Ever since Kelvin and Helmholtz first estimated the age of the Sun to be less than 100 million years, assuming that gravitational contraction was its prime energy source, there has been a tension between stellar age estimates and estimates of the age of the universe. In the case of the Kelvin-Helmholtz case, the age of the sun appeared too short to accomodate an Earth which was several billion years old. Over much of the latter half of the 20th century, the opposite problem dominated the cosmological landscape. Stellar ages, based on nuclear reactions as measured in the laboratory, appeared to be too old to accomodate even an open universe, based on estimates of the Hubble parameter. Again, as I shall outline in the next section, the observed expansion rate gives an upper limit on the age of the Universe which depends upon the equation of state, and the overall energy density of the dominant matter in the Universe.
There are several methods to attempt to determine stellar ages, but I will concentrate here on main sequence fitting techiniques, because those are the ones I have been involved in.
The basic idea behind main sequence fitting is simple. A stellar model is constructed by solving the basic equations of stellar structure, including conservation of mass and energy and the assumption of hydrostatic equilibrium, and the equations of energy transport. Boundary conditions at the center of the star and at the surface are then used, and combined with assumed equation of state equations, opacities, and nuclear reaction rates in order to evolve a star of given mass, and elemental composition.
Globular clusters are compact stellar systems containing up to 10^{5} stars, with low heavy element abundance. Many are located in a spherical halo around the galactic center, suggesting they formed early in the history of our galaxy. By making a cut on those clusters with large halo velocities, and lowest metallicities (less than 1/100th the solar value), one attempts to observationally distinguish the oldest such systems. Because these systems are compact, one can safely assume that all the stars within them formed at approximately the same time.
Observers measure the color and luminosity of stars in such clusters, producing color-magnitude diagrams of the type shown in Figure 2 (based on data from [10].
Next, using stellar models, one can attempt to evolve stars of differing mass for the metallicities appropriate to a given cluster, in order to fit observations. A point which is often conveniently chosen is the so-called main sequence-turnoff (MSTO) point, the point in which hydrogen burning (main sequence) stars have exhausted their supply of hydrogen in the core. After the MSTO, the stars quickly expand, become brighter, and are referred to as Red Giant Branch (RGB) stars. Higher mass stars develop a helium core that is so hot and dense that helium fusion begins. These form along the horizontal branch. Some stars along this branch are unstable to radial pulsations, the so-called RR Lyrae stars mentioned earlier, which are important distance indicators. While one in principle could attempt to fit theoretical isochrones (the locus of points on the predicted CM curve corresponding to different mass stars which have evolved to a specified age), to observations at any point, the main sequence turnoff is both sensitive to age, and involves minimal (though just how minimal remains to be seen) theoretical uncertainties.
Dimensional analysis tells us that the main sequence turnoff should be a sensitive function of age. The luminosity of main sequence stars is very roughly proportional to the third power of solar mass. Hence the time it takes to burn the hydrogen fuel is proportional to the total amount of fuel (proportional to the mass M), divided by the Luminosity - proportional to M^{3}. Hence the lifetime of stars on the main sequence is roughly proportional to the inverse square of the stellar mass.
Of course the ability to go beyond this rough approximation depends completely on the on the confidence one has in one's stellar models. It is worth noting that several improvements in stellar modeling have recently combined to lower the overall age estimates of globular clusters. The inclusion of diffusion lowers the age of globular clusters by about 7% [11], and a recently improved equation of state which incorporates the effect of Coulomb interactions [12] has lead to a further 7% reduction in overall ages. Of course, what is most important for the comparison of cosmological predictions with inferred age estimates is the uncertainties in these and other stellar model parameters, and not merely their best fit values.
Over the course of the past several years, I and my collaborators have tried to incorporate stellar model uncertainties, along with observational uncertainties into a self consistent Monte Carlo analysis which might allow one to estimate a reliable range of globular cluster ages. Others have carried out independent, but similar studies, and at the present time, rough agreement has been obtained between the different groups (i.e. see [13]).
I will not belabor the detailed history of all such efforts here. The most crucial insight has been that stellar model uncertainties are small in comparison to an overall observational uncertainty inherent in fitting predicted main sequence luminosities to observed turnoff magnitudes. This matching depends crucially on a determination of the distance to globular clusters. The uncertainty in this distance scale produces by far the largest uncertainty in the quoted age estimates.
In many studies, the distance to globular clusters can be parametrized in terms of the inferred magnitude of the horizontal branch stars. This magnitude can, in turn, be presented in terms of the inferred absolute magnitude, M_{v}(RR)of RR Lyrae variable stars located on the horizontal branch.
In 1997, the Hipparcos satellite produced its catalogue of parallaxes of nearby stars, causing an apparent revision in distance estimates. The Hipparcos parallaxes seemed to be systematically smaller, for the smallest measured parallaxes, than previous terrestrially determined parallaxes. Could this represent the unanticipated systematic uncertainty that David has suspected? Since all the detailed analyses had been pre-Hipparcos, several groups scrambled to incorporate the Hipparcos catalogue into their analyses. The immediate result was a generally lower mean age estimate, reducing the mean value to 11.5-12 Gyr, and allowing ages of the oldest globular clusters as low as 9.5 Gyr. However, what is also clear is that there is now an explicit systematic uncertainty in the RR Lyrae distance modulus which dominates the results. Different measurements are no longer consistent. Depending upon which distance estimator is correct, and there is now better evidence that the distance estimators which disagree with Hipparcos-based main sequence fitting should not be dismissed out of hand, the best-fit globular cluster estimate could shift up perhaps 1, or about 1.5 Gyr, to about 13 Gyr.
Within the past two years, Brian Chaboyer and I have reanalyzed globular cluster ages, incorporating new nuclear reaction rates, cosmological estimates of the ^{4}He abundance, and most importantly, several new estimates of M_{v}(RR). The result is that while systematic uncertainties clearly still dominate, we argue that the mean age of the oldest globular clusters has increased about 1 Gyr, to be 12.7^{+3}_{-2} (95%) Gyr, with a 95% confidence range of about 11-16 Gyr [7]. It is this range that I shall now compare to that determined using the Hubble estimates given earlier.
As alluded to earlier, in a Friedman-Robertson-Walker Universe, the age of the Universe is directly related to both the overall density of energy, and to the equation of state of the dominant component of this energy density. The equation of state is parameterized by the ratio = p / , where p stands for pressure and for energy density. It is this ratio which enters into the second order Friedman equation describing the change in Hubble parameter with time, which in turn determines the age of the Universe for a specific net total energy density.
The fact that this depends on two independent parameters has meant that one could reconcile possible conflicts with globular cluster age estimates by altering either the energy density, or the equation of state. An open universe, for example, is older for a given Hubble Constant, than is a flat universe, while a flat universe dominated by a cosmological constant can be older than an open matter dominated universe.
If, however, we incorporate the recent geometric determination which suggests we live in a flat Universe into our analysis, then our constraints on the possible equation of state on the dominant energy density of the universe become more severe. If, for existence, we allow for a diffuse component to the total energy density with the equation of state of a cosmological constant ( = - 1), then the age of the Universe for various combinations of matter and cosmological constant are shown below.
_{M} | _{x} | t_{0} |
1 | 0 | 9.7±1 |
0.2 | 0.8 | 15.3 ± 1.5 |
0.3 | 0.7 | 13.7 ± 1.4 |
0.35 | 0.65 | 12.9 ± 1.3 |
Clearly, a matter-dominated flat universe is in trouble if one wants to reconcile the inferred Hubble age with the lower limit on the age of the universe inferred from globular clusters. In fact, if one took the above constraints at face value, such a Universe is ruled out on the basis of age estimates and the Hubble constant estimates. However, I am old enough to know that systematic uncertainties in cosmology often shift parameters well outside their formal two sigma, or even three sigma limits. In order to definitely rule out a flat matter dominated universe using a comparison of stellar and Hubble ages, uncertainties in both would have to be reduced by at least a factor of two.