An extensive recalibration of the Alpha Proton X-ray Spectrometer (APXS) was performed to improve the data evaluation of the Mars Pathfinder (MPF) rock and soil in situ measurements. Many samples and geostandards were measured with an MPF identical spare APXS under simulated Martian atmospheric pressure and low night temperature conditions. Precise calibration curves for major and minor elements were obtained. Using an improved spectral evaluation method for the original MPF spectra, refined and new elemental compositions of the MPF rocks and soils were derived. However, all initial conclusions about soils and rocks remained valid. The MPF soil composition revealed the mafic nature of the surface as inferred from the results of Viking 1 and 2 data. However, the MPF rocks reflect a felsic composition. For all samples, linear correlations of sulphur with all other elements were noticed, reflecting the fact that the rocks were covered with varying degrees of soil, which compared to the rocks is rich in sulphur. Extrapolating to low sulphur abundance (0.3%) the composition of a soil-free rock was calculated. The MPF rock composition is high in Si and K and low in Mg and Fe compared to the soil. Using elemental data from MPF and Martian meteorites, a global estimation of the Martian crust composition was derived. The crust is of basaltic nature with a high abundance of incompatible elements (K, Rb, Nd, U, Th) and volatile elements (S, Cl), while carbonates are lower or absent.
 The upcoming two Mars Exploration Rover (MER) missions (launch mid-2003) with their unique robotic arms for unprecedented multi-instrument measurements awakes renewed interest in the previous Mars Pathfinder mission at the Ares Vallis landing site. An Alpha Proton X-Ray Spectrometer (APXS) was attached to the Sojourner rover to measure in situ soils and rocks on the surface of Mars and derive their elemental compositions. Many sampling sites, soils, and rocks around the lander were approached, such as the rock “Barnacle Bill,” which was the first rock ever analyzed on Mars. Sojourner drove a traverse of ∼100 m during its longer than expected mission lifetime and extended the radius of investigations to a distance of ∼12 m from the lander [Golombek et al., 1999]. Solar power during the day and nonrechargeable batteries during the night kept the rover alive. Finally, the failure of the rover batteries terminated the nightly measurements of the APXS, while the rover still could operate during the day. The X-ray detector of the APXS needed cold night temperatures (below −50°C) to achieve its then unsurpassed high-resolution performance.
 Shortly after the mission was over, a preliminary elemental analysis of the Pathfinder rocks and soils was reported [Rieder et al., 1997b]. To improve the analysis, a thorough recalibration of the instrument was carried out [Brückner et al., 2001]. In the wake of the renewed interest, details about the reanalysis are presented.
2. Instrument Description
 The Alpha Proton X-Ray Spectrometer (APXS) is a small instrument to determine the elemental composition of a sample. The sensor head contains the detectors and the radioactive curium-244 sources (Figure 1). The electronics box, which was located inside the warm compartment of the rover, held the analog and digital boards. The APXS sensor head had a diameter of 52 mm and a length of 80 mm; the overall weight (sensor and electronics) was 0.6 kg, and the power consumption came to only 0.4 W.
 The sensor head was attached to the APXS deployment device located on the rear side of the rover. The deployment mechanism used one motor and spring loaded support rods to press the contact ring of the sensor head properly against sample surfaces, permitting automatic adjustment to the normal of the surface within 20°. Approximately vertical rock surfaces were preferred because of a seemingly lower dust load. For measurements of Martian soils the sensor head was rotated by 90° with respect to its normal (horizontal) position.
 The APXS has two modes: (1) X-ray spectroscopy (X-ray mode) and (2) Rutherford backscattering (alpha mode). The surface of a sample is bombarded by 5.8-MeV alpha particles emitted by Cm-244 and monochromatic X rays, mainly emitted by the Pu-244 daughter of Cm (L lines). In X-ray mode, two different X-ray production mechanisms take place: (1) alpha particle induced X-ray emission (PIXE) responsible for mainly low Z elements (Na to Ca) X-ray excitation and (2) X-ray fluorescence by Pu lines responsible for high Z elements (Ti to Ni) excitation. The X rays are detected by a silicon PIN diode that provides an energy resolution of ∼260 eV at 6.4 keV when cooled down to temperatures below −50°C. At temperatures above −20°C, the resolution degrades to a level, where complex X-ray spectra cannot be evaluated sufficiently well. The backscattered alpha particles are recorded by a thin alpha detector (thickness of 35 μm). Both detectors are collimated to limit the field of view. The alpha detector is covered on its backside by a thick (700 μm) silicon detector forming a detector sandwich that suppresses the charged-particle cosmic ray background. In addition, alpha-induced protons emitted by the sample are registered by a coincidence circuit of the sandwich detectors. However, the obtained proton signal is very weak, and no additional information is provided compared to the strong X-ray and alpha signals.
 The X-ray mode is very sensitive for major elements, such as Mg, Al, Si, K, Ca, and Fe and to a lower degree also for minor elements, including Na, P, S, Cl, Ti, Cr, Mn, and Ni. The alpha mode is sensitive to C and O and to a lesser extent to major elements with higher Z in vacuum; however, in gas of low pressure (<15 mbar), the sensitivity is reduced because the gas is an additional layer of matter that the alpha particles have to penetrate. On the Martian surface, there is an atmosphere containing 95% CO2 at a pressure of 5–15 mbar depending on season and altitude. The measured C and O alpha signals must be corrected for significant atmospheric interferences.
 The curium alpha sources were specifically designed for alpha backscattering measurements. They have to satisfy the following conflicting requirements: (1) very small spread of the original alpha particle energy (5.8 MeV) to obtain distinctive slopes in the spectra requiring very thin sources and (2) high alpha particle output to reduce the measurement times requiring thick sources. For a given source area a required specific activity (determined by the half-life of the isotope) results in a certain thickness, which in turn increases the energy spreading. In order to keep the curium stable, interelement compounds have to be used. The Pathfinder sources were made from curium silicides on semiconductor-grade silicon, which resulted in a full width at half maximum (FWHM) of 2.3% for the 5.8-MeV alpha peak. Nine sources with a specific activity of 15 mCi/cm2 (0.56 GBq/cm2), each, provided a total activity of 45 mCi (1.7 GBq). However, such source strength leads to rather long counting times.
 The sampling depth of the incoming and outgoing radiation is very shallow (on the order of 1 to several 10 μm) and depends strongly on the energy of the element. For X rays the Na lines result from ∼1 μm, while Fe lines sample more than 10 μm. Care has to be taken that the composition of the first 10 μm surface layer of a sample is representative of the bulk elemental composition, otherwise, the measured APXS results are misleading.
 The detection limit of the X-ray measurements depends on the excitation mechanism (alpha or X ray), cross section for K ionization (Z of element), and counting time for low concentrations. In addition, the 8 μm thin beryllium window in front of the X-ray detector lowers the sensitivity of elements below Si (absorption as function of energy), notably Na and Mg. In general, the detection limit varies between 0.1 to 1 wt % depending on the element.
 The APXS is rather insensitive to small variations of the geometry of the sample surface because all major and minor elements can be determined. Under perfect geometrical conditions, the oxides of all elements should sum to 100 wt % (within a small error). If the docking of the sensor head was not perfect or the surface not flat, the concentration can be normalized to 100% (closure to 100). More details of the instrument can be found in Rieder et al. [1997a].
2.1. X-Ray Spectra
 The APXS uses dual excitation processes: Alpha particles and X rays. The cross sections of the excitation reactions depend on energy of the exciting particle and atomic number of the target nucleus. Alpha particle excitation has a high K shell ionization cross section for low Z elements and strongly decreases with increasing Z. For X-ray excitation, the ionization cross section increases with increasing Z. The minimum of the two excitation cross sections is around the element Ti. The APXS has high sensitivities both for low Z (Si) and high Z (Fe) elements, a matchless feature resulting from its PIXE and X-ray fluorescence excitations. This feature is well reflected in the X-ray spectra of rocks (Figure 2), where Si produces the highest peak and Fe the second highest.
 The logarithmic scaled X-ray spectra of the Martian rock Yogi (A-7) and the soil Mermaid Dune (A-15), Figure 3, reveal the smaller peaks from minor elements, such as S, Cl, Ar, Ti, and Cr (only for Mermaid Dune). The Ar peak results from the thin layer of Martian atmosphere between detector and sample. No Ni peak was observed in rock and soil samples on Mars. Note that plots of all spectra are normalized to a measurement time of 1000 s for ease of comparison.
2.2. Alpha Spectra
 A typical alpha spectrum of a Martian rock, in this case Barnacle Bill, the first in situ measured Martian rock, is shown in Figure 4. An increasing channel number corresponds to increasing alpha energy. However, the absolute energy value of the alphas is of no importance; therefore the x axis is always labeled “Channel.” There are two distinctive structures in the alpha spectrum: peaks and steps. According to the rules of Rutherford, alpha particles, which are back-scattered (at an angle of ∼180°) by a nucleus, have a distinct energy: The higher Z of an element, the higher is the energy of a backscattered alpha. If the sample is very thin (<1 μm), a peak is produced in the spectrum. If the sample is infinitely thick, the alphas can be scattered in different layers of the sample. The sum of all alphas scattered at a particular element form a step, because the alphas lose energy by entering and leaving the sample according to the stopping power of the sample. The steps are described by the height of their plateau and the position of their edge. Their edge position is characteristic for an element, while their height reflects its concentration. In Figure 4, two edges are labeled as Si and Fe, while a strong peak is labeled as C. The carbon peak results from alpha particles scattered in the thin CO2 atmosphere (there is no C in the sample). The oxygen signal is an overlay of a step (O of the sample) and a small peak (O in CO2). The C peak is so large because of additional resonant alpha scattering (also strongly energy-dependent).
 The difference between spectra taken in vacuum and thin gas is demonstrated for a terrestrial rock sample (Figure 5). The alpha spectra were recorded by an instrument identical to the MPF APXS. In gas the position of the edge is somewhat lower compared to its counterpart taken in vacuum. The thin layer of gas reduces the mean energy of the incoming and backscattered alpha particles. The prominence of the C peak illustrates the strong interference of atmospheric carbon to potential carbon in the sample.
 In comparison to the terrestrial alpha spectra, which were commonly measured over several days to obtain good counting statistics, the Martian alpha spectrum could be measured in only several hours. A 6.5-hour alpha spectrum of the Martian rock “Wedge” is shown in Figure 6. Idealized spectral components are included for better understanding. The vertical edges of the steps are deformed by the spectral response function of the instrument. Important to note is that left of the atmospheric C peak an additional step of carbon (of the sample) on top of the oxygen plateau could be located. In case of Wedge, removal of the C peak interference revealed that no carbon step was present.
2.3. Measurement Chamber
 For the recalibration of the Pathfinder APXS, an identical sensor head was placed into a special “Mars simulation chamber.” The different samples to be measured could be positioned in a well-defined, reproducible distance from the detectors. The chamber could be evacuated below 10−3 mbar or filled with a gas at low pressure (around 10 mbar).
2.4. Instrument Performance
 To control precision work of the APXS in the laboratory, checks of the instrument performance were done from time to time. Two main degradations of performance can occur: degradation of energy resolution and degradation of efficiency. A tantalum plate test sample was measured regularly to assess the performance of the instrument under replicate conditions.
 Degradation of energy resolution can be the result of increased noise level, produced either in the detector or the signal electronics chain. Energy resolution is also a function of detector temperature. To assess energy resolution changes for the X-ray mode, a reference peak, the 8.1-keV Lα line of Ta was measured.
 Degradation of efficiency can occur if the signal strength is weakened by contamination of the exit windows of the alpha sources, the beryllium windows of the X-ray detector, or the surface area of the alpha detector. For the X-ray mode, efficiency was assessed by examining the intensity ratio of a low- and a high-energy line of Ta, knowing that a contaminating film absorbs the low-energy line more than the high-energy one.
 On Mars, no test sample was available to obtain independent data on the instrument performance. However, the X-ray and alpha spectra can be assessed to provide basic information on the performance. The overall shapes of the X-ray and alpha spectra of all samples were compared with each other and no substantial changes in resolution and position of peaks and steps were observed. As an outstanding example, the energy resolution of the Fe peak and its position in the X-ray spectrum was monitored for all samples over the mission period. Both parameters agreed very well within the error of determination, an excellent performance of the instrument under the harsh Martian temperature conditions. As the temperatures varied between −5° and −65°C during day and night, the temperature stabilized gain control worked very well.
 Meaningful measurements during the day were not possible because of largely increased noise in the X-ray spectra. The X-ray detector was covered by an 8 μm thick beryllium window, thick enough to bloc incident daylight completely, which was tested on ground successfully. However, on Mars, it seemed that scattered light hit the X-ray detector surface somehow and induced excess noise. During night operations no excess noise was observed. It was never found out what caused the unexpected sensitivity for daylight. The overall outcome was that meaningful measurements could be performed only during night hours. Therefore the APXS measurements were tied to the power of the rover batteries. After a lifespan much longer than expected, these batteries died due to repetitive exposure to very cold temperatures. The rover continued to operate during the day using its solar cells, but worthwhile APXS measurements could not be obtained anymore.
3. Elemental Composition Determination
 The main objective of the recalibration was to improve the determination of elemental composition of the Pathfinder samples. A set of geostandards, measured in artificial Martian atmosphere, was used to determine the accuracy of the instrument. The outcome of the recalibration effort was a set of improved calibration curves for each element under question. These curves describe the relationship between measured peak areas of each element and their corresponding concentrations for the X-ray spectra.
3.1. Recalibration Procedure
 The following elements were measured during the recalibration effort of the X-ray mode: Na, Mg, Al, Si, P, S, Cl, K, Ca, Ti, Cr, Mn, and Fe. As the gas argon is only found in the Martian atmosphere, no calibration was carried out for that element.
 Different types of samples were used for different purposes. Single-element samples and two-element compounds were mainly used for peak shape analysis. To minimize matrix effects of different compositions, multielement samples of similar matrix were chosen for concentration measurements. Siliceous material (like a rock) contains roughly 50 wt % oxygen and ∼20 wt % silicon, which adds up to 70%, leaving only 30% for the other elements. To establish elemental calibration curves certified geological rocks, validated terrestrial rocks, and validated meteorites (simply referred to as “geostandards”) were selected for measurements. The geostandards were the basic tool to ascertain the accuracy of the calibration.
 Such geostandards were selected for the recalibration that did span the range of elemental concentrations expected for Martian igneous rocks. A listing of the used geostandards and their elemental concentrations are provided in Table 1.
Table 1. Preferred Elemental Concentrations in Weight Percent for Used “Geostandards”a
The values of the highest and lowest concentration, which bound the concentration span, are indicated by H and L, respectively.
 Not all geostandards used were certified by a bureau. Therefore the remaining ones had to be validated. Validation in this context meant that two or more independent analytical methods were applied to determine the elemental composition of these samples and that their results agreed within the error limits. The rocks and meteorites used for the validation were measured in the Mainz laboratory (MPICh), Germany, the laboratory of the Institute for Geochemistry of the University of Frankfurt, Germany, and other quoted labs. Certain certified rock standards were treated as unknown samples. Their concentrations agreed very well within the error limit with the certified values.
 The following analytical methods were used to determine the concentration of elements in geostandards: Mainz laboratory: Instrumental Neutron Activation Analysis (INAA), Lecoautomat RC-412 for determination of carbon and H2O, and Carbon-Sulfur-Analyzer 5003; Frankfurt laboratory: X-Ray Fluorescence analysis (XRFA).
 For most of the measured samples it was found that the concentrations obtained by different analytical techniques agreed well within the error limits for most of the elements. In these cases the obtained measured concentration values were taken as “true values.”
 Treatment of each powdered sample began with thorough grinding, followed by sieving through a 50-μm mesh. The sieved powder was then subjected to a long duration mechanical mixing process before extracting an aliquot for measurements. The powdered sample was filled into the depression of a copper plate that provided a sufficiently large surface diameter, and the surface was smoothed to avoid geometric effects.
 A key feature of all APXS measurements is the microscopic penetration depth of the exciting radiation. A special process of sample preparation was developed to minimize moisture adsorption on the sample surface, which could degrade the quality of the analysis. The powdered samples were poured into the depression of a copper plate and placed into a vacuum oven. The gas pressure of the oven was slowly reduced while the sample was closely monitored through a window. Sudden unexpected gas releases by the sample could occur, when the degassing process was too fast, resulting in spilled powder all over. When a pressure of ∼10−3 mbar was reached, a heater was switched on to drive off all remaining adsorbed moisture. A temperature of 110°C in vacuum was maintained for at least several hours until the Mars simulation chamber was ready for the next sample. Then, the hot copper plate containing the sample was quickly transferred into the chamber. Again, the gas pressure was slowly reduced in the chamber to preserve the integrity of the sample surface. When a good vacuum was achieved, a gentle gas flow was established to achieve a continuous gas pressure of 6.5 mbar.
 The samples were measured in the Mars simulation chamber. The simulated Martian atmosphere consisted of CO2 (95%), N2 (3%), and Ar (2%). The gas pressure in the chamber could be lowered to any value above 1 mbar using a membrane pump or below 10−3 mbar by an oil-free turbomolecular pump. A steady flow of gas was applied to avoid static conditions under low pressure. A feedback system guaranteed a constant pressure over a time period of several days. The 6.5-mbar atmospheric pressure measured during the Mars Pathfinder mission was normally applied to the samples.
 To simulate the nightly conditions on the Martian surface during the mission, the detectors had to be cooled, in particular the X-ray detector whose performance was very dependent on the operating temperature. Therefore in the simulation chamber there was a provision to cool down the alpha and X-ray detectors to temperatures of −60° to −70°C.
3.2. Spectra Processing
 The X-ray mode of the APXS is sensitive to major elements and to many geochemical important minor and trace elements including chlorine, chromium, manganese, and phosphorus. Owing to the limited energy resolution of the used X-ray detector, the spectral unfolding, especially in the low-energy region, had to be improved. Originally, the X-ray spectra were fitted using simple Gaussian peak shapes as first approximation. Now, the exact peak shapes were derived from the measurements of single- and two-element samples and a library spectrum of Kα and Kβ peaks was created for each element. Potential small variations of the Kα/Kβ peak area ratios were neglected. The Martian spectrum under investigation was divided in three major regions of interest and each region was fitted separately. A special algorithm using the library spectra for each element and a general background spectrum fitted the measured Martian spectral region. The fit provided goodness of fit by chi-square, peak area, and error for each element and the residuum between data and fit. This allowed the operator to add or remove weak peaks depending on the overall quality of the fit.
 In particular, the evaluation of Na was very difficult, because the energy of Na (1.04 keV) was very close to the lower threshold of the spectrum and its intensity was low due to the absorption by the entrance window of the detector. Estimations had to be used in some cases, where the fitting had problems with the weak Na peak. The deconvolution of the Na, Mg, Al, Si, P, and Cl peaks required very careful fitting because in this region the peaks were all smeared out due to the limited energy resolution. A resolution of 230 eV at 1.7 keV is still not sufficient for this crowded region of interest. Some weak peaks were strongly distorted by large peaks, e.g., the phosphorus peak sitting on the huge flank of the Si peak (Figure 7).
 Several peaks had to be corrected for interferences. In silicon detectors, so-called escape lines that are 1.74 keV lower compared to the parent peak are produced. If the parent peak is large, then the escape peak can contribute significantly to the peak area of a weak peak at the same position. The Ca escape peak at 1.95 keV interfered with P at 2.1 keV and the Fe escape peak at 4.66 keV with Ti at 4.50 keV. The corrections were done by using measured escape peaks adjusted to the size of the parent peak.
 In addition, the peaks revealed a nonperfect Gaussian shape when cooled to below −40°C. Careful inspection showed that at the low-energy side of a peak a non-Gaussian shoulder existed that changed shape as function of energy (Figure 8). This shoulder resulted from incomplete charge collection in the detector volume. A weak peak in front of a large one could contain a considerable amount of counts resulting from the shoulder of its huge neighbor. Using the measured shape of a peak for fitting resolved this problem for all cases. In particular, the Fe shoulder contributed ∼50% of all counts in the Mn peak.
3.3. Foil Correction
 As a general protection of the environment against recoil sputtered debris of the curium sources (a well-know process during radioactive alpha decay), a very thin alumina/VYNX foil was positioned in front of the alpha sources. This thin foil sticking on a support grid had a limited lifetime. To protect the foil during cruise phase, a shutter was mounted between source and grid. Only during operation of the APXS should the shutter have been opened; otherwise it had to be closed. During testing of the Pathfinder APXS at Jet Propulsion Laboratory a failure occurred in the electronic driver circuit for the motor, which actuated the shutter in front of the alpha sources. As this failure remained unexplained, the project decided to discard this circuit and fly the instrument with an open shutter. Consequently, the thin foil and its support grid were removed.
 However, all calibrations were carried out with this foil, together with its support grid, to prevent contamination of calibration samples. The support grid reduced the flux of alpha particles by ∼15%, whereas the attenuation of the X-ray flux from the source was almost negligible. This reduction of alpha and X-ray fluxes during calibration had been neglected in the preliminary analyses [Rieder et al., 1997b]; a foil correction factor was introduced for the new evaluation of Pathfinder spectra.
3.4. Calibration Method
 To improve the former calibration [Rieder et al., 1997b], exact duplicates of the MPF APXS sensor head and main electronics were reactivated and operated for the recalibration program. The set of alpha sources that was used in the laboratory was manufactured in the same batch as the flight sources that went to Mars. Therefore the source strength and the energy resolution were similar. After having completed the reanalysis of the Martian samples, it turned out that the sums of all oxides were not too far off 100 wt %; some were above 100, others below. Remembering that the docking of the APXS on Mars was never perfect, i.e., the mean distance of the detectors to the samples was always larger than the nominal distance in the lab, the new data revealed that the source strength of the laboratory sources was really comparable to those flown.
 The recalibration focused on careful measurements of complex geostandards that were preferred over simple elemental standards. Evolved fundamental parameter methods to correct numerically for matrix effects were not applied, because the shape of the obtained calibration curves confirmed the experimental approach.
 The known concentrations of each element for different samples were plotted versus measured peak areas obtained from the X-ray spectra. The fit of these data points is called a “calibration curve.” It turned out that in most cases the curve is a straight line that can be extrapolated closely to the origin of the coordinate system. If the curve is very close to the origin, the fit of the straight line was forced to pass through the origin emphasizing the fact that an element with zero concentration should not produce a signal (Figures 9–11). These curves should not be extrapolated to much higher than measured concentrations, because nonlinearities will occur for samples with completely different matrices. In the case of silicon, nonlinearities appear above and below our measured range (no linear extrapolation to zero). This reflects a stronger matrix effect for silicon compared to other elements. The fact that most of the data sets can be fitted by straight lines passing through zero reflects the similarity of the matrix of the selected geostandards. Within the scatter of the data, no noticeable matrix effects can be seen, except for Si. A simple numerical modeling of the expected intensities for major elements confirmed this observed trend.
 The Ti curve deviates from the origin: for 0.2% TiO2 weight percent the peak area goes to zero. This can be explained by iron interference: The Fe escape line of 4.66 keV interferes seriously with the 4.51 keV Ti line. This Fe interference was corrected for, apparently it was somewhat overdone removing too much from the weak Ti signal. Therefore the Ti peak area reaches zero before the concentration is zero.
 To compare the derived calibration curves, the slope of the linear fits can be plotted versus Z (Figure 12). In fact, the slope is the number of measured counts per time and per weight percent of each element, i.e., the sensitivity of the detection. A large number means a strong signal is obtained per unit time and unit concentration. Looking at the curve of Figure 12, several observations can be made. Between Si and Ti the curve declines reflecting the fact that the K ionization cross section of alpha excitation decreases with increasing Z. Beyond Ti, the curve rises again, because the ionization cross section of X-ray excitation increases with Z. Between Na and Si, the curve steeply falls as the result of attenuation of X-rays in the detector entrance window (8 μm thick). The lower the Z, the lower is the X-ray energy and the higher the attenuation.
 There are a few data points that seem to deviate from the general trend. The peak area of P was difficult to evaluate because of the strong Si interference leading to an overestimation. For Cl concentrations, only one standard was available leading to a bias. However, no supplementary standards could be measured, because the work had to be terminated. The foils on the alpha sources started to deteriorate producing an uncontrolled attenuation of the alpha particle energy. Therefore P and Cl can be considered as outliers. The Mn data point also seems to be a little bit off. This may stem from the strong interference of the low-energy Fe shoulder resulting in an overcompensation of the Fe counts. The values for P, Cl, and Mn are the best we have. The otherwise smooth trend of the sensitivity curve is a good indication of a reliable calibration. This curve indicates that the detection limit is different for each element.
3.5. Elemental Composition of Samples
Table 2 contains the elemental composition of the MPF soil and rock samples normalized to 100%. The normalization is necessary because of the nature of the APXS docking process and the unknown surface morphology. The average error for each element is derived from the errors of peak areas and calibration curves. The large peaks, i.e., high concentrations, have the smallest errors, while elements that are close to the detection limit of the instrument, show large errors up to 50%.
Table 2. Concentration of Mars Pathfinder Samples in Weight Percent, Measured by APXS, Based on Recalibration, Normalized to 100%, and Corrected for Foilsa
Average error is in relative percentage and holds for soil and rock samples. Note that iron is quoted as Fe3+ for soils and as Fe2+ for rocks. The soil-free rock is derived from the measured rocks by assuming an S concentration of 0.3%. APXS, Alpha Proton X-ray Spectrometer.
A-8, Scooby Doo
A-3, Barnacle Bill
A-18, Half Dome
Rel. Error (%)
 As a result of the recalibration, the Si concentrations of the MPF samples fell by ∼10%, while the Fe concentrations rose by ∼25% compared to the preliminary data by Rieder et al. [1997b]. The modification of the composition was sufficiently small that the major observations of Rieder et al. [1997b] remained valid.
 No Ni was detected on the Martian surface. Our detection limit of ∼0.1 wt % is the upper limit of the amount of meteoritic component on the surface.
4. Martian Rocks and Soils
 As evident from the MPF panorama camera images, the surfaces of the rocks at the landing site are covered to varying degrees with adhering dust [Golombek et al., 1999]. The APXS method cannot discriminate between rock surface and adhering dust as everything within the field of view of the sensor head contributes to the measured signal. To determine the amount of soil that was transported by the wind to the rock surfaces, several steps were taken.
 The analysis of the Pathfinder soils showed a high concentration of sulphur, similar to the Viking landing sites. As most rocks do not contain much sulphur, S was used as a marker for the amount of soil seen by the APXS. Therefore the concentrations of elements of rocks and soils were plotted against the sulphur content. Most data sets could be easily fitted with a straight line (Figures 13 to 17). Mg, Al, Si, Cl, and K show strong correlations, while Ca, Ti, and Fe show weak correlations. Phosphorus does not have any correlations. The differences in the correlations of the measured elements with sulphur was first emphasized by McLennan  and interpreted as complex mixing and transport of mineral phases. For example, sedimentary transport could cause separation of heavy minerals like ilmenite or magnetite from the soil. Furthermore, a possible transport of Fe2O3-rich dust cannot be ruled out.
 The camera observations indicate that airborne soil is sitting in varying amounts on the rock surfaces and suggest a physical mixing between soil and rocks. Since no close-up images of the measured samples are available, nothing can be concluded about their texture. However, there is the fact that the compositions of the rocks are different from the soils. Keeping in mind the small penetration depth of a few micrometers for the radiation, the surface coverage of the rocks was not a contiguous layer of soil; rather soil patches were trapped in little pores or openings, while other areas were clean of soil.
 The mean soil shows a sulfur concentration of 2.7 wt %, while the S concentration of the Pathfinder rocks varies from 0.8 to 2.0% and exceeds by far the concentrations accommodated in magmas or igneous rocks. As shergottites (the most abundant type of Martian meteorites) contain S between 0.13 to 0.32%, and as the Pathfinder rocks seem to be more fractionated than shergottites, an upper S concentration of 0.3% was adopted for the Pathfinder rocks. Using this value, a plausible soil-free composition of the rocks could be calculated from the linear regressions of the rock and soil compositions versus sulphur (Figures 13–17). Of course, the uncertainty of the amount of S in the rocks remains. A sulphur value of zero would lead to a very low concentration of 1% for Mg, indicating that some S is present inside the rocks. Whatever reasonable S concentration is adopted, the composition of the rocks is different from the soils.
 On Mars, very slow weathering processes over millions of years should produce some debris around old rocks. Therefore the soils should reflect the elemental composition of the local rocks. The different composition of the rocks elucidate that the global dust storms and local dust devils (observed by the Pathfinder and Mars Global Surveyor cameras) are powerful enough to create a rather homogenous soil composition independent of geologic provinces.
5. Search for Carbonates
 The main contribution of the alpha mode to the general analysis is its ability to detect and determine carbon, which cannot be measured by the X-ray detector because of its thin (8 μm) beryllium window and the low-energy threshold. All other elements, except O, are much better determined by the X-ray mode. Therefore the alpha mode was mainly used to search for carbon.
 As already pointed out, the Martian CO2 atmosphere produces a strong interference for C and O. Oxygen content is sufficiently high in all siliceous samples, while high C content is rarely found except for carbonates. The interference on C and O was systematically investigated in the laboratory.
5.1. Carbon in Samples
 To determine the detection limit of the carbon signal in the alpha mode, two carbon-bearing meteorites were used. The carbonaceous CM chondrite Murchison containing organic substances, which are volatile, has a carbon concentration of 1.71 wt %, as determined in our Mainz laboratory (in comparison to 1.85% by Jarosewich ). An aliquot of the material was used for APXS measurements. This sample was heat treated as described in the paragraph on the preparation of geostandards. To be sure that these treatments do not affect the C concentration by loss of volatile components, an aliquot was taken from the APXS sample after the measurement was completed. The C content of this aliquot was 1.93%. The mean value of the three values is 1.83% with an error of 5.0%, which gives sufficient confidence in the C analysis.
 The second meteorite, the carbonaceous CV chondrite Allende, contains less C than Murchison. Jarosewich  quotes 0.29% C; our first analysis gave 0.29%; and our second analysis of an aliquot of the APXS sample after the measurement resulted in 0.32%, again with an error of 5.5%.
 To duplicate the interference correction procedure of Pathfinder samples, the Murchison and Allende samples were measured under simulated Martian atmosphere. After removal of the interfering carbon peaks, whose shape was obtained by comparing measurements in vacuum and gas, the two spectra are shown in Figure 18. The residual spectrum of Murchison shows a signal from 1.8% carbon above the horizontal black line (derived from the oxygen plateau), while for Allende no additional signal is found. The 0.3% carbon of Allende is already below the detection limit, even in the favorable laboratory case. On the basis of the counting statistic of the residual spectra a detection limit of 0.8% was estimated for carbon.
5.2. Carbon in Martian Soils and Rocks
 A similar procedure to search for carbon as done in the laboratory was used for the Mars Pathfinder samples. No vacuum and gas spectrum of the same Pathfinder sample was available. Therefore a possible link between Mars and Earth samples was examined. A sample, called SSK1.1, which is an andesitic rock [Pearce et al., 1995], turned out to have a very similar elemental composition as the Pathfinder rocks, especially to the rock Shark. Shark had the least surface contamination by Martian soil compared to the other rock samples. Its alpha spectrum was assumed to be very similar to SSK1 and of no carbon content, which later was verified recursively.
 The CO2 corrected SSK1 spectrum was used to derive the Martian atmospheric C interference. The derived C and O peak shapes were used to remove the interference from the Pathfinder rock and soil spectra. Owing to the subtraction small dips did arise at the locations of the sharply rising and falling slopes of the C peak. However, they did not disturb the left part of the C plateau (C signal from sample). Two examples are given in Figure 19, which show the rock Barnacle Bill and the soil A-10 alpha spectra corrected for CO2. No carbon signal from the sample could be found as for all other samples. At the Mars Pathfinder landing site the carbon content of rocks and soils must be below the detection limit of 0.8%.
6. Martian Surface Chemistry and Implications for the Crust
 Data obtained by the APXS yielded a huge compositional difference between soils and rocks at the Mars Pathfinder landing site, whereas the soils at the landing sites of Viking 1 and 2 and Pathfinder are chemically identical [Rieder et al., 1997b]. The refined analysis (Table 2) confirmed many observations obtained from the previous work. Although the soils are compositionally similar to those at the Viking landing sites, the rocks, with their high Si and low Mg concentrations, represent highly differentiated crustal material similar to what is found on Earth. The felsic compositions of the MPF rocks are in contrast to the concept of a rather mafic Martian surface that was inferred from Viking soil data and the composition of Martian meteorites. Martian meteorites are a special group of meteorites, the so-called SNCs (Shergottites, Nakhlites, Chassigny, and Orthopyroxenite). These meteorites originate from Mars with very high probability.
 The Pathfinder rocks can be classified as andesite, or more precisely as icelandite [McSween et al., 1999]. However, it does not mean that they are subduction-related by origin, it articulates more the notion that Si is enriched compared to basaltic composition. Alternative formation scenarios are also possible. Wänke et al.  suggested that melts similar to Pathfinder soil-free rock composition could have been produced during the fractional crystallization of Shergotty-like primitive magmas. Crushing the Pathfinder rocks cannot account for the chemistry of the soils, even if weathering and addition of evaporates like sulfates and chlorides are taken into account. The most plausible interpretation for the high concentrations of Cl and S in the Martian soil is the formation of sulfates and chlorides by the interaction of volcanic gases with the surface material [Clark, 1993].
 The most striking feature of the preliminary soil analysis was the similarity of the Pathfinder and Viking 1 and 2 soils, a fact that is unaffected by the refined data. As Figure 20 shows, the general pattern of concentrations is like a fingerprint. Al, Cl, Ca, and Fe concentrations are very similar, while the MPF Mg and Ti concentrations are somewhat higher and the Si and S values lower compared to Viking 1 and 2. Only for potassium is there a substantial difference. The investigators of Viking (XRF spectrometer) reported an upper limit of 0.12% K [Clark et al., 1982], whereas Pathfinder found 0.5% K, a value four times higher. Assuming the Viking detection limit is valid, the K concentration of the Pathfinder soil is indeed enriched. The reason could be the weathering of the local Pathfinder rocks, whose soil-free K concentration is 1.1%. Another possibility is that Clark et al.  might have overestimated the sensitivity of their Viking X-ray detector for K. It could well be that the K concentration of the Martian surface is in certain regions higher than at the Viking sites. The gamma ray spectrometer of the Phobos 2 mission measured values of 0.3% to 0.6% K [Trombka et al., 1992; Surkov et al., 1994]. Recent K maps obtained by the Mars Odyssey gamma ray spectrometer [Boynton et al., 2003] also show large variations in K on the Martian surface ranging from nearly 0.2 up to 0.7% using the Pathfinder soil K value as ground truth.
 There are observations from the orbit by the Thermal Emission Spectrometer (TES) of the Mars Global Surveyor spacecraft [Bandfield et al., 2000] that lead to the classification of two specific surface types: called type 1 and type 2. Type 1 is compatible with basaltic rocks, while type 2 is enriched in Si and K [Boynton et al., 2003]. There are several interpretations about the nature of type 2 rocks: andesite [Bandfield et al., 2000] or weathered basalt [Wyatt and McSween, 2002]. Since gamma rays sample a rock to tens of centimeters depth in contrast to the APXS, which measures only the very upper layer of a rock, e.g., the weathering rind, it can be concluded that the weathered material is rather thick. Vast layers of sediments that formed in a paleo ocean cannot be excluded. Further evidence is needed to really understand type 2 material.
 In this context, it should be emphasized that the MPF rock composition has no similarity to known Martian meteorites, which would be related to type 1 material. Pathfinder rocks could be related to type 2 material.
 The composition of soils was interpreted by Brückner et al.  and Wänke et al.  as a mixture of weathered local rocks and a more mafic component compositionally similar to that observed in Martian meteorites. The mixing calculation was consistent with five components that are necessary to explain the composition of the MPF soil. The major component consists of felsic material followed by mafic material and ∼20% of salts and iron-bearing phases (Figure 21).
 Similar chemical composition of soils at Viking and Mars Pathfinder landing sites and the equatorial region that was analyzed by the gamma spectrometer of the Phobos 2 spacecraft [Surkov et al., 1989, 1994] strongly indicates a good mixing of surface material on a global scale. Therefore the average soil composition can be taken as the mean composition of the near surface crust on Mars [McLennan, 2001].
 On Earth, well-mixed average sediment is remarkably similar to upper continental crust estimates for the major elements [Taylor and McLennan, 1981]. On Mars, the similarity of the chemical composition of soils from Chryse, Ares Vallis, and Utopia may well serve the same purpose and permit an estimation of the average Martian near surface crustal composition.
 As a first step we suggest the mean soil data from Ares Vallis as an approach to the prediction of a Martian crust composition. Neglecting the extremely high S and Cl concentrations of the soil, the similarity in the chemical composition of the soil and the basaltic shergottites is remarkable as shown in Table 3. However, the estimated Martian crust has a higher K content indicating an enrichment of incompatible elements in the crust similar to Earth. This higher K concentration on the Martian surface might suggest an enrichment of Rb in the crust, which should result in more radiogenic Sr in the Martian crust, as found in the shergottites [Jagoutz, 1991]. Assuming a chemically homogeneous Martian crust, we estimated the Rb and Sr content of the crust from observed K/Rb and K/Sr ratios in Martian meteorites and the postulated crustal K content (Figure 22). A good correlation between the highly incompatible elements K and Rb was found for all Martian meteorites, independent of their rock type.
Table 3. Comparison of Mars and Earth Chemistry Showing Refined Data (Table 2) of MPF Soil-Free Rock, Average Soil, Average Soil Minus SO3 and Cl, Concentrations of Basaltic Martian Meteorites, Average Composition of Terrestrial Continental Crust, and Composition of an Andesite from the South Sandwich-Islanda
Oxide, wt %
MPF “Soil-Free” Rock
MPF Soil Average
MPF Soil Minus S and Cl
Basaltic Martian Meteorites
Earth Continental Crust
MPF, Mars Pathfinder.
 The SNC K/Rb ratio of 230 matches the terrestrial “main trend” (K > 1%) in unmetamorphosed rock types [Shaw, 1968]. However, for terrestrial rocks with low K concentrations (<1%), Rudnick et al.  found higher K/Rb ratios compared to the main trend. In Martian meteorites and soils, sampled so far, the K content is below 1% and the K/Rb ratios match the terrestrial main trend.
 Suggesting the same K/Rb systematic for the Martian crust as observed for the SNCs, we calculated with the known K concentrations from the Pathfinder and Phobos missions the respective Rb contents (Table 4). Furthermore, Sr and Nd abundances in the crust can also be derived from the observed K/Sr and K/Nd systematic in SNCs and the K contents of the Martian surface (Table 4).
Table 4. Estimated Concentrations of Some Elements in Martian Crust, K in Percent, and Rb, Sr, and Nd in ppma
 In this model of the Martian crust, the abundances of large ion lithophile (LIL) elements like Rb and Nd depend on the assumed crustal K content (0.3–1.1%) and the thickness of the crust. Geophysical data determined by Mars Global Surveyor derived crustal thickness values of ∼50 km [Zuber et al., 2000], 60 ± 24 km [Wieczorek and Zuber, 2002], and 30–100 km [Nimmo and Stevenson, 2001]. A mass balance model based on Nd isotopic compositions and rare earth element (REE) abundances in Martian meteorites by Norman  gives a value of 20–30 km crustal thickness and an Nd concentration of ∼34 or 23 ppm, respectively. This would imply that 55% of the total Nd is in the crust. In a revised model, implied by an ultradepleted mantle with an Nd isotopic composition like that of the Martian meteorite QUE 94201, Norman  postulated a crustal thickness of 85 km. Such a thick crust with a composition of radiogenic heat producing elements (K, U, and Th) similar to those derived by Mars Pathfinder (this paper) and Mars Odyssey [Taylor et al., 2003] would contain more than 100% of these elements. Therefore Norman  suggested a multistage formation of the Martian crust with geochemically enriched and depleted components as a possible scenario.
Table 4 reports our calculations of the concentrations of K, Rb, Sr, and Nd for a 30 km thick crust. For these estimates the bulk Mars composition of [Wänke and Dreibus, 1988] was used. It can be assumed that U and Th have a similar enrichment in the crust as estimated for K and Th [McLennan, 2001, 2002]. As a consequence, all the radiogenic heat producing elements would be stored in the K-rich (1.1%) crust of 30 km thickness for 4.5 Ga, which was also discussed by McLennan  recently. Assuming for Mars a similar distribution of heat producing elements between crust and mantle of ∼50%, as found for the Earth, the estimated crustal composition based on the MPF soil provides reasonable values, such as 0.5% K and 17 ppm Nd.
 Recently, Nyquist et al.  and Borg et al.  favored a more evolved Martian crust, which should be partly assimilated by the basaltic shergottites during their emplacement. In their assimilation model, Nyquist et al.  proposed that the basaltic shergottite Los Angeles is a mixture of 20% crust plus 80% of a basaltic component similar to the shergottite QUE 94201. Their estimates are also listed in Table 4. Compared to our work, Nyquist et al.  calculated a high enrichment of Nd in the crust. The extremely high Nd content of 11.8 ppm, which they determined in their aliquot of Los Angeles, could be the reason for their high crust estimate. A possible explanation for the high Nd content could be a higher portion of phosphates in the aliquot of Nyquist et al.  compared to others.
 A good correlation exists between Al and P for most Martian meteorites and MPF soil (Figure 23). Compared to terrestrial values in Table 3, the Martian basalts (shergottites), MPF soils, and MPF rocks are rich in P. In Martian meteorites, the content of phosphates goes parallel with the content of feldspar. The deviation of certain aliquots of the same meteorite from the correlation line, such as Zagami or Los Angeles, must be a sampling effect of small aliquots and depends on the specific modal composition of the analyzed aliquot. The MPF soil plots nicely on the Al-P correlation line as shown in Figure 23, providing an additional link between SNCs and Mars.
 The APXS measured similar P concentrations in MPF rocks and soils [Dreibus et al., 2000]. The identical concentrations of P in soils and rocks at Ares Vallis (Table 2) would be more in accordance with a sedimentary origin of these rocks rather than an igneous origin. However, the solubility of apatite in a melt depends on its SiO2 content [Harrison and Watson, 1984]. Hence it might be that the solubility limit of apatite confines the P concentration in the SiO2-rich Pathfinder rocks.
 The APXS data of the Pathfinder landing site allowed us a global estimation of the Martian surface and crust composition. The combination of in situ measurements of the Martian surface and element correlations among the Martian meteorites imply a basaltic crust with high abundances of incompatible elements (K, Rb, Nd, U, Th) and volatile (S, Cl) elements. Carbonates are absent or low in soils and rocks (Figure 19) with an upper limit of 5% MgCO3 [Brückner et al., 1999]. In fact, carbonates should not be expected in Martian soil because of the dominance of SO3 as it was emphasized by Wänke and Dreibus  and recently explained in more detail by Clark .
 We would like to thank J. Huth for invaluable support of measurements.