The kinematics of ionized gas in Lyman-Break Analogs at z ~ 0.2
Thiago S. Gonçalves, Antara Basu-Zych, Roderik Overzier, D. Christopher Martin, David R. Law, David Schiminovich, Ted K. Wyder, Ryan Mallery, R. Michael Rich, Timothy H. Heckman
DDraft version: October 25, 2018
Preprint typeset using L A TEX style emulateapj v. 11/10/09
THE KINEMATICS OF IONIZED GAS IN LYMAN-BREAK ANALOGS AT Z ∼ Thiago S. Gonc¸alves, Antara Basu-Zych, Roderik Overzier, D. Christopher Martin, David R. Law, DavidSchiminovich, Ted K. Wyder, Ryan Mallery, R. Michael Rich, Timothy H. Heckman Draft version: October 25, 2018
ABSTRACTWe present results for 19 “Lyman Break Analogs” (LBAs) observed with Keck/OSIRIS with anAO-assisted spatial resolution of less than 200 pc. We detect satellites/companions, diffuse emissionand velocity shear, all with high signal-to-noise ratios. These galaxies present remarkably high velocitydispersion along the line of sight( ∼
70 km s − ), much higher than standard star-forming spirals inthe low-redshift universe. We artificially redshift our data to z ∼ . Subject headings: galaxies: kinematics and dynamics - galaxies: starburst - galaxies: evolution INTRODUCTION
Our understanding of galaxy formation has changedconsiderably over the course of the last two decades. Re-markable progress has been made with numerical simula-tions that reproduce the growth of the large-scale struc-ture in the universe, and the results from these simula-tions agree well with studies of galaxy clusters and thecosmic microwave background (e.g. Springel et al. 2005;Benson & Bower 2010). However, the small-scale, non-linear baryonic physics that goes into forming the galac-tic structure remains an open question. The so-called“gastrophysics”, comprising AGN feedback and super-nova winds among other processes, is still poorly under-stood. Simulations rely on ad hoc recipes, which are inturn based on observational results and are purely phe-nomenological; the underlying physical processes are notyet known.The traditional paradigm of galaxies forming fromslowly cooling shock-heated gas (e.g. White & Rees(1978); Mo et al. (1998); Baugh (2006) and referencestherein) does not seem to apply in many cases. An ele-vated fraction of galaxies at high redshift display clumpystructures (Elmegreen et al. 2008), which might formfrom internal instabilities (Noguchi 1999; Immeli et al.2004; Bournaud et al. 2007) or, alternatively, from merg-ers of subgalactic gas clumps (Taniguchi & Shioya 2001),in agreement with the idea of hierarchical galaxy forma-tion in LCDM models. Furthermore, recent numericalsimulations indicate that star formation at high redshift [email protected] California Institute of Technology, MC 278-17, Pasadena,CA 91125 NASA Goddard Space Flight Center MPA UCLA Columbia University Johns Hopkins might be fed through cooling flows supplying the centersof dark matter haloes directly with gas at just below thevirial temperature (Dekel & Birnboim 2006; Dekel et al.2009; Kereˇs et al. 2009). In this context, it becomesimportant to analyze the kinematics in these galaxies,and to confront the relative contributions from orderedrotation, random motions and merger-induced featureswith predictions from the aforementioned models. Be-cause stellar kinematics at high redshift are largely be-yond reach of current instruments and telescopes, thebright nebular emission line gas is often used as a tracerfor the underlying kinematics.In an early attempt to study kinematics of star-forminggalaxies at z ∼ −
3, Erb et al. (2006b) analyzed long-slitspectra of H α emission in UV-selected galaxies (Steidelet al. 2003, 2004), detecting significant velocity shears in12% of the objects in their sample. In all cases, velocitydispersion in the ionized gas was high in comparison withthe observed velocity shears, with v c /σ ∼
1. These ob-servations are challenging, since they are seeing-limitedand slit-alignment plays an important role in actuallydetecting any shears (Erb et al. 2006b; Law et al. 2006).More recently, Law et al. (2009) improved on this re-sult, with spatially resolved kinematics of the gas fromadaptive-optics (AO) assisted integral-field spectroscopyof 12 star-forming galaxies at redshift z ∼ .
5. This tech-nique has the advantage of not depending on alignmentchoice, detecting velocity shears all across the extent ofthe galaxy, while the AO system resolves features at sub-kpc scales. The authors detect, again, high velocity dis-persion values of σ (cid:39) −
70 km s − . In most cases thereis no evidence for ordered rotation across the galaxy, andin general the gas dynamics appear to be dominated byrandom motions. The authors also find a mild trendof rotational properties with stellar mass, with massivegalaxies typically displaying more pronounced velocityshears. a r X i v : . [ a s t r o - ph . C O ] S e p Gon¸calves et al.In a similar study, F¨orster Schreiber et al. (2009) stud-ied a large sample of 62 star-forming galaxies at simi-lar redshifts with the SINFONI instrument. This workdiffers from Law et al. (2009) in that most observationsare seeing-limited, with spatial resolution elements of ap-proximately 4 kpc. In addition, most galaxies in thissample were drawn from the BzK sample of Daddi et al.(2004), and are typically two times as massive as the UV-selected galaxies. The authors found that their samplecan be subdivided into three groups: rotation-dominatedobjects, with pronounced velocity shears and v c /σ val-ues of up to 4; dispersion-dominated objects, with littleto no velocity shear across the major axis; and mergers,with multiple components or peculiar velocity profiles.In addition, F¨orster Schreiber et al. (2009) also found atrend of properties with stellar mass, with more massivegalaxies presenting higher v c /σ ratios and larger sizes.A number of observations at intermediate and high red-shifts also support the hypothesis of extreme starburstsbeing protodisks resulting either from minor mergers orsmooth accretion from the intergalactic medium (Bouch´eet al. 2007; Cresci et al. 2009; Wright et al. 2009).Jones et al. (2010) also studied the kinematics of theionized gas in high-redshift star-forming galaxies, buta sample of strongly lensed objects was used instead.The authors were then able to reconstruct the kinematicstructure by applying models of the gravitational lens,achieving much higher spatial resolution ( ∼
100 pc) alongone spatial dimension. Out of a sample of 6 objects, 5display characteristics of rotating gas disks, again withtrends in velocities as a function of size and dynamicalmass. Although the results help us understand the dy-namical structures of such galaxies, it is challenging toconstruct a statistically significant sample of lensed ob-jects. Additionally, in many cases the major axis is notaligned with the lens shear, in which case the velocityshear comprises few resolution elements in the data.Studies to date explore complementary regions of pa-rameter space.The difference in the prevalence of dif-ferent kinematics observed is probably a function pri-marily of parent sample, compounded with differences inthe sensitivity regime of different techniques. The ben-efit of AO is that it obtains greater spatial resolutionbut is not sensitive to low surface brightness features (ifpresent), while non-AO probes lower surface brightnessesand larger radii but with less fidelity. In both cases obser-vations are technically challenging, due to the distance tothe galaxies, which results in low intrinsic spatial resolu-tion and cosmic surface brightness dimming. Therefore,it is advantageous to observe similar galaxies at lowerredshifts in order to assess whether certain features de-rived from observations at high redshift are intrinsic orbiased due to observational effects.Heckman et al. (2005) have selected a sample of UV-bright galaxies in the low-redshift universe ( z ∼ . L ∼ L ∗ z =3 , where L ∗ z =3 is the characteristic luminosity of LBGs at z ∼ z ∼ −
4, theirmorphologies are remarkably similar to LBGs at theseepochs (e.g. Giavalisco et al. 1996; Papovich et al. 2005;Lotz et al. 2006; Law et al. 2007b), while the subtle, lowsurface brightness merger features tend to disappear evenin the deepest rest-frame UV or optical imaging data.This implies that on the basis of morphologies alone, itcannot be ruled out that LBGs grow through clumpyaccretion and mergers, perhaps together with rapid gasaccretion through other means (Overzier et al. 2010).Furthermore, strong hydrogen lines and compact sizesmake them ideal candidates for IFU spectroscopy. InBasu-Zych et al. (2009a), we presented preliminary re-sults of the IFU survey discussed here for three LBAs,showing how these galaxies resemble the kinematic struc-tures of high-redshift star-forming galaxies. In this work,we expand the sample to investigate the ionized gas kine-matics of 19 LBAs, observed with spatial resolution downto ∼
200 pc. The paper is divided as follows: in section(2), we describe the data acquisition and analysis, includ-ing target selection and how we artificially redshift ourdata to z = 2 . H = 70 km s − Mpc − , Ω m = 0 .
30 andΩ Λ = 0 . OBSERVATIONS AND DATA REDUCTION
Sample Selection
We investigate a subsample of the ultraviolet-luminousgalaxies (UVLGs). These objects were first defined byHeckman et al. (2005) to have far-ultraviolet (FUV) lu-minosities ≥ × L (cid:12) , which is roughly halfway be-tween the characteristic luminosity of present-day galax-ies and that of higher redshift Lyman-Break Galaxies(LBGs).As described in the previous section, Hoopes et al.(2007) later expanded the analysis of these objects andFU Spectroscopy of LBAs 3subdivided the sample in terms of average FUV surfacebrightness ( I ), using the SDSS u -band half-light ra-dius as proxy for the UV size of the galaxies. The samplewas divided in three categories: large UVLGs ( I ≤ L (cid:12) kpc − ), compact UVLGs ( I > L (cid:12) kpc − )and supercompact UVLGs ( I > L (cid:12) kpc − ). Thelatter represents the aforementioned LBAs.The LBAs are compact systems undergoing intensestar formation; in fact, they are among the most star-forming galaxies in the low-redshift universe. The ob-served physical properties, such as metallicity, dust at-tenuation, UV/optical morphologies and star formationrates, are remarkably similar to those of high-redshiftLBGs. We further discuss the analogy between low- andhigh-redshift objects in subsequent sections. Observations and Data Reduction
LBAs are selected to have high surface brightness val-ues, which translate into small physical sizes, rangingfrom 0.4 to 1.9 kpc half-light radii in the ultraviolet(Overzier et al. 2010). Together with the high star for-mation rates up ∼
100 M (cid:12) yr − (Hoopes et al. 2007),which translates into extremely bright nebular hydrogenemission lines, LBAs are highly suitable targets for adap-tive optics (AO) assisted integral field spectrography.We have used OSIRIS in the Keck II telescope (Larkinet al. 2006). OSIRIS is an integral field unit (IFU) avail-able solely for use with AO. It provides a spectral reso-lution of R ∼ α emission line (rest wave-length λ = 1875 . ∼ α line, depending on gas tem-perature (Osterbrock & Ferland 2006). In all cases thisis redshifted into the redder half of the K-band, with ob-served wavelength varying between 2055 nm (cid:46) λ obs (cid:46) α lineat each redshift and lesser impact of space command clo-sures (when observers are prevented from using the laserdue to possible collisions with artificial satellites). There-fore, no biases were introduced in the data beyond theoriginal LBA selection.The properties of individual objects are shown in Table1 along with observing information. The stellar masseswere taken from the SDSS/DR7 MPA-JHU value-addedcatalog . These masses were calculated by fitting a large grid of spectral synthesis models from Bruzual & Charlot(BC03,2003) to the SDSS u (cid:48) , g (cid:48) , r (cid:48) , i (cid:48) , z (cid:48) photometry. Thelack of near-infrared data and TP-AGB stars from thesynthesis library should introduce an uncertainty of ∼ α andMIPS-24 µ m data; they typically present an uncertaintyup to 0.3 dex (Overzier et al. 2009). For an in-depthdiscussion of properties of LBAs and comparison withhigh-z galaxies, see Hoopes et al. (2007) and Overzieret al. (2009, 2010).Given the limited physical size of the detector, thereis a trade-off between spatial coverage and wavelengthcoverage; since we are interested in a single emission line,we have chosen to use the narrowband mode for mostgalaxies in order to maximize the spatial coverage of thedata. In most cases we observed with the 50 mas spaxelscale; the UV sizes of the remaining objects were largerand we chose to use the 100 mas scale with double theFOV.In many cases, the object occupies a significant por-tion of the FOV of the instrument. Because appropriatesky subtraction is crucial for a reliable detection of emis-sion lines in the data, we have ensured an exclusive skyframe was taken in conjunction with each science frame.The best strategy to maximize on-target telescope timewas to observe in 45 minute blocks of science-sky-scienceframes, with 15 minute exposures in each case. Weatherranged from acceptable to excellent in all cases, withuncorrected seeing (in V band) varying from ∼
1” inmoderate conditions to 0.5” in the best cases. Weatherconditions directly affect spatial resolution in our data,since the quality of AO corrections depend on the stabil-ity and brightness of the laser guide star and the tip-tiltstar.Data were reduced with the OSIRIS pipeline, whichsubtracts the sky frames and translates the two-dimensional detector image into a 3D datacube, com-posed of two spatial dimensions and one wavelength di-mension (for details, see Wright et al. 2009). In addition,we have written custom IDL code to further subtract skyemission residuals still present in the datacube. This isdone for each galaxy simply by fitting the 1D spectrumat spaxels where we believe no signal from the observedgalaxy exists; this is then subtracted from all spaxels inthe datacube.
Kinematic Maps
Gon¸calves et al.
Table 1
Summary of LBA ObservationsName z Observing Spaxel Exposure AO FWHM SFR (H α + 24 µ m) R l a log M ∗ date (UT) scale (mas) (s) (mas) ( M (cid:12) yr − ) (kpc) (M (cid:12) )005527 0.167 Oct 01, 2007 50 900 90 55.4 0.36 9.7015028 0.147 Oct 20, 2008 50 2400 82 50.7 1.34 10.3021348 0.219 Oct 19, 2008 100 2100 177 35.1 0.38 10.5032845 0.142 Jan 24, 2010 50 1800 103 8.7 0.86 9.8035733 0.204 Oct 20, 2008 100 1800 116 12.7 1.00 10.0040208 0.139 Sep 13, 2009 50 2100 80 2.5 0.80 9.5080232 0.267 Jan 24, 2010 100 1800 115 30.4 3.01 10.7080844 0.096 Feb 06, 2010 100 1200 187 16.1 0.08 9.8082001 0.218 Jan 25, 2010 50 2400 69 40.0 2.78 9.8083803 0.143 Feb 05, 2010 50 1800 105 6.2 1.02 9.5092600 0.181 Feb 26, 2008 50 1800 101 17.0 0.68 9.1093813 0.107 Feb 06, 2010 50 1800 77 19.8 0.65 9.4101211 0.246 Feb 06, 2010 50 1200 96 6.2 N/A 9.8113303 0.241 Jan 24, 2010 50 2400 76 7.7 1.36 9.1135355 0.199 Feb 05, 2010 50 2100 68 19.4 1.45 9.9143417 0.180 Feb 26, 2008 50 2700 98 20.0 0.90 10.7210358 0.137 Oct 20, 2008 50 1500 65 108.3 0.44 10.9214500 0.204 Sep 13, 2009 50 2400 70 16.4 1.13 9.9231812 0.252 Sep 13, 2009 100 1800 130 63.1 2.77 10.0 a UV half-light radius from HST data
In order to produce velocity moment maps, we fitgaussian functions to the emission lines detected at eachspaxel. In most cases, our LBA spectra do not show anycontinuum, only the Pa- α line emission. The zero-pointof the fit is the center of a gaussian fit to the integratedone-dimensional spectrum of the collapsed datacube.We smooth every datacube spatially with a kernel of1 . − S/N ) ratios shown are obtainedby dividing the area of the gaussian fit to the emis-sion line in each spaxel by the sum of the noise fluc-tuation over the same wavelength range. The noise isdetermined from a region of the sky with no emissionline detection. We introduce a minimum threshold of
S/N = 6 for a fit to be deemed acceptable; anythingsmaller is discarded. This minimizes the presence of ar-tifacts in the final maps. This
S/N threshold representsa detection limit in star formation surface density of or-der Σ
SFR ∼ . (cid:12) yr − kpc − , comparable to surfacebrightness limits determined in F¨orster Schreiber et al.(2009) and an order of magnitude deeper than the datapresented in Law et al. (2009). The velocity-dispersion( σ ) maps, corrected for instrumental broadening, alwaysshow values greater than the intrinsic instrumental res-olution of ∼
35 km s − , with the exception of some lowsurface brightness spaxels.Figure 1 shows the recovered kinematic maps for eachof the objects in our sample. In each case, the two leftpanels show the HST images of the galaxy, with line emis-sion contours overlaid. The third panel shows the zero-th moment of the fit, which is simply the total intensity ineach spaxel, shown as the signal-to-noise of the fit in eachspaxel. The fourth panel shows the velocity maps, andthe final panel shows the velocity dispersion maps. Wealso show the resolution element, given by the FWHM ofa star observed before the galaxy, in the exact same con-figuration (band filter and pixel scale). Also shown is ahorizontal bar indicating a physical size of one kpc at theredshift of the galaxy. Three of these galaxies (092600,143417 and 210358) have been previously analyzed inBasu-Zych et al. (2009a). Comparison with HST morphologies
Figure 1 shows the HST images for each galaxy in rest-frame optical (left panel) and UV (second-to-left). Im-ages are scaled at logarithmic (black) and linear(blue)stretch, to distinguish between low surface brightnessstructures and more compact ones. Pa- α flux contours,in red, typically enclose 1/3 of the rest-frame optical flux,and above 60% of the UV flux. In general we are ableto detect emission where the bulk of the stellar mass ispresent, unless no significant star formation is present(e.g. the southeast components in 080844 and 210358).Comparison between both bands in HST shows moreextended structures in the rest-frame optical, in partic-ular at low surface brightness (black). This might in-dicate an underlying older stellar population in whichstar-forming regions exist. A complete discussion of LBAmorphologies in both bands can be found in Overzieret al. (2009, 2010). Simulation to High Redshift
As briefly discussed in section 1, LBAs have been de-fined on the basis of UV luminosity and surface bright-ness thresholds as appropriate for high-redshift Lymanbreak galaxies. Previous studies have supported theanalogy, finding both apparent and physical propertiesconsistent with those of their high-z counterparts. Inthis section we investigate the parallel in terms of gaskinematics of LBAs compared to LBGs.FU Spectroscopy of LBAs 5
Figure 1.
We show here the velocity moment maps for all galaxies observed for this work. The two leftmost figures show the HST rest-frame optical (left) and UV (right) morphologies, with logarithmic (black) and linear (blue) stretches. The Pa- α S/N levels are overlaidin red. There is no UV image available for 101211. The following images show, from left to right, the signal-to-noise ratios, line-of-sightvelocity in km s − and line-of-sight velocity dispersion, also in km s − . For the latter two we overplot S/N contours in white. The axesshow the angular scale in arcsec; orientation is the same in every panel, with north pointing up and east to the left. We indicate in eachpanel the FWHM of a point source as a proxy for spatial resolution and the physical scale corresponding to 1kpc at the redshift of eachgalaxy. Gon¸calves et al.
Figure 1. continued.
In order to allow for a direct comparison between kine-matics of LBA- and LBG-type systems, we have artifi-cially redshifted all our galaxies to z ∼ . , and reob-served them with the simulated IFU prescriptions de-scribed by Law et al. (2006). At this redshift, thesegalaxies would be observed in H α . We scale our ob-served Pa- α flux maps to the total H α fluxes determinedby SDSS. One should notice that the code used to arti-ficially redshift our sample represents the exact same in-strument, observational setup and reduction software asin Law et al. (2009), and has been shown to appropriatelyreproduce actual OSIRIS observations. This ensures therobustness of comparisons between the LBA sample andthat of LBGs presented in Law et al. (2009)We have also artificially redshifted our data and sim-ulated observations with the SINFONI instrument, in This precise redshift was chosen to avoid major OH emissionlines. non-AO mode. In this case, the optimal hydrogen line-emission surface brightness detection limits in F¨orsterSchreiber et al. (2009) is comparable to our sample: onone hand the instrument is more sensitive, H α is brighterand there is no loss due to the adaptive optics system;on the other hand, cosmological surface brightness dim-ming would make sources up to 200 times fainter persolid angle unit. Therefore, we simply degrade our spa-tial resolution with a 0.5” gaussian kernel, rebinning ourdatacubes to the nominal 0.125” pixel scale of SINFONI,while simultaneously reducing the total angular size ofthe galaxy as determined by the ratio of angular diame-ter distances at z = 2 . Figure 1. continued.
Gon¸calves et al.
Figure 1. continued. the loss in spatial resolution causes different star-formingregions to be confused into one larger clump. This mightlead to misinterpreting multiple clumps with velocity dif-ferences as one larger, smoother rotating disk, with im-plications for inferences about its formation mechanism(see sections 4, 5). This is particularly true for the simu-lated SINFONI data, in which case many LBAs are noteven spatially resolved.These simulations will be used below when comparingkinematical measurements of LBAs and actual high red-shift galaxies observed. These comparisons, along withimplications for the analogy between LBAs and star-bursts at high redshift, will be discussed in detail in sec-tion 4. ANALYSIS OF INDIVIDUAL OBJECTS
In the following sections we briefly describe each ob-ject in more detail. Two of these objects (021348 and080232) are not resolved even with adaptive optics. Theypresent dominant central objects (DCOs) as discussed inOverzier et al. (2009). A third object (101211) is toofaint, and no extended structure is detected. We excludethese three objects from the kinematic analysis in subse-quent sections.
This is the only object observed in broadband mode.Velocity dispersion is rather uniform across the wholegalaxy, at about 100 km s − . The optical morphologyis evidently much more extended than the Pa- α emit- ting region, which might indicate an underlying, moreextended, older stellar structure. This is an object showing two clearly distinct star form-ing regions. There is also a clear velocity shear in theeast-west direction, which is not aligned with the axisconnecting the two bright clumps. Velocity dispersionis higher in the eastern half of the galaxy. In addition,there is some additional emission to the south, at highervelocity than the rest of the galaxy; it is unclear whetherthis represents a spiral arm or a tidal tail from an ongoinginteraction.
This is the faintest object observed, and we have onlybeen able to detect an unresolved point source in thecenter of the galaxy, in addition to a low S/N region(
S/N <
10) to the south. It is the first of five objectsobserved with OSIRIS that were classified as having aDCO, according to Overzier et al. (2009). Since we can-not make any inferences about the resolved kinematicstructure of the galaxy, we have excluded it from anyfurther analysis.
Figure 2.
Velocity maps of 214500 at its intrinsic redshift ( left ) and artificially redshifted to z ∼ .
2, as observed by OSIRIS ( center ) andSINFONI ( right ). Legends are the same as in Figure (1). As expected, spatial resolution is lower, and low surface brightness features areharder to distinguish. shows an antenna-like structure, with distinct nuclei, inwhat appears to be a merger.
We have been able to detect not only the brightestcomponent, but the faint companion to the east, whereline emission is evidently weaker. A comparison withthe HST image shows a much more extended structurethan what is seen here. The western region, however, isclearly defined, and shows a definite velocity shear acrossits major axis, resembling a rotating disk, but still withline-of-sight velocity dispersion values of approximately70 km s − , close to the value of the velocity shear acrossthe major axis. The companion to the east is at the samesystemic velocity as the main component. This is one of the faintest galaxies we have observed(SFR= 2 . M (cid:12) yr − ), therefore the signal-to-noise ratiois considerably smaller. There are a number of star-forming regions northeast of the main component, andthe velocity offset between them is rather small. This is another DCO, like 021348. Again, we detectvery little emission besides a bright point source in thecenter of the galaxy. This object is also excluded fromfurther analysis.
This is another DCO, but in this case we were ableto detect emission from the companion to the southeast.There is little velocity structure within the main compo-nent, but the companion is offset more than 200 km s − from the point source. This object shows a main emission region larger thana kpc across, with little velocity structure. In addition, we were able to detect emission from a fainter structureto the south, with a velocity offset from the main com-ponent of ∼
50 km s − . This structure is also seen in theHST image. This is another example of an LBA with a compan-ion structure, also evident in the HST image. The com-panion presents a ∼
50 km s − shift with respect to themain structure. Also evident is a velocity shear acrossthe main region itself, albeit small – ∼
50 km s − – es-pecially when compared to the velocity dispersion of ap-proximately ∼
100 km s − found in the galaxy. This isthe least massive of our objects (log M ∗ /M (cid:12) = 9 .
1) andhas also been described in Basu-Zych et al. (2009a).
This is one of the galaxies with strongest line emissionin our sample (the Pa- α line is detected at S/N (cid:38) − between them. Showing signs of a recentor ongoing strong merger event in the HST optical data,the velocity dispersion seems higher where the merginggalaxies appear to meet, to the west, where Pa- α emis-sion is strongest. The emission is weak, and little structure is detectedbeside a faint companion to the northeast. Due to lesserdata quality in comparison with other galaxies in oursample, we do not use this object for our subsequentanalysis.
This galaxy shows a remarkable lack of velocity struc-ture within the main component, with a shear of a fewtens of km s − , comparable to the instrument resolutionitself. However, we were able to detect some faint emis-sion from a component to the southwest, offset from themain region at approximately 100 km s − . This object presents two clearly distinct regions of starformation, along the east-west axis. The regions are atdistinct velocities with respect to each other. In addition,we detect fainter emitting regions to the north and north-west, at very different velocities from the two brightestregions. These two regions are part of much more elon-gated structures, as can be seen in the HST image, whichshows strong signs of an ongoing interaction. This hasalso been discussed in Basu-Zych et al. (2009a).
This is the most massive object we have observed, andone with very unique features. It is one of the DCOobjects as described in Overzier et al. (2009), and weconfirm the existence of a bright, unresolved region inthe center of the galaxy. This region has high Pa- α sur-face brightness, with values above 10 − erg s − cm − arcsec − . This galaxy presents the strongest velocityshear across its major axis, v shear ∼
250 km s − . This isthe third object presented in Basu-Zych et al. (2009a). This galaxy presents high velocity shear across its ma-jor axis, uncommonly so for its low stellar mass (see sec-tion 4.2). However, its structure is not smooth, and thereare undetected stellar components to the south, seen inthe HST image. Likewise, the velocity dispersion map isnot as well structured as other disk-like galaxies. Thismay indicate a recent merger event.
This is one of the largest galaxies in our sample, andtherefore was observed with the 100 mas spaxel scale tomaximize its field of view. It shows a bright componentwith fainter structure to the south and west. The star-forming region to the south has a velocity offset of ∼ − from the brightest part of the galaxy. RESULTS
In this section we discuss some of the analytic resultsobtained from our observations, describing the method-ology used to calculate each of the quantities presented.
Sizes
Previous studies of LBAs and high-redshift star-bursts infer kpc-scale sizes for the star forming regions,from rest-frame UV continuum as observed with HST(Overzier et al. 2010, and references therein) and theemission line regions as observed with IFU instruments(Law et al. 2009; F¨orster Schreiber et al. 2009). Here wepresent our calculations for sizes of LBAs as seen withOSIRIS, comparing these figures with results from theabove-mentioned studies.We replicate the method described in Law et al.(2007a), to allow for a direct comparison with resultsfor LBGs utilizing the same instrument. This comprisescounting the number of spaxels N above a certain S/N threshold to represent the size of the star-forming region.We use the same threshold, namely
S/N >
6. To de-termine a radius, we assume galaxies are approximatelycircular, and therefore calculate a radius in spaxels as r = ( N/π ) / . This number is corrected for the PSFsize in each case (see section 2.3) and later convertedto a physical size at the corresponding spaxel scale andredshift of each object. We only use contiguous spaxelsconnected to the brightest region of the galaxy in orderto exclude companions. Finally, we repeat the process forour simulated high-redshift observations of LBAs. Errorsare typically 0.1 kpc (low-z), 0.4 kpc (OSIRIS) and 2.0kpc (SINFONI), given by half the PSF size in each case.The results are presented in Table 2 and shown in Fig-ure 3 along with actual measurements for high-redshiftgalaxies. Most LBAs present sizes between 1 and 2 kpc,consistent with findings from Overzier et al. (2010).From Figure 3 we notice this method yields smallersizes at high-redshift in AO-assisted observations; this iscaused be surface brightness dimming of objects, whichprevents detection of emission at the outer radii of galax-ies. Sizes for our OSIRIS simulated observations are re-markably similar to those found in Law et al. (2009) forLBGs, which have been calculated in an identical man-ner; a two-sided Kolmogorov-Smirnov test yields a 97%probability of both samples being drawn from the sameparent population.Non-AO observations, however, produce different re-sults. Galaxies are apparently larger, in many casesdue to blending of different components, combined withsomewhat improved sensitivity at the outer radii, asidefrom obvious loss of resolution. Still, simulated LBAslook smaller than galaxies in the SINS survey (F¨orsterSchreiber et al. 2009) - many, in fact, show sizes smallerthan the inferred uncertainty, which means they areessentially unresolved. When comparing HST sizes ofLBAs with those found for BzK galaxies, Overzier et al.(2010) conclude both samples have similar rest-frame UVsizes, but the latter has larger rest-frame optical sizes;therefore it is not unreasonable to assume galaxies inthe SINS survey might be intrinsically larger than bothLBAs and LBGs. Kinematics and Dynamics of Star FormingGalaxies
The ionized gas in LBAs exhibit very high velocity dis-persions , with median ∼
67 km s − and some galaxiesreaching values above 100 km s − . This is much higherthan those observed in ordinary local star forming galax-ies (typical gas velocity dispersions of 5–15 km s − , e.g.,Dib et al. 2006) but analogous to the increased velocitydispersions observed in local (ultra-)luminous infraredgalaxies (e.g. Arribas et al. 2008; Monreal-Ibero et al.2010). These values are also in good agreement withhigh-redshift star-forming galaxies, as observed both insingle-slit spectroscopy (e.g. Pettini et al. 2001; Erbet al. 2006b) and the aforementioned integral field studies(Law et al. 2009; F¨orster Schreiber et al. 2009). The global velocity dispersion σ measured for the each galaxyis an average of each spaxel, weighted by flux. This allows for amore accurate measurement than simply measuring the velocitydispersion of the whole cube, since it does not incorporate theintrinsic velocity shear within the galaxy. FU Spectroscopy of LBAs 11
LBAs @ z~0.2 LBAs @ z~2 (OSIRIS) Law+ (2009)
LBAs @ z~0.2 LBAs @ z~2 (SINFONI) FS+ (2009) F r ac ti on Figure 3.
Distribution of sizes for LBAs. Black filled histogramsrepresent intrinsic values, while red hashed histograms representsimulated values at high redshift as would be detected with SIN-FONI ( top ) and OSIRIS ( bottom ). Blue histograms indicate sizedistributions of actual high-redshift starburst galaxies as measuredwith corresponding instruments by F¨orster Schreiber et al. (2009)and Law et al. (2009).
We also measure the velocity shear within each galaxy.Since we cannot always precisely define an axis of ro-tation, we simply determine the difference between themaximum and minimum velocities observed within themain body of the galaxy (excluding companions in or-der to probe for intrinsic rotation of one star-formingregion). We determine v max and v min as the median ofthe 5-percentile at each end of the velocity distribution,so that outliers and artifacts are excluded. The velocityshear is then simply defined as v shear = ( v max − v min ).The values vary between a few tens of km s − and over200 km s − . These measurements are presented in Ta-ble 2. In many cases, the velocity shear is not causedby actual rotation of the whole galaxy, since there is nota significant velocity gradient observed across the entireobject.There is a strong trend of velocity shear with stellarmass: more massive objects tend to show greater velocitydifferences between distinct regions of ionized gas. Thiscan be seen in detail in Figure (4). Velocity dispersion σ ,also correlates with stellar mass, albeit with a shallowerslope. For comparison, we also show in Figure 4 the localTully-Fisher relation derived in Bell & de Jong (2001),corrected for an average inclination factor of (cid:104) sin i (cid:105) =0 .
79 (see Appendix in Law et al. 2009). Although aninference for such a relation for LBAs is not reasonable, since these objects are not necessarily rotating disks, thisserves as a comparison with velocity shear in local spirals.These values are slightly smaller for a given stellar mass,especially at lower masses (up to a factor of 2). Alsoshown is the derived relation for star-forming galaxies at z ∼ . v circ values than spirals in the present day; however, in theformer, the galaxies studied are more massive ( M ∗ (cid:38) − × M (cid:12) ), and were pre-selected to look like rotatingdisks. * /M Ο • L og v s h ea r , σ ( k m s - ) Figure 4.
Velocity shear v shear ( blue circles ) and velocity disper-sion σ ( green squares ) as a function of stellar mass. The plot showsclearly how more massive galaxies show a stronger velocity shearthan less massive ones, particularly the ones above ∼ M (cid:12) .The same trend, albeit weaker, exists for velocity dispersion σ .Dashed line shows a power-law fit to our data, while the solid lineis the Tully-Fisher relation at z ∼ z ∼ Due to the difference in slopes, the ratio between veloc-ity shear and velocity dispersion ( v/σ ) is also a functionof stellar mass (black triangles in Figure 5). A Spear-man’s ρ correlation test shows a ∼
6% null-hypothesisprobability of M ∗ and v/σ not being correlated. Thisindicates that more massive LBAs have a stronger com-ponent of rotational support against gravitational col-lapse, as opposed to less massive ones, which are moredispersion dominated.When artificially redshifted, the v/σ ratio decreases,from a combination of two effects: on one hand, surfacebrightness dimming causes the high-velocity values at theoutskirts of the galaxy to be undetected - this is particu-larly true for the artificial OSIRIS high-z data (shown asred downward triangles in Figure 5). On the other hand,loss of spatial resolution, especially for non-AO obser-vations performed with instruments such as SINFONI(blue downward triangles in Figure 5), causes blendingof features and inner velocity values to dominate, due tohigher signal-to-noise. The net result is lower v shear val-ues. Although our observed v/σ values are higher than2 Gon¸calves et al. Table 2
Kinematic data for LBAsName r Pa − α v shear σ v/σ r Pa − α, hiz v shear , hiz σ hiz r Pa − α, hiz v shear , hiz σ hiz log M dyn K asym K asym , hiz K asym , hiz (OSIRIS) (OSIRIS) (OSIRIS) (SINF) (SINF) (SINF) (M (cid:12) ) (OSIRIS) (SINF)005527 1.2 42 104 0.41 1.0 35 89 2.5 18 122 10.2 0.77 0.46 0.93015028 1.5 78 74 1.05 1.0 57 73 1.7 31 82 10.0 0.21 0.19 0.09032845 1.5 73 68 1.08 0.6 13 46 3.9 101 78 9.9 1.60 0.63 0.59035733 1.4 50 66 0.76 1.9 28 47 1.5 19 62 9.9 0.27 0.26 0.25040208 0.6 53 50 1.06 N/A N/A N/A 0.7 23 35 9.3 0.89 N/A 0.44080844 1.2 27 92 0.30 < . < . high-redshift ones (open circles and squares in Figure 5),when artificially redshifted these galaxies look very sim-ilar to high-z star-forming galaxies, with 72% chance ofbeing drawn from the same parent population accordingto a standard Kolmogorov-Smirnov test. We present allrelevant values in Table 2, along with measurements attheir real redshift. We caution the reader, however, tothe fact that the observed ratios at low redshift are stillmuch smaller than found in local spiral galaxies, whichhave v/σ values of 10-20.The main kinematic difference when comparing LBAsand local spirals comes from gas velocity dispersions, in-dicating that LBAs have a dynamically thick structure,disk or otherwise. We find it unlikely that the dynam-ics in all of the LBAs is actually dominated by rotation,given low overall v/σ values. Instead, the trend withstellar mass might simply indicate a colder, less randomdynamical structure in the process of forming a disk fromthe dynamically hot gas in more massive galaxies.Another quantity one can infer from gas kinematics isthe dynamical mass of the galaxy, assuming the velocitydispersion in the nebular gas is dominated by randommotions within a gravitational field. In that case, M dyn = Cσ rG , (1)where G is the gravitational constant, r represents thesize of the galaxy and C is a proportionality constantrelated to the geometry of the galaxy; for a disk, C = 3 . C = 5 (Erb et al. 2006b).The geometry is not always well determined in LBAs, butwe assume dispersion-dominated dynamics with C = 5to allow for a direct comparison with LBGs as discussedin Law et al. (2009). We use Pa- α radius determined insection 4.1. The results are presented in Table 2).Dynamical masses are well correlated with stellarmasses. M dyn agrees with stellar masses within a fac-tor of two (0.3 dex) in 63% of the galaxies in our sample,and they agree within a factor of three for 81% of theobjects. This means that for most LBAs, the high ob-served velocity dispersions can be explained simply byrandom motion of the gas given the observed masses. InFigure 6, we present the ratios between dynamical massand stellar mass as a function of stellar mass. We no- tice that these ratios are larger for less massive galaxies,with the implied dynamical masses an order of magni-tude smaller than the observed stellar mass for the mostmassive objects. This supports the hypothesis that thedynamical support offered by the rotating disk is moresignificant in the most massive star forming galaxies. It isalso interesting to notice the same trend, with very sim-ilar dynamical and stellar mass values, for high-redshiftstar-forming galaxies, as indicated by red squares. Thisreinforces the analogy between LBAs and LBGs. Kinemetry Measurements
Another way of assessing the presence of a rotationalcomponent within the dynamics of the gas in each galaxyis provided by the kinemetry method, as introduced byKrajnovic et al. (2006). The method comprises a decom-position of the velocity moment maps into its Fouriercomponents, that is, for a given ellipse: K ( ψ ) = A + A sin( ψ ) + B cos( ψ ) + A sin(2 ψ ) ++ B cos(2 ψ ) + ..., (2)where ψ is the azimuthal angle along which one measuresa given velocity moment K (in our case, velocity v orvelocity dispersion σ ). Written in another way, K ( r, ψ ) = A ( r ) + N (cid:88) n =1 k n ( r ) cos [ n ( ψ − φ n ( r ))] , (3)where the expansion terms have been redefined as k n = (cid:112) A n + B n and φ n = arctan ( A n /B n ). For a detaileddiscussion of the method, see Krajnovic et al. (2006) andShapiro et al. (2008).For an ideal rotating disk, one would expect the veloc-ity profile to be perfectly antisymmetric, that is, the B term would dominate the Fourier expansion. Likewise,the velocity dispersion map is expected to be perfectlysymmetric, and therefore all terms with the exception of A would vanish.Shapiro et al. (2008) have used this method to ana-lyze the dynamics of high-redshift star-forming galaxiesobserved with the SINFONI instrument. In quantifyingthe asymmetry of the velocity moment maps, they haveFU Spectroscopy of LBAs 13 * /M Ο • v s h ea r / σ LBAs @ z~0.2LBAs @ z~2 (OSIRIS)LBAs @ z~2 (SINFONI)FS+ (2009)Law+ (2009)
Figure 5.
Ratios between velocity shear and velocity dispersion v shear /σ as a function of stellar mass ( black triangles ). The dashed lineshows a fit to the v shear /σ data at their intrinsic redshift. We see a moderate trend, indicating more massive galaxies have a strongerrotational dynamical component than less massive ones. Also shown as downward triangles are v shear /σ values for galaxies artificiallyredshifted to z ∼ z = 2. * /M Ο • -1.5-1.0-0.50.00.51.01.5 L og M dyn / M * LBAs
Law+ (2009)
Figure 6.
Ratio between dynamical and stellar masses as a func-tion of stellar mass. Black filled circles indicate the LBAs in thispaper, while the hollow red squares are the high-z LBGs from Lawet al. (2009). The dashed line indicates M dyn = M ∗ . More massivegalaxies, both at low and high redshifts, present lower ratios, with M dyn ∼ / M ∗ for galaxies with log M ∗ ∼ M (cid:12) . defined the quantities v asym = (cid:28) k avg , v B ,v (cid:29) r (4)and σ asym = (cid:28) k avg ,σ B ,v (cid:29) r . (5)By using local galaxies and numerical models as if ob-served at high redshift as templates for disk versusmerger events, they have found the threshold of K asym = (cid:113) v + σ = 0 . K asym = 0 . M ∗ /M (cid:12) = 9 .
9. According to a standard Kolmogorov-Smirnov test, there is a 0.7% probability that stellarmasses from K asym > . K asym < . K asym = 0 . . (cid:46) K asym (cid:46) .
0, which might in-dicate instead a transition region between disk galaxiesand mergers in the kinemetry plot. This region is wheremost star-forming galaxies at high-redshift lie, as is thecase for the LBAs. Ο • ) 01230.11.0 K a s y m Figure 7.
Kinemetric asymmetry measurements as a functionof galactic stellar mass. Left y-axis shows values of K asym , whileright y-axis shows quantities for histograms. Gray histogram showsnumber of galaxies that would be classified as mergers in Shapiroet al. (2008), while the green hatched histogram shows the num-ber of galaxies that would be classified as disks. Galaxies withhigh K asym are predominantly less massive, but the lowest valueof stellar mass for a galaxy with K asym < . Finally we compare our results with the high redshiftsimulations from Section 2.5. In Figure 8, we showthe measurements based on the simulations and com-pare them to the “intrinsic” values measured in the orig-inal (i.e. low redshift) data. The dashed lines show thesame threshold of K asym = 0 . K asym , i.e. they appearmore symmetric than they actually are. One-third of thegalaxies would be classified differently at high redshift(lower-right quadrant). The net effect is that the per-centage of galaxies classified as mergers drop from ∼ ∼ asym (Low-z data)0.11.0 K a s y m ( L B A s @ z ~ ) Figure 8.
Kinemetry measurements for our high-redshift simula-tions as a function of ”intrinsic” values measured at low redshift.Dashed lines show the same threshold of K asym = 0 .
5. Red pointsrepresent OSIRIS-AO simulations, while blue points represent theSINFONI non-AO simulations. The gray shaded area indicates theregion of the plot where one finds LBAs having high-asymmetryvalues at low redshift but low values at z ∼ . DISCUSSION
In this section we briefly discuss some of the currentmodels for galaxy formation, and how they may explainthe observed properties of both LBAs and LBGs in termsof their morphologies and kinematics. In addition, wediscuss some of the implications of the stellar mass de-pendence of observables discussed in the previous section.
Ionized gas kinematics as a diagnostic for galaxyformation mechanisms
In light of new techniques and integral-field instru-ments, recent studies of the kinematics of ionized gas athigh redshifts have been used as diagnostics for galaxyformation models attempting to explain the distinctiveproperties observed in star-forming galaxies at z ∼ − α images show we are tracingregions that contain approximately a third of the totalstellar mass in the galaxy (section 2.4). Furthermore,the gas in these regions is subject to a number of lo-cal feedback effects from stellar winds and turbulence,and thus may not always represent motion of the bulkof the dynamical mass within. Lehnert et al. (2009) ar-gue that the high velocity dispersion values could not besustained simply by cosmological gas accretion; instead,self-gravity drives the early stages of galaxy evolution un-til dense clumps collapse, at which point star formationis self-regulated by mechanical output of massive stars.Although the inferred dynamical masses from velocitydispersion can be explained for the most part as a resultof random motions within the potential well of the galaxy(section 4.2), there is a second-order effect in LBAs thatsupports the influence of star formation-driven turbu-lence; in Figure 9 we show v shear and σ as a function ofstar formation rates. Since all variables correlate withstellar mass, we present the residuals of a power-law fitfor all of them with respect to M ∗ , in order to exclude anyinduced correlations. v shear is independent of star forma-tion rates (52% null-hyphotesis probability according toSpearman’s ρ test), but more star-forming galaxies showstronger velocity dispersion (2% null-hypothesis proba-bility), supporting the idea that star formation has aneffect on σ values, generating turbulence from the en-ergy output in processes related to star formation. -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6SFR, res -0.4-0.20.00.20.4 v s h ea r , r e s -0.2-0.10.00.10.2 σ , r e s Figure 9.
Velocity shear and velocity dispersion as a functionof star formation rates. Values are the residuals from a fit withrespect to stellar mass ( see text ). Another relevant point in this discussion is that rotat-ing kinematics do not exclude the possibility of a merger-triggered starburst. In a hydrodynamical simulation of agas-rich major merger, Robertson et al. (2006) show that6 Gon¸calves et al.for wet mergers rotating disks may form approximately ∼
100 Myr after the final coalescence. At this point, alarge, rotating gaseous disk is formed, with kinemetricasymmetry indices that would in principle rule out themerger scenario (Robertson & Bullock 2008). The signa-ture for interactions might be as subtle as small displace-ments of the σ map peak from the center of the galaxy(Flores et al. 2006). Puech (2010) have used similar argu-ments through IFU studies of clumpy, intermediate red-shift ( z ∼ .
6) galaxies to claim that interactions mightbe responsible for driving star formation at all redshifts.In these cases, high gas fractions are a fundamental in-gredient in the formation of the disk. Indeed, a num-ber of recent CO observations indicate that star-forminggalaxies at high redshifts have a much larger gas fractionthan standard spirals in the local universe (Tacconi et al.2010; Daddi et al. 2010).It is revealing to compare kinematics of some LBAs inour sample with their optical morphologies as observedwith HST (Overzier et al. 2009, 2010). In Figure 10 wereproduce the velocity map of 210358 and 135355 andcompare them with their optical image. Although theyappear as rotating disks in Pa- α , their optical morpholo-gies indicate recent major merger events, with quantita-tive classification supporting that view. When simulatedat higher redshift, both the morphology and the veloc-ity structure appear much more regular (see Fig. 1 inOverzier et al. (2010) and Figs. 2 and 10 in this paper).Also, on larger scales LBAs seem to support the ideaof mergers as triggers for the high star formation ratesobserved in these galaxies, as these galaxies tend to pairwith other galaxies more strongly than a random sampledoes (Basu-Zych et al. 2009b). Some studies at high-redshift have concluded that the galaxy number densityis not high enough to account for mergers in all observedstarbursts (Conroy et al. 2008; Genel et al. 2008), whilesome morphological studies indicate the merger fractionis high, up to 50%, with M ∗ > M (cid:12) galaxies under-going ∼ z > The dependence of rotational properties on stellarmass
It has been shown that stellar mass in star-forminggalaxies at high-redshift correlates with a number ofphysical properties, such as metallicity, star formationrates and age (Erb et al. 2006b,a; Magdis et al. 2010).Some results from kinematic studies of the H- α emis-sion at high redshift also indicate this dependence, withmore massive objects being more extended and present- ing higher v/σ ratios (Law et al. 2009; F¨orster Schreiberet al. 2009).In this work, we have shown that massive galaxies aremore likely to present disk-like features, as evidenced byhigher v/σ ratios and higher levels of symmetric kine-matics, while gas kinematics in less massive objects isdominated by random motions, as indicated by higherdynamical-to-stellar mass ratios as inferred from veloc-ity dispersion measurements. This distinction is particu-larly important when taking into account the stellar massfunction of LBGs at z ∼ −
3. Reddy & Steidel (2009)have found that the stellar mass function is particularlysteep at these redshifts, which means an elevated contri-bution from less massive galaxies. That in turn wouldsuggest more random dynamics for the majority of star-forming galaxies in the early universe, which are respon-sible for a significant fraction of the stars observed today– ∼
45% of the present-day stellar mass have formed ingalaxies with L bol < L (cid:12) (Reddy & Steidel 2009).The dependence on mass is predicted even in moretraditional star formation models. From a large N -body/gasdynamical simulation, Sales et al. (2009) haveshown that the angular momentum in a z ∼ α traces a re-gion containing a fraction of the stellar mass. It would beinteresting to determine whether these relations still holdfor the stellar mass contained within that small region,but for accurate measurements we need high-resolutionnear-infrared imaging, in order to trace stellar mass dis-tribution at sub-kpc scales. Alternatively, longer expo-sures or more sensitive instruments capable of tracingstellar dynamics instead of nebular gas could probe thekinematic properties at low surface brightness regions.Evidently, this is more difficult at high redshift, wherecosmological dimming decreases surface brightness val-ues by a factor of up to 200.It is also unclear whether the trend with stellar massrepresents an evolutionary effect or simply distinct for-mation scenarios. It is tempting to assume these galaxieskeep forming stars for a period of time, increasing stel-lar mass while at the same time settling onto a rotatingdisk. However, this would mean that LBAs would nec-essarily keep elevated star formation rates for a periodover 1 Gyr. Dynamical times of objects containing mul-tiple star-forming regions, however, are too short (on theorder of few tens of Myr), and the galaxy would coalescemuch more rapidly. Therefore, a continuous inflow of gasor a sequence of minor mergers feeding star formation inthese galaxies would be necessary to keep the observedstar formation rates. Alternatively, it is possible thatmore massive galaxies have experienced more violent starformation episodes in the past, after which the dynami-cal structure has cooled down. An in-depth comparisonwith hydrodynamical simulations, with careful examina-tion of star formation histories in LBAs, is required toexamine each hypothesis in detail. SUMMARY AND CONCLUSIONS
FU Spectroscopy of LBAs 17
Figure 10.
Velocity maps at low (left) and high (center) redshifts for 210358 and 135355. On the right we show the optical morphologiesof each object as seen by HST, combining optical(orange) and ultraviolet(blue) data. High-z simulated map for 210358 is for OSIRIS data,while for 135355 this is the simulated SINFONI data. In the top case we see a galaxy for which a disk is apparent even at low redshift,while in the second case we notice the effect of loss of spatial resolution. Both these galaxies are classified as mergers through quantitativemorphological analysis of the optical images.
We have performed adaptive-optics assisted observa-tions of 19 Lyman Break Analogs (LBAs) with theOSIRIS spectrograph at the Keck telescope. By study-ing spatially resolved Pa- α emission in these objects, weare able to draw the following conclusions:(1) All galaxies show high velocity dispersions, indi-cating gas dynamics with a strong random component.Most galaxies show velocity shears of the same order ofmagnitude as velocity dispersions along the line of sight.This is consistent with our general picture of LBAs asdynamically young, starburst-dominated galaxies thatare frequently undergoing mergers as shown by our HSTdata;(2) The kinematics in LBAs are remarkably similar tohigh-redshift LBGs, which have also been the target ofIFU studies. This is demonstrated by artificially red-shifting the LBA sample to z ∼ z ∼ α kinematical data with accurate rest-frame op-tical morphologies that can now be measured with theIR channel on Wide Field Channel 3 aboard HST. Fur-thermore, ALMA will allow detection of molecular gas8 Gon¸calves et al.in a large number of star-forming galaxies at redshifts z ∼ −
3, which should shed more light on the issueof gas-rich mergers. At low redshifts, ALMA will allowhigh-resolution measurements of molecular gas distribu-tion and kinematics, providing deeper understanding ofthe conversion of gas into stars. Finally, the upcoming20- and 30-m class telescopes, which should be opera-tional at the end of the decade, will allow IFU studies ofLBGs with higher sensitivity and resolution levels com-parable to what is available now to LBAs, while the latterwill be resolved at scales of giant molecular clouds, andwe will be able to study the physical processes of starformation in situ . In all cases, the LBA sample studiedin this paper offers a unique low redshift dataset usefulfor contrasting and comparing with starbursts at highredshift.The authors would like to thank Jim Lyke, Al Conrad,Randy Campbell and Hien Tran for invaluable assistancewith the laser observations. We also thank the anony-mous referee for useful comments regarding dynamicalmasses. TSG thanks Brant Robertson for useful discus-sions concerning theoretical modeling of galaxy forma-tion at z ∼
2. We would also like to thank those ofHawaiian ancestry for hospitably allowing telescope op-erations on the summit of Mauna Kea.2. We would also like to thank those ofHawaiian ancestry for hospitably allowing telescope op-erations on the summit of Mauna Kea.