] ILs the bound states were likely from the cationic ring. From N 1s resonant X-ray emission spectroscopy for [C 2 C 1 Im][NTf 2 ], N 2p-based occupied valence state to N 2p-based unoccupied valence state for the [NTf 2 ] + anion was 9.5 eV (403.5 À 394.0 eV). 114 E i (IL) = 9.4 eV for [C n C 1 Im][NTf 2 ], suggesting that the transition measured using N 1s resonant X-ray emission spectroscopy was to an unbound state.
These measurements of valence XPS for liquid phase halometallate anions and dianions serve as an excellent complement to gas phase measurements of halometallate anions and dianions; many of the anions and dianions studied here would not be stable enough to be studied in the gas phase, e.g. [FeCl 4 ] 2À . 115

Relationships of E i with other IL properties
Comparing our data to electrochemical stability, our ILs with very large E i (IL,pred), e.g.  4 ], are used as supporting electrolytes, which need to be very electrochemically stable. 116 Further comparisons to electrochemical stability data are challenging at this stage, given the tricky task of finding an experimental electrochemical dataset to test against, as the IL selection needs to be sufficiently diverse as a test set, but also a relatively inert electrode must have been used. We believe we have produced an excellent experimental dataset of IL valence electronic structures for which comparisons can be made in the future. It is a similar story with respect to comparisons of experimental electronic structure and thermal stability; is there an experimental thermal stability dataset of sufficient IL diversity to provide a high-quality test of our electronic structure data? One significant challenge is quantifying thermal stability; there are a number of different metrics, e.g. onset temperature at a certain % of mass loss, activation energy. 117,118 A pyridinium-based cation in an IL can act as an electron donor to a neutral dye solute. 119 This study demonstrates that the cation has been considered as an electron donor in ILs, but the full potential and importance has not yet been considered.

Conclusions and future work
We have successfully measured valence electronic structure descriptors for 60 ILs, most importantly, E i and the HOMO identity. Measuring E i for such a structurally diverse set of ILs represents a significant step forward in the understanding of the valence electronic structure of ILs. The structurally diverse range of cations and anions studied allow us to provide qualitative design rules linking ion structure to valence electronic structure. The electronic influence of the countercation on the anion valence electronic structure (and vice versa) was demonstrated to be dominated by non-specific, electrostatic interactions; the largest effect was 0.6 eV, although most effects were much smaller than that. Given that the cation-anion effects were relatively predictable, we were able to make predictions of both E i and the HOMO identity for a further 516 ILs.
B10% of the ILs have the cation rather than the anion as the HOMO. The cation must be considered as a possible electron donor (or partial electron donor, when donating electron density rather than a formal electron pair) in such ILs in particular, especially for neutral solutes where electrostatic ion-solute interactions are expected to be less dominant.
Adding new anions to the dataset should be relatively facile if studied on a standard lab apparatus, given most anions dominate non-resonant XP spectra recorded at hn = 1486.6 eV; suitable charge referencing is achievable for any new IL. Adding new cations to the dataset will prove far more of a challenge, given the multiple experimental difficulties, especially those caused by the normally dominant anionic contributions to nonresonant XP spectra recorded at hn = 1486.6 eV.
Given our significant experimental and predicted data of valence electronic structure descriptors, the development of models linking experimental valence electronic structure descriptors to other IL properties, e.g. electrochemical stability and thermal stability, is now possible. Furthermore, our dataset will provide a very valuable benchmark for validation of electronic structure calculations.
Both qualitative comparisons (e.g. visual) and quantitative comparisons (e.g. peak E B separation) of liquid phase and gas phase photoelectron spectra have great potential to provide insight into the effect of solvation on electronic structure. In the gas phase, a standard [C][A] IL has only one counterion, whereas in the liquid phase each ion is fully solvated. ILs that gave the cation as the HOMO, e.g. [C n C 1 Im][FAP], would be ideal candidates, given the cationic contributions to the valence electronic structure can be readily identified along with the anionic contributions, allowing any phase-related E B shifts to be observed.

Conflicts of interest
There are no conflicts to declare.