Comparison of GC/Time-of-Flight-MS with GC/Quadrupole-MS for halocarbon trace gas analysis

We present the application of time-of-flight mass spectrometry (TOF MS) for the analysis of halocarbons in the atmosphere, after cryogenic sample preconcentration and gas chromatographic separation. For the described field of application, the quadrupole mass spectrometer (QP MS) is 5 the state-of-the-art detector. This work aims at comparing two commercially available instruments, a QP MS and a TOF MS with respect to mass resolution, mass accuracy, stability of the mass axis and instrument sensitivity, detector sensitivity, measurement precision and detector linearity. Both mass 10 spectrometers are operated on the same gas chromatographic system by splitting the column effluent to both detectors. The QP MS had to be operated in optimised single ion monitoring (SIM) mode to achieve a sensitivity which could compete with the TOF MS. The TOF MS provided full mass range in15 formation in any acquired mass spectrum without losing sensitivity. Whilst the QP MS showed the performance already achieved in earlier tests, the sensitivity of the TOF MS was on average higher than that of the QP MS in the “operational” SIM mode by a factor of up to 3 reaching detection limits of 20 less than 0.2 pg. Measurement precision determined for the whole analytical system was up to 0.2% depending on substance and sampled volume. The TOF MS instrument used for this study displayed significant non-linearities of up to 10% for two third of all analysed substances. 25


Introduction
With increasing evidence that anthropogenic chlorinated and brominated hydrocarbons can be transported into the stratosphere and release chlorine and bromine atoms that can de-plete ozone in catalytic cycles (Molina and Rowland, 1974;Farman et al., 1985;Solomon, 1990), the production and use of such species were regulated under the Montreal Protocol in 1987.Most of these fully halogenated compounds are declining in the atmosphere (Montzka and Reimann, 2011).However, many partially halogenated compounds are still increasing in the atmosphere (Montzka and Reimann, 2011), as are some newly detected fully halogenated species (Laube et al., 2014).Also, many fluorocarbons which do not destroy stratospheric ozone and are thus not regulated under the protocol show increasing trends in the atmosphere (Laube et al., 2012;Ivy et al., 2012;Vollmer et al., 2011).Although these fluorocarbons do not destroy ozone, many of them are strong greenhouse gases with long atmospheric lifetimes, resulting in increased radiative forcing of the troposphere.Therefore, the need persists for continuous measurements to identify new compounds in the atmosphere and monitor and document their atmospheric trends.The mass spectrometric instrument commonly used for halocarbon analysis is the quadrupole mass spectrometer (QP MS) (Cooke et al., 2001;Aydin et al., 2002;Miller et al., 2008;Sala et al., 2014).Besides the QP MS, the use of high mass resolving and extremely sensitive sector field MS has also been reported (Lee et al., 1995;Laube et al., 2014).Time-of-flight mass spectrometry (TOF MS) has only been applied sporadically for measurements of atmospheric trace gases (Kim and Kim, 2012;Kundel et al., 2012;Watson et al., 2011;Jordan et al., 2009) and in particular not with focus on halocarbons.The main advantage of coupling a TOF MS to a gas chromatograph (GC) over using the QP MS is the intrinsic full mass range acquisition and the better mass resolution and accuracy.The identification of unknown peaks is signif-icantly facilitated by these advantages and the use of more narrow mass intervals is expected to reduce interferences and background noise.In addition, much higher data acquisition rates are possible using TOF MS, which is an advantage for fast chromatography.A TOF MS instrument can measure more than 10000 mass spectra per second.They are added up and averaged over a certain time period to yield the desired time resolution.The possibility of operating the TOF MS at high data rates is also of high interest for fast chromatography and narrow peaks, for which the operating frequency of quadrupole instruments (especially when measuring several ions) can be a limiting factor.The maximum time resolution for the TOF MS used in this study is 50 Hz.An increase in the data frequency will lead to decreased signal-to-noise levels.The data frequency must therefore be optimised to provide a sufficient number of data points per chromatographic peak while keeping the signal-to-noise level as high as possible.In contrast, a QP MS is a mass filter and will only measure one mass at a time.It needs to scan many individual masses sequentially to register a full mass spectrum.To achieve high sensitivity, QP MS are therefore often operated in single ion monitoring (SIM) mode in which the instrument is tuned to only one or a few selected ion masses and all other ions do not pass the quadrupole mass filter.Regardless of these limitations of the QP MS, it is widely used in analytical chemistry due to its stability, ease of operation, high degree of linearity, good reproducibility as well as sensitivity.Especially for atmospheric monitoring the advantage of obtaining the full mass information from the TOF instrument might allow retrospective quantifications of species which were not target at the time of the measurement.For this purpose the TOF MS must be well characterised (in particular with respect to linearity) and the calibration gas used during the measurements must contain measurable amounts of the retrospective substances and be traceable to an absolute scale.
In this paper, a comparison of a state-of-the-art QP MS and a TOF MS is presented, with both mass spectrometers being coupled to the same gas chromatographic system.The instrumental setup is described in Sect. 2. The GC QP MS system was characterised and used before for studies by Laube and Engel (2008); Brinckmann et al. (2012) and showed consistent results in the international comparison IHALACE (International Halocarbons in Air Comparison Experiment) with the NOAA (National Oceanic and Atmospheric Administration) network (Hall et al., 2013).We discuss the use of TOF MS in atmospheric trace gas measurements, in particular for the detection and quantification of halocarbons, focusing on four substances: CFC-11, CFC-12, Halon-1211 and Iodomethane.These four substances cover the boiling point and typical concentration range of a total of 35 substances analysed.The six key parameters for atmospheric trace gas measurements discussed in this paper are (1) mass resolution and (2) mass accuracy of the detectors, (3) stability of the mass axis and instrument sensitivity, (4) detector sensitivity represented by the limits of detection (LOD), (5) repro-

Preconcentration unit
Atmospheric mixing ratios (mole fractions) of halocarbons are very low, i.e. in the parts per trillion (ppt) to parts per quadrillion range (ppq).To achieve signals clearly distinguished from noise in GC MS analysis, a sample preconcentration procedure is required.In this work, the method of sample preconcentration on adsorptive material followed by thermodesorption prior to gas chromatographic separation was used.Figure 1 shows a schematic of the preconcentration unit; an explanation follows.A similar setup was described by (Sala et al., 2014).A 1/16 inch stainless steel tube (sample loop, ID = 1 mm, length = 15 cm) packed with HayeSep D (10 mg) adsorption material was cooled to a temperature of −80 • C for sample preconcentration.The sample flow during preconcentration was adjusted to 50 mL min −1 controlled by a needle valve.For cooling, a Stirling cooler was used (Global Cooling,Inc.,model M150).The sample loop was placed inside a cooled aluminium cylinder (cooling head) and was thermally and electrically isolated with two layers of glass silk and one layer of Teflon shrinking hose.The cooling head was thermally isolated towards ambient air with two layers of Aeroflex-HF material.All sample components which were not trapped on the adsorption material were collected in a 2 L stainless steel flask equipped with a pressure sensor.The pressure difference between beginning and end of the preconcentration phase was recorded to calcu-late the preconcentration volume.After the preconcentration phase, the sample loop was heated resistively to +180 • C in a few seconds for instantaneous injection of the trapped analyte fraction onto the GC column.Desorption temperature was maintained for 4 min to clean the sample loop from all remaining compounds.All tubing (stainless steel) used for sample transfer between sample flask and preconcentration unit as well as preconcentration unit and GC was heated to 80 • C to avoid loss of analytes to the tubing wall.

Gas chromatograph
An Agilent Technologies 7890A GC with a Gas Pro PLOT column (0.32 mm inner diameter) was used for separation of analytes according to their boiling points.The column had a total length of 30 m, divided inside the GC oven into 7.5 m pre-column (backwards flushable) and 22.5 m main column.Purified helium 5.0 (Alphagaz 1, Air Liquide, Inc.) was used as carrier gas.The GC was operated with constant carrier gas pressure on both pre-and main column.5) The final temperature of 200 • C was kept for 2.95 min.The resulting runtime was 17.95 min.The pre-column was flushed backwards with carrier gas after 12.6 min to avoid contamination with highboiling substances.The gas chromatographic column was connected to the QP MS and the TOF MS using a Valco three-port union and two fused silica transfer lines.The transfer line to the QP MS had a total length of 0.70 m with an inner diameter of 0.1 mm, and the transfer line to the TOF MS had a total length of 2.10 m with an inner diameter of 0.15 mm.Based on the length, temperatures and inner diameters of the transfer lines, a split ratio of 63 : 37 (TOF MS : QP MS) was calculated.Using the ratios of the peak areas of the quadrupole when receiving the entire sample (TOF transfer line plugged) to those obtained in the split mode, a spilt ratio of 66 : 34 was calculated.We have adapted this latter value as it is based on actual measurements rather than calculations.All parts of the transfer lines outside the GC oven were heated to 200 • C.

Mass spectrometer
The two mass spectrometers in comparison were (1) an Agilent Technologies 5975C QP MS and (2) a Markes International (former ALMSCO) Bench TOF-dx E-24 MS.Both MS were operated in electron ionisation mode with an ionisation energy of 70 eV and ioniser temperatures of 230 • C. The QP MS was operated in SIM and SCAN mode (see Table 2 for more information).As the GC was operated in constant pressure mode, i.e. the head pressure of the columns were kept constant, the carrier gas flow into the two MS therefore varied according to the temperature ramp during each gas chromatographic run.Pressures inside the ion flight tubes of the MS therefore also varied; the TOF MS had a pressure range from 1.8 × 10 −6 to 1.6 × 10 −6 hPa and the QP MS had a pressure range from 2.1 × 10 −5 to 1.8 × 10 −5 hPa.The Bench TOF-dx uses a direct ion extraction technique with an acceleration voltage of 5 kV.In contrast to many other TOF instruments, the ions are accelerated directly from the ion source into the drift tube instead of extracting them from the ion source and then accelerating them orthogonally to the extraction direction (orthogonal extraction).The direct extraction method in combination with the high acceleration energy orients the instrument towards a high sensitivity, especially for heavier ions (five technologies GmbH, G. Horner and P. Schanen, personal communication, 2014).The TOF MS was set up to detect mass ranges from 45 to 500 m / z; higher and lower m / z were discarded.The reason to discard ions with m / z ratio below 45 was to eliminate a large part of the CO 2 which is trapped by our preconcentration method and can lead to saturation of the detector.A schematic of the Bench TOF-dx is given in Fig. 2. The spectra extraction rate was adjusted to 4 Hz to get a data acquisition rate comparable to that of the QP MS.

Experimental
All characterisation experiments were conducted using a high-pressure air sample (50 L Aluminium flask, 70 bar) filled in 2007 at Jungfraujoch, Switzerland.Prior to preconcentration, the air sample was dried using a heated (70 • C) Mg(ClO 4 ) 2 water trap.Halocarbon mixing ratios were assigned to this reference gas by calibration against an AGAGE (Advanced Global Atmospheric Gas Experiment) gas standard (H-218).Table 1 shows reference gas mixing ratios of specific substances discussed in this paper.

Measurement procedure and data evaluation
To ensure measurement quality, both MS were tuned in regular intervals (autotune by operating software) at least every 2 months but especially before sample measurements and/or characterisation experiments.Autotune options of both mass spectrometers were used without further manual adjustments.To increase the sensitivity and linearity of the TOF MS, its detector voltage was increased by 30 V, as described in Sect.4.6.Additionally, a zero measurement (evacuated sample loop), a blank measurement (preconcentration of purified Helium 5.0) and two calibration gas measurements were conducted to condition the system before every measurement series.At the end of every measurement series, another blank measurement was added.Every measurement series itself consisted of a calibration measurement followed by two sample measurements (same sample).This sequence of three measurements was repeated n times depending on the type of experiment and then terminated by a calibration measurement.For characterisation experiments both calibration and sample measurements were taken from the same gas cylinder (reference gas, see description above) but treated differently in data evaluation, e.g. as a calibration or sample measurement.Chromatographic peaks were integrated with a custom designed software written in the programming language IDL.The peak integration is based not on a standard baseline integration method commonly used in chromatographic applications but on a peak fitting algorithm.
For the results shown here Gaussian fits were used for peak integration.This software was also used for data processing by Sala et al. (2014) and described there.Noise calculation was performed on baseline sections of the ion mass traces of interest.The noise level was determined as the 3-fold standard deviation of the residuals between data points and a second degree polynomial fit through these data points.This approach accounts for a drifting non-linear baseline.Otherwise, a non-linear baseline would cause an overestimation of the noise level.The integrated detector signal was divided by the preconcentration volume to get the detector response per sample volume.To account for detector drift during measurement series, the calibration measurements bracketing the sample pairs were interpolated linearly.Thereby, interpolated calibration points are generated for each sample measure-ment.The response for each sample was then derived by calculating the quotient between sample and corresponding interpolated calibration point.Experiments were conducted to analyse six key parameters (Sect.3.2 to 3.7) important for measurements of halogenated trace gases in the atmosphere: mass resolution, mass accuracy, limits of detection, stability of the mass axis and instrument sensitivity, measurement precision and reproducibility as well as detector linearity.

Mass resolution
The mass resolution (R) is defined as follows: with m being the full width at half maximum (FWHM) of the exact mass m of the ion signal.
The mass resolution determines whether two neighbouring mass peaks can be separated from each other.It is considered an instrument property, i.e. influenced only by internal factors like instrument geometry, ion optics, etc.The mass resolution of the TOF MS was calculated with its operating software ProtoTOF in a mass calibration tune.The QP MS was operated with MS Chemstation (Agilent Technologies, Inc.) which only processes unit mass resolution, independent of mass range.

Mass accuracy
The mass accuracy (δa) defined as and quantifies the deviation between a measured ion mass m m and the according expected exact mass m of each fragment.Like mass resolution, it is considered an instrument property.In this work, so called 1 amu centroid mass spectra are used to calculate mass accuracy.The exact mass is thus taken as the maximum intensity of the mass spectrum within a certain window (±0.5 u) around the nominal mass.Mass accuracy was calculated for four different ion masses of four different substances: HFC-134a (CF + 3 , 68.995 u), CFC-12 (CF 35  2 Cl + , 84.866 u), CFC-11 (CF 35 Cl + 2 , 100.936 u,) and methyl iodide (CH 3 I + , 141.928 u), which cover most of the mass range of the substance peaks in our chromatogram.Individual values for the mass accuracy were taken at the maximum of each chromatographic peak.Data from reproducibility experiments (see Sect. 3.6) as well as regular sample measurements were analysed to gain information about mass accuracy for the four exemplary ion masses.Only measurements taken under well-equilibrated conditions were used for this analysis.As the first two measurements of a measurement day often show enhanced variability they were excluded from the analysis of the mass accuracy.

Stability of the mass axis and instrument sensitivity
To evaluate the stability of the two mass spectrometers with respect to sensitivity and accuracy of the mass axis, a reproducibility experiment was used.The relative difference between the minimum and maximum detector response of the day and the 1σ standard deviation of all measurements over this day were taken as measures of the drift.For drift in mass accuracy over the day, the mean value and the 1σ standard deviation are given for the main masses for the following four compounds: HFC-134a (CF + 3 , 68.995 u), CFC-12 (CF 35 2 Cl + , 84.866 u), CFC-11 (CF 35 Cl + 2 , 100.936 u,) and Iodomethane (CH 3 I + , 141.928 u).To evaluate the stability of the mass accuracy over a longer time period, the mass accuracy was calculated on measurement days with different time differences since the last mass calibration tune.

Limits of detection
The lowest amount of a substance that can reliably be proven is considered to be its LOD and serves as a measure for the sensitivity of the analytical system.Based on the assumption that a molecule fragment (f ) can be detected when its detec- tor signal height (H f i ) is equal to or higher than 3 times the signal noise (N f i ) on the adjacent baseline (signal-to-noise level (S / N) > 3), a limit of detection for a fragment (f i ) from an analyte substance (S i ) with a mass (m S i ) in the injected sample can be calculated as For comparison with the QP MS, the LOD of both instruments were calculated from calibration gas measurements by linear downscaling.Possible detector non-linearities were omitted in this case.The LOD error was considered to be the standard deviation of 10 calculated limits of detection.Different settings of the QP MS (SCAN mode (1), optimised (opti.)SIM mode (2) and operational (oper.)SIM mode (3)) were applied.In the SCAN mode (1), the quadrupole MS scanned from 50 to 500 u (comparable to the mass range of the TOF MS) with a dwell time of ≈ 3.7 ms ion −1 and a scan rate of 1.66 scans per second.In the optimised SIM mode (2), the quadrupole MS measured only one ion with a dwell time of 310 ms with ≈ 3 scans per second.In the operational SIM mode (3) the quadrupole MS measured several masses (up to six) in one scan with individual dwell times given in Table 2 and ≈ 3 scans per second.
The LOD in pg and ppq were calculated for 0.28 L sample volume with respect to the split ratio (see Sect. 2.2) and then extrapolated to 1 L of ambient air.

Reproducibility and measurement precision
The measurement precision describes the repeatability of a measurement.We determine the precision from the reproducibility (i.e. the standard deviation) of the measurements.The mean reproducibility is derived from dedicated multiple experiments designed to assess measurement precision (reproducibility experiment).Reproducibility was analysed over five measurement series, conducted on 5 different days, to give the mean measurement precision.Every experiment followed the procedure described in Sect.3.1, with a total of 19 evaluated measurements of the same ambient air sample.A subset of the samples was treated as standard, the other part as unknown samples (two samples bracketed by two standards).Every individual measurement of these five series was conducted with a preconcentration volume of 0.28 L of the reference gas.Two additional reproducibility experiments were conducted with a higher preconcentration volume of 1 L to assess the possible dependence of the reproducibility on the preconcentrated sample volume.For each sample pair, the standard deviation of the relative response was calculated, summed up over all pairs and divided by the number of pairs to form the sample pair measurement reproducibility of that measurement series.The described procedure was applied to all analysed substances and reproducibility experiments.The mean value of measurement reproducibilities is considered to be the measurement precision of the system for the respective substance and volume.

Detector linearity
Detector linearity was analysed in two linearity experiments by varying the default preconcentration volume of 0.28 L by factors of 0.33, 0.66, 1.25 and 2 (sample positions in the measurement sequence, see Sect.3.1).As calibration measurements, the default preconcentration volume was used.For comparison, detector responses were calculated as the ratio of the area of a chromatographic peak (A) to the preconcentration volume (V ).All detector responses were normalised to 1 (relative detector response) by dividing them by the mean A/V of the calibration measurements.An ideally linear detector would show a relative response of 1 for any preconcentration volume used.The errors for the linearity measurements were derived as the 3-fold standard deviation given from reproducibility experiments.

Mass resolution
If mass resolution is sufficiently high, it is possible to separate mass peaks of equal unit mass but differing exact mass.This separation drastically enhances the possibility to identify specific molecule fragments and to reduce crosssensitivity.For halocarbon analysis, it is interesting to separate halogenated molecule fragments with exact masses typically below unit mass from other fragments with exact masses typically at or slightly above unit mass (e.g.hydrocarbon fragments).It could then be possible to reduce background noise generated by interfering ion signals or even compensate co-elution of non-target species from the GC column.For quantitative analysis the separation of adjacent mass signals implicates a possible loss of signal area when both mass peaks are not fully separated.The imposed error, i.e. the peak area lost due to separation, should not decrease measurement precision and should therefore be lower than the targeted measurement precision, in our case 0.1 %.
For this purpose, the definition of a qualitative and a quantitative separating resolution R Sep is introduced (see Fig. 3 for an illustration).Assuming a Gaussian peak shape (normal distribution) of the ion signal on the mass axis, a separation of two neighbouring signals m 1 and m 2 (with m 2 > m 1 ) by 8σ (SD, 4σ per peak) is considered a quantitative separation (less than 0.01 % loss of peak area) while a separation by less than 8σ is considered to be only a qualitative separation.Further assuming that 1σ is approximately 1/2 FWHM (or 1/2 m respectively) and that m 1 is not significantly different from m 2 , one can estimate R Sep (at m 1 or m 2 ) for a known (m 2 − m 1 ) difference: For a value of n σ = 8, Eq.( 4) gives the quantitative separating resolution, while for a value of n σ = 2 it gives a qualitative separating resolution.Table 3 shows some examples for qualitative and quantitative separating resolutions required for separation of halogenated mass fragments from hydrocarbon molecule fragments with slightly different masses.
To separate e.g. the CClF + 2 ion signal from the C 6 H + 13 ion signal qualitatively, a resolution of 600 is necessary.For a quantitative separation, the mass resolution has to be R = 3700 according to the definition of 8σ separation (see above).For the Bench TOF-dx, the calculated mass resolution was R = 1000 at mass 218.985 u for the fragment C 4 F + 9 in a mass calibration tune by the software Pro-toTOF.This allows a qualitative separation of two neighbouring mass peaks like the ones listed in Table 3, e.g. the separation of mass 84.966 u to mass 85.102 u.An example of a mass spectrum centred around 85 u is shown in Fig. 4 for a chromatogram of a typical ambient air sample at a retention time of 11.35 minutes.Two mass peaks, one centred at 84.943 u (CH 35 Cl 37 Cl + ), a fragment of the Trichloromethane (CHCl 3 ) molecule and one with a mass slightly above unit mass, can be clearly distinguished.The higher mass is the result of an unidentified hydrocarbon peak eluting shortly before the Trichloromethane peak.
The resulting chromatogram centred at 11.3 minutes is shown in Fig. 5. Three different mass ranges were extracted from the raw data, the nominal mass range from 84.5 u to 85.5 u, the lower mass range from 84.7 u to 85.0 u and the higher mass range from 85.0 u to 85.3 u.When extracting the information centred around the unit mass range, a double peak is observed.An extraction of the lower mass range of the 85 u signal yields a much lower signal in the earlier eluting peak yet the signal cannot be reduced to baseline level.An extraction of the higher mass range of the signal gives a larger signal for the earlier eluting peak, but again the signal does not drop to baseline level.
This shows that the mass resolution of the Bench TOF-dx is sufficient to qualitatively show that two different fragments are present but that the resolution does not allow the separatation of these fragments in a way sufficient for quantifications.For a quantitative separation as defined above, the mass resolution of the Bench TOF-dx is not sufficient without further data processing steps like a peak deconvolution.

Mass accuracy
While sufficient mass resolution is necessary for an unambiguous separation of two mass peaks, mass accuracy is in addition needed for chemical identification of the detected ion.The better the mass accuracy, the lower the number of possible fragments that might be the source of the mass signal.The mass accuracy for the Bench TOF-dx was found to be in a range of 50 to 170 ppm for a mass range from 69 u to 142 u.Mass accuracies for the analysed target masses were determined as follows: (100±60) ppm for mass 68.995 u, (80±50) ppm for 84.966 u, (120±50) ppm for 100.936u and (130±40) ppm for 141.928 u.A correlation between the displayed masses is observed: when the accuracy of one mass is decreased, the others are, too.There is no correlation given by the proximity of target masses to tuning compound (PFTBA, e.g.68.995 u) masses.A suspected reason for the instability of the mass axis is the instrument temperature and resulting changes in material elongation.This is, however, speculation.At a mass resolution of R = 1000 at ion mass 85 u and an accuracy of 100 ppm, the mass difference between measured and exact mass would be 10 % of the FWHM of this mass peak (or 5 % at 50 ppm).The stability and absolute accuracy in the determination of the exact mass is thus not a significant additional limitation in the ability of the Bench TOF-dx to separate different ions (see Sect. 4.1).

Stability of the mass axis and instrument sensitivity
A reproducibility experiment was used to evaluate the stability of two detectors over a measurement series (typically 10 h).For that purpose, the minimum and maximum value of the detector response relative to all recorded responses and the 1-fold relative standard deviation of all recorded responses were used (see Table 4).
For the substances CFC-11 and CFC-12 the drift of the sensitivity of the TOF MS and QP MS are on the same level.For the low concentrated substances, the drift of the TOF MS is higher than that of the QP MS.
For evaluating the stability of the mass axis, the drift over a day was calculated as mean accuracy and standard deviation (1σ ).The stability over a long time period was observed over different days away from a mass accuracy tune.As shown in Sect.4.2 the mass accuracy of the Bench TOF-dx was observed to be on the order of 50-170 ppm.Within this uncertainty no drift of the mass axis with time could be observed for periods of up to 19 days after the mass axis calibration.The stability and absolute accuracy in the determination of the exact mass is thus not a significant additional limitation in the ability of the Bench TOF-dx to separate different ions (see Sect. 4.1).

Limits of detection
For halocarbon measurement, sensitivity is an important issue as atmospheric concentrations can be below 1 pg L −1 of ambient air, especially for newly released anthropogenic species.Table 5 shows the calculated LOD for the QP and the TOF MS for the four selected species with different measurement settings of the quadrupole MS detector.
For the QP MS, the signal-to-noise level of a certain m / z depends on the concentration and dwell time.The dwell time represents the time interval in which the quadrupole mass filter is tuned to the specific mass-to-charge ratio m / z before switching to the next mass setting.Lower dwell times will decrease sensitivity but allow for more different mass filter settings per scan, resulting in more different m / z monitored per time.Higher dwell times increase the detector sensitivity towards specified m / z ratios but reduce the number of m / z monitored per time.For this work, data based on three different instrument settings were used for LOD calculation (see Table 2).The SCAN mode of the QP MS was chosen for a direct comparison with the TOF MS (scan range from 45 u to 500 u) and is shown in Table 2 (1).Higher and lower m / z ratios were discarded.Reducing the scan range will result in better detection limits for the QP MS and theoretically also for the TOF MS as long as no significant amounts of ions heavier than the chosen upper scan limit are produced in the ion source.Remaining ions in the TOF MS flight tube from a preceding extraction would result in unambiguous detector signals.The optimised SIM mode monitors only one m / z of the respective substance, Table 2 (2).In measurements of ambient air, several m / z are usually monitored simultaneously (operational SIM mode (3)).The dwell times are optimised for the different substances.For substances with high concentration shorter dwell times are chosen, while the dwell time is increased for substances with low concentrations in order to increase the sensitivity.Only one ion is measured for most species in order to reach optimum sensitivity.As a consequence, limits of detection are higher in such measurements as in the optimised SIM mode.Respective LOD for the discussed dwell time settings are shown in Table 5.
In comparison to the QP MS, the TOF MS is up to 12 times more sensitive than the QP MS in the SCAN mode.In the optimised SIM mode with increased dwell times (2) for specific ion masses, limits of detection in quadrupole MS and time-of-flight MS are similar.During routine measurements (operational SIM mode (3)), the limits of detection of the TOF MS were up to a factor of 3 lower than those of the QP MS.

Reproducibility
A high measurement precision is required as it is of great importance to detect very small variability of halocarbons in the atmosphere, e.g. to characterise trends of highly persistent substances (Montzka and Reimann, 2011;Montzka et al., 2009;Vollmer et al., 2006).Table 6 shows exemplary reproducibilities for both instruments based on a preconcentration volume of 0.28 L. The reproducibility is rather similar for both MS, with values below 1 % for the species with high ambient air concentrations and therefore high signal-to-noise levels .For the species with lower concentration and lower signal-to-noise levels the reproducibility of the TOF seems to be slightly but not significantly better (see Table 6).
The reproducibilities shown in Table 6 are based on measurements with a relatively small sample volume.Larger preconcentration volumes should result in better reproducibilities as signal-to-noise levels are increased and error sources during sample preparation should become smaller relative to the sample volume.Therefore, two reproducibility experiments with a larger preconcentration volume of 1 L were performed.The results are shown in Table 7.
The increase of the preconcentration volume to 1 L yields a significant improvement of the measurement precision.The high signal-to-noise species CFC-12 and CFC-11 now show reproducibilities below 0.3 % for the QP and for the TOF.For the low signal-to-noise species Halon-1211 and CH 3 I the reproducibilities are improved by a factor of up to 4 for the TOF MS and by a factor of up to 3 for the QP MS, with the TOF instrument showing better reproducibilities.As for the TOF MS, the detector itself was found to be a limitation to higher preconcentration volumes as it showed saturation effects for some analysed ions already at 0.5 L preconcentrated sample.For example, CFC-12 had to be evaluated on mass 87 u (relative abundance: 32.6 %) and CFC-11 on mass 103 u (relative abundance: 65.7 %) (NIST, 2014) as both main quantifier ion masses (85 and 101 u) showed satu-ration in the respective retention time windows.This saturation reflects the limited dynamic range of the analog to digital converter (memory of 8 bits) used in the Bench TOF-dx.

Linearity
For the calculation of the mixing ratio of a measured substance, its detector signal has to be correlated with the signal of the same substance in a calibration measurement with known mixing ratio.If the detector behaves linearly, this correlation is linear and the calculation of the mixing ratio is straight forward.As mixing ratios in different air samples might vary to a great extent (e.g.diurnal variations of shortlived substances) (Sala et al., 2014;Derwent et al., 2012;Law and Sturges, 2011), a linear detector simplifies data evaluation to a great extent.Furthermore, retrospective analysis of substances that were not identified at the time of measurement is possible without an unknown error due to detector non-linearity.Figures 6 and 7 show linearity plots for the QP MS for the CFC-11 and CFC-12 based on two linearity experiments.The QP MS showed a linear behaviour within the measurement errors (3-fold measurement reproducibility for the respective substance).This linearity test includes possible effects of the preconcentration unit (quantitative adsorption and desorption) as well as the determination of the preconcentration volume, the GC and data processing (signal integration).Figures 6 and 7 illustrate results from the two linearity experiments for the TOF MS.For CFC-11 (Fig. 6) a deviation from linearity for small preconcentration volumes of nearly 10 % is observed, while detector behaviour is close to the ideal value for high preconcentration volumes.The red curve was derived based on the standard detector voltage of −2244.8V.An decrease of the detector voltage by −30 V brought slight improvements but did not solve the issue.Figure 7 shows a linearity plot for the substance CFC-12.For CFC-12 the detector is considered to be linear within the error bars.Both detectors compared in this work depend on the same sample preparation and separation steps before detection.As measurement reproducibilities of QP MS and TOF MS were not significantly different, the direct comparison is possible without limitations.The examples displayed for the QP MS and the TOF MS are two of 35 substances measured and analysed.The QP MS showed linear behaviour for all substances within the uncertainty range.The non-linearity of the TOF-MS was highest for the low preconcentration volume (33 %, 0.09 L) with deviations of −10 to +20 % compared to a standard preconcentration volume of 100 % (0.28 L).For most substances the instrument showed a similar behaviour as observed for CFC-11 (decreased sensitivity for low amounts of analyte) while some species showed the opposite behaviour (increased sensitivity with decreasing amount of analyte).Reasons for this conflicting behaviour are still subject to further investigations.Proportionality of detector signal against the amount of analyte in the sample over the given concentration range was thus   found for the QP MS but only for some species in the TOF MS.If the detector does not behave linearly, the relationship between the integrated peak area and the atmospheric concentration has to be approximated by a fit function.In order to generate this fit function, additional measurements with varying preconcentration volumes are necessary before each measurement series.This procedure was found to be necessary for the TOF MS.It lengthens measurement series, implies an additional error source and requires additional time for data processing.

Conclusions
A Markes International Bench TOF-dx was compared to an Agilent Technologies 5975 QP MS with respect to the measurement of halogenated trace gases in the atmosphere.Both detectors ran in parallel (66:34 split) after cryogenic preconcentration and gas chromatographic separation of the air sample.The comparison included the mass resolution, mass accuracy, the limit of detection, the measurement precision (reproducibility) and the detector linearity.The TOF MS showed a resolution of 1000 and a m of 0.071 at mass 219.995 u with a mass accuracy of 50 to 170 ppm.Therefore it is able to qualitatively separate ion signals at different exact mass but equal unit mass (for example the mass 84.966 u from the mass 85.106 u by a m of 0.136).This qualitative mass separation of the TOF MS could be sufficient for improved substance identification and is an advantage over the QP MS.The QP MS does not allow for separation of exact masses as the mass resolution of QP MS instruments is generally too low (R ≈ 200) for that purpose.The analysis of detection limits showed that the TOF MS is generally more sensitive than the QP MS (despite using selected ion monitoring mode).The LOD of the QP in the SCAN mode are up to a factor of 12 higher than the LOD of the TOF MS.LOD of the TOF MS are lower by factors of up to 3 (Table 5) in comparison to the QP MS with operational SIM mode settings used for routine measurements.In the SIM mode with only one quantifier (optimised SIM mode) the TOF MS is similar to the QP MS.In that respect, the TOF MS with its very high sensitivity and full mass range information provides a considerable advantage compared to a QP MS.The reproducibility of both instruments was found to be on an equal level with slightly better reproducibilities of the QP MS at high signal-to-noise levels and slightly better reproducibilities of the TOF MS for low-concentrated species.Regarding detector linearity, the Bench TOF-dx in its current configuration could not compete with the QP MS.A high degree of linearity is, however, necessary for high accuracy measurements in trace gas analysis.The encountered non-linearities necessitate a correction which adds an error source, especially when there is a large concentration difference between sample and calibration measurement.It furthermore complicates measurements as well as data evaluation.For other applications where concentration variability is significantly higher than the non-linearity of the detector, the observed detector non-linearities might not be of such high relevance.In conclusion, the TOF MS does show advantages with respect to mass resolution and sensitivity without losing the full mass spectra information.Persisting non-linearities are a big disadvantage but might be conquered in the future by developments in detector electronics.With reduced non-linearities, TOF MS could well be the technology of the future for the analysis of halogenated trace gases in the atmosphere, despite the significantly higher costs of the TOF MS in comparison to QP MS instruments.These conclusions are only valid for the Markes International Bench TOF-dx E-24 MS and atmospheric trace gas measurements and might turn out differently for another field of research or another TOF MS.

Figure 1 .
Figure 1.Schematic of the cooling head.The aluminium cylinder which contains the sample loop is placed on top of the Stirling cooler's cold end.Electric connectors are located at each end of the sample loop for resistive heating.
The temperature program of the GC consisted of five phases.(1) For the first 2 min, the temperature was kept at 50 • C. (2) Then the oven was heated at a rate of 15 • C min −1 up to 95 • C, (3) from thereon at 10 • C min −1 up to 135 • C and (4) then at a rate of 22 • C min −1 up to 200 • C. (

Figure 2 .
Figure 2. Scheme for the direct ion extraction of the Bench TOF-dx direct extraction (five technologies GmbH, G. Horner and P. Schanen, personal communication, 2014).The red dotted line represents a typical ion path.

Figure 3 .
Figure 3. Schematic display of two different mass resolutions (blue and black curves).Two signals on masses 84.966 and 85.102 u with equal intensities demonstrate the mass separation with R = 600 (blue curve) and R = 3700 (black curve).Assuming Gaussian peak shapes for the signals, R = 3700 separates both peak by 8σ (quantitative separation) and R = 600 separates them by only 2σ (qualitative separation).

Figure 4 .
Figure 4. So-called 0.01 u mass spectrum of the substance Trichloromethane.Two mass peaks are shown.The higher one by mass 84.9 u is identified as the molecule fragment (CH 35 Cl 37 Cl + ) and the other one by mass 85.1 u is an unidentified hydrocarbon peak.

Figure 5 .
Figure 5.A chromatogram of an unidentified hydrocarbon peak (smaller one) eluting slightly earlier than the higher Trichloromethane peak.The nominal mass 85 u (black) shows a double peak.By choosing the lower mass range (84.7 u to 85.0 u; red) a lower signal for the unidentified hydrocarbon peak is observed, and by choosing the higher mass range (85.0 u to 85.3 u, blue) a lower signal for the Trichloromethane peak is observed.

Figure 6 .
Figure 6.Linearity graphs of CFC-11 (CFCl + 2 fragment) based on two different linearity experiments (red and black plots in each graph).Primary x axis (lower): mass on column in ng.Secondary x axis (upper): preconcentration volume variation in % versus a default preconcentration volume of 0.28 L (dashed line).Y axis: deviation from the normalised relative detector response versus the detector response of the default preconcentration volume).For every preconcentration volume, the relative response should be one in case of a linear detector behaviour (dashed line).The error bars show the three-fold measurement precision, on the left-hand side for the QP MS and on the right-hand side for the TOF MS.The second linearity experiment (black) of the TOF MS was conducted with an decreased detector voltage (−2274.8V instead of −2244.8V).

Table 1 .
Mixing ratios in ppt in the reference gas used in this work for the discussed substances.

Table 2 .
Dwell time settings for given substance fragments in QP MS modes with a data frequency of ≈ 3 Hz.SCAN mode (1): QP scanned from 50 to 500 u with 1.66 scans per second and a dwell time of 3.7 ms.Optimised (opti.)SIM mode (2): settings used for measurements on which LOD calculation was based, with 310 ms dwell time per ion and a scan rate of 3 scans per second.Operational SIM mode (3): default settings, used for reproducibility and linearity experiments with 3 scans per second.

Table 3 .
Three exemplary halocarbon/hydrocarbon fragment pairs with equal unit mass but differing exact mass.The qualitative separating resolution (qual.R sep ) with n σ = 2 and the quantitative separating resolution (quan.R sep ) with n σ = 8.

Table 4 .
The difference of the minimal (Min) and maximal (Max) values in % in one reproducibility experiment for the relative response is shown with a 1σ relative standard deviation (RSD) over all measurements (20) on this day.In the comment line the trend of the calibration gas over the day is given.

Table 5 .
The limit of detection (LOD) in ppq and pg of the substances CFC-12, CFC-11, Halon-1211 and Iodomethane in 1 L of air sample per detector.The dwell times and settings for the QP MS are given in Table2.The given errors are 1σ standard deviation.

Table 6 .
The reproducibility (REP) for the QP MS and the TOF MS as a mean value of five measurement series with 20 measurements each and a preconcentration volume of 0.28 L. The given errors are 1σ standard deviation over five reproducibility experiments.

Table 7 .
The reproducibility (REP) for the QP MS and the TOF MS as a mean value of two measurement series with 20 measurements each and a preconcentration volume of 1.00 L. The given errors are 1σ standard deviation over two reproducibility experiments.