The constancy of galactic cosmic rays as recorded by cosmogenic nuclides in iron meteorites

We measured the He, Ne, and Ar isotopic concentrations and the 10Be, 26Al, 36Cl, and 41Ca concentrations in 56 iron meteorites of groups IIIAB, IIAB, IVA, IC, IIA, IIB, and one ungrouped. From 41Ca and 36Cl data, we calculated terrestrial ages indistinguishable from zero for six samples, indicating recent falls, up to 562 ± 86 ka. Three of the studied meteorites are falls. The data for the other 47 irons confirm that terrestrial ages for iron meteorites can be as long as a few hundred thousand years even in relatively humid conditions. The 36Cl‐36Ar cosmic ray exposure (CRE) ages range from 4.3 ± 0.4 Ma to 652 ± 99 Ma. By including literature data, we established a consistent and reliable CRE age database for 67 iron meteorites. The high quality of the CRE ages enables us to study structures in the CRE age histogram more reliably. At first sight, the CRE age histogram shows peaks at about 400 and 630 Ma. After correction for pairing, the updated CRE age histogram comprises 41 individual samples and shows no indications of temporal periodicity, especially not if one considers each iron meteorite group separately. Our study contradicts the hypothesis of periodic GCR intensity variations (Shaviv 2002, 2003), confirming other studies indicating that there are no periodic structures in the CRE age histogram (e.g., Rahmstorf et al. 2004; Jahnke 2005). The data contradict the hypothesis that periodic GCR intensity variations might have triggered periodic Earth climate changes. The 36Cl‐36Ar CRE ages are on average 40% lower than the 41K‐K CRE ages (e.g., Voshage 1967). This offset can either be due to an offset in the 41K‐K dating system or due to a significantly lower GCR intensity in the time interval 195–656 Ma compared to the recent past. A 40% lower GCR intensity, however, would have increased the Earth temperature by up to 2 °C, which seems unrealistic and leaves an ill‐defined 41K‐K CRE age system the most likely explanation. Finally, we present new 26Al/21Ne and 10Be/21Ne production rate ratios of 0.32 ± 0.01 and 0.44 ± 0.03, respectively.


INTRODUCTION
An important question in climate change studies is the effect of cosmic rays. Svensmark and Friis-Christensen (1997) and Svensmark (1998) claimed that the Earth's cloud cover is correlated with the cosmic ray flux. In these pioneering studies, the authors concluded that about 3-4% of the global cloud cover is correlated with the intensity of cosmic rays, which is itself inversely correlated with solar activity. Their study is based on cloud data from only one solar cycle and they considered only oceanic cloud cover data. Kristj ansson et al. (2002) added data for a second solar cycle, performed a reevaluation, and found no significant correlation. Later, Laut (2003) argued that the correlations between cloud coverage and the galactic cosmic ray (GCR) flux were obtained erroneously. Kirkby (2007) argued that, although the fundamental physical and chemical processes are not understood, there is a close relationship between GCR variations and climate changes, influencing to some extent cloud coverage, tropical rainfall, and the location of the Intertropical Convergence Zone. Erlykin et al. (2010) argued that there is a negative correlation between GCR intensity and cloud coverage at low altitudes and a positive correlation of solar irradiance with cloud coverage at middle altitudes.
By assuming that there is a causal link between GCR intensity and global warming, Erlykin et al. (2009) determined that the long-term variation is expected to be less than 0.07°C since 1956, which would be less than 14% of the observed global warming. Other studies have confirmed these results (e.g., Sloan andWolfendale 2008, 2013a;Sloan 2013). Ormes (2017) on the other hand, concluded that there is no identifiable correlation between cloud coverage and GCR intensity.
Although there is no consensus on a correlation, some studies have continued to assume a relationship between GCR intensity and cloud cover. Frigo et al. (2013) argued for a possible influence of GCRs on climate on a regional scale and Artamonova and Veretenenko (2011) concluded that GCRs may influence the process of cyclone and anticyclone formation. Shaviv (2002Shaviv ( , 2003 proposed that the occurrence of ice-age epochs on the Earth was influenced by the solar system crossing through the spiral-arms of the Milky Way. In the diffusional model of Shaviv, variations in GCR fluence between the regions within and outside of the spiral arms may be up to a factor of three. The GCR flux is higher in galactic spiral arms than in interarm regions due to higher supernova rates and/or closer proximity to supernovae explosions. He proposed, by assuming four galactic spiral-arms for the Milky Way, a spiral crossing period for the solar system of~140 Ma. Shaviv also argued that this periodicity is seen in the GCR exposure age histogram for iron meteorites. Although some studies were supportive (e.g., Wallmann 2004;Gies and Helsel 2005), many subsequent investigations have questioned the original hypothesis. First, Sloan and Wolfendale (2013b) calculated GCR intensity variations between spiral-arm and interarm regions to be in the range 10-20% and not more than 30%, that is, far less than the factor of three assumed by Shaviv (2002Shaviv ( , 2003. In addition, the calculated periodicity depends on the number of galactic spiral-arms. While there are indeed some arguments that the Milky Way has four spiral-arms (e.g., Vall ee 2017), there are also some arguments for a two-armed spiral in the Milky Way (Drimmel 2000). If true, the periodicity of the GCR intensity variations would be closer to 280 Ma and would no longer be in accord with periodic Earth climate changes. Using new data on the structure of the Milky Way, Overholt et al. (2009) argued for a nonsymmetric galaxy and concluded that the timing of the crossing of our solar system through galactic spiral-arms is most likely irregular. A second and motivating rationale for the present study are questions about the interpretation of the meteorite data by Shaviv (2002Shaviv ( , 2003. For example, Rahmstorf et al. (2004) argued that Shaviv misinterpreted the iron meteorite data (see also Wieler et al. 2013). Going one step further, Alexeev (2016) used the same data set as Shaviv (2002Shaviv ( , 2003 but reinterpreted the data by fully considering uncertainties for the cosmic ray exposure (CRE) ages and using the now accepted chemical grouping for iron meteorites. Doing so, he found no indications for a periodicity of 140 Ma in the CRE age histogram but he found slight indications of a periodicity in the range 400-500 Ma. Jahnke (2005), using the same data set, found no indications for any periodic signal in the CRE age histogram for iron meteorites.
This discussion illustrates the difficulties of studying periodic structures in the CRE age histogram of iron meteorites using existing data, which mainly consist of the 41 K-K CRE ages from H. Voshage (e.g., Voshage andFeldmann 1978, 1979). The main reason lies in the sometimes large uncertainties of up to 100 Ma for individual ages. The large uncertainties make it difficult or even impossible to judge whether two meteorites from the same chemical group were produced in the same ejection event. This pairing correction must be applied before the CRE age histogram can be used to identify periodic structures and/or meteorite delivery mechanisms. Any under-or overcorrection for pairing can either produce (apparent) peaks in the CRE age histogram or smear them out. This is the motivation for the present study: to improve the quality of the CRE ages of iron meteorites and to then identify possible structures in the CRE age histogram, if they exist. Shaviv (2002Shaviv ( , 2003 proposed that periodic galactic fluence variations produce peaks in the CRE age histogram for iron meteorites. Wieler et al. (2013) demonstrated that periodic GCR intensity variations produce periodic structures in the CRE age histogram if the meteorite delivery rate is constant. Their theoretical approach is based on CRE ages calculated via the 40 K-K dating system, which is based on a cosmogenic radionuclide with a half-life of 1.251 Ga. For our work, we use the 36 Cl-36 Ar dating system, which is based on a radionuclide with a half-life of 301,000 years. Figure 1 shows artificial CRE age histograms of 36 Cl-36 Ar CRE ages for iron meteorites assuming periodic changes in the GCR intensity. The histogram has been constructed as follows: First, the time axis for meteorite production ranging from zero up to 1050 Ma has been divided into 10 Ma bins. For each bin, a nominal GCR fluence has been assigned. For the first 150 Ma, we assigned arbitrary a GCR fluence of 1; for the time interval 150-300 Ma, we assigned a GCR fluence of either 1, 1.5, or 3; for the time interval 300-450 Ma, we again assigned a value of 1, and so on. The upper two panels are for a constant GCR intensity (always 1), the middle two panels are for periodic GCR fluence variations of a factor of 1.5, that is, roughly in accord with the proposal by Sloan and Wolfendale (2013a), and the lower two panels are for periodic GCR fluence variations of a factor of 3, as proposed by Shaviv (2002Shaviv ( , 2003. As a period we always use 150 Ma, close to the value proposed by Shaviv (2002Shaviv ( , 2003. Second, for each bin, we calculated the total number of cosmogenic 36 Ar atoms. For example, for the bin 750-760 Ma, we calculated the total number of 36 Ar atoms produced for the entire irradiation of the meteorite, from its production 760 Ma ago up to its fall on the Earth (T = 0). By combining the thus calculated 36 Ar concentrations with the 36 Cl data, which are always for the recent GCR fluence of 1, we then calculated the apparent age of the meteorite, which is higher than the real age due to enhanced production of 36 Ar in the time period 150-300 Ma, 450-600 Ma, and 750-760 Ma. In this special example, the real age is 760 Ma and the apparent age is 1380 Ma. Doing this type of calculation for all 10 Ma bins, we then constructed a CRE age histogram for the apparent ages. Note that the real-time axis is always from 0 to 1050 Ma and that the periodic GCR intensity changes are always in the same time intervals, that is, low at 0-150 Ma, 300-450 Ma, 600-750 Ma, and 900-1050 Ma and enhanced at 150-300 Ma, 450-600 Ma, and 750-900 Ma. Therefore, the number of peaks visible in the histograms is always the same, only their magnitude and their period depends on the assumed enhancement of the GCR fluence. For the data shown in the left panels, we assume a constant rate for the production of meteorites and an unchanging probability of transport to and capture by the Earth. For constructing the diagrams in the right panels, we divided the time axis into 10 Ma bins (as for the panels on the left) and we used a random number generator to assign to each bin how many meteorites have been produced and delivered to the Earth. We do not consider here the expected exponential decease of number of iron meteorites with increasing CRE age, that is, we assume a constant probability of the Earth capture and disregard competing mechanisms for meteoroid destruction (e.g., Alexeev 2016). This assumption does not compromise the following discussion and conclusions. What does affect the CRE age histogram are the assumptions we made concerning uncertainties and bin sizes. For the artificial histograms shown here, we neglect any uncertainties, that is, the ages have an uncertainty of zero. With the assumed flux variations of factors 1.5 and 3.0 and the chosen bin size of 30 Ma, there is no piling up in the histogram, there is only a periodic compression. With larger flux variations, and/or larger bin sizes, and/or including uncertainties, events pile up in the histogram. The histograms are normalized to the same number of meteorites, which explain some of the non-integer numbers in the right hand panels.

THEORETICAL BASICS
There are two important findings; first, periodic GCR fluence variations indeed produce time-dependent structures in the CRE age histogram. This is true for both cases, for constant (left column) and for stochastic (right column) production of iron meteorites. The solid lines indicate the results of a peak search algorithm. Other than the identification of noise, there are no clear indications for periodic structures in the upper two panels (constant GCR). Time-dependent structures are clearly visible for a constant meteorite delivery and DGCR = 1.5 (middle left panel). For the case DGCR = 1.5 and stochastic meteorite production, there are hints for time-dependent structures, but they are not well constrained. The time-dependent structures become much more robust for DGCR = 3.0 (lower panels). Second, the time dependence is not strictly periodic. For example, for DGCR = 1.5, the peaks in the CRE age histogram are at 90, 450, 870, and 1190 Ma; the period ranges from 320 to 420 Ma. For periodic GCR intensity changes of DGCR = 3.0, the peaks are at 70, 670, 1270, and 1850 Ma; the periodicity is in the range 600 Ma. The periodicity seen in the CRE age histogram not only depends on the period of the GCR fluence change but also on the magnitude of this change. This finding, though never discussed so far, is actually expected; during times of low GCR intensity, the "cosmic ray clock" is running slow, and at times of high GCR intensity, the clock is running fast. Therefore, GCR intensity variations stretch and compress the time-axis in the histogram. The histogram in Fig. 1 is set up in a way that the average GCR fluence used in the upper two panels (GCR constant) corresponds to the lowest fluence assumed for the other two cases with periodic GCR fluence variations. The case DGCR = 1.5, The left panels are for constant production of iron meteorites in the asteroid belt. The right panels are for stochastic production of iron meteorites. The upper two panels are for a constant GCR flux, the middle two panels are for GCR flux variations of a factor of 1.5, and the bottom two panels are for GCR flux variations of a factor of 3. The solid line is the result of a simple peak-search algorithm applied to the constructed CRE age histograms. For more information, see text.
therefore, stands for an average GCR fluence at the low end and a 1.5 times higher GCR fluence at the high end. Assuming a periodicity always increases the ages and expands the time-axis in the CRE age histogram. This can also be seen by the finding that for modeling we assume continuous meteorite production from present up to 1 Ga, but the apparent ages, calculated with the periodic GCR fluence variations, range up to 2 Ga, that is, the entire time-axis is stretched.
To summarize, provided that a high quality CRE age histogram for iron meteorites can be established, it would not only be possible to study transit processes for iron meteorites, it would also help answering the question of whether the CRE ages for iron meteorite support the assumption of periodic fluence variations.

Samples
We selected a total of 56 iron meteorites from six different groups; 42 are IIIABs, seven are IIABs, three are IVAs, one IC, one IIA, one IIB, and one ungrouped. Table A1 in Appendix A summarizes some of the relevant information: names, source collections, and the total collected mass. All samples were analyzed for the noble gases He, Ne, and Ar and the lighter cosmogenic radionuclides (CRN) 10 Be, 26 Al,36 Cl,and 41 Ca.
The samples were first cut into smaller pieces with masses in the range 60-200 mg. They were cleaned with ethanol to remove any contamination from cutting and were carefully studied under a binocular microscope to check for troilite and schreibersite inclusions. Samples with visible inclusions were not used for further studies. This preselection reduces the problem caused by inclusions (see below); we cannot completely exclude minor inclusions in our samples (see also Ammon et al. 2008).

Noble Gases
The He, Ne, and Ar isotopic concentrations were measured in the noble gas laboratory at the University of Bern. For a detailed description of the procedure used, see Ammon et al. (2008, 2011) and Smith et al. (2017. The samples were loaded into an all-metal (except for a glass window) noble gas extraction line and were preheated in vacuum at~80°C for 1-2 days to remove atmospheric surface contamination. The samples were degassed in a Mo crucible held at 1800°C for~30-40 min. A boron-nitride liner inside the Mo crucible prevents corrosion of the latter. We regularly performed second extractions at slightly higher temperatures to check whether the samples have been completely degassed. In addition, we regularly performed blank measurements. Both re-extractions and blanks contributed less than~1% to the measured sample gas amounts for He and Ne isotopes and less than 10% for Ar isotopes. After extraction, the gases were first cleaned on various getters (SAES Ò ) working in the temperature range between room temperature and 280°C. Subsequently, an Ar fraction was cryogenically separated from the He-Ne fraction using activated charcoal held at the temperature of boiling liquid nitrogen (LN 2 ). Helium and Ne isotopes were analyzed using an in-house made sector field mass spectrometer. During He and Ne measurements, charcoal (at LN 2 ) was used to reduce interfering species such as Ar, water, and some hydrocarbons. The separated Ar fraction (the same sample) was routed to an in-house made tandem mass spectrometer optimized to reduce baseline variations on 36 Ar and especially on 38 Ar due to scattered 40 Ar ions. All measurements were performed in peak-jumping mode and both spectrometers were regularly calibrated using standard gases. All standard gases are of atmospheric isotopic ratios except for He, which is enriched in 3 He relative to air.

Cosmogenic Radionuclides
The procedures used for chemical separation were adopted from those developed earlier by Merchel and Herpers (1999). The measurements were divided into eight batches (excluding one batch for the big iron meteorite Twannberg, cf. Smith et al. 2017), each consists of seven samples and one blank. Meteorite samples with masses ranging from 56 to 214 mg were dissolved in HNO 3 (2 M) at room temperature. After adding carrier solutions of natural 35 Cl/ 37 Cl, AgCl was precipitated and treated separately in a dedicated 36 Cl laboratory for isobar reduction and further cleaning. The remaining sample solution was transferred to a normal laboratory and further stable isotope carriers were added ( 9 Be, 27 Al, nat Ca, 55 Mn) to allow chemical separation. To determine the target elements relevant for the production of the cosmogenic nuclides of interest here (Fe and Ni), we took aliquots of 2-3% for inductively coupled plasma mass spectrometry (ICP-MS). For the further separation of Be, Al, Ca, Fe, and Mn fractions, anion and cation exchange and repeated precipitations were performed. The separated fractions were further purified from their respective isobars and finally converted to oxides (BeO, Al 2 O 3 , MnO 2 , Fe 2 O 3 ) and fluorides (CaF 2 ). The 10 Be/ 9 Be, 26 Al/ 27 Al,36 Cl/ 35 Cl, and 41 Ca/ 40 Ca ratios were measured at the Dresden AMS facility (DREAMS) at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR, Germany; Akhmadaliev et al. 2013;Pavetich et al. 2016;Rugel et al. 2016).

Noble Gases
Helium The 4 He/ 3 He ratios and 3 He concentrations are given in Table A2 in Appendix A. We cannot give He data for the meteorites Bristol, Brownfield, Calico Rock, Fort Pierre, Gan Gan, Mapleton, Norfolk, Rowton, Treysa, and Verkhne Udinsk due to technical problems during the measurements. Figure 2 depicts the 4 He/ 3 He ratios, which range between 1.63 AE 0.09 and 12.89 AE 0.73. Thirty-eight of the selected meteorites have 4 He/ 3 He ratios in the range 3-7 (indicated by the two horizontal lines in Fig. 2), well within the range expected for cosmogenic production (e.g., Ammon et al. 2009). Four samples have 4 He/ 3 He ratios lower than expected for cosmogenic production and two (Braunau and Squaw Creek) have 4 He/ 3 He ratios of 9.2 and 12.8, that is, significantly higher than cosmogenic. The higher 4 He/ 3 He ratios for Braunau and Squaw Creek can either be due to remaining atmospheric contamination ( 4 He/ 3 He~10 6 ) or radiogenic 4 He from decay of uranium and thorium; both are known trace elements in troilite and/or schreibersite inclusions. An additional possibility is the loss of tritium before its decay to 3 He caused by solar heating in orbits close to the sun. Atmospheric 4 He contamination would likely be accompanied by an even more significant 40 Ar contamination. The 40 Ar/ 36 Ar ratios of 143 and 207, respectively, for Braunau and Squaw Creek are relatively high, but they are comparable to ratios of other meteorites. We conclude that atmospheric contamination most likely is not the reason for the high measured 4 He/ 3 He ratios. If the studied samples from Braunau and Squaw Creek were to have been compromised by 4 He from troilite and schreibersite, there would be higher than normal 22 Ne/ 21 Ne ratios (see below). The measured 22 Ne/ 21 Ne ratios for both meteorite samples are not exceptionally high. This seemingly leaves 3 He deficits due to tritium loss the most likely explanation. However, there are also convincing arguments against the tritium loss hypothesis. The 4 He cos values given for Braunau and He deficits due to tritium losses are also not a viable explanation for the high 4 He/ 3 He ratios. Note that Hampel and Schaeffer (1979) Fig. 2), indicating a small pre-atmospheric radius. Indeed, the recovered masses for the four meteorites in question are less than 21.5 kg (see Table A1 in Appendix A). In Fig. 2, the four meteorites follow the general trend of decreasing 4 He/ 3 He with decreasing 4 He/ 21 Ne ratios. Therefore, there is no reason to doubt the data and the conclusion might be that the model predictions overestimate 4 He/ 3 He ratios for small meteorites and/or close to the surface.

Neon
The measured 20 Ne/ 22 Ne and 22 Ne/ 21 Ne ratios and 20 Ne gas amounts after correction for fractionation and interferences but before blank correction are given in Table A2 in Appendix A. We decided not to make any blank corrections at this point because the blank is atmospheric and can be best corrected using a twocomponent deconvolution (see below). The 20 Ne/ 22 Ne ratios range between 0.867 and 1.273, clearly indicating that Ne is dominantly cosmogenic with minor additional contributions. We determined cosmogenic 21 Ne concentrations and 22 Ne/ 21 Ne ratios (index cos) assuming that measured Ne is a mixture of cosmogenic Ne with 20 Ne/ 22 Ne = 0.867, which is the lowest ratio measured for the samples, and atmospheric contamination ( 20 Ne/ 22 Ne = 9.78). The decomposed values are given in Table 1. The corrections for 21 Ne cos are all negligible, <1%. The ( 22 Ne/ 21 Ne) cos ratios range between 1.025 and 1.098. As discussed earlier ( 22 Ne/ 21 Ne) cos in the metal phase (hereafter labeled FeNi) range between 1.02 and 1.04 (cf. Ammon et al. 2008); ratios higher than this indicate contributions from phosphorous and/or sulfur. We recently measured ( 22 Ne/ 21 Ne) cos of 1.21 AE 0.04 for two schreibersite inclusions from the large iron meteorite Twannberg (Smith et al. 2017). For our samples, most of the ( 22 Ne/ 21 Ne) cos ratios indicate contributions from sulfur and/or phosphorous. We corrected the 21 Ne cos concentrations for these contributions using the procedure developed by Ammon et al. (2008): where 21 Ne m and R M are the measured 21 Ne concentrations and 22 Ne/ 21 Ne ratios, respectively. The ( 22 Ne/ 21 Ne) cos ratios for troilite/schreibersite inclusions and pure FeNi metal are R s and R FeNi , respectively. The 21 Ne FeNi values corrected in this way are given in Table 1. The corrections are~40% for Lombard and less than 1% for Bristol. On average,~7% of the measured 21 Ne cos is due to contributions from troilite and/or schreibersite, that is, the corrections are typically minor. The 21 Ne cos production rates from troilite and schreibersite are on average 10 times higher than from pure metal (e.g., Ammon et al. 2008;Smith et al. 2017), which makes the 21 Ne production rates from sulfur and/ or phosphorous 30-50 times higher than from iron and nickel. The measured 22 Ne/ 21 Ne ratios indicate sulfur and/or phosphorous concentrations in the range of 0.1-0.2 wt%, which is reasonable considering that we preselected the samples using only optical means. If we only consider the data for the IIIAB iron meteorites, the corrections for 21 Ne are~6%. In contrast, the average correction for the iron meteorites from the other groups is slightly higher, that is, in the range of 10-12%. The 21 Ne FeNi concentrations (in 10 À8 cm 3 STP per g) range from 0.066 for Squaw Creek (IIAB) to 9.02 for Sandtown (IIIAB), that is, the variation among the studied samples is almost a factor of 137.

Argon
The 40 Ar/ 36 Ar and 36 Ar/ 38 Ar ratios and 38 Ar concentrations are given in Table A2 in Appendix A. The measured 40 Ar/ 36 Ar and 36 Ar/ 38 Ar ratios before blank correction range between 7-256 and 0.64-4.53, respectively, indicating that besides cosmogenic Ar, there is an additional component, most likely residual atmospheric contamination. We corrected the Ar data for the contamination assuming that all 40 Ar is atmospheric. The 36 Ar cos and 38 Ar cos concentrations are given in Table 1. The ( 36 Ar/ 38 Ar) cos ratios range from Both meteorites have exceptionally low Ar concentrations, which make the measurements highly uncertain and which might explain the strange ratios. It is possible that both Ar fractions were accidentally lost during gas extraction. These data are not discussed any further.
The Isotopic Ratio ( 4 He/ 38 Ar) cos According to model predictions (Ammon et al. 2009), the ( 4 He/ 38 Ar) cos ratios are in the range of 35-115 for iron meteorites with pre-atmospheric radii between 10 and 120 cm. In contrast, the range of isotopic ratios found in this study varies from 3.2 for Greenbrier County to 226 for Gibeon, the range is much wider. The low ratio for Greenbrier County is most likely due to 4 He (and 3 He) deficits because not only is ( 4 He/ 38 Ar) cos much lower than the range given by the model predictions but also the ( 3 He/ 38 Ar) cos and ( 4 He/ 21 Ne) cos ratios are lower than the range defined by the other studied meteorites. Since the measured 3 He and 4 He concentrations for Greenbrier County are well within the range of concentrations measured for the other studied meteorites (Table A2 in Appendix A), we can exclude analytical problems and conclude that the 3 He and 4 He deficits occurred while the meteoroid traveled in space. Greenbrier County also exhibits a low ( 4 He/ 3 He) ratio of 1.95, further confirming that the He data are peculiar (see also above). The same argument holds for the iron meteorite Joel's Iron ( 4 He/ 38 Ar = 4.68), the measured 3 He and 4 He concentrations are not exceptionally low but the ratios ( 4 He/ 38 Ar) cos , ( 3 He/ 38 Ar) cos , and ( 4 He/ 21 Ne) cos are low. The ( 4 He/ 3 He) ratio for Joel's Iron is 3.31, which is within the range of the other meteorites, though at the lower end. The data for Braunau are similar; all three isotopic ratios are low but in contrast to Greenbrier County and Joel's Iron, the 3 He and 4 He concentrations are also low. The perplexing nature of these observations leads us to reluctantly accept the possibility of unrecognized problems during the measurement. Additionally, the measured 4 He/ 3 He ratio is high, that is, 12.9, and we had to calculate 4 He cos = 3 He cos 9 3.96 (see above). It might be possible that for Braunau, the value of 3.96 is too low, which then results in too low 4 He cos and consequently in too low 4 He/ 38 Ar and 4 He/ 21 Ne ratios. This, however, cannot explain the low ( 3 He/ 38 Ar) cos ratio. Anyway, ignoring the data for Greenbrier County, Joel's Iron, and Braunau reduces the measured range to 30-226, that is, the lower limit is in agreement with the model predictions. The lowest ratio is now for the iron meteorite Benedict, which has rather normal 3 He/ 38 Ar and 4 He/ 21 Ne ratios. The high 4 He/ 38 Ar ratio of~226 for Gibeon is most likely due to the very low cosmogenic Ar concentration, which is either due to Ar loss during gas extraction and/or cleaning or it is due to the fact that the cosmogenic Ar concentration is extremely low due to the large pre-atmospheric size. Since the ( 21 Ne/ 38 Ar) cos ratio for Gibeon is slightly higher than the average for the other meteorites, we argue that the reason for the high ( 4 He/ 38 Ar) cos ratio is in a low 38 Ar concentration. Another explanation for high ( 4 He/ 38 Ar) cos ratios might be a complex exposure history. The depth dependency for production of 4 He cos and 38 Ar cos are different; 4 He cos is produced at greater depths than 38 Ar cos . A sample that was originally buried deep in a large parent body might contain some 4 He cos but no 38 Ar cos . After further breakup, there is production of 4 He cos and 38 Ar cos , but the ( 4 He/ 38 Ar) cos ratio will always be too high due to the inherited 4 He cos . Anyway, also excluding Gibeon from the data set reduces the measured ( 4 He/ 38 Ar) cos ratios to the range 30-101, that is, in agreement with the model predictions. The cosmogenic components are labeled "cos." The given uncertainties (1r) include the uncertainties for (1) measurements of the ion currents, (2) corrections for interfering isotopes, (3) corrections for instrumental mass fractionation, and (4) corrections for trapped components, and (5) corrections for contributions from S and/or P on 21 Ne, that is, 21 Ne corr . Systematic uncertainties in the calibration standard are not included; they amount to 4% for concentrations and 1% for isotope ratios.
The Isotopic Ratio ( 4 He/ 21 Ne) cos The ( 4 He/ 21 Ne) cos ratios range between 157 for Schwetz and 675 for Gibeon. Using equation 16 from Honda et al. (2002), we can connect the ( 4 He/ 21 Ne) cos ratios to the pre-atmospheric shielding depth: Using corrected 21 Ne cos concentrations and considering only reliable 4 He cos concentrations, we calculate shielding depths in the range~34 tõ 1460 g cm À2 , which corresponds to depths-and accordingly-minima radii of 4.3 cm for Schwetz to 186 cm for Gibeon. The inferred minimum preatmospheric radius for Gibeon agrees with previous studies that inferred pre-atmospheric radii in the range 2-3 m Honda et al. 2008). The low ( 4 He/ 21 Ne) cos ratio for Schwetz and low minimal radius of 4.3 cm is in accord with the recovered mass of only 21.5 kg, which requires a pre-atmospheric radius of at least 8 cm. The large ( 4 He/ 21 Ne) cos ratio of 675 for Gibeon is expected due to its large pre-atmospheric size. The large iron meteorite Aletai (previously named Xinjiang, renamed in 2016), with a pre-atmospheric radius of more than 1 m, has an average ( 4 He/ 21 Ne) cos ratio of 593 AE 55 (Ammon et al. 2011). We measured ( 4 He/ 21 Ne) cos ratios in the range 295-756 for Twannberg, an extremely large iron meteorite; this ratio is slightly higher than that of Gibeon (Smith et al. 2017). In a systematic study of 12 different fragments from the large iron meteorite Canyon Diablo, Michlovich et al. (1994) measured ( 4 He/ 21 Ne) cos ratios that vary from~280 to~480. The average shielding depth, that is, the average minimum radius, for all meteorite samples studied by us is 284 g cm À2 or 36 cm. This average radius corresponds to an average mass of 1.5 t.
The Correlation ( 4 He/ 38 Ar) cos as a Function of ( 4 He/ 21 Ne) cos Figure 3 depicts the element ratios ( 4 He/ 38 Ar) cos as a function of ( 4 He/ 21 Ne) cos for the studied meteorites. There is a reasonable linear correlation with R 2 = 0.98. The slope of the correlation line, if forced through the origin, is 0.198 AE 0.005 (solid black line). The gray shaded area shows the 95% confidence interval. This finding confirms earlier results that the ( 4 He/ 38 Ar) cos ratios can together with or in addition to the ( 4 He/ 21 Ne) cos ratios also serve as a shielding indicator (e.g., Ammon et al. 2009). The ( 4 He/ 38 Ar) cos ratios can often be determined more reliably than ( 4 He/ 21 Ne) cos ratios because the latter are often compromised by contributions from sulfur and/or phosphorous to 21 Ne cos . If such contributions are not properly subtracted, the ( 4 He/ 21 Ne) cos ratios are too low, which results in an underestimation of the shielding conditions, that is, the size of the meteorite and shielding depth of the sample. Based on Equation 3 and by using the linear correlation between ( 4 He/ 38 Ar) cos and ( 4 He/ 21 Ne) cos , we propose a new equation for calculating pre-atmospheric shielding depths for iron meteorites based on ( 4 He/ 38 Ar) cos : Isotope Ratios 38 Ar/ 21 Ne The ratios 38 Ar cos / 21 Ne FeNi for samples, except Trenton and Carthage, range between~3 for Gibeon (IVA) and 51.3 for Joel's Iron (IIIAB), the grand average value is~6.65. The ratio for Joel's Iron is much higher than the other ratios and must be considered as an outlier. The reason for this anomalously high ratio is most likely the low 21 Ne FeNi concentration, which is among the lowest in our database. In contrast, the 38 Ar cos concentration is well within the range defined by the other studied meteorites. If we exclude the 38 Ar cos / 21 Ne FeNi ratio for Joel's Iron from the database, the ratios range between~3 and~6.4 and the average value is~4.8. For Twannberg, we measured slightly higher 38 Ar cos / 21 Ne FeNi ratios betweeñ 2.8 and~7.9 (Smith et al. 2017). Also, the average value of~6.3 for Twannberg is slightly higher than the average value obtained here for a large selection of different iron meteorites.  ( 4 He/ 38 Ar) cos as a function of ( 4 He/ 21 Ne) cos for the studied iron meteorites. There is a linear correlation between both elemental ratios, indicating that ( 4 He/ 38 Ar) cos can also be used as an indicator for shielding.

Cosmogenic Radionuclides
The cosmogenic 10 Be, 26 Al,36 Cl, and 41 Ca activities (dpm kg À1 ) are given in Table A3 [Ni]. Some radionuclides in some samples, for example 10 Be, 26 Al,and 36 Cl in Gibeon, could not be quantified due to very low radionuclide concentrations, either due to high shielding depths, low chemical yield (Gibeon), handling losses, or other. The results are labeled "below detection limit (bdl)" in Table A3 in Appendix A. Among the measured radionuclides, 41 Ca is the most challenging to measure . First, some samples have very low concentrations. Second, the chemical yields for some samples were low; for example, in the range of~40% for Davis Mountains or Kenton County or even as low as~15% for Henbury. All radionuclide data for Henbury are very close to the detection limit (see below). The major reason, however, for the substantially larger uncertainties of the 41 Ca concentrations of~17% (average) compared to the other radionuclides is a significant isobar ( 41 K) contamination in the ion source, as described in Rugel et al. (2016) (from Cs used for sputtering or from migration from stainless steel pins used as backing), which compromised some of the measurements and/or interpretation of the results. Before we discuss the radionuclides in detail, we first need to correct the data for radioactive decay during terrestrial residence.

Terrestrial Ages
Most of the iron meteorites in this study are finds, so the radionuclide data for these meteorites must be corrected for decay during terrestrial residence (T terr ) before discussing production rates. We chose the 36 Cl and 41 Ca radionuclide pair because the commonly used 36 Cl/ 10 Be dating method is occasionally compromised by contributions from 0.1 to 0.2 wt% of P-and/or S-rich inclusions. In addition, 41 Ca has a shorter half-life compared to 10 Be (1.37 9 10 6 years for 10 Be versus 0.995 9 10 5 years for 41 Ca); it is more suitable for calculating terrestrial ages in the range typical for iron meteorites -from zero up to about 500 ka (Table 2, see also Smith et al. 2017). The basic equation is:  (Nica et al. 2012).
For the radioactive decay of 41 Ca, we use the new decay constant of k 41 = (6.97 AE 0.11) 9 10 À6 yr À1 from J€ org et al. (2012), which is 4.7% higher than the earlier value of k 41 = (6.66 AE 0.32) 9 10 À6 yr À1 by Kutschera et al. (1992). Using the new value of k 41 = (6.97 AE 0.11) 9 10 À6 yr À1 is not without problems because most of the earlier data are based on the old value (k 41 = [6.66 AE 0.32] 9 10 À6 yr À1 ), which makes comparisons with earlier data difficult.
The 41 Ca/ 36 Cl terrestrial ages are given in Table 2. The uncertainties are dominated by the uncertainties for the 41 Ca concentrations (17% on average) and by the uncertainty for the 41 Ca/ 36 Cl production rate ratio (~4.3%). Since the differences are well within typical uncertainties, the choice of the half-life has only a minor influence on T terr , the CRE age, and on the CRE age histogram (see below). We cannot give T terr for Arispe, Gan Gan, Gibeon, and Piñon due to missing 36 Cl and/or 41 Ca data. For Benedict, Brownfield, Catalina 107, and Mapleton, the measured 41 Ca/ 36 Cl ratios are slightly higher than the assumed production rate ratio of 1.157. However, there is still agreement within the (sometimes large) 1r uncertainty. For these meteorites we assume T terr 0. For the meteorites Boxhole, Fort Pierre, and Schwetz, the measured 41 Ca/ 36 Cl ratios are sometimes significantly higher than the assumed production rate ratio, though the discrepancy is still within the 2r uncertainties. For the three meteorites, we also assume T terr 0. For the four meteorites Braunau, Costilla Peak, Davis Mountains, and Zerhamra, the terrestrial ages are all <10 ka with uncertainties in the range 10-  Lavielle et al. 1999). The very high T terr for Verkhne Udinsk of more than 1.2 Ma is suspect considering the rather normal 36 Cl concentration of 22.76 AE 0.50 dpm kg À1 , which is close to the saturation value. We consider the 41 Ca concentrations too low but errors, during chemical extraction or AMS measurements cannot be an explanation. Since the 36 Cl concentration is rather normal, we assume T terr 0. The 53 terrestrial ages obtained by us are in good agreement with literature data that are usually obtained using the less reliable 36 Cl/ 10 Be method. The terrestrial age for Grant of 13.2 +14.9 À13.2 ka is in agreement with ages obtained by Lavielle et al. (1999) and Shankar (2011) of 22 AE 20 ka and 43 AE 49 ka, respectively, that is, all three studies indicate a short T terr . The T terr for Norfolk of essentially zero is in agreement with the very short T terr = 35AE20 ka determined by Lavielle et al. (1999). For Picacho, we determined T terr = 96.5 AE 16.3 ka, which is in agreement with the ages of 76 AE 29 ka and 129 AE 44 ka determined by Shankar (2011) and Lavielle et al. (1999), respectively. The age obtained by us for Sacramento Mountains of 148 AE 17 ka is in perfect agreement with the age of 148 AE 25 ka given by Shankar (2011).
The terrestrial ages range from essentially zero for Benedict, Boxhole, Brownfield, Catalina 107, Fort Pierre, Kayakent, Mapleton, and Schwetz to 562 AE 86 ka for North Chile. Twelve meteorites out of the 53 for which we have data have T terr larger than 100 ka, long for T terr in nonpolar regions. Long T terr for some large iron meteorites were also reported by Aylmer et al. (1988) and Chang and W€ anke (1969). For example, using the 10 Be/ 36 Cl radionuclide pair, Chang and W€ anke (1969) determined for the iron meteorite Tamarugal (Chile) a terrestrial age of~2.7 Ma. In addition, long terrestrial ages were also observed by Nishiizumi et al. (2002) for a Martian and a lunar meteorite from a hot desert region.
We recently published a long terrestrial age of 202 AE 34 ka for the large IIG iron meteorite Twannberg (Smith et al. 2017). By restudying the age using not only the new 41 Ca half-life but also applying an isochron technique based on the five measured Twannberg samples, we calculated a new 41 Ca/ 36 Cl ratio of 0.508 AE 0.050, which corresponds to an age of 176 AE 19 ka, slightly lower than our former result. While this long T terr is still surprising considering the humid conditions in Switzerland, it is well within the range of T terr values determined here. For example, T terr for Twannberg is very similar to T terr of Forsyth County (187 AE 19 ka) and Djebel-in-Azzene (192 AE 22 ka). While Djebel-in-Azzene was found in Algeria and experienced more arid conditions, Forsyth County, Kenton County, and Bristol, which all have T terr similar to or even longer than Twannberg, were found in the east of the USA, that is, also in humid conditions.
Combining all data, we conclude that T terr for iron meteorites can be as long as a few hundred ka, even in humid environmental conditions. As we already speculated for the discussion of the Twannberg data, there might be a process responsible for avoiding or slowing down weathering processes in iron meteorites.

Production Rates and Production Rate Ratios
Here, we discuss the 10 Be, 26 Al,and 36 Cl production rates, that is, the concentrations at time of fall. With Table 2. Continued. Terrestrial ages ( 36 Cl/ 41 Ca), 10 Be, 26 Al,and 36 Cl production rates (dpm kg À1 ), and 36 Cl/ 36 Ar cosmic ray exposure (CRE) ages (Ma). The uncertainties for the 26 Al and 36 Cl production rates, that is, 26 Al(0) and 36 Cl (0), are calculated using Gaussian error propagation and considering the uncertainties for the radionuclide activities (Table A2 in Appendix A) and the uncertainties for the terrestrial ages but not the uncertainties for the decay constants. For 26 Al(0), we also consider uncertainties introduced by the correction for contributions from S and/or P (see text). a Data are from Shankar et al. (2011) the known terrestrial ages, it is straightforward calculating production rates for 36 Cl. For 10 Be and 26 Al, the situation is more complicated because both radionuclides can be compromised by contributions from S and/or P. Consequently, 10 Be and 26 Al production rates not only depend on shielding but also on the trace element concentrations of the studied sample. The correction procedure as well as how we treated the uncertainties is described in detail in Appendix B.
The corrections for 26 Al are <10%; exceptions are Charcas (correction factor 1.10), Sikhote-Alin (1.16), Boxhole (1.18), Piñon (1.25), Squaw Creek (1.27), Arispe (1.43), and Lombard (1.66). The corrections for 10 Be are smaller than the corrections for 26 Al. The 10 Be, 26 Al (corrected), and 36 Cl concentrations at the time of fall, that is, the production rates, are given in Table 2. Production rates for 41 Ca are not given because they are no longer independent; 41 Ca production rates are simply given by 1.157 times the 36 Cl production rates.
The 10 Be, 26 Al, and 36 Cl production rates for Cape York, Casas Grandes, Henbury, and Lombard are all low, indicating a large pre-atmospheric size. For Lombard and Casas Grandes, this finding is in accord with high 4 He/ 21 Ne ratios of 600 and 569, respectively. In contrast, the 4 He/ 21 Ne ratios of 309 for Henbury and 290 for Cape York are not exceptionally high and are therefore in contradiction to the low radionuclide production rates. The low 4 He/ 21 Ne ratio for Cape York is surprising because with a recovered mass of more than 58 t Cape York was obviously a large object. Since the CRE age for Henbury is unreasonably high due to the low 36 Cl concentration (see below), we consider all radionuclide production rates for Henbury as unreliable; they are not considered any further.
The 10 Be production rates range between 0.09 dpm kg À1 for Gibeon and 5.96 dpm kg À1 for Calico Rock; the average is 3.60 dpm kg À1 . The data are well within the range typical for iron meteorites. For example, Aylmer et al. (1988), Lavielle et al. (1999), and Xue et al. (1995) determined 10 Be production rates in the range 3-6 dpm kg À1 . In a recent study, Shankar et al. (2011) measured 10 Be production rates in the range 0.36-6.00 dpm kg À1 . Figure 4 depicts the 10 Be production rates as a function of ( 4 He/ 21 Ne) cos . As expected, the 10 Be production rates decrease with increasing ( 4 He/ 21 Ne) cos ratios, that is, with increasing shielding. Also shown is the correlation line given by Aylmer et al. (1988), which is only slightly different from the correlation line given earlier by Chang and W€ anke (1969). Our data are in accord with the correlation line, despite the fact that the data by Aylmer et al. (1988) and Chang and W€ anke (1969) are, first, based on different 10 Be half-lives and, second, their 21 Ne data are not corrected for contributions from S and/or P. While this all indicates consistency, this correlation is not useful. The substantial scatter precludes the determination of the shielding conditions only via 10 Be.
The (corrected) 26 Al production rates measured here range from 0.10 dpm kg À1 for Cape York to 4.65 dpm kg À1 for Djebel-in-Azzene; the average value is 2.46 dpm kg À1 . Again, the results are in agreement with literature values that are usually in the range 2-5 dpm kg À1 (e.g., Aylmer et al. 1988;Xue et al. 1995;Lavielle et al. 1999). In a recent study, Shankar et al. (2011) determined 26 Al production rates between 0.27 and 5.07 dpm kg À1 .
To check the data for reliability, we calculated the 10 Be/ 26 Al production rate ratio. Published ratios are 1.3 AE 0.1 (Xue et al. 1995) and 1.4 AE 0.2 (Lavielle et al. 1999). The average for our data is 1.48 AE 0.26, that is, in agreement with literature data. There are, however, outliers that have 10 Be/ 26 Al production rate ratios that differ from the average by more than the standard deviation. The ratio for Nazareth of~0.44 is too low. In contrast, the ratio for Verkhne Udinsk is too high compared to the average value; the difference is~3 times the standard deviation. By not considering the outliers, we calculated a weighted average value of 1.45 AE 0.02 (using the uncertainties as weights). Here we want to stress that our average value is based on 49 individual data, which makes our value most likely more accurate than the earlier literature data.
The 36 Cl production rates range between 0.5 dpm kg À1 for Cape York and 29.3 dpm kg À1 for Greenbrier County. The low value for Cape York is in accord with the 10 Be and 26 Al production rates, which are also low for this meteorite. The value of 29.3 AE 0.9 dpm kg À1 for Greenbrier County is unexpectedly high. According to our improved model calculations (Smith et al. 2017), the upper limit for the 36 Cl production rate is~25 dpm kg À1 , which is well in accord with our data. For example, the second highest 36 Cl production rate is 25.3 AE 0.8 for Elyria. Currently, however, we have no reason to consider the data for Greenbrier County as unreliable and we might speculate that some of the excess 36 Cl has been produced in Clbearing inclusions.

CRE Ages
We calculated the CRE ages using the 36 Cl-36 Ar dating system, which is very reliable because almost 90% of the cosmogenic 36 Ar is from the isobaric decay of 36 Cl. Consequently, the 36 Cl production rate is an excellent proxy for the 36 Ar production rate, which makes the system very reliable and almost independent on shielding conditions and chemical composition, provided the cosmic ray flux is constant. Using this system, the CRE age is given by (e.g., Begemann et al. 1976;Lavielle et al. 1999 (7) where P( 36 Cl)/P( 36 Ar) is the production rate ratio. The cosmogenic nuclide concentrations are 36 Ar cos (10 À8 cm 3 STP g À1 ) and 36 Cl (0) (dpm kg À1 ), and k 36 (2.301 9 10 À6 yr À1 ; Nica et al. 2012) is the 36 Cl decay constant. The factor 511 corrects for the different units and includes the decay constant. For the production rate ratio, we use the value of 0.835 AE 0.040 from Lavielle et al. (1999). The CRE ages range between~4 Ma for Squaw Creek and~640 Ma for Grant (Table 2). Our results agree well with the available literature data. For example, Lavielle et al. (1999) and Shankar (2011) determined for Brownfield CRE ages of 207 AE 6 and 194 AE 33 Ma, in agreement within two sigma with our result of 243 AE 18 Ma. Published CRE ages for Grant range from 426 AE 9 Ma to 640 AE 100 Ma (cf. Lipschutz et al. 1965;Lavielle et al. 1999;Ammon et al. 2008;Shankar et al. 2011;Schaeffer and Heymann 1965). Here, we determined a CRE age of 642 AE 39 Ma. The published CRE ages for Picacho of 453 AE 32 Ma (Lavielle et al. 1999) and 430 AE 13 Ma (Shankar 2011) agree well with our age of 433 AE 27 Ma. For Sacramento Mountains, we determined a CRE age of 202 AE 13 Ma, that is, in agreement with the 209 AE 14 Ma given by Lavielle et al. (1999). Also, our CRE age of 385 AE 24 Ma for Sikhote-Alin is in agreement with an earlier result of 320 AE 150 Ma (Lipschutz et al. 1965).
There are, however, also some discrepancies. The CRE age for Lombard of 57 AE 6 Ma is significantly lower than the value of 301 AE 18 Ma given by Lavielle et al. (1999). The reason for this discrepancy is in the 36 Ar cos concentrations. While we measured a low 36 Ar cos concentration of only~0.4 9 10 À8 cm 3 STP g À1 , Lavielle et al. (1999) measured a 36 Ar cos concentration of 2.4 9 10 À8 cm 3 STP g À1 ; that is, about six times higher. In contrast, the 36 Cl production rates are almost identical, 3.2 AE 0.2 dpm kg À1 in our study and 3.43 9 0.20 dpm kg À1 for Lavielle et al. (1999). Despite the finding that the noble gas ratios 3 He/ 38 Ar and 21 Ne/ 38 Ar measured by us for Lombard fit well into the range given by the other meteorites, we consider our data as unreliable. From our data, we determined 26 Al/ 21 Ne and 10 Be/ 21 Ne production rate ratios (atom/ atom) of 1.5 AE 0.8 and 3.1 AE 1.2, which are unreasonably high compared to expected values of 0.35 and 0.55 (e.g., Lavielle et al. 1999) and therefore also indicate a too low CRE age. For further discussion, we use for Lombard the CRE age given by Lavielle et al. (1999). For Treysa, we calculate a CRE age of 332 AE 21 Ma, significantly different from the age of 530 Ma given by Lipschutz et al. (1965). However, these authors used the 21 Ne-26 Al dating system, which we consider as unreliable due to interfering contributions from S and/or P. The CRE ages of 430 AE 40 Ma and 370 AE 30 Ma, based, respectively, on the 36 Cl-36 Ar and 39 Ar-38 Ar pairs, given by Schaeffer and Heymann (1965) agree slightly better with our result. Note that Schaeffer and Heymann (1965) used the differences between the 36 Cl-36 Ar and 39 Ar-38 Ar CRE ages for Treysa to discuss cosmic ray intensity variations.
We use the production rate ratios 26 Al/ 21 Ne and 10 Be/ 21 Ne as a proxy for the reliability of the CRE ages. The 21 Ne production rates were calculated by dividing the corrected 21 Ne concentrations by the CRE ages. Therefore, incorrect CRE ages would lead to incorrect 21 Ne production rates. By considering all data, we calculated a weighted average 26 Al/ 21 Ne ratio of 0.35, that is, in agreement with the value of 0.38 given by Hampel and Schaeffer (1979). There are, however, some outliers. The 26 Al/ 21 Ne production rate ratio of 2.71 AE 0.26 for Greenbrier County is too high. The reason is most likely in the too low 21 Ne FeNi concentrations as it is indicated by a low ( 21 Ne/ 38 Ar) cos ratio of 0.02 as opposed to typical ratios of 0.2 for most of the other iron meteorites. We assume that the problem is in the 21 Ne FeNi concentration and not in the 36 Ar cos concentration and consider the CRE age for Greenbrier County as reliable. The same argument holds for Joel's Iron. The measured ( 21 Ne/ 38 Ar) cos ratio is about 10 times lower than the average, making the 21 Ne FeNi data questionable. Again, we consider the CRE age as reliable. Finally, the 26 Al/ 21 Ne production rate ratio for Verkhne Udinsk of 0.16 AE 0.02 is too low. However, the ( 21 Ne/ 38 Ar) cos ratio of 0.16 is reasonable. Remember that the terrestrial age for Verkhne Udinsk was set to zero (see above). If this assumption is not true, this would explain the low ratio. However, a nonzero terrestrial age would increase the 36 Cl production rate and therefore reduce the CRE age, which in turn would reduce the 26 Al/ 21 Ne production rate ratio even more. Consequently, there is no coherent explanation for the low 26 Al/ 21 Ne production rate ratio for Verkhne Udinsk. Considering only the reliable data, for example, excluding Greenbrier County, Joel's Iron, and Lombard (see above), we calculate a weighted 26 Al/ 21 Ne production rate ratio of 0.32 AE 0.01 (n = 46), that is, slightly lower than the value used before. Using the same procedure for the 10 Be/ 21 Ne production rate ratio we calculate 0.44 AE 0.03 (n = 48), that is, lower than the value of 0.55 as used before (e.g., Lavielle et al. 1999). There is no dependency of the production rate ratios 10 Be/ 21 Ne and 26 Al/ 21 Ne on the ( 4 He/ 21 Ne) cos ratios, that is, on the shielding conditions. Figure 5 depicts the CRE age histogram for iron meteorites. In addition to the data from Table 2, we also included the CRE age for Twannberg of 193 AE 43 Ma, which differs slightly from the CRE age given by Smith et al. (2017). This change is due to a new terrestrial age (176 ka instead of 202 ka, see above) that changed the 36 Cl production rate and consequently the CRE age. In total, we determined CRE ages for 48 iron meteorites. In addition, we also considered the data from Lavielle et al. (1999) for Ainsworth,Bendego,Bohumilitz,Carlton,Huizopa,Lombard,Merceditas,Morradal,Nelson County,Norfolk,Surprise Springs,Yanhuitlan,and Yardymly (n = 13). For Brownfield, Grant, and Picacho, we used our data. For the discussion of the Lombard data, see above. From the study by Shankar et al. (2011), we used the CRE ages for Cape of Good Hope, Charlotte, Gibeon, Henbury, Hoba, and Tlacotepec (n = 6). We decided to also use the data from the two other studies because they are consistent with ours and it is therefore not expected that any bias is introduced.

THE CRE AGE HISTOGRAM
In total, the CRE age histogram consists of 68 iron meteorites (including Twannberg).
We have chosen a bin size of 60 Ma, which corresponds to about twice the average uncertainty of the CRE ages. The argument for the relatively wide bin size is as follows. First, with a normal distribution of the individual uncertainties of the CRE ages, the probability of assigning an age into a wrong bin is more than 60% if the bin size corresponds to 1r but is only 40-50% if the bin size is 2r. Second, a range of CRE ages from essentially zero up to 827 Ma and a bin size of 60 Ma results in 14 bins. With 68 iron meteorites in total, we have 4.9 meteorites per bin for a uniform distribution. Although this is still a low number, it might be sufficient to detect time-dependent structures.
At first glance, the CRE age histogram indicates that the delivery of iron meteorites to the Earth is not continuous but that most of the (studied) iron meteorites were produced in just a few major collisions. Based on the data shown in Fig. 5, one could argue for at least two peaks, one at 400 and at 630 Ma. For example, the 13 IIIAB iron meteorites Benedict, Boxhole, Cape York, Casas Grandes, Davis Mountains, Djebel-in-Azzene, Joel's Iron, Nazareth, Picacho, San Angelo, Tamentit, Verkhne Udinsk, and Merceditas have CRE ages that agree within their uncertainties. It is therefore possible that all 13 meteorites were ejected in one single large impact event on the IIIAB parent body, that is, that they are all launch paired. Typically one distinguishes launch pairing and fall pairing. Meteorites that are launch paired have been ejected In addition to the 48 data from this study, we also show data from the study by Lavielle et al. (1999) and Shankar et al. (2011). The data are not corrected for pairing.
from the same parent body and have been produced in the same ejection event. The fall on the Earth, however, could have been at different times; only the sum of the CRE age and the terrestrial age, typically called ejection age, is the same for launch paired objects. In contrast, fall paired meteorites originate from the same preatmospheric meteoroid that broke up into smaller fragments during atmospheric entry. These fragments have the same CRE age and the same terrestrial age.
Since the terrestrial ages for iron meteorites are usually much shorter than the CRE ages, we consider meteorites launch paired if they come from the same parent body, that is, the same chemical group, and if their CRE ages are identical to within 30 Ma or within the individual uncertainties (whatever is larger). With this assumption, we consider the following iron meteorites as paired (group, weighted average of CRE Using the individual uncertainties as weights for calculating averages is not without problems because some of the uncertainties are correlated (through the uncertainties of the production rates). However, since these uncertainties influence all data in the same way, they give similar weights and therefore do not compromise the calculated average. The CRE age histogram after correction for pairing is shown in Fig. 6. After considering pairing, there are 15 individual ejection events for IIIAB meteorites, seven individual events for IIAB iron meteorites, four individual events for IVA iron meteorites, and 15 individual ejection events for all the other studied iron meteorites. At first glance, there are still some indications for peaks in the CRE age histogram. However, the small number of events precludes statistically significant assessments; most of the peaks are not statistically significant, especially not if one considers each of the four groups separately. There is, however, an interesting result: the large number of iron meteorites (n = 13) that are likely ejected from only one impact event at~425 Ma (second group) indicate that this impact must had been massive. All other pairing cases contain fewer than five samples and, in most cases, fewer than three or two samples.

PERIODIC OR SUDDEN GCR VARIATIONS?
The major goal of this study is to explore whether or not the CRE age histogram for iron meteorites (Fig. 6) can provide any information about periodic and/or sudden GCR fluence variations. We first checked CRE age data for a periodicity of 134 Ma as proposed by Shaviv (2002Shaviv ( , 2003. In Fig. 7, we plot the CRE ages as a function of the CRE with a modulo 134. In such a plot, it means that for data having a period of 134 Ma, the data group around a given x-value, which depends on the phase of the GCR oscillation, that is, the data group close to this x-value but spread over the entire range of CRE ages on the y-axis. The data, however, indicate differently; there is a large spread of data over the entire x-axis, indicating that there is no apparent periodicity. We found the same result for periods of 147 Ma (Scherer et al. 2006) and for 400 and 500 Ma (Alexeev 2016). Consequently, from our data, there is no indication for a periodicity in the CRE age data, which is in clear contradiction to the proposals by Shaviv (2002Shaviv ( , 2003, Scherer et al. (2006), andAlexeev (2016). Furthermore, our result nicely confirms the earlier findings by Rahmstorf et al. (2004) and Jahnke (2005).
Going one step further, we can now compare our calculated CRE ages to available literature values for Fig. 6. Cosmic ray exposure age histogram for 67 iron meteorites corrected for pairing. After pairing correction, the data indicate 41 individual ejection events. The CRE ages are from this study, Lavielle et al. (1999), andShankar et al. (2011). 41 K-K CRE ages. In a previous study, Lavielle et al. (1999) highlighted a~28% discrepancy between 36 Cl-36 Ar and 41 K-K CRE ages. The upper panel of Fig. 8 Lavielle et al. (1999) and Shankar et al. (2011). The 41 K-K CRE ages are from Voshage (1967). Unfortunately, from the 68 iron meteorites with reliable 36 Cl-36 Ar ages and the about 80 iron meteorites with 41 K-K ages, there are only 20 iron meteorites where we have both 36 Cl-36 Ar and 41 K-K ages. The solid black line is the 1:1 correlation, that is, data plotting close to this line are for meteorites with identical or very similar 36 Cl-36 Ar and 41 K-K ages.
The lower panel in Fig. 8 shows the ratio of 41 K-K CRE age to 36 Cl-36 Ar CRE as a function of the 36 Cl-36 Ar CRE age. Most of the ratios are higher than 1, indicating that the 41 K-K CRE ages are higher than the 36 Cl-36 Ar CRE ages. The average value considering all ratios is 1.41 AE 0.07, that is, significantly different from unity. The finding of systematic discrepancies can have two reasons: first, there might be an offset between both age systems. Remember that the 41 K-K CRE age system is based on relatively crude assumptions about the production rate ratios of potassium isotopes. The production rates of potassium isotopes are difficult to model because there is little experimental data and nuclear model codes often fail to accurately describe production of magic or doubly magic isotopes, like the magic proton number 39 K, 40 K, and 41 K and the doubly magic nucleus 40 K. Second, it might also be that the observed systematic discrepancy indicates a change in the GCR fluence. The production rate ratio P( 36 Cl)/P ( 36 Ar) = 0.835 used here to determine the CRE ages is based on the assumption of the same GCR fluence for 36 Cl and 36 Ar (e.g., Lavielle et al. 1999;Ammon et al. 2009). Radioactive 36 Cl reaches saturation after about 1.5 Ma and, therefore, only records the GCR fluence over the last 1.5 Ma. In contrast, 36 Ar production is over the entire CRE. A low 36 Cl-36 Ar CRE age can therefore indicate that the GCR fluence over the exposure time of the iron meteorite was lower than the GCR fluence over the last 1.5 Ma. The increase in the last 1.5 Ma would have essentially no effect on the 36 Ar budget considering that CRE ages are all in the range of a few hundred Ma. The data may therefore indicate that the GCR fluence over the time interval studied by us, that is, from 195 to 656 Ma, was lower by up to 40% compared to the currently assumed GCR fluence. This finding is in accord with the results by Lavielle et al. (1999). However, our value is slightly higher, that is, 40% compared to 28%, but is based on a significantly larger database for CRE ages.
From Fig. 8 one could also argue that there was a change in the cosmic ray flux 400 Ma ago. The ratio below 400 Ma is 1.52 AE 0.15, that is, different from unity by 3.4r and the ratio above 400 Ma is 1.38 AE 0.07, that is, significantly different from one by 5.4r. Although both values agree within the uncertainties, the ratios below 400 Ma scatter more than the values above 400 Ma. The meteorites with very high ratios, that is, large differences between 41 K-K and 36 Cl-36 Ar CRE ages are Bendego (2.77), Huizopa (1.74), Treysa (1.90), and Charlotte (1.52). We consider the 36 Cl-36 Ar CRE ages for all four meteorites reliable and there are also no indications for questioning the 41 K-K CRE ages. If the ratios were different for meteorites with short or long CRE ages, it might be possible to further pin down the time of the GCR fluence change (if there were any). With the current database, however, the uncertainties for the 41 K-K CRE ages are too large to firmly establish exactly when the GCR fluence changed.
We conclude that there are no periodic GCR intensity variations but that it is possible that the GCR intensity over the time interval from 195 Ma to 656 Ma was up to 40% lower than the GCR intensity over the last few Ma. Erlykin et al. (2009)  GCR flux corresponding to a higher temperature. Assuming a linear anticorrelation between GCR intensity and Earth climate, a change of 40% in the GCR intensity could change the Earth temperature by up to 2°C. This being the case, the data would indicate that the Earth temperature in the time interval 195-656 Ma was~2°C higher than it was during the last few Ma (besides anthropogenic global warming). Temperature changes that large seem very unrealistic. We therefore speculate that the most likely explanation is a systematic bias between the 36 Cl-36 Ar and the 41 K-K dating systems.

SUMMARY AND CONCLUSIONS
We measured the He, Ne, and Ar isotopic concentrations in 56 iron meteorites from six different groups (42 IIIABs, 7 IIABs, 3 IVAs, 1 IC, 1 IIA, 1 IIB, 1 ungrouped). In aliquots, we measured the Ni concentrations using ICP-MS and the 10 Be, 26 Al,36 Cl, and 41 Ca radionuclide concentrations using AMS. From 41 Ca and 36 Cl activities together with model calculations (see Smith et al. 2017), we calculated terrestrial ages ranging from zero up to 562 AE 86 ka. The data confirm earlier results that terrestrial ages for iron meteorites can be as long as a few hundred thousand years even in relatively humid climate conditions (e.g., Chang and W€ anke 1969;Aylmer et al. 1988;Smith et al. 2017).
The CRE ages were calculated using 36 Cl production rates, that is, the activity at the time of fall, and the cosmogenic 36 Ar concentrations. The ages range from 4.3 AE 0.4 Ma to 652 AE 99 Ma. By also considering recent data for the large iron meteorite Twannberg (Smith et al. 2017) and the data from Lavielle et al. (1999) and Shankar et al. (2011), we established a consistent and reliable CRE age database for 68 iron meteorites. At a first glance, the CRE age histogram shows peaks at about 400 and 630 Ma.
We corrected the data for pairing and concluded that, for example, 13 of the studied IIIAB iron meteorites were ejected in one single large impact event on the IIIAB parent body. After pairing correction, the database consists of 41 individual ejection events: 15 for the IIIAB iron meteorites, 7 for the IIAB meteorites, 4 for the IVAs, and 15 individual ejection events for all the other studied iron meteorites. The updated CRE age histogram shows no indications for periodic structures, especially not if one considers each group separately. Our data therefore confirm earlier results also arguing that the CRE age histogram for iron meteorites shows no indications for periodic GCR intensity variations (e.g., Rahmstorf et al. 2004;Jahnke 2005). In this respect, our data together with the other studies clearly contradict the hypothesis of periodic GCR intensity   Lavielle et al. (1999), and Shankar et al. (2011). The 41 K-K CRE ages are from Voshage (1967). variations as proposed by Shaviv (2002Shaviv ( , 2003. However, the number of events is still too low to allow for statistically significant conclusions. The 36 Cl-36 Ar ages determined by us and others are systematically lower than the 41 K-K CRE ages (e.g., Voshage 1967). The discrepancy is in the range 40%, which is in accord with earlier findings by Lavielle et al. (1999) and Ammon et al. (2009). A possible reason can simply be an offset in the 41 K-K dating system, which is not well constrained (e.g., Ammon et al. 2009). Another possible explanation is that there was a significant change in the GCR intensity with a 40% lower GCR flux in the time interval 195-656 Ma than over the last few Ma. Erlykin et al. (2009) argued that GCR intensity changes in the range AE1.5% can change the Earth temperature in the range AE0.07°C. By assuming a linear dependence between GCR intensity and Earth temperature, a 40% lower GCR intensity could result in an Earth temperature about 2°C higher. Temperature changes that large during this time period seem unrealistic. We conclude that the discrepancy, or at least a significant part of it, between the 41 K-K and 36 Cl-36 Ar CRE ages is caused by incorrect assumptions in the 41 K-K dating system. Before we can safely conclude whether or not there was a GCR intensity change, there is a need to better understand and probably also to improve the 41 K-K dating system. The given uncertainties are only the uncertainties due to ion counting statistics, blank corrections, interference corrections, and to the extrapolation of the measured signal to the time of gas inlet into the spectrometer. Since noble gas mass spectrometers are optimized for isotope ratio measurements, the uncertainties caused by the extrapolation are very often much smaller for isotope ratios than for individual isotopes.

APPENDIX B CORRECTING COSMOGENIC 10 Be AND 26 Al FOR CONTRIBUTIONS FROM SULFUR AND PHOSPHOROUS
The light cosmogenic nuclide concentrations in iron meteorites, for example, 10 Be, 26 Al, can be compromised by contributions from S and/or P. Consequently, 10 Be and 26 Al production rates not only depend on shielding but also on the trace element concentrations of the studied sample. Since the latter are usually not known, the interpretation of 10 Be and especially 26 Al production rates is notoriously difficult and sometimes unreliable. While we were able to correct contributions from S and P for the production of cosmogenic Ne isotopes, such corrections are not possible for 10 Be and 26 Al. A logical step would be to apply the results obtained from the Ne data also to 10 Be and 26 Al, that is, to assume that the degree of contamination determined from Ne also apply, with some scaling factors, to the radionuclides. This, however, is not correct. First, noble gases and radionuclides were determined on different aliquots. Second, the extraction procedure for noble gases, that is, total melting of the sample, likely degasses the metal together with the troilite and schreibersite inclusions. In contrast, the extraction procedure for the radionuclides is optimized to dissolve the metal; traces of troilite and schreibersite are very likely not be completely dissolved. Therefore, the S and P concentrations determined from cosmogenic Ne are expected to be higher than the ones for 10 Be and 26 Al. Despite being not entirely correct, we nevertheless use the results from Ne for correcting 26 Al.
Doing so, we apply the percentage of the correction used for 21 Ne also for 26 Al, that is, if we had to correct the 21 Ne data by 5%, we also correct the 26 Al data by 5%. The data for 10 Be are not corrected because the mass difference between 10 Be and S/P is much larger than the mass difference between 21 Ne or 26 Al and S/P. Consequently, the same contaminating level of S and/or P induces a significantly smaller contribution to 10 Be than it does for 21 Ne and 26 Al. The thus applied corrections for 26 Al are usually below 10%, exceptions are Charcas (correction factor 1.10), Sikhote-Alin (1.16), Boxhole (1.18), Piñon (1.25), Squaw Creek (1.27), Arispe (1.43), and Lombard (1.66). The 10 Be, 26 Al (corrected), and 36 Cl concentrations at the time of fall, that is, the production rates, are given in Table 2. The given uncertainties include the uncertainties of the measured radionuclide concentrations and the uncertainties of the terrestrial ages. For 26 Al, we added half of the value of the correction to the final uncertainties. For example, the 26 Al production rate for Avoca after correction for decay during terrestrial residence but before correcting for contributions from S and/or P is (3.06 AE 0.08) dpm kg À1 . After correction for contributions from S and/or P, the value reduces to 2.91 dpm kg À1 , that is, 0.15 dpm kg À1 is from S and/or P. We subsequently added half of this value, that is, 0.075 dpm kg À1 to the final uncertainties, that is, the final data are (2.91 AE 0.15) dpm kg À1 (Table 2). Production rates for 41 Ca are not given because they are no longer independent. Due to the way we calculated the terrestrial ages, 41 Ca production rates are simply given by 1.157 times the 36 Cl production rates. The given uncertainties include the uncertainties from the AMS measurements and the blank corrections but not the uncertainties of the 10 Be, 26 Al,36 Cl, and 41 Ca decay constants and the uncertainties of the carrier masses, the latter two are usually negligible. The uncertainties for the Ni concentrations are typically less than 1%. bdl = below detection limit, n.d. = not determined.