## New Research In

### Physical Sciences

### Social Sciences

#### Featured Portals

#### Articles by Topic

### Biological Sciences

#### Featured Portals

#### Articles by Topic

- Agricultural Sciences
- Anthropology
- Applied Biological Sciences
- Biochemistry
- Biophysics and Computational Biology
- Cell Biology
- Developmental Biology
- Ecology
- Environmental Sciences
- Evolution
- Genetics
- Immunology and Inflammation
- Medical Sciences
- Microbiology
- Neuroscience
- Pharmacology
- Physiology
- Plant Biology
- Population Biology
- Psychological and Cognitive Sciences
- Sustainability Science
- Systems Biology

# Searching for the missing baryons in clusters

Edited by Marc Davis, University of California, Berkeley, CA, and approved January 10, 2011 (received for review July 8, 2010)

## Abstract

Observations of clusters of galaxies suggest that they contain fewer baryons (gas plus stars) than the cosmic baryon fraction. This “missing baryon” puzzle is especially surprising for the most massive clusters, which are expected to be representative of the cosmic matter content of the universe (baryons and dark matter). Here we show that the baryons may not actually be missing from clusters, but rather are extended to larger radii than typically observed. The baryon deficiency is typically observed in the central regions of clusters (∼0.5 the virial radius). However, the observed gas-density profile is significantly shallower than the mass-density profile, implying that the gas is more extended than the mass and that the gas fraction increases with radius. We use the observed density profiles of gas and mass in clusters to extrapolate the measured baryon fraction as a function of radius and as a function of cluster mass. We find that the baryon fraction reaches the cosmic value near the virial radius for all groups and clusters above . This suggests that the baryons are not missing, they are simply located in cluster outskirts. Heating processes (such as shock-heating of the intracluster gas, supernovae, and Active Galactic Nuclei feedback) likely contribute to this expanded distribution. Upcoming observations should be able to detect these baryons.

Clusters of galaxies, the largest virialized systems in the universe, are powerful tools in constraining cosmology and tracing the large-scale structure of the universe (1–4, and references therein). The large mass of clusters (∼10^{14} to ) implies that their contents—dark and baryonic matter—have been accreted from very large regions of ∼10 comoving Mpc, and therefore should be representative of the mean matter content of the universe; on these large scales there are no clear mechanisms to separate dark and baryonic matter (e.g., refs. 5 and 6). The strong gravitational potential of clusters also implies that baryons cannot easily escape from these systems. Therefore, clusters are expected to retain the cosmic baryon fraction, the relative fraction of baryons to total matter on large scales. This basic expectation of a representative baryon fraction in clusters was used in 1993 (6) to suggest that the mass-density of the universe must be low, since the observed baryon fraction in clusters was considerably larger than expected for a critical density universe. Most of the baryons in clusters reside in the X-ray emitting hot intracluster gas, which approximately traces the cluster gravitational potential dominated by dark matter. The rest of the baryons are in the luminous galaxies and in isolated stars that comprise the small amount of faint diffuse intracluster light (ICL).

A puzzle has been raised, however, over the last few years: Detailed X-ray observations from Chandra, XMM-Newton, and ROSAT suggest that the cluster baryon fraction (gas plus stars relative to total mass) is considerably lower than the cosmic value. The cosmic baryon fraction is well determined both from Big-Bang nucleosynthesis (7, 8) and from observations of the cosmic microwave background to be *f*_{b} = 0.1675 ± 0.006 (WMAP7: 9). The cluster gas fraction has been reported by observations (10–15) to be only 60–80% of the cosmic value, with stars contributing only a small (∼10%) additional amount of baryons. Whereas the baryon fraction appears to approach the cosmic fraction in the richest clusters, it is still systematically below the cosmic value. This baryon discrepancy, especially the gas fraction, is observed to increase with decreasing cluster mass (14, 15). This raises the questions: Where are the missing baryons? Why are they “missing”?

Attempted explanations for the missing baryons in clusters range from preheating or other energy inputs that expel gas from the system (16–22, and references therein), to the suggestion of additional baryonic components not yet detected [e.g., cool gas, faint stars (10, 23)]. Simulations, which do suggest a depletion of cluster gas in the inner regions of clusters, do not yet contain all the required physics [stellar and Active Galactic Nuclei (AGN) feedback, cosmic ray heating, magnetic fields, etc.] for providing accurate comparisons to the data. Therefore, we use only observations in this work; the collective data should help shed light on which physical processes are most essential.

In this paper we investigate the possibility that the “missing baryons” are not missing at all, but are rather located in the outskirts of clusters where few detailed observations have yet been made. The missing baryons problem is typically observed within the central regions of clusters, generally within a radius of *R*_{500} (where the enclosed mass-density is 500 times the critical density). This radius is ∼0.5 of the virial radius of the cluster [where the enclosed density is ∼100 times the critical density for the current Lambda Cold Dark Matter (LCDM) cosmology (24, 25)]. Thus for a virial radius of ∼1.5 Mpc, the typical missing baryon problem is observed only at ∼0.75 Mpc from the cluster center.

Observations show that the gas density profile in the outer parts of clusters decreases with radius slower than the mass profile in these regions. Using gravitational lensing, the latter has been observed out to large radii (11, 26, 27) and is consistent with the expected Navarro, Frank, White (NFW) profile (28). Whereas the cluster mass density declines with radius approximately as *r*^{-2.6} in these outer regions, the gas density declines only as *r*^{-2.2}. This implies that the gas is more extended than the total mass, and therefore the gas fraction increases with radius beyond the observed radius of *R*_{500}. A shallow slope of the gas profile (as compared with the mass profile) is indeed expected if gas heating occurs in the clusters (e.g., from shock-heating of the gas, supernovae, and AGNs). The heating makes the gas less bound relative to the dark matter potential, and spreads it out to larger radii.

Here we use the observed slopes of the gas-density and mass-density profiles in the outer regions of clusters to extrapolate the observed gas fraction from *R*_{500} to larger radii, up to the virial radius [*R*_{vir} = *R*_{100} (24, 25)]. We add the observed stellar fraction to the extrapolated gas fraction to find the baryon fraction at large radii. We perform this extrapolation as a function of cluster mass from groups to rich clusters, and as a function of radius from *R*_{500} to *R*_{vir}. Note that this analysis is based entirely on observations.

We find that the baryon fraction increases systematically with radius, and show that there is no missing baryon problem in rich clusters when the data is extrapolated to near the virial radius, where the baryon fraction becomes consistent with the cosmic value. Most of the missing baryons are therefore expected to be in the outskirts of clusters, between *R*_{500} and *R*_{vir}. This result can be tested with upcoming observations of the Sunyaev–Zeldovich (SZ) effect in clusters [e.g., South Pole Telescope (SPT) (29); Atacama Cosmology Telescope (ACT) (30)] as well as with more sensitive X-ray observations.

Observations have shown that the missing baryon problem at *R*_{500} becomes more severe for lower mass clusters and groups of galaxies than for rich clusters; the observed gas fraction decreases considerably with decreasing cluster mass. This too would be expected if the heating processes expand the gas: the lower gravitational potential of the smaller systems will not be able to hold on to their gas as well as the higher mass clusters. The gas-density profile in small groups is indeed observed to be shallower than the gas-density profile in massive clusters, suggesting that the gas in low-mass systems is more spread out. We extrapolate the observed baryon fraction as described above for clusters as a function of their mass—from groups to rich clusters. We find that for this entire mass range the baryon fraction within *R*_{vir} is flat and is consistent with the cosmic value.

In the next section we describe the observations and analysis. A discussion of the results and our conclusions follow. We use a LCDM cosmology with *h* = 0.72 and Ω_{m} = 0.258.

## Observations, Analysis, and Results

The gas fraction in clusters of galaxies has been measured for a relatively large sample of groups and clusters within *R*_{500}. The total baryon fraction is obtained by adding the stellar mass fraction observed for these systems within the same radius. This baryon fraction is systematically lower than the cosmic value measured by WMAP7; the discrepancy increases, especially for the gas fraction, with decreasing cluster mass (refs. 14 and 15 and references therein). Here we investigate the possibility that the missing baryons are spread out to larger radii, beyond *R*_{500}. We investigate this possibility by extrapolating the observed gas fraction to larger radii, from *R*_{500} up to the virial radius, using the mean observed gas and mass density profiles in these outer regions; these density profiles have been measured up to the virial radius (and occasionally beyond). The observed stellar mass fraction, including the small contribution from the ICL, is added to the gas fraction to yield the total baryon fraction. The baryon fraction is then investigated as a function of radius, from *R*_{500} to *R*_{vir}, and as a function of mass, from groups to rich clusters.

### Gas Fraction at *R*_{500}.

Although some observations extend to *R*_{200}, the gas fraction has been accurately measured for a sufficiently large sample of nearby clusters only out to *R*_{500}. We use X-ray observations of the gas fractions for 39 nearby clusters observed with Chandra and XMM-Newton (12–14). These authors use similar methods of data reduction and analysis. We use the compilation by Giodini et al. (15) of these groups and clusters above the mass of (where *M*_{500} is the mass within *R*_{500}). After conversion to a common cosmology, the three samples have been binned into four logarithmically spaced mass bins, from groups to rich clusters (15). (We do not include the lowest mass bin at that contains only two groups with large error bars.) Our cluster sample has a mass range of and a redshift range of 0.012 < *z* < 0.23. The mean observed gas fraction for each mass bin is listed in Table 1 and shown in Fig. 1. The error on the mean is the rms standard deviation divided by . The horizontal bars are the mass ranges for the bins. Also presented in Fig. 1 is the cosmic baryon fraction observed by the WMAP7 microwave background measurements. One can see that the cluster gas fraction at *R*_{500} is significantly lower than the cosmic baryon fraction. The gas fraction decreases significantly from rich to poor clusters; whereas rich clusters contain about 12% gas within *R*_{500}, the gas fraction in poor clusters and groups is only ∼6–7%.

### Stellar Fraction.

The galactic stellar fraction has been measured in nearby clusters using multiband optical and infrared surveys combined with stellar population models. We use the results obtained from the COSMOS survey (15) and 2MASS survey (31) for nearby clusters. The combined COSMOS and 2MASS sample covers our entire cluster mass range, . We bin the observed stellar fraction into the same four logarithmic mass bins as the gas fraction sample. The mean stellar fraction declines with cluster mass as *M*^{-0.37±0.04} (see ref. 15 and Fig. 5), exhibiting an opposite trend to that of the gas fraction.

To determine the contribution of the diffuse intracluster light to the stellar fraction, we use observations by Zibetti (32), who stacked hundreds of clusters from the Sloan Digital Sky Survey to reach unprecedented depth and cluster-centric distances. They find the ICL is centrally concentrated, and that on average the ICL contributes ∼10% of the stellar light within the central 500 kpc for all cluster masses. We add this 10% contribution to the galactic stellar fraction discussed above for all clusters. We note that the ICL contribution may decline to less than 10% when extending to larger cluster radii; but since the ICL is a very small contribution to the total baryon fraction, this effect has negligible consequences (see *Discussion*). The total stellar fraction for the four mass bins is summarized in Table 1. It is added to the gas fraction to obtain the total average baryon fraction for each mass range. The baryon fraction within *R*_{500} is listed in Table 1 and plotted as a function of mass in Fig. 1. The deficiency of baryons within *R*_{500} relative to the cosmic value is clearly seen in Fig. 1; the deficiency becomes more severe for lower mass clusters.

### Gas and Mass-Density Profiles.

Extrapolating the observed gas fraction to larger radii beyond *R*_{500} requires the knowledge of the observed gas and mass-density profiles in these regions. The gas-density profile has been measured well in the outer parts of clusters (*R*_{500} - *R*_{vir}) using X-ray observations of nearby clusters with ROSAT, Chandra, XMM-Newton, and Suzaku. The observed gas profile in the outer regions fits well to a beta-model with a density slope of 3*β*_{gas} = *α*_{gas} (where *ρ*_{gas} ∝ *r*^{-αgas}). We use observations that have small uncertainties on the gas-density slope at large radii and that cover our entire cluster mass range (12, 33–36).

The weighted average observed β-values of the gas-density slope are presented as a function of cluster temperature in Fig. 2. The data include measurements for 51 clusters as well as average slopes for stacked samples of hundreds of optical clusters; the slopes are binned in temperature. The error bars on the bin-averaged gas slopes are the standard deviation divided by . The ROSAT observations of 39 *z* < 0.25 clusters by Vikhlinin et al. (33) cover the full outer regions of clusters, from 0.3*R*_{500} to 1.5*R*_{180} (∼*R*_{vir}). The best-fit slope to these outer regions is determined for a wide range of cluster temperature, from 2 keV to over 10 keV. The Chandra observations of 10 massive clusters by Vikhlinin et al. (12) provide β-fits to gas-density slopes near *R*_{500}; their weighted average values are consistent with the trend shown by the slopes of the previous sample (33). For the lowest mass bin () we use the observed density slope from ROSAT by Dai et al. (34; their richness class 1), who obtain integrated X-ray gas profile for stacked samples of hundreds of low-mass optical clusters out to *R*_{vir}. We also include Bautz et al. (35) who measure the gas profile of Abell 1745 to *R*_{200} using Suzaku observations.

The mean observed beta slopes presented in Fig. 2 are consistent with each other, and show a shallower slope for lower mass systems than for massive clusters. Similar results were noted by refs. 37 and 38. For our four mass bins (Table 1 and Fig. 1) we find the following mean gas-density slopes: *α*_{gas} = 3 × *β*_{gas} = 1.8 ± 0.2 for Bin 1, 1.9 ± 0.07 for Bin 2, 2.1 ± 0.02 for Bin 3, and 2.3 ± 0.02 for Bin 4. The error bars are the standard deviation of the average β-values in each bin divided by . When comparing a slope at a given temperature bin in Fig. 2 to a given mass bin in Fig. 1, we use the observed *M*_{500} - *T* relation (12); the results are not sensitive to the exact conversion because of the slowly varying *β*_{gas}(*T*) relation.

The final piece required for extrapolating the gas (and baryon) fraction to large radii is the total mass-density profile. This has been observed to follow the NFW profile (28) to large radii, by using weak lensing observations with the Sloan Digital Sky Survey and other observations (11, 26, 27, and references therein). We use the average observed value of the concentration parameter *c*_{200} = 5 for our mass range (26). The NFW profile has a mass-density slope of *α*_{m} = 2.6 in the radius range from *R*_{500} to *R*_{200}, and *α*_{m} = 2.7 from *R*_{200} to *R*_{vir} (where *ρ*_{m} ∝ *r*^{-αm}). These slopes are considerably steeper than the corresponding gas-density slopes, thus yielding an increasing gas fraction with radius in cluster outskirts.

### Extrapolation and Results.

Using the observed gas-density and mass-density slopes at large radii, we extrapolate the observed gas fraction from *R*_{500} as a function of radius up to the virial radius. The gas-fraction increases with radius as:

The extrapolated gas fraction from *R*_{500} to *R*_{200} and to *R*_{vir} is presented as a function of cluster mass in Fig. 3*A*. Because the gas-density slope is shallower in groups than in rich clusters, the increasing trend of gas fraction with mass becomes weaker at the outer radii.

The baryon fraction is presented for the different radii—*R*_{500}, *R*_{200}, and *R*_{vir}—as a function of mass in Fig. 3*B*; the baryon fraction within these radii is the sum of the gas fraction (Fig. 3*A*) and the stellar mass fraction discussed above, including the 10% ICL. We assume that this fraction remains constant with increasing radius; the main results do not change significantly if this assumption is changed because of the relatively small contribution of the stellar component (see *Discussion*). The error bar on the extrapolated gas fraction is the propagated errors of the gas fraction at *R*_{500} and the gas slope used for extrapolation. The error bar on the baryon fraction is the combined errors of the gas and stellar fractions.

The results in Fig. 3*B* show that the baryon fraction flattens considerably as a function of cluster mass when extrapolated to larger radii; this is due to the combined effect of the shallower gas-density profile in groups, which results in more gas in their outskirts, plus the larger observed stellar mass fraction in groups, which adds more baryons in the smaller systems. In fact, at the virial radius we find that the baryon fraction is essentially flat from groups to rich clusters, at a level consistent with the cosmic baryon fraction. This suggests that there are no missing baryons: Most of the missing baryons are likely located in the outskirts of clusters, extending to nearly the virial radius.

The extrapolated baryon fraction is presented as a function of radius for two of our mass bins (Bins 2 and 4) in Fig. 4; the results show the slow but steady increase in the baryon fraction with radius.

## Discussion

The gas fraction at *R*_{500} is often cited to claim that clusters contain fewer baryons than the universal baryon fraction and therefore exhibit a missing baryon problem. Here we show, based purely on observational results, that the gas (and baryon) fraction increases substantially with radius beyond *R*_{500}. Using the observed gas and mass-density profiles, we extrapolate the observed baryon fraction (gas and stars) as a function of radius from *R*_{500} to the virial radius. Since the gas density is observed to decline more slowly with radius than the total mass, we find that the average baryon fraction increases with radius for clusters of all masses; it reaches the cosmic baryon fraction near the virial radius (*R*_{100}). This suggests that baryons are not missing in clusters, they are simply located in cluster outskirts.

Recent observations of the SZ effect in 15 massive clusters using the South Pole Telescope (29) measure the gas-density pressure profile in these clusters as a function of radius up to the virial radius (and beyond); they are well fit with a beta-model in the outer regions. The detection of the gas to these large radii, and their observed beta-model slope (when corrected for the temperature profile), are nicely consistent with our results and the conclusion that the baryons are out between *R*_{500} and *R*_{vir}, although the gas-density slope cannot yet be accurately determined from the SZ measurements. Similarly, George et al. (39) use Suzaku X-ray observations to trace the gas-density profile in the massive cluster PKS0745-191 up to the virial radius, observing the baryons at the cluster outskirts and measuring a shallow gas-density profile in these outer regions (with lower resolution).

The gas-density profile is observed to be even shallower in low-mass clusters than in massive clusters. This is qualitatively consistent with the gas being heated via shocks and feedback (e.g., from supernovae and AGN). This feedback will be more significant in low-mass systems when compared to the binding energy of the gas. If star formation is more efficient in groups than in clusters (e.g., ref. 31), this will further increase the gas entropy in these systems, because the star formation removes the lowest-entropy gas from the intracluster medium, leaving behind gas with higher average entropy (40). A higher stellar fraction also implies more supernovae/AGN activity per unit mass. This explains why, at *R*_{500}, groups have a lower baryon fraction than clusters. However, by the virial radius, we are able to account for all the baryons expected from the cosmic value.

The stellar fraction has been observed for a large number of clusters by Gonzalez et al. (41), who report a somewhat higher stellar fraction for low-mass clusters and a slightly lower stellar fraction for high-mass clusters than the observations by Giodini et al. (15) and Lin et al. (31). The higher stellar fraction in groups (41) is likely due to a selection bias toward systems with dominant Brightest Cluster Galaxies in the small groups; the large, centrally located galaxy in such systems dominates the stellar fraction. Their observed trend of stellar fraction with mass is therefore somewhat steeper, log *f*^{stars} = (7.57 ± 0.08) - (0.64 ± 0.13) log *M*_{500}. Using this stellar fraction in our analysis does not change our result significantly; the baryon fraction at the virial radius decreases by ≈5%, to 0.164, for the most massive bin, and increases by ≈7%, to 0.179, for the lowest mass bin. These are within our 1-*σ* error bars.

We use the observed 10% contribution of the ICL to the stellar fraction (32). If the ICL fraction changes at the outer radii, it will not affect our results since the ICL contribution is only ∼0.2–0.4% of total mass. Similarly, we assume in our analysis that the stellar fraction does not change when extrapolated from *R*_{500} to *R*_{vir}, since no observations of the stellar fraction have been made beyond *R*_{500}. If the stellar mass fraction decreases somewhat with radius, then the baryon fraction will slightly decrease. If we assume, for example, a 20% decrease in stellar fraction from *R*_{500} to *R*_{vir}, we find that the baryon fraction is lowered by only 2% for the massive clusters and 8% for the low-mass groups (at *R*_{vir}); these values are within 1-*σ* of our baryon fraction results even for the lowest mass groups.

For the mass-density profile, we use the mean observed value of the concentration parameter of the NFW profile, *c*_{200} = 5, as observed from weak lensing (26) for our mass range. As Mandelbaum et al. (26) discuss, this value is slightly lower than results from simulations and some previous studies. If *c*_{200} is larger than 5, the mass profile will be more concentrated; i.e., fall off even steeper with respect to the gas profile at the outer radii. This would cause the gas fraction to increase even more with radius. The effect is small, however, and a change of *c*_{200} to 7 induces a change in baryon fraction at the virial radius of < 5%.

The observations presented above are qualitatively consistent with an energy input in clusters that heats the intracluster gas and makes it less bound than the dark matter. Hydrodynamic simulations of cluster formation that include some of the relevant physics are in qualitative good agreement (within ∼10%) with the picture presented here. Using simulations without cooling, star formation, or feedback, the baryon fraction is roughly constant, at 90% of the cosmic value, from *R*_{500} to *R*_{200} (42). When these processes are included (with feedback coming from AGN), then the baryon fraction is found to be lower at *R*_{500}, but instead of being flat it increases with radius (43, 44). Similar results, based on cluster energetics, are found in the models of Bode et al. (22). However, simulations do not yet contain all the physics needed for accurate comparison with the observations (including sources of nonthermal pressure and possible nonequipartition effects).

## Conclusions

We investigate the missing baryon problem in clusters of galaxies. Observations show that the baryon fraction (gas plus stars) measured within a radius of *R*_{500} in groups and clusters is significantly below the cosmic value; this baryon discrepancy increases with decreasing cluster mass. This gives rise to the puzzle: Where are the missing baryons? Why are they missing? We show that the baryons may not be missing at all, but rather are spread out to larger radii, beyond *R*_{500}, and can be found in the cluster outskirts. Based entirely on observations, we investigate the dependence of the baryon fraction on radius for clusters of different masses, from groups to rich clusters. We use the mean observed gas and mass-density profiles in clusters to extrapolate the observed gas fraction as a function of radius from *R*_{500} to the virial radius. Since the gas-density profile is significantly shallower than the mass-density profile, the gas-fraction increases with radius; it increases more rapidly for groups than for rich clusters because of shallower gas-density slope in groups. We add the observed stellar fraction and the diffuse intracluster light to the gas fraction to obtain the total baryon fraction. We find that the average baryon fraction for all groups and clusters with increases steadily with radius, reaching the cosmic value and becoming flat as a function of mass when measured within the virial radius (for the LCDM cosmology). This suggests that baryons are not missing in clusters, but are simply located in cluster outskirts. This picture is qualitatively consistent with heating processes (such as shock-heating of the intracluster gas, as well as supernovae and AGN feedback) causing the gas to expand to the cluster outskirts. Upcoming observations in the X-rays and SZ should be able to detect these baryons.

## Acknowledgments

B.R. thanks her research adviser, collaborator, and friend Neta Bahcall for her judicious and delightfully rewarding stewardhip of her senior thesis, which culminated in this paper. B.R. also thanks Princeton University for enabling this experience. The authors thank Rachel Mandelbaum, Greg Novak, David Spergel, Michael Strauss and Alexey Vikhlinin for their helpful comments. Computational work was performed at the TIGRESS high performance computer center at Princeton University, which is jointly supported by the Princeton Institute for Computational Science and Engineering and the Princeton University Office of Information Technology.

## Footnotes

Author contributions: B.R., N.A.B., and P.B. designed research, performed research, contributed new reagents/analytic tools, analyzed data, and wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

## References

- ↵
- ↵
- Bahcall NA,
- Ostriker JP,
- Perlmutter S,
- Steinhardt PJ

- ↵
- ↵
- ↵
- Peacock JA

- ↵
- White SDM,
- Navarro JF,
- Evrard AE,
- Frenk CS

- ↵
- Walker TP,
- Steigman G,
- Kang H-S,
- Schramm DM,
- Olive KA

- ↵
- ↵
- Jarosik N,
- et al.

- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- Cavaliere A,
- Fusco-Femiano R

- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- Eke VR,
- Cole S,
- Frenk CS

- ↵
- ↵
- Mandelbaum R,
- Seljak U,
- Hirata CM

- ↵
- ↵
- Navarro JF,
- Frenk CS,
- White SDM

- ↵
- ↵
- ↵
- ↵
- Zibetti S

- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵
- ↵

## Citation Manager Formats

## Sign up for Article Alerts

## Article Classifications

- Physical Sciences
- Astronomy