Table Of ContentPreprinttypesetusingLATEXstyleemulateapjv.05/12/14
SPATIALLY EXTENDED AND HIGH–VELOCITY DISPERSION MOLECULAR COMPONENT IN SPIRAL
GALAXIES:
SINGLE-DISH VS. INTERFEROMETRIC OBSERVATIONS
Anahi Caldu´-Primo 1 Andreas Schruba 2 Fabian Walter 1 Adam Leroy 3 Alberto D.Bolatto 4 Stuart Vogel 4
(Accepted to AJ December 6, 2014)
ABSTRACT
Recent studies of the molecular medium in nearby galaxies have provided mounting evidence that
the molecular gas can exist in two phases: one that is clumpy and organized as molecular clouds
5
and another one that is more diffuse. This last component has a higher velocity dispersion than the
1
clumpy one. In order to investigate these two molecular components further, we compare the fluxes
0
2 andlinewidthsofCOinNGC4736andNGC5055,twonearbyspiralgalaxiesforwhichhigh–quality
interferometricaswellassingle–dishdatasetsareavailable. Ouranalysisleadstotwomainresults: 1)
n
Employing three different methods, we determine the flux recovery of the interferometer as compared
a
to the single–dish to be within a range of 35–74% for NGC4736 and 81–92% for NGC5055, and 2)
J
when focusing on high (SNR≥5) lines of sight, the single–dish line widths are larger by ∼(40±20)%
2
thantheonesderivedfrominterferometricdata; whichisinagreementwithstackingalllinesofsight.
2
These results point to a molecular gas component that is distributed over spatial scales larger than
30(cid:48)(cid:48)(∼1kpc), and is therefore filtered out by the interferometer. The available observations do not
]
A allow us to distinguish between a truly diffuse gas morphology and a uniform distribution of small
clouds that are separated by less than the synthesized beam size (∼3(cid:48)(cid:48) or ∼100pc), as they would
G
both be invisible for the interferometer. This high velocity dispersion component has a dispersion
. similar to what is found in the atomic medium, as traced through observations of the Hi line.
h
p Keywords: galaxies: ISM — ISM: molecules — ISM: clouds — radio lines: galaxies
-
o
r 1. INTRODUCTION gestiveadditionalmoleculargascomponentfromanout-
t
s In the classical picture of the interstellar medium side perspective.
a Molecular gas is preferentially observed through CO
(ISM), molecular gas is thought to be organized mostly
[ emission, which can be mapped both with single–dish
inside giant molecular clouds (GMCs), typically traced
telescopes and with interferometers. Single–dish tele-
1 by CO emission, with temperatures of 10 − 20 K and
v internal velocity dispersions of ∼ 2 − 8km s−1 (e.g., scopes, especially after the development of multi-beam
6 Bolatto et al. 2008). This simple picture, however, heterodynereceiverarrays,providealargeinstantaneous
4 does not hold at the Galactic level, were different stud- field of view and good sensitivity, although at limited
6 ies have shown that diffuse CO emission is pervasive spatial resolution. Interferometers, on the other hand,
5 providehigherspatialresolution,withthedrawbackthat
within our Galaxy (Polk et al. 1988; Liszt & Lucas
0 notallspatialscalescanberecovered. Thelargestspatial
1998; Goldsmith et al. 2008, and references therein).
. scale to which an interferometer is sensitive to, is deter-
1 Based on studies of molecular gas in our Galaxy, molec-
minedbytheshortestbaselines(i.e.,theclosestprojected
0 ular clouds are generally classified as either diffuse
5 (A ∼0.2, 30(cid:46)T(K)(cid:46)100), translucent (A ∼1–2, distance between two individual telescopes in the plane
1 15V(cid:46)T(K)(cid:46)50), or dense (A ∼5–10, 10(cid:46)TV(K)(cid:46)50) perpendicular to the line of sight). Thus, any emission
: (e.g., Snow & McCall 2006). VOn the one hand, studying arising from structures larger than this spatial scale is
v ‘invisible’totheinterferometer. Ifmostgasinsidegalax-
theISMintheGalaxymightseemadvantageousfromthe
Xi sensitivityandresolutionpointsofview. Galacticstudies iesweredistributedinstructures(e.g.,molecularclouds)
that are smaller than this largest spatial scale (and that
r can carry out observations of many different molecules,
a isotopologues, and transitions; while extragalactic stud- are separated from each other by at least the synthe-
sizedbeamsize),theneitherinstrumentwoulddetectthe
ies typically detect only the most abundant molecule in
same gas component, only with different spatial resolu-
its brightest transition. On the other hand, due to pro-
tion. Therefore, the flux recovered by the interferometer
jection effects and uncertainties in distance determina-
depends on the gas distribution and the molecular cloud
tions,itisdifficulttoquantifytheamountofgaspresent
spatialscalespresentintheobservedgalaxyascompared
in this diffuse molecular component within our Galaxy.
to the specific resolution of the interferometer’s configu-
Nearby galaxies offer the opportunity to study this sug-
ration.5
Pety et al. (2013) conducted a study of the molec-
1Max-Planck-Institut fu¨r Astronomie, K¨onigstuhl 17, 69117
ular gas phase as traced by the CO(1–0) emission in
Heidelberg,Germany;[email protected]
2Max-Planck-Institutfu¨rExtraterrestrischePhysik,Giessen- M51. They used single–dish IRAM 30m data, Plateau
bachstr. 1,85748Garching,Germany de Bure Interferometer (PdBI) data, and a combination
3National Radio Astronomy Observatory, 520 Edgemont
RoadCharlottesville,VA22903,USA
4Department of Astronomy, University of Maryland, College 5 During the image deconvolution the flux recovery will also
Park,MD20742-2421,USA stronglydependonthenoiseproperties(e.g., Helferetal.2002).
2 Caldu´-Primo et al.
of both data sets, to compare flux recovery and veloc- discussion and summary is provided in Section 5.
ity dispersions measured in these three data sets. They
show that the interferometer recovers only (50±10)% 2. DATAANDSAMPLE
of the total flux, indicative of a molecular component
In order to carry out the analysis, the data need to
that is missed by the interferometer. In addition, they
meet three conditions: high sensitivity, a clear contrast
found that the purely single–dish velocity dispersions
betweenthescalesrecoveredbytheinterferometerandby
are typically twice as large as the interferometric veloc- the single–dish, and available Hi velocity maps. Even
itydispersions. Asstatedbefore,moleculargasmightbe
though there is a large survey, BIMA SONG (Helfer
present in different cloud types, with different tempera-
et al. 2003), in which the CO(1–0) molecular emission
tures, densities, and spatial distribution. These different
in nearby galaxies has been mapped with an interferom-
physical configurations would modify the expected ve-
eter, its sensitivity and resolution render it not suitable
locity dispersions present in the gas. Thus, their result
for this specific study. Therefore, we limit our study to
implies not only that ∼50% of the gas is missed by the
the 2 galaxies for which high–sensitivity and resolution
interferometer, but that this gas gives rise to larger ve-
CARMA interferometric maps exist.
locity dispersions compared to the velocity dispersions
measuredforthemoleculargas(i.e.,COemission)inside 2.1. CO(1–0) Data
the GMC sample (which the interferometer is sensitive
2.1.1. Interferometric Data
to).
In a previous paper, Caldu´-Primo et al. (2013) inves- The interferometric CO data come from the CARMA
tigated the differences in velocity dispersion of atomic interferometer. Both galaxies are mapped in the
(Hi) gas (from the THINGS survey, Walter et al. 2008) CO(1–0) line covering the inner ∼ 3(cid:48) diameter of the
and of molecular gas (from the HERACLES CO survey, targets via a 19–point mosaic. Observations were taken
Leroyetal.2009)inasampleof12nearbyspiralgalaxies. between January 2007 and July 2008 using the C, D,
Onspatialscalesof∼1kpcwemeasuredsimilarvelocity andEconfigurationsresultinginasynthesizedbeamsize
dispersions for both Hi and CO (as measured by a sin- of 3.51 × 2.97(cid:48)(cid:48) for NGC4736 and of 3.24 × 2.81(cid:48)(cid:48) for
gle–dishtelescope),andinterpretedthisastheexistence NGC5055. The correlator was setup with two overlap-
of a high–dispersion molecular gas component. ping 62MHz spectral windows to yield a spectral resolu-
Based on these two previous results, we deem impor- tion of 1.95MHz (or 5.08km s−1 at 115 GHz)6.
tanttorigorouslyinvestigatethepresenceofthemolecu- The CARMA data were calibrated using MIRIAD
lar gas component giving rise to these large velocity dis- (Sault et al. 1995), weighted by noise variance, and ap-
persions and determine its mass fraction as compared to plied robust weighting. The data were cleaned down to
themoleculargasinsidemolecularclouds. Theexistence a cut–off of 1.5 times the theoretical rms noise in each
of such a molecular gas component would require that pixelusingMOSSDI2,whichperformsasteerCLEANon
CO single–dish and interferometric observations be an- a mosaicked image (Steer et al. 1984). The images are
alyzedandinterpretedusingdifferentassumptionsabout primary–beamcorrected. Thefluxscalewasdetermined
theoriginoftheemission. Forexample,studiesofthere- by observations of flux standards, including Uranus and
lationshipofmoleculargasandrecentstarformationrate Mars. In La Vigne (2010) a full description of the data
inbothnearbyordistantgalaxies(e.g.,reviewbyKenni- reduction can be found. In their analysis they compare
cutt&Evans2012)implicitlyassumethatmoleculargas the CARMA observations with the BIMA SONG data
as traced by CO observations is inside molecular clouds (Helfer et al. 2003). After verifying the flux scales of
and participates in the current star formation process; both datasets they concluded that no flux rescaling was
whereastheassumptionofasignificantdiffusemolecular required. The rms noise sensitivity at 10km s−1 reso-
componentwouldallowforsomeoftheobservedmolecu- lution is 27mJybeam−1 (237.5mK)7 for NGC4736 and
largasbeingnotdirectly(oratall)relatedtothecurrent 19mJybeam−1 (193.2mK) for NGC5055.
star formation activity. In this study we will further in-
2.1.2. Single–dish Data
vestigate the presence of a wide–spread, high–velocity
dispersion molecular gas component by comparing CO We also use single–dish CO(1–0) mapping from the
line widths as measured from interferometric and sin- “Nobeyama CO Atlas of Nearby Spiral Galaxies” (Kuno
gle–dish observations, as well as the fluxes measured by et al. 2007), which encompasses 40 nearby spiral galax-
meansofbothinstruments. Unfortunately,highsensitiv- ies at distances smaller than 25Mpc. The observations
ity interferometric maps of nearby galaxies are still very were carried out with the Nobeyama 45m telescope and
scarce, a situation that will however soon change with typically cover most of the optical disk with a spatial
the advent of ALMA. Thus, we concentrate our current resolution of 15(cid:48)(cid:48).
study on two galaxies: NGC4736 and NGC5055. We InthecaseofNGC4736,thereceiverusedwasBEARS,
comparesingle–dishCOdatafromtheHERACLESSur- which consists of 25 beams (Sunada et al. 2000). In
vey (Leroy et al. 2009) and the “Nobeyama CO Atlas of this case digital spectrometers were used as backends.
Nearby Spiral Galaxies” (Kuno et al. 2007) to interfero- The total bandwidth and frequency resolution were
metric observations taken by La Vigne (2010) using the 512MHz and 605kHz, which at 115GHz correspond to
CombinedArrayforResearchinMillimeterwaveAstron- 1331km s−1 and 1.57km s−1, respectively.
omy (CARMA).
6 ThenowpubliclyavailabledataforNGC4736has10kms−1
The paper is organized as follows: In Section 2 we
since the original data were unfortunately lost, and cannot be re-
discussthemaincharacteristicsoftheemployeddatasets covered.
andthetwogalaxiesstudied. InSection3wediscussour 7 All sensitivities in K are stated using the beam size of the
methodology, to finally show the results in Section 4. A originaldata.
Single-Dish vs Interferometric CO Velocity Dispersions 3
In the case of NGC5055, a 4–beam SIS receiver 3. METHODOLOGY
was used, together with acousto–optical spectrometers
3.1. Data Homogenization
(AOS) used as backends. This configuration yields a
frequency resolution and channel spacing of 250kHz For our analysis we convolve the data to match the
and 125kHz, respectively, providing a total bandwidth spatialandspectralresolutioninallcases. Thiscommon
of 250MHz. At 115GHz this corresponds to veloc- resolutionisdefinedbythepoorestresolutionofourdata
ity resolution and velocity coverage of 0.65km s−1 and sets: the spectral resolution is set by the CARMA data,
650km s−1, respectively. and the spatial resolution is set by the Nobeyama data.
In both cases the spectra were then smoothed to Afterhanningsmoothingthedatatoacommonspectral
5km s−1 (Nakai et al. 1994) as part of the data re- resolution(whichisdiscussedbelow), weusetheREGRID
duction. This is the spectral resolution of the data task in MIRIAD (Sault et al. 1995) to put all the data
we use. At 10km s−1 resolution, the rms noise sen- sets on a common grid. Finally, we convolve the data
sitivity is 69mJybeam−1 (28.3mK) for NGC4736 and sets to the 15(cid:48)(cid:48) Nobeyama limiting spatial resolution.
172mJybeam−1 (70.7mK) for NGC5055. The limiting spectral resolution is different for the
two galaxies. NGC4736 has a CARMA–limited
10km s−1 spectral resolution, while NGC5055 has a
2.2. CO(2–1) Single–dish Data CARMA/Nobeyama–limited 5km s−1 spectral resolu-
We use CO(2–1) data from the HERACLES sur- tion. Therefore,forNGC5055wefirstcarryouttheanal-
vey (Leroy et al. 2009). This survey used the IRAM ysis at a 5km s−1 spectral resolution and then compare
30m telescope to map molecular gas from 48 nearby the results when using a 10km s−1 spectral resolution.
galaxies inside the optical radius (r25). This survey Wefindthatourresultsdonotchange. Thus,forsakeof
used the HERA (a 9–beam dual–polarization hetero- uniformity,weherepresenttheanalysisusingacommon
dyne receiver array) together with the Wideband Line spectral resolution of 10km s−1 for both galaxies.
Multiple Autocorrelator (WILMA) backend. WILMA InordertocorrectlyNyquistsamplethedata, wecon-
consists of 18 units of 2MHz channel width each struct a hexagonal grid with a 7.5(cid:48)(cid:48) spacing (half of the
and 930MHz bandwidth, yielding a velocity resolu- beam size).
tion of 2.6km s−1 in a 1200km s−1 velocity range at Due to the attenuation of the primary beam in inter-
230GHz. The spatial resolution obtained at this fre- ferometricobservations,thesensitivitydropstowardsthe
quencyis13(cid:48)(cid:48) aftergridding. Thermsnoisesensitivityat edgesofthemap. Weconstructasensitivitymask(after
10km s−1 is of 90mJybeam−1 (11.5mK) for NGC4736 primary– beam correction of the cube) from the inter-
and 94mJybeam−1 (12mK) for NGC5055. ferometricnoisemapandapplyittoalldatasets. Inthis
way we make sure: 1) to use the same individual lines–
2.3. Hi Data of–sight (LOS) in all data sets, and 2) that we only use
LOSwhichhavehomogeneousnoisepropertiesinthein-
For our analysis we use Hi intensity weighted veloc-
terferometricdata. Todoso,wederivethegalactocentric
ity maps. These maps come from the THINGS survey
distance at which the noise in the CARMA observations
(Walteretal.2008). Thissurveyencompasses34nearby
is increased by 30%, and only use this region for further
galaxies at distances of 2−15 Mpc. The observations
analysis. For clarification, we show this region as a solid
for the two galaxies we analyze have high spectral (5.2
blackcircleinFigure2foreachgalaxyontopoftheHE-
km s−1) and spatial (∼11(cid:48)(cid:48) for natural weighting) reso-
RACLES IRAM 30M integrated intensity map in gray
lutions.
shades. Insidethecirclewealsoshowourhexagonalgrid
points (blue x symbols). Finally, we also show the 15(cid:48)(cid:48)
2.4. Sample radial bins used later in this paper as red ellipses.
As mentioned in the introduction, our sample is FromFigure2wenotethatthesensitiveregiondefined
limited by the availability of sensitive, wide-field inter- bytheCARMAprimarybeamresponseisindependentof
ferometric data sets. We thus focus on two galaxies theinclinationofthegalaxy. However,theradialbinsare
only, NGC4736andNGC5055. Theyhavethefollowing ellipsesbecauseoftheinclination. Thisresultsinhaving
properties: lower covering factors within the ellipses for the outer-
most radial bins. To quantify this effect, we present in
NGC4736 is an isolated spiral galaxy of type SAab Table1 thenumberofLOSfromthehexagonalgridthat
(Kennicutt et al. 2003). It is at a distance of 5.20 ± fallwithineachradialbinandwithinourmask(thirdcol-
0.43Mpc(Tonryetal.2001),hasaninclinationof41◦(de umn), ascomparedtothenumberofLOSthatwouldbe
Blok et al. 2008), and an optical radius R ≈ 5arcmin present without taking into account thesensitivitymask
25
(∼7.5kpc) (Kennicutt et al. 2003). (second column).
NGC5055 is a flocculent spiral galaxy classified as
SAbc (Kennicutt et al. 2003). Its distance is 7.8 ±
2.3Mpc (Masters 2005), its inclination is 59◦ (de Blok
3.2. Measuring Line Widths for Individual LOS
et al. 2008), and its optical radius is R ≈ 6arcmin
25
(∼13kpc) (Kennicutt et al. 2003). WeestimatethespectrallinewidthsusingsingleGaus-
Moreinfomationonthegalaxypropertiescanbefound sianfits. Inordertoobtainreliablemeasurementsforin-
inKennicuttetal.(2003),Walteretal.(2008),andLeroy dividual lines of sight (LOS), we require that they have
etal.(2009). WeshowtheintegratedintensityCOmaps high significance. Therefore, we begin by computing the
of the three different surveys at a common resolution of line widths for the LOS which comply with the follow-
15(cid:48)(cid:48) in Figure 1. ing characteristics: a)lie within the CARMA sensitive
4 Caldu´-Primo et al.
NGC 4736
CARMA CO(1-0) HERACLES CO(2-1) Nobeyama CO(1-0)
09'
0) 08'
0
0
2
c (J
e
D 07'
+41°06'
51m00.72s 54.24s 12h50m47.76s 51m00.72s 54.24s 12h50m47.76s 51m00.72s 54.24s 12h50m47.76s
RA (J2000)
NGC 5055
CARMA CO(1-0) HERACLES CO(2-1) Nobeyama CO(1-0)
03'
0)
0
20 02'
c (J
e
D
+42°01'
55.20s 50.40s 13h15m45.60s 55.20s 50.40s 13h15m45.60s 55.20s 50.40s 13h15m45.60s
RA (J2000)
Figure 1. COintegratedintensitymapsofNGC4736(top)andNGC5055(bottom)at15(cid:48)(cid:48)resolution(thebeamsizeisshownatthelower
left corner). The first column shows the CARMA CO(1–0) interferometric data, the second column shows the HERACLES IRAM 30m
CO(2–1)maps,andthethirdcolumnshowstheNobeyama45mCO(1–0)maps. ThethreemapswereconstructedusingtheHERACLES
mask (details in Section4.5). For both galaxies the intensity range goes from 0 to 38 Kkms−1, and the 6 contours show intensity levels
separated by 6 Kkms−1, starting at 6 Kkms−1. The black circle shows the region where the CARMA cube has homogeneous noise
properties,i.e.,wherethermsiswithin30%ofthevalueatthecenterofthemap.
region (black circle in Fig.1), b)have peak SNR> 58, moves LOS with significant negative structures (‘bowls’)
and c)h√ave integrated intensities of at least 10σ, where in the interferometric data9.
σ =∆v nrms and n is the number of measured points The errors are computed as discussed in detail in
used to fit the line. For each of these high SNR LOS Caldu´-Primoetal.(2013): foreachLOSweaddrandom
we use the Hi first moment value,i.e., the local inten- noise to the original spectrum and redo the line width
sity–weightedvelocity, asaproxyforthevelocitywhere measurement. We repeat this procedure a 1000 times.
we expect to find the corresponding CO line (We take The adopted error in the line width measurement is the
the Hi data from the THINGS survey (Walter et al. dispersion between the results of these 1000 iterations.
2008)). This is based on our earlier findings that the
Hi and CO line–of–sight velocities agree well at the 3.3. Measuring Line Widths for Stacked Spectra
resolutionweareworkingat(Schrubaetal.2011;Caldu´-
The main purpose of the stacking is to increase the
Primo et al. 2013). We attempt to fit a single Gaussian
SNR in our data as well as to include all available LOS
toeachspectrumusingMPFIT(IDLprocedurefromCraig
in our analysis. The stacked spectra are the most mean-
Markwardt) inside a window of 100km s−1 centered in
ingful measurements, as they represent the luminosity–
the Hi first moment value. If the fitted Gaussian is
weighted FWHM including all LOS inside a tilted ring
broaderthan100km s−1,weproceedtorefitwithadou-
(and inside the CARMA sensitive region). In order to
ble–horn profile (a Gaussian scaled by a symmetric sec-
stack individual LOS coherently, we first remove the cir-
ond order polynomial Saintonge (2007)). Double–horn cular motions of the four data sets using the Hi first
profiles are mostly encountered in the inner regions of
moment value. The THINGS survey has very high sen-
galaxies, wherebulkmotionsorbeamsmearingcouldbe
sitivity, resulting in reliable velocity maps. Even in the
dominant. These will be discarded for our analysis (see caseofNGC4736,whichhaslowcontentofHiinthecen-
below). Theintegratedintensityselectioncriteria(b)re-
9 These negative bowls are expected to be present in interfero-
8ThisSNRiscomputedinthe10kms−1spectralresolutionand metricdatabecauseoftheextrapolationrequiredtoovercomethe
15(cid:48)(cid:48) spatialresolutioncubes. incompletenessoftheu–vcoverage.
Single-Dish vs Interferometric CO Velocity Dispersions 5
09'
03'
) 08' )
0 0
0 0
0 0 02'
2 2
(J (J
c c
e e
D 07' D
+42°01'
+41°06'
51m00.72s 54.24s 12h50m47.76s 54s 48s 13h15m42s
RA (J2000) RA (J2000)
Figure 2. HERACLESIRAM30MCOintegratedintensitymapsofNGC4736(left)andNGC5055(right)at15(cid:48)(cid:48) resolution(grayscale).
Theblacksolidcirclerepresentstheareawheretheinterferometricprimarybeamsensitivityiswithin30%ofthevalueofthatatthecenter
of the map. The blue points inside the circle show the hexagonal grid points that are used for sampling the data. The red ellipses show
the 15(cid:48)(cid:48) radial bins used for stacking. From inside out and in units of r25 they are, for NGC4736: 0.04, 0.10, 0.17, 0.23, 0.29, 0.35, 0.41,
and0.46;andforNGC5055: 0.03,0.07,0.11,0.15,0.19,0.23,0.27,and0.30.
Table 1
Fillingfactorswithineach15(cid:48)(cid:48) radialbin
#gridpoints #gridpointsin #ofsignificantgridpoints,SNR>5b
R25 inradialbin CARMA–sens. region(%)a CARMA(%) HERACLES(%) Nobeyama(%) THINGS(%)
1 2 3 4 5 6 7
NGC4736
0.04 10 10 (100) 4 (40) 9 (90) 3 (30) 6 (60)
0.10 35 35 (100) 28 (80) 32 (91) 20 (57) 32 (91)
0.17 57 57 (100) 44 (77) 57 (100) 40 (70) 57 (100)
0.23 73 73 (100) 34 (47) 73 (100) 43 (59) 73 (100)
0.29 101 101 (100) 12 (12) 72 (71) 24 (24) 101 (100)
0.35 116 99 (85) 0 (0) 43 (43) 0 (0) 99 (100)
0.41 142 64 (45) 1 (2) 25 (39) 0 (0) 61 (95)
0.46 165 10 (6) 0 (0) 3 (30) 0 (0) 8 (80)
NGC5055
0.03 9 9 (100) 6 (67) 9 (100) 4 (44) 9 (100)
0.07 23 23 (100) 7 (30) 20 (87) 16 (70) 20 (87)
0.11 34 34 (100) 4 (12) 34 (100) 33 (97) 34 (100)
0.15 54 54 (100) 12 (22) 54 (100) 47 (87) 54 (100)
0.19 66 54 (82) 11 (20) 54 (100) 48 (89) 54 (100)
0.23 86 44 (51) 9 (20) 39 (89) 27 (61) 39 (89)
0.27 89 27 (30) 5 (19) 27 (100) 18 (67) 27 (100)
0.30 117 26 (22) 11 (42) 24 (92) 14 (54) 26 (100)
a NumberofgridpointswithintheCARMAsensitiveregioninaparticularbin. Inparenthesistheisthe%ofthesegridpointscompared
tothetotalnumberofgridpointswithineachradialbin(i.e. comparedtoColumn2).
b ForeachinstrumentandforeachradialbinweshowthenumberofLOSwithSNR>5andwithintheCARMA–sensitiveregion(black
circleinFig.1). Inparenthesisisthe%ofthesepointscomparedtoColumn3.
tralparts,thepeakSNRislargerthan5σin97.3%ofthe spectrathathavebeenstacked(whichweassessthrough
LOS inside 0.5R . Once all spectra are centered in the bootstrapping following Caldu´-Primo et al. 2013).
25
same velocity frame (in this case zero velocity) they can
4. RESULTS
be stacked coherently. The stacking procedure, together
with the error determination, are described in detail in 4.1. FWHM as function of Galactocentric Distance
Caldu´-Primoetal.(2013). Afterthestackingisdone,we We can select which LOS we want to stack. Since
compute the line widths as discussed above. The total many galactic properties depend on the distance to the
error in the determination of line widths has contribu- center of the galaxy, we choose to stack the individ-
tions both due to noise (as modeled for the individual ual LOS inside tilted rings of 15(cid:48)(cid:48) width within each
LOS above) as well as due to the specific selection of galaxy. As a first step, we stack only individual LOS
6 Caldu´-Primo et al.
with SNR>5. We then compare the measurements ob- only. A possible explanation is that the faint LOS have
tained from these (high SNR) stacked spectra, to the systematicallywiderlinewidthsthanthehighSNRLOS
median values measured from the individual high SNR and when included in the stacks, they widen the result-
LOS. We do this comparison to test the validity of the ing spectral line width. Since high SNR LOS are less
stacking method. In Figure3 we plot the FWHM mea- than 20% of the total LOS in the CARMA data (which
surements as function of galactocentric distance for each account for ∼60% of the interferometric flux), we can-
of the 4 different data sets (we include Hi for sake of not rely on these measurements as being representative
completeness). The FWHM measured when fitting high of the underlying weaker emission.
significance(SNR>5)individualLOSareshowninsmall
greysymbols. TheFWHMmeasuredwhenonlystacking 4.2. Comparison of FWHM from Different Instruments
the grey points (i.e., individual LOS with SNR> 5) are
A visual inspection of Figures 3 and 4 already in-
shown in medium–sized blue symbols, whereas the me-
dicates that the CARMA data give the smallest line
dian FWHM of the individual high SNR LOS are shown
widths of all data sets (in those areas not affected by
inblacksmallsymbols(Theredpointsareexplainedbe-
beam smearing). Figure5 shows the ratio of the FWHM
low). Weseethatatsmallradii(R(cid:46)0.2R )weobserve
25 linewidthsmeasuredbythesingle–dishtelescopes: HE-
a rapid increase in the measured FWHM, however, here
RACLES (diamonds) and Nobeyama (squares) divided
theobservedspectraaresignificantlybroadenedbybeam
by the line widths measured by CARMA for our two
smearing (see Caldu´-Primo et al. 2013, for a thorough
galaxies: NGC4736 (Fig.5, left column) and NGC5055
discussion on this effect). The shaded regions highlight
(Fig.5, right column). The figure is divided in two. In
the radial range where beam smearing broadens the in-
Fig.5(a) we show the results for the stacked spectra in-
trinsiclineprofilebymorethan30%. Caldu´-Primoetal.
cludingallLOS.InFig.5(b)weshowtheresultsinvolving
(2013) show that for a galaxy with characteristics simi-
high SNR LOS: on top the stacked spectra, and on the
lar to NGC4736 beam smearing accounts for >30% of
bottom the median values of the individual LOS. For
the measured line width at R(cid:46)0.15R and for more
25 comparison, we also show the ratio of the line widths
than 10% out to 0.25R . In NGC5055, beam smear-
25 of the THINGS Hi data (triangles) with the CARMA
ing accounts for >30% of the observed line widths at
CO(1–0) data. Open symbols show spectra fitted by a
R(cid:46)0.2R , and >10% at 0.4R . At large radii out to
25 25 double–hornprofile(typicallyatsmallradiiwherebeam
0.5R , where line profiles are only mildly or not at all
25 smearing is dominant), and solid symbols correspond
affectedbybeamsmearing,wefindroughlyconstantCO
to Gaussian fits (typically in the galactic disk)11. The
velocitydispersions,ashavebeenfoundalreadyinearlier
shaded regions mark the galactocentric distance where
studies (e.g., Tamburro et al. 2009; Caldu´-Primo et al.
beam smearing is larger than 30% for NGC4736 and
2013). In the regions of interest (where beam–smearing
NGC5055, respectively. Beam smearing tends to repre-
is not dominant), the measurements obtained for LOS
sentthelarge–scalegaskinematicsinsteadoftheintrin-
with SNR>5 are in agreement.
sic line profile. As this affects all data sets equally (after
In Table1 (columns 4–7) we show the number of
LOS with SNR larger than 5 within each 15(cid:48)(cid:48) radial bin convolutiontoacommonresolution),itdrivesanyratios
of line widths toward unity, thus concealing any intrin-
andwithintheCARMA–sensitiveregionforeachofthe
sic differences. This is obviously the case at small radii
datasets(inparenthesisweshowwhichpercentageofthe
(inside the shaded regions). At larger radii, where beam
total number of LOS within the CARMA–sensitive re-
smearingisnotimportant, theCOlinewidthsmeasured
gion these high SNR LOS represent). It is important to
fromsingle–dishdatasets(butalsofortheinterferomet-
note that we select the individual LOS with high SNR
ric Hi data) are larger than the line widths measured
independently for each data set,i.e.,a high SNR LOS in
from the interferometric CARMA data.
the HERACLES cube does not necessarily have such a
high SNR LOS counterpart in the CARMA cube10(see In Table 2 we list the mean values for the ratios of
thesingle–dishlinewidthsfromHERACLESandNobe-
Table1). It is clear that by taking only high SNR mea-
maya to the interferometric line widths from CARMA,
surements we are excluding a large percentage of LOS
for data only mildly affected by beam smearing,i.e.,
from the analysis.
outside 0.15R for NGC4736 and outside 0.2R for
Therefore,wethenproceedtostackall LOSinsidethe 25 25
15(cid:48)(cid:48) wide tilted rings. The resulting stacked spectra are NGC5055. The main finding of this paper is that the
average ratio of single–dish to interferometer line width
shown in Figure4 for NGC4736 and NGC5055, respec-
is1.4±0.2whenlookingattheCO(1–0) transitionand
tively. In these figures we show the normalized stacked
taking into account only the high SNR LOS. This value
spectra with peak SNR higher than 5 in all data sets.
is in agreement with the ratio found when taking into
The corresponding FWHM obtained for these stacked
account all LOS (1.5±0.1). Thus single–dish observa-
spectraareplottedinFigure3inlargeredsymbols. The
tionstypicallytracemoleculargaswithlargerlinewidths
analysis of the stacked spectra (red symbols) gives the
thaninterferometricobservations. Thispointstotheex-
line widths for all LOS within a radial annulus in a lu-
istence of a molecular gas component that is missed by
minosity–weighted sense, thus also including faint LOS.
the interferometer and which has larger line widths.
We find that the line widths of the stacked spectra in-
cludingallLOSaretypicallylarger(by2–15%)thanthe
valuesobtainedwhenstackingindividualhighSNRLOS 4.3. Tests
10 This problem does not exist when stacking all LOS. In that 11 InFigure5(b)(bottom)weplotonlyfilledsymbols. Thereis
case we do not perform a preselection, we simply stack all LOS noinformationaboutlineprofilesinthisplot,sinceweareshowing
fallingwithinthecorrespondingradialbin. themedianofindividualmeasurements.
Single-Dish vs Interferometric CO Velocity Dispersions 7
Figure 3. ThemeasuredFWHMasafunctionofgalactocentricdistanceforthefourdifferentinstruments: CARMA(topleft),HERACLES
(topright),THINGS(bottomleft),andNobeyama(bottomright). ThesmallgreysymbolsshowindividualLOSwithpeakSNR>5. The
largeredsymbolsshowthevaluesobtainedwhenstackingallLOSwithinagiventiltedringbin,themediumbluesymbolsshowtheresults
whenstackingspectrawithSNR>5only,andtheblacksmallsymbolsshowthemedianvaluesforthegreypoints,i.e.,themedianvalues
oftheindividualhighSNRLOSmeasurements. Insidetheshadedareabeamsmearingcontributesbymorethan30%tothemeasuredline
width.
8 Caldu´-Primo et al.
(a) NGC4736
(b) NGC5055
Figure 4. Stackedspectrabygalactocentricdistanceinbinsof15(cid:48)(cid:48) width. Thespectraforthe4datasetsareshowntogether,therefore
fluxes have been rescaled to peak intensities of unity. The black line (labeled ‘C’) is the CARMA data, the red dashed line (‘H’) is the
HERACLESdata,andthegreendash-dottedline(‘N’)istheNobeyamadata. ForcomparisonwealsoshowtheTHINGSHidataasthe
blackdashedline(labeled‘T’).
4.3.1. Radial bin filling factor We compare the “true” FWHM (i.e.,the FWHM ob-
tained when stacking 100% of the LOS inside the radial
As noted before, in the case of NGC5055 we face in-
bin) to the mean FWHM obtained for each filling fac-
completeness for the outermost radial bins in this study
tor. FortheCARMAdatathevariationisatmost2%of
(see Figure5(a)). We test how representative the mea-
theoriginalvalue. FortheHERACLES,Nobeyama, and
sured values are, taking into account the incompleteness
THINGS data the variations are at most 2%, 3%, and
in these points. To do this we take the last complete
6%,respectively. Forthesethreedatasets,thevariations
radial bin in NGC5055, which goes from 45–60(cid:48)(cid:48) (cen-
rise to 5%, 12%, and 4%, respectively in the case where
tered at ∼0.15R ). We then take a grid of filling fac-
25 thefillingfactorisofonly10%. Lookingateachdataset
torsgoingfrom90–10%(i.e., howmanyindividualLOS
individually we note, as expected, that the dispersion in
within the radial bin are taken into account). We se-
the measurements increases with decreasing filling fac-
lect the corresponding percentage of random individual
tor, going from ∼1km s−1 for the 90% filling factor, up
LOS inside the radial bin and we stack them. Finally
to ∼4km s−1 for the 10% filling factor. In all cases, the
we measure the FWHM from the resulting stacked spec-
dispersion is of the same order as the errors calculated
trum. We perform this same procedure 1000 times, in
for the stacked spectra (∼2–4km s−1).
ordertoseehowmuchthemeasurementvariesbyselect-
ing random individual LOS. At the end we calculate a
4.4. FWHM dependance on azimuthal location
meanvalueoftheFWHMcomputed1000timesforeach
filling factor value and the dispersion among individual Sincethecircularvelocitieshavealargergradientalong
measurements. the minor axis (due to projection effects), we also test
Single-Dish vs Interferometric CO Velocity Dispersions 9
(a) StackingallLOS
(b) ResultsinvolvingindividualLOSwithSNR>5
Figure 5. RatiooftheFWHMcomputedforthedifferentinstrumentsovertheFWHMcomputedfromtheCARMAdataforNGC4736
(left column) and NGC5055 (right column). Top: Stacked spectra using all LOS inside radial bins of 15(cid:48)(cid:48). Center: Stacked spectra
of high SNR LOS. Bottom: Median values of the high SNR individual LOS. The different symbols correspond to the different ratios:
HERACLES/CARMAindiamonds,THINGS/CARMAintriangles,andNobeyama/CARMAinsquares. Theshadedareascorrespondto
theregionsinsidewhichbeamsmearingcontributesmorethan30%tothemeasuredlinewidthmeasurements. Thedashedlinerepresents
unity. ThecolorscorrespondtothecolorsinFigure3.
10 Caldu´-Primo et al.
whether the angular distance to this axis affects our
FWHM measurements. For each galaxy we mask out Table 2
LinewidthRatios
a wedge of X degrees around the minor axis, where X
ranges from 5 to 40 degrees in steps of 5 degrees. We
HERACLES / Nobeyama /
then stack the remaining LOS and compare the FWHM CO(2–1) CO(1–0)
CARMA CARMA
to the value obtained when using no mask. For both CO(1–0) CO(1–0)
galaxiesandforallanglesthedifferencesinFWHMmea- Stacked,allLOSa 1.2±0.1 1.5±0.1
surements are less than a channel width (<2.5km s−1) Stacked,SNR>5b 1.3±0.1 1.4±0.2
Median,SNR >5c 1.3±0.2 1.3±0.2
which is 5–10% of typical FWHM measurements, and
therefore insignificant. a MeanofthevaluesmeasuredwhenstackingallLOS.
b Mean of the values measured when stacking individual LOS with
SNR >5.
4.4.1. Pointing uncertainties
c Median of the values measured for the individual LOS with SNR
Atthisstagewetestwhetherthesingle–dishpointing >5.
uncertainties could be responsible for the excess in the
FWHM measurements. We convolve each single–dish use to blank the noise–dominated regions in the cubes.
cube with a Gaussian of width equal to the correspond- We construct these 3D–masks by finding regions with
ing pointing uncertainty12. We compare the resulting peak SNR larger than 5σ in at least two consecutive
FWHMtotheFWHMmeasuredfromtheoriginalcubes. channels. We then expand these masks to masks con-
In all cases the differences between both measurements structed in the same way, but using a SNR cut of 2σ
are less than 2%. The pointing uncertainties are thus instead. Finally, we expand the mask by half a beam
insignificant in the determination of FWHM values. size and include adjacent velocity channels to capture
all emission. We construct two masks for each galaxy;
4.4.2. Filtering of extended emission one for the interferometric CARMA data set and an-
other one for the single–dish data sets. For the sin-
Inthissubsectionwesimulatewhattheinterferometer
gle–dishdatasetsweusetheHERACLEScubestocon-
would detect by using the task simobserve in the Com-
struct the masks, as it has higher SNR as compared to
mon Astronomy Software Applications package (CASA)
the Nobeyama data sets. The final mask is applied to
(McMullin et al. 2007). As our galaxy template, we
each cube and the integrated intensity is calculated.
use the HERACLES single–dish cube, which contains
We compare the fluxes in three ways: A) We com-
all emission (clumpy and/or diffuse). We simulate
pute the integrated intensity map for each data set in-
what CARMA’sE–configuration UV–coverage would
dependently (each data set with its own 3D–mask). We
recover. WeherechooseCARMA’smostcompactconfig-
then measure the flux inside the region limited by the
urationsincethisconfigurationrecoversthelargestemis-
CARMA–sensitiveregion(indicatedinFigure1). B)We
sion scales. After constructing the interferometric cube
computetheintegratedintensitymapforeachdatacube,
usingsimobserve,weanalyzethesimulatedinterferomet-
butusetheCARMA3D–maskinallcases. C)Wecom-
ric data in the same way as we did with the real data.
pute the integrated intensity map for each data cube,
Finally we compare the FWHM measured in the sim-
butusetheHERACLES3D–maskinallcases. Wethen
ulated interferometric observations to the HERACLES
measure the flux inside the CARMA–sensitive region.
observations. We find that the HERACLES values are
The uncertainties we state in Table3 only take into ac-
larger by a similar amount (∼50%) to the values mea-
count the canonical 10% flux calibration uncertainty for
suredinthesimulatedinterferometricdata. Thisfurther
each measurement, which is likely a lower limit to the
supports the idea that the information filtered by the
actual uncertainties.
interferometerresultsinnarrowerlinewidths,i.e.theex-
Thepercentageoffluxrecoveredbytheinterferometer,
tended emission filtered out by the interferometer has
compared to the flux measured by the single–dish tele-
also larger velocity dispersion.
scope, for the three different approaches are presented
Finally, we caution that CARMA and Nobeyama ob-
in Table 3. Method A tells us how much flux is recov-
servedtheCO(1–0)transitionwhereasHERACLESob-
ered by each instrument, even if the flux is originating
served CO(2–1). The mean ratio of the Nobeyama
in not exactly the same position, but in the area where
to CARMA line widths is ∼25% larger than the cor-
CARMA has sufficient sensitivity to pick up emission.
responding mean ratio of HERACLES to CARMA line
MethodBisconstrainedtolookforemissionwherethere
widths. This difference might be due to different excita-
is CARMA emission, whereas Method C is constrained
tiontemperatureswithintheclumpyordiffusegasordue
tolookforemissionwherethereisHERACLESemission.
todifferentopticaldeptheffects. Withtheavailabledata
These results show that even with Method B, where we
we cannot differentiate between those two mechanisms.
measure flux in precisely the interferometric 3D–mask,
theinterferometerrecovers∼74–81%(dependingonthe
galaxy) of the flux recovered by the single–dish. Using
4.5. Flux Comparison Method A, which looks for flux in the whole region were
We compare the fluxes recovered by the single–dish CARMA has good sensitivity, we find that in the case
and by the interferometric data sets by means of inte- of NGC4736 the flux recovery by the interferometer is
grated intensity maps (i.e., zeroth-moment maps). To ∼52%only,whileforNGC5055itismuchhigher(92%).
construct these maps we first create 3D–masks that we With Method C, which by construction looks for emis-
sion using the single–dish 3D–mask, the interferometer
12TheIRAM30mtelescopehasapointingaccuracyof∼2”and recovers ∼35% of the flux in the case of NGC4736 and
theNobeyama45mtelescopeasapointingaccuracyof(cid:46)7”. ∼92%inthecaseofNGC5055. InthecaseofNGC5055
