THE ASYMPTOTIC GIANT BRANCH AND THE TIP OF THE RED GIANT BRANCH AS PROBES OF STAR FORMATION HISTORY: THE NEARBY DWARF IRREGULAR GALAXY KKH 98

HST observations of KKH 98 were obtained in both the F475W and F814W filters of ACS as part of the ANGST survey (see Dalcanton et al. 2009 for the details). Three observations of the galaxy were made in each filter. The combined images have an effective exposure time of 2265 s and 2280 s, respectively, and are shown in Figure 1. The ACS field of view (202'' × 202'') encompasses the bulk of the galaxy (see Figure 1).

Zoom In Zoom Out Reset image size

On the night of 2007 November 24, we obtained K'-band (2.12 μm) Keck LGS AO imaging of KKH 98. We used the wide-field (40'' × 40'') camera of the NIRC2 instrument to achieve the largest field of view. With a pixel scale of 004 pixel⁻¹, the wide-field camera undersamples the AO point-spread function (PSF; 006 resolution), limiting the effective resolution to ∼01. However, as is discussed below, the loss of resolution did not affect our ability to measure the photometry of the uncrowded IR-bright stars in KKH 98.

Individual exposures were 1 minute, with a dither applied after two successive exposures. During the observations, the laser was fixed to the center of the field. Thus, as the telescope dithered, the laser dithered as well. This method was preferred over fixing the laser to the sky because of significantly lower overhead. In addition, as was demonstrated in Steinbring et al. (2008), this method provides a more uniform AO correction over the field. An R = 14.2 tip/tilt star was located ∼20'' north of the galaxy center (Figure 1). The AO system used this star to track and correct for image motion, while the laser spot was used for higher order wave front corrections.

We followed the image-reduction method outlined previously in Melbourne et al. (2005). Sky and flat-field frames were created from the individual science frames after masking sources. Frames were then flat fielded and sky subtracted. We corrected frames for known NIRC2 camera distortions. We aligned images by centroiding on sources, and combined them with a clipped mean algorithm. The final reduced image has an effective exposure time of 45 minutes. Figure 1 shows the resulting AO image of KKH 98. It covers the central 40'' × 40'' region of the galaxy.

We observed the UKIRT photometric standard star FS3 immediately after the galaxy observations. However, the night was not photometric. Therefore, we set the zero point for the AO data based on two bright unsaturated stars in the actual science frame which have Ks-band magnitudes from the Two Micron All Sky Survey (Skrutskie et al. 2006). Ultimately, we compared our photometry to stellar isochrones produced in the K-band system (Bessell & Brett 1988). However, the transformation to the K-band system is negligible, typically ≲0.02 mag for red stars (including the transformation from Ks to K (Carpenter 2001) and the color terms to transform from K' to Ks; Wainscoat & Cowie 1992). Therefore, for simplicity, we heretofore report all AO magnitudes in the K-band Vega system.

The standard star observations of FS3 were obtained with tip–tilt AO correction to stabilize image motion, but without higher order LGS AO correction from the laser spot. This strategy produced a high signal-to-noise ratio (S/N) observation of the profile of the wings of the AO PSF (which are primarily set by seeing and tip–tilt). We used these standard star observations in the reconstruction of the AO PSF as outlined in the following section.

To perform accurate photometry, we need a model of the AO PSF. Unfortunately, the AO correction degrades spatially with distance from the laser spot (anisoplantism) and the tip–tilt star (anisokinetisism), producing a spatially varying PSF. We model the PSF using two components: (1) an AO-corrected core with a FWHM of roughly the diffraction limit of the telescope and (2) a halo with a profile width set by the atmospheric seeing and AO tip–tilt correction. While the shape of the PSF halo is relatively stable across the image, the core varies significantly with position.

To get an accurate representation of the profile of the PSF core, we identify luminous, isolated stars within 25 equal-area sub-regions of the image. After subtracting neighbors, we register and co-add the stars in each sub-region, and trace the PSF profile of the co-added image.

Unfortunately, the PSF halo is spread over hundreds of pixels, and is much lower S/N than the diffraction-limited core. Therefore, its profile cannot be accurately traced even for the stacked PSFs described above. Instead, we used images of the very bright flux standard star, FS3, observed with tip–tilt AO correction on, and high-order AO correction off. With high-order correction off, light that would have been corrected into the PSF core by the AO system, remained in the halo. The resulting PSF has a very high S/N profile out to large distance (∼2'') from the PSF center, and is a good match to the halo of the AO PSF (Figure 2).

Zoom In Zoom Out Reset image size

To produce a final PSF for each of the 25 sub-regions, we splice together the halo and core images, scaling the halo image to map smoothly onto the image of the core. As was done in Melbourne et al. (2008), we select the splice point to be the radius at which the flux in the azimuthally averaged core profile is 9σ above the sky noise, as this is a good match to bright AO-corrected stars where the halo is easily observed. Figure 2 shows that the PSF profile is relatively independent of the particular choice of splice point, for splice points larger than several times the diffraction limit (e.g., ∼02). This figure also shows that the halo PSF profile from the standard star is a good match to the few AO-corrected stars in the science frame that are bright enough to adequately detect the halo.

The stellar photometry of the AO image was performed with small apertures on the high S/N core of the AO PSF. We chose a 4 pixel radius aperture (016). This aperture is 50% larger than the typical FWHM of the stellar PSF, and maximizes the S/N of the aperture photometry. We convert the small aperture photometry to total flux with aperture corrections calculated from the 25 model PSFs, one for each sub-region of the image. Total flux for the model PSFs was measured with a curve of growth technique out to 2''.

The aperture correction varies from 0.6 mag at the center of the field to ∼1 mag near the edge, where a larger fraction of the light remains uncorrected. Figure 3 shows the variation of the aperture correction across the field. A linear fit to the data gives aperture correction as a function of distance from the image center for an arbitrary stellar position. The fit provides a more finely sampled measure of the aperture correction across the field and improves the estimates in the outskirts of the image where the PSF is poorly measured due to a lack of stars. We use this functional form for the aperture corrections in the final photometry.

Zoom In Zoom Out Reset image size

The deep HST images of KKH 98 identify the locations of stars in the galaxy. To translate the HST positions onto the Keck AO image, we first selected a set of 157 bright stars that are easily visible in both the HST and AO images. We then used the IRAF CCTRAN program to transform the HST coordinates into the AO image coordinates, using a quadratic transformation. Since the geometric distortions in the AO image were removed during image processing, this transformation worked well across the entire frame.

We measured the K-band magnitude for each star in the HST star list that was located on the AO frame. For each star, bright neighbors were identified with Sextractor. Scaled model PSFs were used to subtract the bright neighbors using a PSF fitting technique. We then measured aperture photometry on the star of interest, applying the aperture correction appropriate for the radial separation of the star from the center of the image (Figure 3). A correction for the local sky was made, using a median sky value measured in an annulus of radius of 1.0–1.2 arcsec. When making the sky measurement, stars identified in the sky annulus by Sextractor segmentation maps and the HST star lists were masked out.

Two types of uncertainty dominate the error budget in these measurements: (1) the photometric uncertainty, set primarily by the Poisson sky noise, and (2) the uncertainty introduced by the spatially varying PSF and aperture corrections. The photometric Poisson uncertainty can be measured directly from the images. For each star, we measure the standard deviation of pixels in the nearby sky to estimate this uncertainty. The uncertainty introduced by the aperture corrections can be measured from the scatter in Figure 3. While the scatter is small near the center of the image, it increases significantly toward the edges of the image.

To estimate any systematic in the measurement methods, we also performed a Monte Carlo simulation, embedding 500,000 artificial stars into the image at random locations with sub-pixel sampling. Artificial stars ranged in magnitude from K = 17 to 24, but were all assumed to be identified in the corresponding HST image. We measured the photometry of the artificial stars in the exact same manner as the real stars in the HST catalog. Figure 4 shows the standard deviation of the photometry as a function of input K-band magnitude. We find to K = 22, the photometry is better than 0.2 mag. At K = 23.5 the photometric uncertainty is about a magnitude.

Zoom In Zoom Out Reset image size

An additional 22 IR-bright stars were identified in the AO image that were not included in the final HST photometry lists, usually because they fell below a blending or S/N threshold in the F475W image. We found that these stars do not alter the conclusions of this paper and therefore do not include them in the final analysis.

We estimate the SFH of KKH 98 by comparing the observed CMDs with CMDs produced from the stellar isochrones of Girardi et al. (2002), with updated AGB models from Marigo et al. (2008). These isochrones were used together with the bolometric corrections and color transformations from Girardi et al. (2008). Sample isochrones, overlayed on the CMDs, are shown in Figure 6. Metal-poor (1/10 solar) isochrones (Figure 6, top row) span the photometric data. In contrast, more metal-rich isochrones (Figure 6, bottom row) tend to be redder than the data, suggesting that the stars in KKH 98 are metal-poor.

We use the MATCH package (Dolphin 2002) to fit the CMD for an SFH, metallicity history, dust reddening, and distance modulus (DM), for an assumed Salpeter IMF and binary fraction of 0.35. MATCH uses a Poisson Maximum Likelihood statistic to determine the best-fit model and the 1σ uncertainties on that model. For a complete description of the MATCH routine, see Dolphin (2002).

Williams et al. (2007) showed that the IMF choice does not affect the relative SFRs (as a function of time), but can change the overall normalization. Our primary goals are to compare the SFH measured from the optical data to the SFH measured from the IR data, and to determine if the SFH is "bursty" or continuous. As these are relative measurements, neither will be significantly affected by choice of IMF or binary fraction.

Additional inputs to MATCH include (1) a range of acceptable distances and reddening values, (2) time step sizes, and (3) ranges and binning sizes in color–magnitude space. In general we, attempted to reproduce the choices made in previous ANGST work (Williams et al. 2007, 2009b). Distance ranges were allowed within ±0.5 mag of the distance modulus in the literature of DM = 26.95 ± 0.11 (Karachentsev et al. 2002). Reddening was allowed to vary from A_V = 0.1–0.7, a reasonable range for a dwarf galaxy with foreground reddening A_I = 0.24 as measured in the Schlegel et al. (1998) Galactic dust maps. We use a series of 12 logarithmically spaced time steps which allow us to explore the SFH in both the recent (Myr) and distant (Gyr) past. The minimum sizes for color–magnitude bins are set by the uncertainties of the stellar isochrones and are roughly 0.05 mag in color and 0.1 mag in magnitude. However, when setting bin sizes, additional considerations include allowing for reasonable number statistics in each bin while providing the resolution necessary to resolve features in the CMDs. For the optical data, we used color bins of size 0.1 mag and magnitude bins of size 0.15 mag. We doubled the sizes of these bins in the NIR to offset the smaller number statistics in the NIR CMDs. Small changes in the bin sizes were not found to alter the results.

We used the latest version of MATCH (version 2.3) which handles the isochrone color transformations for Carbon and Oxygen AGB stars independently. In older versions of MATCH, C/O ratios were not tracked and the color transformations produced AGB stars significantly redder (2–3 mag) than expected. The color transformations of C stars in the updated version of MATCH were derived from Loidl et al. (2001). Because the information content of the IR CMD is biased toward the AGB populations, it was important for us to use the most realistic prescriptions for AGB photometry. In the optical CMDs, these considerations were less vital because the information content at young and intermediate ages primarily arises from the main-sequence turnoffs (MSTOs).

For the optical data, we chose to model the CMD with photometry brighter than the 12% uncertainty limits and 50% completeness limits (F475W < 27.6 and F814W < 27.0) to avoid fitting areas of the CMD with poor photometry or large completeness corrections. These limits result in photometry for 2539 stars in the area overlapping the AO data. For the IR CMD, we chose stars with K < 23.5, the point at which the K-band photometric uncertainty is about 1 mag. These cuts provide 592 stars. While the photometric precision at the fainter end of K is poor, the long-wavelength baseline between the F814W and K bands means that there is still statistically interesting information at these fainter limits. Table 1 gives a summary of the MATCH-derived results which are discussed in detail below.

Figures 7 and 8 show Hess diagrams of the CMDs in the optical and IR (top left). Also shown are the best-fit model CMDs derived by MATCH (top right), the residual difference between the model and the data (bottom left), and the residuals scaled by the statistical significance of bin (bottom right). The residual differences between the data and MATCH-derived models are small for both the optical and IR CMDs, except for AGB stars, especially in the optical data where the models predict 130 AGB stars compared to 42 observed (with the caveat that the AGB must be brighter than the TRGB, and redder than F475W − F814W = 1.75). In the IR, the models predict 92 AGB stars (brighter than the TRGB). We return to these discrepancies in the following section and elaborate in the discussion section. In contrast, the numbers of stars in other evolutionary stages, such as the RGB and main sequence are better reproduced by the models. Table 2 gives the measured and predicted numbers of stars in different regions of the KKH 98 CMD.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

4.1.1. Star Formation Histories

The optical and IR-derived SFHs of KKH 98 are plotted in Figure 11. For the bulk of cosmic time, the SFRs from the two CMDs agree to within the uncertainties. For ages older than 3 Gyr, both data sets show evidence for a modest, roughly constant SFR of 5 × 10⁻⁴ M_☉ yr⁻¹, or 2 × 10⁻⁴ M_☉ yr⁻¹ kpc⁻². From 1 − 3 Gyr ago, both show a drop in star formation to roughly negligible rates. More recently, both show renewed star formation for ages <0.5 Gyr ago. However, in these youngest time bins, there are differences in the details. At the youngest ages (age <0.1 Gyr), the optical photometry suggests ongoing star formation, while the IR data set does not. This difference is not alarming, as the near-IR data are not very sensitive to hot (blue) main-sequence stars produced by recent star formation.

Another difference is that the optical data set suggests a burst of star formation, at age ∼0.5 Gyr (Figure 11). In contrast, the IR data suggest SFRs roughly a factor of 4 smaller at that time. The evidence for a burst in the optical data comes from stars at the well-populated MSTO at the faint end of the optical CMD. This is shown in Figure 9 where a realization of the predicted model CMD divided into different time bins. At log(age) = 8.6 − 8.8 [yr], large numbers of stars are predicted at the MSTO. In contrast the IR data do not reach deep into the main sequence, and therefore the IR-derived SFH is not sensitive to the MSTO feature. However, the IR data should contain information on populations of that age from the AGB sequence (see Figure 10). Therefore, if the burst is real, evidence for it should exist in the AGB population, assuming the AGB models are accurate. In the discussion section, we examine possible explanations for the measured differences in the SFR at these ages. Here we discuss the effect of time binning.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Time binning may play a role in the differences between the optical and IR-derived SFHs. Logarithmic time bins have been selected to fully explore the SFH on both recent and older timescales. However, especially for the IR data, which have only 592 stars, the time binning may be more fine than the data warrant. The third panel of Figure 11 shows the IR results now binned with larger time steps. These bins were selected by first binning the data on small timescales and then tracking the goodness of fit parameters as bins were dropped from the sample. If the fits did not change by more than the measured uncertainties, a dropped bin would be incorporated with its neighboring bin. Ultimately, this resulted in four bins. We caution, however, that this method is likely to miss the importance of bins with little star formation such as those from 1–3 Gyr ago. For more on optimal binning techniques, see Williams et al. (2010). The optical data have been binned to match the IR bins. Binned this way, there are still some differences between the two measured SFHs at the youngest ages.

The SFR uncertainties shown in Figure 11 are a combination of (1) the systematic uncertainty between the models and the KKH 98 photometry, and (2) the random uncertainty associated with small number statistics of stars in any given color–magnitude bin. The first uncertainty is estimated by the MATCH routine from the maximum likelihood statistic as it attempts to fit model parameters within the adopted ranges. The second uncertainty is estimated by a Monte Carlo bootstrap simulation of 100 realizations, where we rerun MATCH but alter the input photometry list, so that it is composed of random draws from the actual photometry list. The systematic uncertainties are roughly equivalent for both the optical and IR data, and are similar in scale to the random uncertainties associated by small number statistics. The error bars shown in Figure 11 are the two uncertainties added in quadrature.

Surprisingly, the optical and IR model uncertainties are similar in scale despite the fact that the optical data contain a factor of 5 more stars. Several factors conspire to lead to this result. First, the color–magnitude bins used in the IR analysis are twice as large as the optical analysis, and thus typically contain as many or more stars as the optical bins. Second, the lack of stars in a feature can be just as important to producing a significant result as the presence of stars. For example the lack of AGB stars in the data forces the IR models to select against a burst of star formation at 0.5 Gyr ago (see Figure 10), resulting in small numbers of AGB stars in both the models and data, and a low uncertainty assigned to this time bin. Whereas in the optical data, the MSTO suggests a burst at that time, and the lack of AGB stars acts to increase the uncertainty of the high SFR measured in this time bin.

4.1.2. Metallicity

Figure 12 shows the MATCH measured metallicity of new stars at every epoch. Again results from both the optical (blue) and IR (red) CMDs are shown, however, in time bins where no star formation was measured, no metallicities are shown (several IR time bins). Both suggest that the chemical history of KKH 98 has been dominated by low-metallicity star formation, producing stars with typical [M/H] ∼ −1.5 to −1.0. There may be a trend to more metal-rich stars at more recent times. MATCH can also be run in a mode where the metallicity is constrained to continually increase, as would be expected in a closed box galaxy model. When run in this mode, the MATCH results do not change appreciably.

Zoom In Zoom Out Reset image size

4.1.3. Reddening and Distance

While the MATCH-derived distances arise from normalizing the entire CMD to models, specific features in the CMD can also be used to estimate the distance. One of the best distance indicators available in the KKH 98 data set is the F814W-band magnitude of TRGB. Figure 13 shows the F814W-band magnitude of the TRGB as a function of metallicity and age, using isochrones from Marigo et al. (2008).⁷ For old, metal-poor stellar populations the TRGB has an absolute magnitude of M^TRGB_F814W = −4.0 (Figure 13). We estimate the apparent magnitude of the TRGB in KKH 98 to be at F814W = 23.2 ± 0.1. Correcting for reddening using A_I = 0.23, this gives a distance modulus of DM = 26.97 ± 0.10, in good agreement with Karachentsev et al. (2002) who found DM = 26.95 ± 0.11 from the I-band TRGB.

Zoom In Zoom Out Reset image size

The apparent K-band magnitude of the TRGB (K = 21.2 ± 0.1 for KKH 98) can also be used to estimate the distance of KKH 98. In Galactic globular clusters with old stars, the K-band magnitude of the TRGB has been shown to be a function of metallicity (Valenti et al. 2004). Gullieuszik et al. (2007) parameterize this empirical function as

$$\begin{equation} M_K^{{\rm TRGB}}=-6.92-0.62\;[M/H]. \end{equation} \tag{ 1 }$$

Assuming the MATCH-estimated KKH 98 metallicity for old stars of [M/H] = −1.4, Equation (1) gives M^TRGB_K = −6.05. Given the apparent TRGB magnitude of K = 21.2 ± 0.1 and negligible dust absorption in the K band, we calculate a KKH 98 distance modulus of DM = 27.25 ± 0.10. This distance measurement is somewhat farther than the distance estimated from the apparent F814W-band magnitude of the TRGB.

However, another estimate of the absolute K-band TRGB can be made from the Marigo et al. (2008) models shown in Figure 13, which are also the models used by MATCH. These models suggest that, at low metallicities and old ages, the K-band magnitude of the TRGB is 0.2 mag fainter than the value from Equation (1). Using the Marigo et al. (2008) models, we find a distance modulus of DM = 27.05 ± 0.10, which is in very good agreement with the DM derived from optical CMD.

Although the number of stars available for analysis in the IR was an order of magnitude smaller than for the deep HST optical data, the bulk of measured properties—including the broad trends in SFH and metal enrichment—agree between the two data sets. There are some differences however. The most obvious is that the IR data set is not sensitive to the youngest stars (i.e., age <0.1 Gyr), because the bluest main-sequence stars emit the bulk of their radiation in the UV and optical. Therefore, HST optical observations are required to constrain the SFHs at the youngest ages.

There are also differences in the details of the SFHs at ages of ∼0.1–1 Gyr ago, precisely when AGB stars are expected to dominate the information content of the IR CMD. For instance, 0.5 Gyr ago the optical data suggest a burst with an SFR four times the prediction of the IR data. The evidence for this burst comes from the well-populated MSTO of the optical CMD at that age (Figure 9).

A burst of this age should produce AGB stars observable in our optical and IR CMDs. Unfortunately, MATCH predicts that this burst should produce roughly the entire observed AGB population, which would only make sense if there were no AGB production at older times (Figure 9). The overproduction of AGB stars by the models is even more striking if we model the burst in the IR. Figure 14 shows a realization of the model IR CMD created with the optically derived SFH. Not only does the model predict large numbers of AGB stars at 0.5 Gyr ago, it also suggests that their colors should be 0.5–1 mag redder than the AGB populations seen in the data. As the SFH derived from the IR data relies heavily on the AGB sequences, it is now clear why the IR data underpredicts the star formation in this time bin; there are few AGB stars in the KKH 98 data that match the luminous, red AGB stars predicted by the models for a burst of that age. Figure 14 also suggests that age/metallicity degeneracies are not producing the discrepancy; the predicted AGB stars are well separated from the observed AGB stars in color space.

Zoom In Zoom Out Reset image size

Not only do the models overproduce young metal-poor AGB populations (and predict redder colors than are seen), they also overproduce old metal-poor AGB populations (Figures 9 and 10). For populations older the 3 Gyr, the models predict ∼57 ± 8 (optical, assuming Poisson uncertainties) and ∼79 ± 9 (IR) AGB stars in KKH 98. These predictions are both larger than the total number of 42 observed AGB stars, which should presumably include some younger AGB stars as well. Taken together, these results suggest issues with models of AGB stars especially for metal-poor populations such as those found in KKH 98.

The modeling of AGB depends crucially on the prescriptions for two difficult-to-model phenomena: mass loss and the efficiency of third dredge-up episodes. To better address the uncertainties related to these processes, modelers resort to synthetic codes in which a few parameters are tuned to fit observational data, typically from the Magellanic Clouds (e.g., Marigo & Girardi 2007).

While synthetic AGB codes have successfully reproduced the numbers, luminosities, and colors of AGB stars in the Magellanic Clouds (Cioni et al. 1999; Nikolaev & Weinberg 2000; Marigo & Girardi 2007), the situation in other galaxies has been less encouraging. For instance, Gullieuszik et al. (2008) showed that models calibrated on the LMC and SMC overpredict the AGB lifetimes of the old, metal-poor dwarf spheroidal, Leo II, by as much as a factor of 6. L. Girardi et al. (2010, in preparation) reach a similar conclusion based on a larger sample of ANGST galaxies.

Perhaps more relevant to the present paper are the conclusions from a similar work from Held et al. (2010): in Leo I dSph, a metal-poor galaxy containing young stellar populations (therefore more similar to KKH 98), they find that the carbon stars predicted by the models are about twice those observed. We may be seeing a similar issue in KKH 98, where we would need to roughly triple the AGB population in the data to match the models, and the difficulties with the AGB seem to be particularly problematic at younger ages. We will potentially be able to better constrain AGB models in this vital time range with observations of the AGB populations of a wide variety of galaxies, as will be possible with the ANGST WFC3 survey.

In KKH 98, the IR-bright stars are well separated, with typical separations of ∼1''. Assuming these separations are common within other galaxies, crowding should not limit the technique until much larger distances (up to 10 Mpc for Keck). The primary limiting factor for future ground-based observations with AO will therefore be sensitivity above the thermal background. Our study was completed with a 45 minute exposure. At larger distances, significantly more exposure time, or better AO correction will be required to reach the depth of our study (∼2 mag below the TRGB). For instance, to reach comparable S/N in a galaxy twice as distant would require 16 times the total exposure time, or 12 hr. There are, however, several ways in which this time requirement could be reduced.

• 1.  
  Improved AO performance. The observing conditions during this run were sub-optimal with only ∼20% Strehl ratio (ratio of peak luminosity in the PSF to theoretical maximum peak luminosity for a diffraction-limited image). With the Keck AO upgrade, typical Strehl ratio's of 30% or more are common. A higher Strehl ratio will dramatically increase the sensitivity above background. In addition, if the Strehl ratio were more uniform across the image, larger numbers of faint stars would be well photometered toward the edges of the image. More uniform AO performance will be possible with upcoming Multi-Conjugate AO systems such as the one being developed for Gemini South, which will have an isoplanatic patch four times the size of the Keck AO system.
• 2.  
  At shorter wavelengths, the thermal IR background is reduced. In the H band, Melbourne et al. (2007) showed that it was possible to photometer an H = 24.0 [Vega] point source to a photometric precision of 17% in an hour-long Keck AO exposure. The H-band sensitivity is significantly deeper than the K-band limit. The drawbacks for using this strategy are: (1) the AO system typically achieves smaller Strehl ratios at shorter wavelengths and (2) there is a smaller wavelength lever-arm in the CMD when using H-band (1.6 μm) data compared with K band (2.2 μm) data. Despite these limitations, using the H band may be a good strategy for galaxies at larger distances than KKH 98.
• 3.  
  With near-IR WFC3 observations from HST, both the thermal background and field-of-view issues can be solved. We expect the planned WFC3 observations of the ANGST sample to reach significantly deeper than our Keck AO observations. These observations will be made in both the H and J bands. The roughly factor of 3 loss in resolution should be relatively unimportant at the distances of the ANGST sample, where we have shown that the IR-luminous stars are typically not crowding limited. In more distant galaxies, crowding will be a problem for HST at these wavelengths.
• 4.  
  Future instruments such as AO on a TMT or JWST should also overcome the limitation of our current data set. These telescopes will potentially resolve individual stars in galaxies out to the Virgo Cluster (18 Mpc) and beyond.