SITELLE Hα Imaging Spectroscopy of z~0.25 Clusters: Emission Line Galaxy Detection and Ionized Gas Offset in Abell 2390 & Abell 2465
Qing Liu, Howard Yee, Laurent Drissen, Suresh Sivanandam, Irene Pintos-Castro, Leo Y. Alcorn, Bau-Ching Hsieh, Lihwai Lin, Yen-Ting Lin, Adam Muzzin, Allison Noble, Lyndsay Old
DDraft version March 1, 2021
Typeset using L A TEX twocolumn style in AASTeX63
SITELLE H α Imaging Spectroscopy of z ∼ Qing Liu,
1, 2
H.K.C. Yee, Laurent Drissen,
3, 4
Suresh Sivanandam,
1, 2
Irene Pintos-Castro, Leo Y. Alcorn, Bau-Ching Hsieh, Lihwai Lin, Yen-Ting Lin, Adam Muzzin, Allison Noble, and Lyndsay Old David A. Dunlap Department of Astronomy & Astrophysics, University of Toronto, 50 St. George St., Toronto, ON M5S 3H4, Canada Dunlap Institute of Astronomy and Astrophysics, University of Toronto, 50 St. George St, Toronto, ON, Canada D´epartement de physique, de g´enie physique et d’optique, Universit´e Laval, Qu´ebec, QC, G1V 0A6, Canada Centre de Recherche en Astrophysique du Qu´ebec, Canada Department of Astronomy & Astrophysics, University of Toronto, 50 St. George St., Toronto, ON M5S 3H4, Canada Department of Physics and Astronomy, York University, 4700 Keele Street, Toronto, Ontario, ON MJ3 1P3, Canada Institute of Astronomy and Astrophysics, Academia Sinica, No. 1, Section 4, Roosevelt Road, Taipei 10617, Taiwan School of Earth and Space Exploration, Arizona State University, Tempe, AZ, 85287, USA European Space Agency, European Space Astronomy Center, Villanueva de la Ca˜nada, E-2691, Madrid, Spain
ABSTRACTEnvironmental effects are crucial to the understanding of the evolution of galaxies in dense environ-ments, such as galaxy clusters. Using the large field-of-view of SITELLE, the unique imaging fouriertransform spectrograph at CFHT, we are able to obtain 2D spectral information for a large and com-plete sample of cluster galaxies out to the infall region. We describe a pipeline developed to identifyemission line galaxies (ELGs) from the datacube using cross-correlation techniques. We present resultsbased on the spatial offsets between the emission-line regions and stellar continua in ELGs from twoz ∼ σ excess for high-velocity galax-ies within the virial radius having the offsets to be pointed away from the cluster center. Assumingthe offset being a proxy for the velocity vector of a galaxy, as expected from ram pressure stripping,this excess indicates that ram pressure stripping occurs most effectively during the first passage of aninfalling galaxy, leading to the quenching of its star formation. We also find that, outside the virialregion, the continuum-normalized H α line flux for infalling galaxies with large offsets are on averagelower than those with small or no measurable offset, further supporting ram pressure as a dominantquenching mechanism during the initial infall stages. Keywords: galaxies: evolution — galaxies: clusters: individual: Abell 2390 — galaxies: clusters:individual: Abell 2465 — galaxies: clusters: intracluster medium. INTRODUCTIONIt is an established consensus that the evolution ofgalaxies is affected by their environment, especially inhigh density regions such as rich galaxy clusters (e.g.,Ellingson et al. 2001, Lewis et al. 2002, Kauffmann et al.2004, Boselli & Gavazzi 2006, Poggianti et al. 2006,Weinmann et al. 2009, Sobral et al. 2011, Muzzin et al.2012, Foltz et al. 2018, Pintos-Castro et al. 2019). Ingalaxy clusters the interplay between galaxies and thehost cluster is likely responsible for the quenching of star [email protected], [email protected] formation in galaxies, i.e., the suppression or cessationof star formation, aside from their internal secular evolu-tions. Such external effects shape the observed correla-tion between the star formation rate (SFR) and the envi-ronmental density (e.g., Balogh et al. 1998, G´omez et al.2003, Peng et al. 2010, Kawinwanichakij et al. 2017),which is closely related to the well-known morphology-density relation (e.g., Dressler 1980, Goto et al. 2003,Postman et al. 2005).One of the breakthroughs of present-day observationaltechniques is 2D imaging spectroscopy, such as the In-tegral Field Unit (IFU) spectroscopy. Several recent, orongoing, IFU surveys such as CALIFA (S´anchez et al. a r X i v : . [ a s t r o - ph . GA ] F e b Liu et al. (cid:48) × (cid:48) , SITELLE pro-vides a unique opportunity to study environmental ef-fects on star formation in galaxy clusters in more details.Specifically, SITELLE allows us to simultaneously ac-quire spatially-resolved spectral information for a large,complete and luminosity-limited sample of emission-linegalaxies (ELGs). Its spatial coverage extends out to thecluster infall region at z ∼ .
25. This is unprecedented;existing spatially-resolved data on galaxy clusters arelimited in either field coverage or spectral resolution.With abundant information, both spectral and spatial,encoded in emission lines, SITELLE offers an excellentopportunity to gain a better understanding of star for-mation in galaxy clusters and investigate the mecha-nisms responsible for the quenching of star formation inhigh density environments.Many possible external quenching mechanisms havebeen investigated in the past two decades: gas strip-ping, starvation, harassment, thermal evaporation, pre-processing, etc. (see Boselli & Gavazzi 2006 for a re-view). Among all the competing quenching mechanisms,the dynamical interaction of the gas in the cluster galaxywith the intracluster medium (ICM), typically in theform of ram pressure stripping (Gunn & Gott 1972), hasbeen considered to be one of the most efficient physi-cal mechanisms to quench star formation activities instar-forming galaxies (SFGs) through the removal ofcold gas. For example, Muzzin et al. (2014) concludedthat ram pressure stripping is the most plausible mecha-nism for satellite quenching by investigating the dynam-ics and quenching timescales of z ∼ α /opticalbands (Koopmann & Kennedy 1999; Yoshida et al. 2008;Smith et al. 2010; Jaff´e et al. 2018). The most extremecases are known as “jellyfish galaxies” (e.g., Ebelinget al. 2014, Poggianti et al. 2017a, Jaff´e et al. 2018),which are nearby SFGs exhibiting conspicuous extendedionized gas tails. Therefore, analyzing the properties ofgas in cluster galaxies is promising for shedding lighton the mechanism of ram pressure stripping and placingconstraints on numeric simulations of their evolutionsin dense environments. For example, many interestingresults have been revealed by the MUSE GASP survey(Poggianti et al. 2017a) about the properties of jellyfishgalaxies and the implications on their evolution histo-ries.This kind of analysis is well suited to 2D spectroscopicobservations using SITELLE. An ongoing project usingSITELLE targeting the H α +[N ii ] lines in clusters atz ∼ ITELLE Cluster ELGs = 70 km s − Mpc − , Ω m =0.3 and Ω Λ = 0.7. OBSERVATIONS2.1.
SITELLE
Our project uses SITELLE (Drissen et al. 2019), thenew imaging Fourier instrument at CFHT. With a broadworking wavelength range from 3500 ˚A to 9000 ˚A,SITELLE is designed to focus on emission-line objects.Band-limiting filters are used to increase spectral reso-lution with finite scanning steps and to reduce the con-tamination from strong sky-lines. The equivalent spec-tral resolution R of IFTS is determined by the totalnumber of mirror steps of the Michelson-type interfer-ometer, which is chosen to resolve the H α and [N ii ] linesand to provide spatially-resolved kinematic information.The primary science drivers for SITELLE include neb-ulae and supernova remnants in the Milky Way, H ii andstar-formation regions in nearby galaxies, and ELGs ingalaxy clusters. The uniqueness of SITELLE in study-ing galaxy clusters is its unparalleled 11 (cid:48) × (cid:48) FOV,allowing for IFU-like 2D spectroscopy of a large andcomplete, luminosity-limited ELG sample of a cluster.The SITELLE data product is a spectral datacube,with an image at each wavelength step (“channel”), af-ter being corrected, transformed and calibrated by theORBS (Martin et al. 2015) pipeline. The pixel scale ofSITELLE is 0.322 (cid:48)(cid:48) /pix, with 2048 × α lines. To reduce sky contam-ination, a narrow window (796 nm – 826 nm) that isless affected by sky-lines is selected for the design of theSITELLE C4 filter, which corresponds to the redshiftrange of H α -[N ii ] lines at 0.21-0.25, allowing us to detectELGs in that redshift range. The window correspondsto the region where the observing efficiency of SITELLE(CCD quantum efficiency, modulation efficiency and op-tical transmission) is the highest (at around 800 nm). https://github.com/thomasorb/orbs Table 1.
Properties of the clusters observed by SITELLE.Cluster Abell 2390 a Abell 2465 b Redshift (z) 0.228 0.245R.A. ( α ) 21 h m s h m s (NE)22 h m s (SW)Decl. ( δ ) +17 ◦ (cid:48) (cid:48)(cid:48) − ◦ (cid:48) (cid:48)(cid:48) (NE) − ◦ (cid:48) (cid:48)(cid:48) (SW) M ( M (cid:12) ) ∼ × ∼ × each R ( Mpc ) 2.1 1.2 σ v ( km/s ) 1100 763 (NE)722 (SW) a M , R and σ v from Carlberg et al. (1997). b M , R and σ v from Wegner (2011). The values listed arefor each subcluster. Nevertheless, there exist some sky-lines that affect thedetectability and measurement of the lines. We presentthe subtraction of the sky in Sections 3.1 and 3.2.Detailed information about SITELLE, including theinstrumentation design, the advantages and drawbacksof IFTS, science capabilities and commission perfor-mance, can be found in Drissen et al. (2019).2.2. Target ClustersThe observations for the first two targets for the sur-vey were carried out on SITELLE in 2017 and 2018 ontwo well-studied galaxy clusters: Abell 2390 and Abell2465 (hereafter, A2390 and A2465).A2390 is a rich massive cool-core galaxy cluster( M ≈ × M (cid:12) ) at redshift of 0.228. It has beenwidely studied in many previous studies for the under-standing of galaxy evolution in dense environment (e.g.,Yee et al. 1996, Abraham et al. 1996, Balogh & Morris2000, Fritz et al. 2005). For example, Abraham et al.(1996) found that the cluster is gradually formed by theaccretion of infalling galaxies with truncation in theirstar formation by investigating the properties (kinemat-ics, colors, morphologies, etc.) of a large sample of clus-ter galaxies using the CNOC data (Yee et al. 1996).Different from A2390, A2465 consists of two merg-ing subclusters, a north-east (NE) clump A2465NE anda south-west (SW) clump A2465SW, at a redshift of0.245 (Wegner 2011). Members of the subclusters havebeen identified based on kinematic energies for the mea-surements of properties for each subcluster in Wegner(2011). The two subclusters have roughly equal virialmasses. A series of detailed multi-wavelength analyseson A2465 can be found in Wegner (2011) and Wegneret al. (2015, 2017), where the authors found interestingresults revealing its collision history and the evolutionhistory of galaxies within it. The redshifts, positions, Liu et al. h m s s m s s ◦ RA (J2000) D ec ( J ) Figure 1.
Deep frames of the Abell 2390 central (C) / south-east (E) / north-west (W) fields observed by SITELLE, overlaidon the SDSS DR12 mosaic. The FOV of a single field is 11 (cid:48) × (cid:48) . The deep frames are constructed using the ORCS piepline(Martin et al. 2015). Emission-Line Galaxies (ELGs) identified in Section 3 are marked by green circles. The brightest centralgalaxy marked by a magenta polygon is also an ELG. Table 2.
Observed information about the individual fields.Field A2390C A2390W A2390E A2465CN stepsa
124 150 150 207Exposure (s) 11,408 13,800 13,800 13,662∆ λ (˚ A ) 5.2 4.3 4.3 3R 1080 1300 1300 1800Seeing( (cid:48)(cid:48) ) 1.1 1.2 1.3 1.1RA ( α ) b h m . s h m s h m . s h m . s Dec ( δ ) b ◦ (cid:48) (cid:48)(cid:48) ◦ (cid:48)(cid:48) (cid:48)(cid:48) ◦ (cid:48) (cid:48)(cid:48) − ◦ (cid:48) (cid:48)(cid:48) a Number of mirror steps to reach the specified R. b Central coordinates of the SITELLE fields.
ITELLE Cluster ELGs h m s s s s − ◦ RA (J2000) D ec ( J ) Figure 2.
Deep frame of the central field of the double clus-ter Abell 2465 observed by SITELLE. Identified ELGs aremarked by green circles. The two brightest central galaxiesare marked as purple polygons which do not have detectedemission. White contours show a rendering of the weak lens-ing mass contours in Figure 7 of Wegner et al. (2017). and physical properties of the two target clusters ob-served by SITELLE are summarized in Table 1.There are three pointings for A2390 and one forA2465. The initial observational design was to use anobservation time of 4 hr per pointing, including overheadbetween steps, with ∼
150 steps, producing a resolution R ∼ α -[N ii ] lines. How-ever, for A2390C, which was observed during a commis-sioning run for the science verification for the C4 filter,only a shorter exposure of 3 hours was obtained, re-sulting in fewer steps and a lower R . The number ofsteps was increased for the subsequent run for A2365C,and future targets, to achieve better spectral resolution.Observational information on the four fields includingthe total number of mirror steps, the exposure time,the average wavelength interval, the average seeing, theapproximate spectral resolution, and the coordinates offield centers are listed in Table 2. EMISSION-LINE GALAXY IDENTIFICATIONThe IFTS spectral datacube is transformed from theoriginal interferometric cube acquired by SITELLE atCFHT and calibrated with the data reduction softwareORBS (Martin et al. 2015). This section introduces theprocedures for detecting and identifying ELGs from theSITELLE spectral datacube.3.1.
Sky-Line Subtraction
While the C4 filter is designed to minimize the num-ber of OH sky-lines in the spectral region, the sky con-tinuum and a few sky-lines still dominate the spectralrange of the filter. Therefore, we first subtract the skybackground in each channel. The background is evalu-ated and subtracted in 2D using the photutils package.The background is locally estimated within a 128 × box using a mode estimator and then subtractedfrom the data of each channel. The spectral axis of thedatacube is clipped into 12100 – 12550 cm − and con-verted into wavelength corresponding to 7970 ˚A – 8265˚A. The wavelength axis is then interpolated to be uni-formly spaced in logarithmic scale. A stacked image isalso created by coadding of all the channels within thisspectral range for the initial source detection describedin Section 3.3.3.2. Fringe Reduction using Low-pass Filtering
The interferometric nature of SITELLE and the pres-ence of strong sky-lines in the C4 filter induce fringesacross the field in specific channels. The fringes are su-perimposed on the uneven large-scale background thathas been subtracted in Section 3.1 . The level of con-tamination from these fringes depends on the channelwavelength, the number of mirror steps, and the posi-tion in the field. In general, channels around strongersky-lines with fewer mirror steps (i.e. lower spectral res-olution) present brighter fringes away from the field cen-ter. The brightness and spatial pattern of these fringesdiffer from channel to channel, which makes it challeng-ing to model them from first principles. The brightnessof fringes vary across the field and could be compara-ble to/brighter than a large number of sources, affectingthe source detection, ELG identification, and potentiallytheir measurements. We adopt an empirical method toreduce the influence of fringes by applying low-pass fil-tering (LPF) on each channel. Details of the LPF pro-cedure and a figure illustrating the fringes and cleaningare presented in Appendix A.3.3.
Source Detection and Spectral Extraction
Although the fringes have been alleviated by the LPFprocedure, small-scale fringe residuals from the sky oc-casionally show up in some channels because of the finitekernel size used. We apply a moving average procedureto further mitigate fringe contamination (Appendix A)and construct a new datacube for source detection. Wenote that this smoothed datacube is used only for sourcedetection but not for analysis. The new datacube is col-lapsed in the spectral axis into an image for the detectionof ELGs (referred to below as the “detection image”)generated by using the maximum value in the spectral
Liu et al. axis at each spatial position, taking advantage of the2D spectroscopy. Using the maxima in wavelength en-hances the possibility of detecting weak narrow peaks.We compare the output candidate list obtained from thedetection image with the one from the mean image ofall channels and it proves that the detection image con-structed in such way offers a more complete sample thatincludes line emitters with weak or zero continuum.The python software photutils is used to detectand deblend sources. We adopt a detection signal-to-noise ratio (S/N) threshold of 2.5, a 5 pixel connec-tion, a multi-thresholding level number of 64, and alocal peak contrast of 0.01. The threshold is locally es-timate based on a 2D background estimate using the
SExtractorBackground class as in Section 3.1. A seg-mentation map is created in this process. An integratedspectrum of each detected source is extracted from thelow-pass-filtered datacube using the corresponding seg-mentation in the map. These spectra are used for sourceidentification and redshift measurement. In total, thedetection algorithm locates ∼ ∼ ∼ Removal of Continuum
For the purpose of automatically identifying ELGs us-ing templates that only contain emission lines, the con-tinua of sources must be subtracted to obtain the resid-ual emission lines. This is accomplished using GaussianProcess (GP) regression, typically used for fitting curvesin a non-parametric manner. The advantage of using GPregression is its flexibility, which allows for slight vari-ations in the continuum such as a mild gradient and isable to smoothly tackle possible irregularity around thefilter edges. We use the
GaussianProcessRegressor realization in the python package scikit-learn . Thekernel is chosen to be a combination of a radial basisfunction (RBF) kernel with a bandwidth of 100 ˚A (suchthat it is wider than the width of H α -[N ii ] lines) whichaccounts for the continuum component, and a white ker-nel accounting for the noise. Possible emission that isone σ above the median value of the spectrum is iter-atively replaced with the median. In the case of a fewbright sources, the sharp drops around the filter edgesmake them poorly fitted by smooth GP kernels and aremasked in the fitting. This procedure is applied on eachof the integrated spectra extracted in Section 3.3.3.5. Construction of Line Template
We use the cross-correlation technique to identifyELGs from detected sources. It is done by cross-correlating the normalized residual spectra with a li- brary of emission-line templates. While the obser-vation is designed to capture the H α -[N ii ] lines inthe galaxy cluster, emission lines other than H α -[N ii ](e.g., [O iii ] λ ∼ ii ] λ ∼ α +[N ii ] lines. Below we describe threesets of emission-line templates used to perform cross-correlation.The primary set includes the H α line and the[N ii ] λλ ii ] λ ii ] λ α and [N ii ] λ iii ] λλ iii ] λ iii ] λ β line is not included in the template set,considering that in most cases it is located out of thefilter or at the filter’s blue edge in the presence of the[O iii ] doublet. As a result, the inclusion of H β might re-duce the optimal S/N of a spectrum with only the [O iii ]doublet in the cross-correlation process, and thereforeit is removed from the template. The third set con-tains a single line, which represents several possible ori-gins including an [O ii ] λ ∼ iii ] λ iii ] λ iii ] λ α line with undetected [N ii ] (e.g., from metal-poor galax-ies, or with low S/N), an isolated H β at z ∼ iii ] λ α emitter at z ∼ T ( λ ) = (cid:34)(cid:88) i δ ( λ − λ ,i ) (cid:35) ∗ Gaussian( λ ,i , σ line )= (cid:88) i A i exp (cid:20) − ( λ − λ ,i ) σ line (cid:21) , (1)with the normalization A i being the line ratios, λ ,i be-ing the line center for the i -th line component, and σ line being the common line width. The line width rangesfrom σ min = ∆ λ , which is the spectral resolution, to σ max = (cid:113) σ min + σ gal , which corresponds to a galac-tic velocity dispersion σ gal =300 km/s. It is notewor-thy that the intrinsic line shape of SITELLE is a sinc-gaussian resulting from an instrumental sinc line shapewith Gaussian broadening (Drissen et al. 2019). How-ever, in practice, introducing sinc-gaussian in a templateadds another degree of freedom and sometimes uninten-tionally magnifies the noise. This is possibly due to ITELLE Cluster ELGs and Martin et al.2015) for SITELLE.3.6. Cross-correlation with Template
Cross-correlation, also known as matched filtering, is atechnique widely used in signal processing that is appliedfor the optimization of the S/N by cross-correlating theobserved signal with templates:
CCF ( v ↔ δλ ) = [ F (cid:63) T ] ( δλ )= (cid:90) λ λ F ( λ ) T ( λ + δλ ) dλ , (2)where F ( λ ) is the input continuum-subtracted spectrumwithin the filter range ( λ , λ ), and CCF is the cross-correlation function. By convention, CCF is in units ofvelocity v , which is converted from wavelength differ-ence δλ . The residual spectra is normalized and linearlyinterpolated with twice as many points as the originalspectra in log scale for a higher precision in peak match-ing. In essence, the cross-correlation procedure accom-plishes two jobs: the detection of a potential signal, andthe localization of its position (redshift). This is run onall the extracted integrated spectra obtained above.The S/N of the CCF, ( S/N ) cc , is calculated as follows:( S/N ) cc = CCF max σ cc , (3)where CCF max is the peak of the CCF and σ cc is therms noise calculated in the region 5 σ line away from anyof the line centers. A wider signal range (8 σ line ) is usedif the matched line ratio H α /[N ii ] λ S/N ) cc >
50) in order to bettermatch bright broad lines. Regions near the filter edges( ∼ S/N ) cc . ( S/N ) cc gives an estimate of the credibility of emission, but itdoes not place constraints on the line width and ratio. http://celeste.phy.ulaval.ca/orcs-doc We further use a significance parameter R defined as: R = ξ · γ = CCF max
CCF (2) max · ( S/N ) cc ( S/N ) cc,max , (4)which multiplies the peak contrast ξ , defined as the ra-tio of the highest peak to the second highest peak of theCCF, by the credibility γ , defined as the ratio of thecomputed ( S/N ) cc and the highest ( S/N ) cc among alltemplates. The best matched template is chosen as theone with maximum R. The parameter R is motivatedby experiments suggesting that maxmizing R typicallyreturns better estimates for the line ratio and the linewidth σ line than simply using ( S/N ) cc,max , while retain-ing the accuracy of the matched redshift. Examples ofCCF are shown in the right panels of Figure 3 with thecorresponding residual spectra shown in the left panels.With the aid of visually inspecting a number of spectraand their CCF, we apply the following S/N criteria toselect potential ELG candidates for detection: • ( S/N ) cc, H α > S/N ) cc, OIII > • ( S/N ) cc,single > S/N ) cc, H α , ( S/N ) cc, OIII and (
S/N ) cc,single arethe resultant ( S/N ) cc using the three template sets.We then calculate quick estimates of W em , the equiv-alent width of the total emission using a 10 σ line widewindow, for the purpose of reducing false detections.An uncertainty of W em , σ ( W em ), is roughly obtainedby perturbing spectra 250 times with the continuumrms. We require the candidate to have W em > A and W em > σ ( W em ) to be considered as a detection. Wefurther require the candidate to be located at least 20pixels away from the field edge and its matched peak atleast 20 ˚A away from the filter edge.Through all the steps above, we achieve a listof ELGs in each of the fields, with 88/80/72 fromthe A2390C/A2390E/A2390W field and 111 from theA2465C field. Repeated detections in overlapping fieldcoverage of A2390 are cross-matched and cleaned whencombining detections among fields. Finally, a visualcheck is done on all the candidates to screen out du-bious detections and to confirm strong [O iii ] emitters.In total, 194 ELGs are found in A2390 and 110 ELGsare found in A2465. ELGs identified in the A2390 andA2465C are marked in Figure 1 and Figure 2, respec-tively, which includes line emitters in the background.In A2390, the central field has the most ELGs, whileA2465C shows an excess close to the center of thesouth-west subcluster. The prevalence of ELGs foundin A2465C suggest enhanced star formation occurring inthe merging clusters, as observed and discussed in Weg-ner (2011) and Wegner et al. (2015). The average 5 σ Liu et al.
Wavelength ( Å ) N o r m e d F l u x [NII]6548H [NII]6584 ID : 936Dz = 0.247
Relative Velocity (km/s) to z = 0.245 CC F S/N : 44.6R : 2.7
Wavelength ( Å ) N o r m e d F l u x [NII]6548H [NII]6584 ID : 1169Dz = 0.251
Relative Velocity (km/s) to z = 0.245 CC F S/N : 6.3R : 2.2
Wavelength ( Å ) N o r m e d F l u x [OIII]4959 [OIII]5007 ID : 2203Cz = 0.64
Relative Velocity (km/s) to z = 0.61 CC F S/N : 15.2R : 5.2
Wavelength ( Å ) N o r m e d F l u x ID : 1451W
Relative Velocity (km/s) to z = 1.16 CC F S/N : 7.5
Figure 3.
Examples of continua-removed integrated spectra (left column) and the corresponding cross-correlation results (rightcolumn). From top to bottom: high S/N H α +[N ii ], low S/N H α +[N ii ], [O iii ] λλ ii ]). In the left column, the continuum-subtracted spectra are shown in blue, whereas the green dashedlines indicate the matched templates; the redshift determined from cross-correlations are shown for the top three rows. In theright column, cross-correlation functions (CCF) are plotted versus velocity shifts with reference to the systematic redshift of thecluster. Shaded areas indicate the range used for computing the signal (light red) and noise (gray). The thick red line showsCCF of the best-matched template. The thin red lines show CCF from individual templates. detection limit is F H α ∼ / / / × − ergs/s/cm inA2390C/A2390E/A2390W/A2465C, estimated by tak-ing the median of the flux of S/N > σ detectionlimit is F H α ∼ × − ergs/s/cm . Adopting the con-version of Kennicutt & Evans (2012), this correspondsto SFR ∼ . M (cid:12) /yr, or ∼ . M (cid:12) /yr if 1 mag dust ex-tinction is applied. ANALYSIS OF SPATIAL OFFSET OF IONIZEDGASWith the acquired ELG list, a simple yet interestinganalysis is to investigate the connection between the spa-tial offset of ionized gas from its parent galaxy and thecluster center in search for ram pressure stripping ef- fects. We present in this section some initial resultsbased on the centroid offsets between emission-line re-gions and stellar continua in the two clusters observedby SITELLE.In Section 4.1 we introduce the methodology of mea-suring the centroids of ionized gas and stellar contin-uum. We then measure the centroid offset and the dif-ference angle between the emission-to-continuum vector and the cluster-centric vector following the analysis ofSmith et al. (2010). We show the centroid analysis re-sults of A2390 and A2465 in Section 4.2.4.1.
Centroid and Angle Measurements
To construct an initial sample of cluster galaxies foreach cluster, we first remove objects with a velocity dif-ference of more than 4 σ v (∆ z (cid:38) ITELLE Cluster ELGs Figure 4.
Example centroid measurements from the A2390C/W/E (top four rows) and A2465C fields (bottom two rows).For each row, the spectrum of the detected ELG is shown in the left panel with the channels used for the creations of theemission/continuum image shown in red/blue. The middle panel shows the H α emission image and the right panel shows thecontinuum image, on which the centroid measurements are performed. The boundary of the distribution of ionized gas emission( I E ) / the stellar continuum ( I C ) and the light-weighted centroid are plotted on each postage stamp in magenta (for emission)/ white (for continuum). The emission-line and continuum centroids are indicated as magenta and white dots with crosses,respectively. The 30%-60%-90% level of the H α emission distribution is overlaid as blue contours on the continuum image aftera mild smoothing by a 3x3 pix Gaussian kernel. The blue arrow indicates the emission-to-continuum offset vector d and theorange arrow indicates the direction to the cluster center. Measured ∆ d and θ d are in units of kpc and degrees, respectively.Galaxies with θ d > − erg/s/cm / ˚ A /pix . Liu et al.
To Cluster Center
Ionized Gas Stellar Continuum θ d Δ d ⃗ d ⃗ r (¯ x E , ¯ y E ) (¯ x C , ¯ y C ) ICM Wind
Figure 5.
Schematic illustrating the centroid analysis.The emission-line region emission (¯ x E , ¯ y E ) and the stellar-continuum centroids (¯ x C , ¯ y C ) are measured from the emis-sion image I E and continuum image I C . The centroid off-set ∆ d and difference angle θ d between the offset vector d and cluster-centric vector r are measured based on the cen-troid positions. The purple arrows illustrate the expecteddirection of the ICM wind arising from the movement of thegalaxy in the ICM halo. tematic velocity of the cluster. We then measure thecentroid of the ionized gas and the centroid of the stel-lar continuum for each cluster galaxy as follow:For each galaxy, a thumbnail datacube centered onthe object is created. The continuum image, I C ( x, y ), isconstructed by taking the mean value of channels in thecontinuum range at each position ( x, y ) after an itera-tive 3 σ -clipping. The continuum range is defined as theregion further than 15 × (1 + z cc ) ˚A away from the [N ii ]lines, but at least 20 ˚A away from the filter edges. Theemission image, I E ( x, y ), is constructed in a narrow-band likewise by taking the mean of the few channelswithin ± × (1 + z cc ) ˚A around the matched H α peak.The continuum image is subtracted from it to obtain thefinal emission image.Another source detection is performed on I C ( x, y ) and I E ( x, y ) using a detection S/N threshold of 2.5, withnearby sources masked using the segmentation map inSection 3.3. The detection of emission or continuumcould fail for a portion of objects in the case of objectbeing : (1) a faint target with low S/N; (2) a backgroundsource with weak or no detectable continuum; (3) a falsepositive peak composed of noise; and (4) close to con-taminants such as stars or diffraction spikes. For objectswith successful detections in both emission and contin-uum, we then measure the flux-weighted centroids (¯ x, ¯ y ) . . . . . . . . . . . Centroid Offset ∆ d [kpc] N u m b e r o f E L G A2390A2465
Figure 6.
The distributions of light-weighted centroid off-set ∆ d between the emission and the stellar continuum forELGs identified in A2390 (left, combining three fields) andthe central field of A2465 (right). The 1kpc threshold usedis indicated by the black dashed line. for the ionized gas or the stellar continuum through(¯ x, ¯ y ) = (cid:18) (cid:80) x i · I ( x i , y i ) (cid:80) I ( x i , y i ) , (cid:80) y i · I ( x i , y i ) (cid:80) I ( x i , y i ) (cid:19) , (5)where the sum is performed on all pixels within the seg-mentation of the distribution { x i , y i } from the sourcedetection for the emission image I E (x,y) or the contin-uum image I C (x,y). Example spectra, continuum im-ages, and emission images are presented in Figure 4.With the emission and continuum centroids, we thenmeasure the offset vector d , defined as the vector fromthe emission-line centroid to the continuum centroid: d = (¯ x E − ¯ x C , ¯ y E − ¯ y C ) . (6)We measure the centroid offset ∆ d , using the flux-weighted centroids:∆ d = (cid:112) (¯ x E − ¯ x C ) + (¯ y E − ¯ y C ) , (7)and the difference angle between d and the cluster-centric vector r , a vector pointing from the galaxy tothe brightest central galaxy (BCG): θ d = (cid:93) ( d , r ) , (8)Given that the offset is much smaller than the dis-tance to the BCG, the object center measured in theprimary source detection is adopted as the galaxy cen-ter. An illustration of the definitions of quantities in themeasurements is shown in Figure 5. The uncertaintypropagation is presented in Appendix B.1. To remove ITELLE Cluster ELGs d > σ ∆ d and exclude objectsnear the field edges (distance <
100 pix). These criteriareject ∼
38% of the sample on average in the four fields.We obtain 117 ELGs in A2390 and 75 ELGs in A2465after the rejection.Figure 6 shows the distributions of ∆ d for A2390 andA2465. The centroid offsets are in general relativelysmall. An empirical assessment for the centroid mea-surements is presented in Appendix B.2 showing thatthese relatively small ∆ d can be measured robustly. Thedistributions of ∆ d extend from 0 to ∼ Distribution of Difference Angles in A2390 andA2465
In the search for evidence of ram pressure effects, itis of great interest to further investigate the correlationbetween the emission spatial offset and the cluster cen-ter, i.e. the distribution of the difference angles, becausethe strength and direction of the ram pressure from theICM on the galaxy are expected to be closely linked tothe infalling galaxy’s orbit and position in the cluster.Because small spatial offsets are prone to be affected bymeasurement uncertainties, we further limit our sampleto ELGs with a cutoff in the measured offset. We choosea physical length of 1 kpc that is higher than the peakvalues of distributions of ∆ d to be the sample threshold,which corresponds to ∼ ∼ d measurements are shown in Figure 4 where the con-tours and centroids for H α emission and continuum aredisplayed. It is interesting to note that the ELGs with∆ d < α emission being largely symmetric with the continuum, while those with larger ∆ d displaya clear non-symmetry.Finally, we obtain 55 ELGs from A2390 and 24 ELGsfrom A2465 out of the parent ELG samples. Hereafter,these ELGs with relatively large emission-to-continuumoffsets (∆ d ) are referred to as the selected ELGs.4.2.1. Abell 2390
Figure 7 shows positions of the cluster ELGs in theA2390 fields and Figure 8 presents their locations on thephase space diagram. The phase space diagram plotsthe distribution of the projected positions and velocitiesof galaxies, normalized by R and the cluster velocitydispersion, respectively. This is often used as an in-dication of the cluster galaxy’s dynamical state in thecluster (e.g., Noble et al. 2013, Muzzin et al. 2014, Jaff´eet al. 2015, Rhee et al. 2017, Yoon et al. 2017). Weuse the phase space diagram to identify cluster mem-bers, i.e., galaxies bound to the gravitational field of thehost cluster, by excluding ELGs falling above the escapevelocity boundary (black solid line in Figure 8).In Figure 7, colored symbols mark the selected ELGsample (∆ d > θ d ≤ ◦ and red solid circles indicatethose with θ d > ◦ . The BCG itself is a strong ELG(Hutchings & Balogh 2000), which has been excluded inour analysis. The direction of the ionized gas offset rel-ative to the stellar continuum (opposite to d ) measuredwith flux-weighted centroids is indicated by arrows. Therest of the cluster ELGs with small offset detected (∆ d > and it maintains a ∼
25% spatial coverageuntil it drops to zero beyond ∼ . We furtherdivide the in-cluster region (below the escape velocityboundary) on the phase space diagram into three sub-regions, according to the typical path followed by aninfalling galaxy during its virialization (see Section 5.2):(I) galaxies outside the cluster virial radius; (II) high-2 Liu et al. . . . . . . . . . RA (J2000) . . . . . D ec ( J ) Abell 2390 R θ d < ◦ , ∆ d > kpcθ d > ◦ , ∆ d > kpc ∆ d < kpc MembersNon-membersMembersNon-members
Figure 7.
Location of the selected ELGs (∆ d > d > θ d < ◦ , while red symbols representselected ELGs in the cluster with θ d > ◦ . For display pur-pose, colored arrows indicate the opposite directions of thedifference vector d , i.e. from continuum to emission. ELGswith velocities above the escape curve (non-cluster members)are shown as open circles. ELGs with small ∆ d ( < R is indicated by the blackdashed circle. The yellow squares indicate the footprints ofthe SITELLE fields. velocities galaxies within the virial radius; (III) low-velocities galaxies within the virial radius. The distri-bution of the selected ELGs on the phase space diagramreveals that the ratio of those with offsets away fromthe cluster center (blue symbols) to those toward thecenter (red symbols) tends to increase from low to highvelocity. This is most significant in region II – a simplenumber count yields 6:2 in region II, compared with 6:4in region I and 5:5 in region III. We will further discussthe physical interpretation in Section 5.2 below.The left histogram in Figure 9 presents the distri-butions of θ d for the selected cluster member ELGs ofA2390. Several patterns can be revealed: first, the dis-tributions of the difference angles show distinct devi-ations from uniformity. We perform the Kolmogorov-Smirnov (K-S) test to test the uniformity of the dis-tributions of θ d . Based on the p-values calculated, we .
00 0 .
25 0 .
50 0 .
75 1 .
00 1 .
25 1 .
50 1 . R / R . . . . . . . | v l o s − v | / σ v IIIII I θ d < ◦ , ∆ d > kpcθ d > ◦ , ∆ d > kpc ∆ d < kpc .
00 0 .
25 0 .
50 0 .
75 1 .
00 1 .
25 1 .
50 1 . R / R . . . . . . C o v e r ag e Figure 8.
The main panel shows the ELGs in Figure 7on the position-velocity phase space diagram of A2390, withthe same markers and color coding. The distances to thecluster center and the velocities relative to the cluster veloc-ity of galaxies are normalized by R and σ v . The blacksolid curve corresponds to | v los /v | ∼ | ( R/R ) − / | ,adopted as the escape velocity boundary. Cluster membersthat fall below the escape velocity curve are shown as solidmarkers. The cluster region is divided into three sub-regions:the outer non-virial region I, the inner non-virial region IIcontaining galaxies near orbit pericenters, and the “virial-ized” region III (Jaff´e et al. 2015). The upper subpanel showsthe fractional field coverage for A2390. can reject the uniformity of θ d at a 95% confidence level( p = 0 . θ d shows a clear peak at the 0 ◦ end. Thehistogram suggests that the emission offset is found tobe preferentially pointed away from the cluster center.This is expected from the effect of ram pressure strip-ping, considering the infalling and quenching process ofcluster galaxies (see discussion in Section 5.2.1). Fi-nally, There is a hint of another peak at the 180 ◦ end,suggesting the emission offsets to be toward the clustercenter. The possible peak mainly consists of objects inregion III that suffer significantly from projection effector might in the backsplash stage of their infall (see dis- ITELLE Cluster ELGs Diffence Angle θ d N u m b e r o f E L G A2390
Diffence Angle θ d N u m b e r o f E L G A2465
Figure 9.
Histogram of difference angles measured fromlight-weighted centroids ( θ d ) for cluster member ELGs inA2390 (top) and A2465 (bottom) with ∆ d > σ θ d . Individual data points are shown as small sticks at thebottoms of the histograms. cussion in Section 5.2.2). Further observations with alarger cluster sample are needed to confirm its presence.In general, the distribution of difference angles inA2390 shows that the spatial offsets of ionized gas incluster ELGs have a preference that is correlated to thecluster center, indicating the impact of ram pressurestripping from the ICM on galactic gas reservoirs.4.2.2. Abell 2465
Figure 10 shows the position of ELGs identified inthe double cluster A2465; while Figure 11 presents theirlocations in the phase space diagram. Symbols andtheir color coding follow the same as Figures 7 and 8.The membership discrimination between the two sub-clusters is simplified by using a straight line that divides . . . . . RA (J2000) − . − . − . − . − . D ec ( J ) Abell 2465
NE SW R θ d < ◦ , ∆ d > kpc θ d > ◦ , ∆ d > kpc ∆ d < kpc Figure 10.
Locations of the selected ELGs (∆ d > the double cluster because the two components havecomparable masses. It can be observed from Figure 10that there are more ELGs in the south-west sub-cluster,with the majority located at the border of the two sub-clusters. However, a large fraction of ELGs at the colli-sion border do not present significant systematic emis-sion offset, possibly caused by the complex conditionof ICM imprinted by shocks and/or galaxy kinematicsthere, while projection effects may also play a role.The right panel of Figure 9 presents the distribution of θ d for cluster member ELGs in A2465 with ∆ d > θ d shows deviationfrom a uniform angle distribution, with a major peakat 0 ◦ and a hint of a weak peak at 180 ◦ . The K-S testindicates the distribution is significantly different fromthe uniform distribution at the confidence level of 95%.The result in A2465 is consistent with the scenariowhere ionized gas spatial offset is induced by ram pres-sure during the infall of gas-rich galaxies into the cluster(see further discussion in Section 5.2). However, it is not In specific, the linear division is trained based on sub-clustermembership presented in Wegner (2011) obtained from kinemat-ics using support vector machine. Liu et al. .
00 0 .
25 0 .
50 0 .
75 1 .
00 1 .
25 1 .
50 1 . R / R . . . . . . . | v l o s − v | / σ v IIIII IIIIII I
A2465 SWA2465 NE θ d < ◦ , ∆ d > kpcθ d > ◦ , ∆ d > kpc ∆ d < kpcθ d < ◦ , ∆ d > kpcθ d > ◦ , ∆ d > kpc ∆ d < kpc Figure 11.
ELGs in Figure 7 on the position-velocity phasespace diagram of A2465. Markers are color coded in thesame way as in Figure 7. The black solid curve correspondsto | v los /v | ∼ | ( R/R ) − / | , adopted as the escapeboundary. Cluster members below the escape velocity curveare shown as solid markers. The region division is the sameas in Figure 8. To match with Figure 8, region I is shownalthough with no data coverage. clear whether such effect is enhanced or suppressed bythe collision of the two clusters. DISCUSSIONIn this section, we discuss the implications of the ob-served spatial offsets of ionized gas, specifically on thenon-uniform distribution of difference angles and the ex-cess in region II on the cluster phase space diagram. Wethen propose a scenario which qualitatively explains theobservation.5.1.
Ionized Gas Centroid Offset: Ram PressureStripping in the Act
It is natural to infer that under ram pressure, the gasoffset in a galaxy moving through the ICM generallyfollows the opposite direction to its velocity relative tothe ICM, given the pressure follows (Gunn & Gott 1972): P ram = ρ ICM v gal , (9)where ρ ICM is the local ICM density and v gal is the rel-ative velocity. Due to the projection effect, it is hard todetermine velocities in 3D and connect them to the 2Dgas offsets. However, observations have revealed thatrecent infall gas-rich galaxies in clusters affected by rampressure are mostly on highly radial orbits (Chung et al.2007, Smith et al. 2010, Vulcani et al. 2017, Jaff´e et al. 2018; see also Ebeling et al. 2014 where such effect is ob-served on galaxies with tangential infall orbits.). Thissuggests a connection between the projected offset ofionized gas emission in these galaxies and their direc-tions to the cluster center. In fact, Smith et al. (2010)observed that ionized gas tails of gas-stripping galaxiesin the nearby Coma cluster predominantly point oppo-site to the cluster center using UV+H α data. Vulcaniet al. (2017) found that ram pressure stripped SFGs outto 0.5 virial radius in intermediate redshift clusters pref-erentially show radial H α offsets away from the clustercenter. Using the IllustrisTNG simulations, Yun et al.(2019) investigated the directions of gas tails of jellyfishgalaxies in galaxy clusters in 3D. They found a nearlyflat distribution with slow drops in the 0 ◦ and 180 ◦ endsfor the angle between gas tail and direction to the hostcenter. This is not unexpected, as, unlike in 2D, in3D only galaxies with purely radial orbits would showgas tails aligned with the direction to the cluster center.However, they did observe a tight correlation betweenthe directions of gas tails and the 3D velocity vector,which should also hold in 2D projection, allowing us touse the offset vector as a proxy for the galaxy’s 2D pro-jected velocity vector.Besides ram pressure stripping, other physical pro-cesses might serve as potential explanations for the ion-ized gas offset such as:(1) Outflows: strong outflows driven by active galac-tic nuclei (AGNs) or starbursts have been observed inmany active galaxies, which could also lead to signifi-cant gas offset (e.g., Comerford et al. 2017, Russell et al.2019). They are considered as an important source ofICM heating to explain the cooling problem in galaxyclusters (e.g., Br¨uggen et al. 2005). Indeed, the high-velocity H α bump in the the north-west of the BCGof A2390 is likely to be driven by an AGN. Because wehave included galaxies with nuclear activities in our ELGsample, it cannot be ruled out that a portion of the spa-tial offsets of ionized gas in our ELG sample are causedby outflows. In fact, recent studies based on simulationsand observations found that galactic nuclear activitiescan be triggered by ram pressure stripping due to en-hanced accretion onto the central black hole (e.g. Pog-gianti et al. 2017b, Ramos-Mart´ınez et al. 2018, Ricarteet al. 2020). However, there has been little evidenceof preference in outflow directions in galaxy clusters.Because outflow is an internal phenomenon controlledby the central black hole in the nuclei region, its sub-sequent interaction with the gas in the host galaxy islikely independent of the global environment. There-fore, gas offsets induced by outflows are not expected tohave a correlation with the cluster center as observed. ITELLE Cluster ELGs ρ ICM is in-creased by compression due to merger shocks. A largersample is needed to construct a sample of mergers withreference to non-mergers to investigate the difference intheir ionized gas distributions.Other systematics such as the presence of AGNs anddust cause dilution effects in our centroid measurementsthat decreases the large ∆ d sample size and increases theuncertainties, but are unlikely to change the observedpatterns. Therefore, we conclude that the non-uniformdistribution of the spatial offset of the ionized gas withrespect to the stellar disk in our observation is primarilycontributed by ram pressure stripping. In comparison,the long-term interplay between a galaxy and the hostcluster, such as strangulation or thermal evaporation,would typically leave behind an undisturbed, symmetricgas distribution.5.2. The Distributions of Ionized Gas Offset in PhaseSpace
As discussed above, we are able to use the offset vec-tor d as a proxy for the projected velocity of an infallinggalaxy, in light of ram pressure stripping acting as themain driver producing the non-uniformity that we ob-serve in distributions of θ d . Combining with the infallhistory of a typical gas-rich cluster galaxy in the phasespace diagram, the distributions of ELGs with large ion-ized gas offset in phase space offer evidence of ram pres-sure stripping as a quenching mechanism in the act.5.2.1. Infall History of a Cluster Galaxy on the PhaseSpace Diagram
The trajectory of an infalling galaxy can be tracedon the cluster phase space diagram. The dark dashedcurve in Figure 12 illustrates the typical infall history ofa cluster galaxy.Generally, an incoming SFG is first found in regionI with a high gas fraction. During its first infall into .
00 0 .
25 0 .
50 0 .
75 1 .
00 1 .
25 1 .
50 1 .
75 2 . R / R . . . . . . . | v l o s − v | / σ v IIIII I
Low High H i g h I n c l . O r b i t L o w I n c l . O r b i t Gas Fraction of Galaxy
Figure 12.
Schematic of the trajectories of an infallinggas-rich cluster galaxy on the velocity-position phase spacediagram. The dark dashed line shows the typical path inwhich it moves from region I to region II, undergoes a fewturnarounds, and eventually being virialized in region III.The light dashed line shows the path of a similar galaxy butwith a low inclination (nearly face-on) orbit. It spends mostof the time in region III from its first passage due to pro-jection effect. The color coding illustrates the gas fractionof the galaxy, with the dark/light-green points indicating or-bit pericenters/apocenters. The black dashed line indicates R . The figure is adapted from Figure 1 in Rhee et al.(2017) (note the original plot is in 3D). the potential well of the cluster, the orbit can be highlyradial (e.g. Vulcani et al. 2017, Jaff´e et al. 2018). Itwill increase its velocity as it approaches the orbit peri-center, moving toward the upper left of the phase spacediagram (region II). The local ICM density also increasesas the galaxy penetrates into the cluster core. Conse-quently, the ram pressure from head-on ICM winds in-creases dramatically. It compresses and strips the coldgas, while having little effect on the stellar components.The SFR may temporarily rise due to the compressionof gas (e.g., Fujita & Nagashima 1999, Poggianti et al.2017a, Vulcani et al. 2018). As star formation proceeds,the neutral gas offset propagates to ionized gas, lead-ing to the difference in the spatial distribution of emis-sion and continuum. Post pericenter, it will then turnaround with decreasing velocity and move out of the cen-tral region, referred to as the “backsplash” stage (e.g.,Gill et al. 2005, Muriel & Coenda 2014). As a result,the backsplash region overlaps with the lower-velocitypart of the first infall region. The galaxy then entersits subsequent infall but with a lower gas fraction. Suchprocess may happen multiple times until the radial ve-6 Liu et al. locity is dissipated as the galaxy spirals in. Eventually,the galaxy falls into the virialized region, with little orno gas reservoir left to maintain its star formation.The fact that cluster galaxies follow typical paths inthe course of their sinking into the cluster potential leadsto their occupations of certain regions on the phase spacediagram. However, the condition becomes more com-plex when orbit orientation and projection effects aretaken into account. The light dashed line in Figure 12illustrates the path of an infalling galaxy with a lowinclination angle orbit, hence, with low line-of-sight ve-locity. While it still takes a few turnarounds to achievethe virialization of the orbit, the galaxy is not able tomove up to region II in the phase space diagram dueto the viewing angle of the orbit. As a consequence,region III also contains galaxies in their first passage.Another concern is that an infalling galaxy physicallyoutside the virial radius may be projected to be within R on the phase space diagram. For an extreme case,a galaxy following an orbit along the line of sight couldhave strong emission-lines detected if there is ongoingstar formation, but it would barely move in the radialaxis. However, in this case the gas offset is also pro-jected, which makes it more likely to have a small ∆ d and thus be excluded from the selected sample.5.2.2. Excess of Gas Offset Vectors toward the ClusterCenter near Orbit Pericenters at First Infall
The number ratios of blue to red symbols in regionII (6:2 in A2390, and 8:1 in A2465 combining two sub-clusters) suggest that first-infalling galaxies have a pref-erence for their ionized gas offsets to point away fromthe cluster center when they approach the orbit pericen-ters. The probability to have such ratios by coincidenceis 11% for A2390, 4% for A2465, and 1% combining thetwo, assuming an equal opportunity (i.e. a binomialdistribution with p=0.5) for the offset vector to pointtoward/away from the center.Such preference can be attributed to the impact ofthe radial component of the ram pressure on the clus-ter galaxy. The result is consistent with the other ob-servational evidence and simulations that suggest rampressure stripping has substantial effects when gas-richgalaxies cross the cluster virial radius and is most effec-tive in their first infall (e.g., Jaff´e et al. 2016, 2018; Lotzet al. 2019). Because of the efficient removal of cold gas,star formation has been mostly suppressed before theincoming galaxies turn around and start their secondinfalls. The typical quenching timescale for these galax-ies is thus suggested to be shorter than the ram pressurestripping timescale ( (cid:46) . ∼ > < ∼ ITELLE Cluster ELGs σ v . Both may contribute the scarcity of ELGswith high velocity ( > σ v ) in Figure 11. Our cur-rent data of A2465 are also restricted within ∼ Additional Support for Quenching by RamPressure Stripping
To find additional evidence of ram pressure strippingaffecting the star formation in cluster galaxies, we in-vestigate the line flux. Here we only look into A2390for two reasons:(1) our current A2465 data do not havecoverage in the outer region, and (2) A2390 is more of atypical rich cluster compared with the merging doublecluster A2465. The H α emission-line flux F H α is ob-tained from the residual spectra yielded in Section 3.4by a trapezoidal integral of ± σ line around the peakof the H α line, where σ line is the best-matched tem-plate line width. To normalize the emission-line flux bygalaxy mass, we divide it by the mean continuum timesa fixed wavelength range of 250˚ A . The normalized lineflux F H α ,n can be used as a proxy for the specific starformation rate (sSFR), defined as the SFR per stellarmass, under the circumstance where the ionization iscaused by star formation. For a composite ionizationof star formation and AGN, F H α ,n should be treatedas upper limits of star formation. F H α ,n also has a de-pendency on the mass-to-light ratio (and thus the starformation history) of the galaxy, but the dependency isminor, and we use it here as a first-order estimate.The continuum-normalized line flux is plotted in Fig-ure 13 versus the distance to the cluster center. ELGswith larger ∆ d are shown by solid colored symbols,whereas those with smaller or undetected ∆ d are repre-sented by gray-filled symbols. Blue, green, orange colorsrepresent ELGs in the low-velocity outer region (region . . . . . . . . R (Mpc) − . − . − . − . − . − . . . . l og ( F H α , n ) R I : ∆ d > kpc I : ∆ d < kpc III : ∆ d > kpc III : ∆ d < kpc II : ∆ d > kpc II : ∆ d < kpc LINERSeyfert
Figure 13.
Emission-line flux normalized by continuum vsdistance to the cluster center for cluster member ELGs inA2390. ELGs with large/small (projected) ionized gas offset∆ d are shown as colored/gray-filled symbols. The face/edge-colors mark ELGs in different regions on the phase space dia-gram (Figure 8). The colored solid/dashed lines indicate thecentral locations of distributions of the colored/gray symbolswith the corresponding color. The colored bands indicatethe 1 σ (16%-84%) uncertainty of the mean. Galaxy withSeyfert/LINER features are marked in diamonds/squares.The magenta dashed-dotted line shows the central locationof the distribution of all markers. The vertical black dashedline marks R . I), the high-velocity inner region (region II), and thevirialized region (region III), respectively, as indicatedin Figure 8. The 1 σ error bars are calculated numer-ically by Monte Carlo simulations, where the spectraare perturbed 250 times using the rms noise of the con-tinua. We use the biweighted location, which is a robuststatistic to represent the central location of a distribu-tion. The biweighted locations of solid/open symbolsare shown as colored solid/dashed lines with the corre-sponding colors. The central location of all markers isshown as the magenta dashed-dotted line. The shadedbands indicate the 1 σ uncertainties of the central lo-cations, calculated by Monte Carlo simulations whereeach data point is randomly perturbed based on its un-certainty. For the initial identification of galaxies withAGN features, we adopt a classification based on theWHAN diagram (Cid Fernandes et al. 2011) using lineratios obtained from cross-correlation. We label themin Figure 13 but do not exclude them from the sample,given that those with LINER features present simulta-neous ionization from star formation in the disk and8 Liu et al.
Seyferts are rare. AGNs in our fields will be investi-gated in detail in future work.Several interesting patterns are revealed in Figure 13.In the outer region (region I), ELGs with smaller ∆ d have higher F H α ,n on average ( ∼ d . The significance is above 3 σ . This isalso true for samples in region II - larger ∆ d objectshave lower F H α ,n , although the significance (at 1.9 σ ) isnot high considering the uncertainties. Region III ap-pears to have statistically consistent average F H α ,n forlarge and small ∆ d objects. Furthermore, comparingobjects within R (region II & III, green & orangesymbols) with objects outside R (region I, blue sym-bols), they have on average lower F H α ,n ( ∼ d objects), although a small fractionof objects within R appear to show strong lines.The lower average line flux for large ∆ d objects inregion I could be the result of them undergoing rampressure stripping, with the star formation being sup-pressed due to gas removal. Some of these could havealready passed through the pericenter of their infall or-bit, moving out to the backsplash stage. Others may bein their first infall, where ram pressure stripping has justbeen turned on not long ago. On the other hand, ob-jects with smaller ∆ d are likely incoming galaxies thathave not experienced significant, if any, ram pressurestripping. They might be farther away from the clus-ter but projected to be closer in 2D, or with smallerimpact parameters and thus less affected by the ICMwinds. Again, we note that galaxies with small ∆ d could also have suffered from ram pressure stripping,but with their small ∆ d being the result of projectioneffect. However, the lower emission line flux observed inlarger ∆ d ELGs clearly serves as additional evidence ofram pressure stripping being responsible for quenching.As discussed, the large ∆ d objects in region II arelikely experiencing strong ram pressure stripping. Thiseffect can consistently explain the difference in line fluxcompared with small ∆ d objects. However, the com-position of the small ∆ d objects is actually complex: itmay contain galaxies (1) in their first passages where gasremoval through ram pressure stripping is in progress;(2) in the backsplash stage, having survived from thedramatic gas removal; and (3) with prominent gas offsetin 3D, but projected to be small. This may explain thelarger scatter in F H α ,n for small ∆ d objects. In the firstcase, the galaxies have higher line fluxes as they are inthe beginning stage of having their star formation under-going suppression, while in the latter two cases, the levelof star formation is expected to be comparable or lower.In the last case, some galaxies that appear within R might actually fall outside it in 3D due to projection and thus would have a smaller amount of star formationsuppression. It would be interesting to discriminate andstudy them using other tracers such as gas fractions.In region III, large and small ∆ d objects have con-sistent line fluxes. This suggests that either the offsetis unassociated with ram pressure stripping (e.g. fromoutflows), or the offsets have different origins but thegalaxies are projected to be there. Galaxies in regionIII suffer from all kinds of projection effects discussedabove, making them hard to interpret in the phase spacealone.The few objects inside R observed with exception-ally high F H α ,n suggest that a small fraction of clustergalaxies suffered from ram pressure stripping could haveenhanced star formation, or nuclei activities, due to gascompression by shocks. The enhancement is suggestedto have short timescales because the fraction of such ob-jects is low. Such enhancement has been reported bymany previous studies (e.g. Fujita & Nagashima 1999,Kronberger et al. 2008, Bekki 2014). More recently,Vulcani et al. (2018) found enhanced SFR in a statisti-cally significant sample of jellyfish galaxies from GASP.However, there is a difference in sample constructiongiven the low occurrence of jellyfish galaxies per clus-ter, whereas our sample mostly consists of mild cases ofgalaxies undergoing ram pressure stripping.Finally, the lower average line flux in regions inte-rior to R serves as an indication of environmentalquenching, suggesting that the sSFR is dependent onthe cluster-centric radius. One of the most competi-tive mechanisms is ram pressure stripping, as discussedabove, while other quenching mechanisms (starvation,thermal evaporation, etc.) could take charge when rampressure stripping is not efficient enough. SUMMARYAn ongoing survey using the IFTS SITELLE at CFHTtargeting the H α -[N ii ] lines in clusters at z ∼ ITELLE Cluster ELGs d and the differenceangle θ d between the emission-to-continuum offset vec-tor d and the cluster-centric vector to investigate thecorrelation between spatial offsets of ionized gas in clus-ter galaxies and their positions in the galaxy cluster.Based on the distributions of ∆ d and θ d , we find (1)ELGs in A2390 and A2465 have ∆ d extending from 0to around 4 kpc with the peak of the offset distributiondeviating from zero, implying some systematic impactcausing the spatial offset of ionized gas. (2) Looking intoELGs with larger emission-line spatial offsets (∆ d > θ d in A2390 and A2465 clearlydeviate from a uniform distribution. The peak of θ d to-ward 0 ◦ in our result is consistent with previous workthat ionized gas preferentially points away from the clus-ter center (e.g. Smith et al. 2010, Vulcani et al. 2017),although we also observe a hint of a minor peak in the180 ◦ , possibly caused by backsplash or projection ef-fects. The p-values from KS tests reject the uniformhypothesis at 95% confidence. This serves as an evi-dence of ram pressure stripping playing an importantrole in shaping and removing the gas reservoirs of clus-ter galaxies, because other mechanisms such as outflowsand tidal stripping are expected to have no directionalrelation to the cluster center while ram pressure strip-ping impacts radially in the course of the infall of clustergalaxies.We further investigate their distributions on theposition-velocity phase space diagram, using the offsetvector d as a proxy for the 2D projected velocity vector.We divide the phase space diagram into three regionsaccording to the infall history of a cluster galaxy: (I)the outskirts region outside the virial radius, (II) thehigh-velocity region inside the virial radius, and (III)the low-velocity region inside the virial radius. We lookinto the ratio of ELGs with large ∆ d having θ d < ◦ and θ d > ◦ (i.e. with emission offset vectors point-ing away and toward the cluster center, respectively).Combining the statistics from both clusters, we find a3 σ excess of galaxies in region II with θ d < ◦ , i.e.,galaxies approaching or close to the orbit pericenters intheir first infalls. The preference suggests star formationin these galaxies is mostly suppressed before they turnaround and start their second infalls. This is consis-tent with conclusions from some numerical studies andobservations of individual gas-stripped cluster galaxiesthat galaxies are quenched at first infall under ram pres- sure stripping after they penetrate into the ICM halo.In addition, we find that the continuum-normalized lineflux of ELGs with large ∆ d in region I (and region II,but with a lower significance), where galaxies are at thebeginning of their first infalls or undergoing backsplash,is lower on average than those with small ∆ d . This sup-ports the scenario that ram pressure stripping is a dom-inant mechanism in the suppression of star formation.No significant difference, however, is found in region III,where projection effects may mix galaxies with differentdynamic states together.Our study is the first analysis of such types of observa-tions and demonstrates the uniqueness and promise ofpanoramic, wide-field 2D spectroscopy on galaxy clus-ters using SITELLE. Data from a larger sample of clus-ters and further detailed analysis will shed more light onthe environmental effects and star formation quenchingmechanisms operating on cluster galaxies.ACKNOWLEDGEMENTWe thank the referee for the useful review that helpedimprove this manuscript. Based on observations ob-tained at the Canada-France-Hawaii Telescope (CFHT)which is operated from the summit of Maunakea bythe National Research Council of Canada, the InstitutNational des Sciences de l’Univers of the Centre Na-tional de la Recherche Scientifique of France, and theUniversity of Hawaii. The observations at the Canada-France-Hawaii Telescope were performed with care andrespect from the summit of Maunakea, which is a sig-nificant cultural and historic site. Based on observa-tions obtained with SITELLE, a joint project betweenUniversit´e Laval, ABB-Bomem, Universit´e de Montr´eal,and the CFHT with funding support from the CanadaFoundation for Innovation (CFI), the National Sciencesand Engineering Research Council of Canada (NSERC),Fond de Recherche du Qu´ebec - Nature et Technolo-gies (FRQNT) and CFHT. Q.L. is supported by an On-tario Trillium Awards. HY’s research is supported byan NSERC Discovery Grant and grants from the Artsand Science Faculty at the University of Toronto. L.D.’sresearch is supported by an NSERC discovery grant. Software:
SExtractor (Bertin & Arnouts 1996), pho-tutils (Bradley et al. 2016), ORBS (Martin et al. 2015),ORCS (Martin et al. 2015), astropy (Astropy Collabo-ration et al. 2013, 2018), numpy (Harris et al. 2020),matplotlib (Hunter 2007), scikit-learn (Pedregosa et al.2011), SAOImage DS9 (Joye & Mandel 2003)REFERENCES
Abadi, M. G., Moore, B., & Bower, R. G. 1999, MNRAS,308, 947 Abraham, R. G., Smecker-Hane, T. A., Hutchings, J. B.,et al. 1996, ApJ, 471, 694 Liu et al.
Acker, A., K¨oppen, J., Samland, M., & Stenholm, B. 1989,The Messenger, 58, 44Astropy Collaboration, Robitaille, T. P., Tollerud, E. J.,et al. 2013, A&A, 558, A33Astropy Collaboration, Price-Whelan, A. M., SipHocz,B. M., et al. 2018, aj, 156, 123Bah´e, Y. M., & McCarthy, I. G. 2015, MNRAS, 447, 969Balogh, M. L., & Morris, S. L. 2000, MNRAS, 318, 703Balogh, M. L., Schade, D., Morris, S. L., et al. 1998, ApJL,504, L75Bekki, K. 2014, MNRAS, 438, 444Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393Boselli, A., Boissier, S., Cortese, L., & Gavazzi, G. 2008,ApJ, 674, 742Boselli, A., Cortese, L., Boquien, M., et al. 2014, A&A,564, A67Boselli, A., & Gavazzi, G. 2006, PASP, 118, 517Boselli, A., Roehlly, Y., Fossati, M., et al. 2016, A&A, 596,A11Boselli, A., Epinat, B., Contini, T., et al. 2019, A&A, 631,A114Bradley, L., Sipocz, B., Robitaille, T., et al. 2016,Photutils: Photometry tools, , , ascl:1609.011Brough, S., Croom, S., Sharp, R., et al. 2013, MNRAS, 435,2903Br¨uggen, M., Ruszkowski, M., & Hallman, E. 2005, ApJ,630, 740Bundy, K., Bershady, M. A., Law, D. R., et al. 2015, ApJ,798, 7Carlberg, R. G., Yee, H. K. C., & Ellingson, E. 1997, ApJ,478, 462Chambers, K. C., Magnier, E. A., Metcalfe, N., et al. 2016,arXiv e-prints, arXiv:1612.05560Chung, A., van Gorkom, J. H., Kenney, J. D. P., &Vollmer, B. 2007, ApJL, 659, L115Cid Fernandes, R., Stasi´nska, G., Mateus, A., & Vale Asari,N. 2011, MNRAS, 413, 1687Coenda, V., Mast, D., Mart´ınez, H. J., Muriel, H., &Merch´an, M. E. 2019, A&A, 621, A98Comerford, J. M., Barrows, R. S., Greene, J. E., & Pooley,D. 2017, ApJ, 847, 41Cortese, L., van de Sande, J., Lagos, C. P., et al. 2019,MNRAS, 485, 2656Dressler, A. 1980, ApJ, 236, 351Drissen, L., Martin, T., Rousseau-Nepton, L., et al. 2019,MNRAS, 485, 3930Ebeling, H., Stephenson, L. N., & Edge, A. C. 2014, ApJL,781, L40Ellingson, E., Lin, H., Yee, H. K. C., & Carlberg, R. G.2001, ApJ, 547, 609 Fillingham, S. P., Cooper, M. C., Pace, A. B., et al. 2016,MNRAS, 463, 1916Foltz, R., Wilson, G., Muzzin, A., et al. 2018, ApJ, 866, 136Fritz, A., Ziegler, B. L., Bower, R. G., Smail, I., & Davies,R. L. 2005, MNRAS, 358, 233Fujita, Y., & Nagashima, M. 1999, ApJ, 516, 619Gill, S. P. D., Knebe, A., & Gibson, B. K. 2005, MNRAS,356, 1327G´omez, P. L., Nichol, R. C., Miller, C. J., et al. 2003, ApJ,584, 210Goto, T., Yamauchi, C., Fujita, Y., et al. 2003, MNRAS,346, 601Graham, M. T., Cappellari, M., Bershady, M. A., & Drory,N. 2019, arXiv e-prints, arXiv:1910.05139Green, A. W., Croom, S. M., Scott, N., et al. 2018,MNRAS, 475, 716Gunn, J. E., & Gott, J. Richard, I. 1972, ApJ, 176, 1Haines, C. P., Pereira, M. J., Smith, G. P., et al. 2013, ApJ,775, 126Harris, C. R., Millman, K. J., van der Walt, S. J., et al.2020, Nature, 585, 357.https://doi.org/10.1038/s41586-020-2649-2Haynes, M. P., Giovanelli, R., & Chincarini, G. L. 1984,ARA&A, 22, 445Hunter, J. D. 2007, Computing in Science & Engineering, 9,90Hutchings, J. B., & Balogh, M. L. 2000, AJ, 119, 1123Jaff´e, Y. L., Smith, R., Candlish, G. N., et al. 2015,MNRAS, 448, 1715Jaff´e, Y. L., Verheijen, M. A. W., Haines, C. P., et al. 2016,MNRAS, 461, 1202Jaff´e, Y. L., Poggianti, B. M., Moretti, A., et al. 2018,MNRAS, 476, 4753Joye, W. A., & Mandel, E. 2003, in Astronomical Society ofthe Pacific Conference Series, Vol. 295, AstronomicalData Analysis Software and Systems XII, ed. H. E.Payne, R. I. Jedrzejewski, & R. N. Hook, 489Kapferer, W., Kronberger, T., Ferrari, C., Riser, T., &Schindler, S. 2008, MNRAS, 389, 1405Kauffmann, G., White, S. D. M., Heckman, T. M., et al.2004, MNRAS, 353, 713Kawinwanichakij, L., Papovich, C., Quadri, R. F., et al.2017, ApJ, 847, 134Kennicutt, R. C., & Evans, N. J. 2012, ARA&A, 50, 531Koopmann, R., & Kennedy, J. D. P. 1999, News Letter ofthe Astronomical Society of New York, 5, 22Kronberger, T., Kapferer, W., Ferrari, C.,Unterguggenberger, S., & Schindler, S. 2008, A&A, 481,337
ITELLE Cluster ELGs Lewis, I., Balogh, M., De Propris, R., et al. 2002, MNRAS,334, 673Li, I. H., Yee, H. K. C., & Ellingson, E. 2009, ApJ, 698, 83Lotz, M., Remus, R.-S., Dolag, K., Biviano, A., & Burkert,A. 2019, MNRAS, 488, 5370Martin, T., Drissen, L., & Joncas, G. 2015, in AstronomicalSociety of the Pacific Conference Series, Vol. 495,Astronomical Data Analysis Software an Systems XXIV(ADASS XXIV), ed. A. R. Taylor & E. Rosolowsky, 327Mayer, L., Mastropietro, C., Wadsley, J., Stadel, J., &Moore, B. 2006, MNRAS, 369, 1021Muriel, H., & Coenda, V. 2014, A&A, 564, A85Muzzin, A., Wilson, G., Yee, H. K. C., et al. 2012, ApJ,746, 188Muzzin, A., van der Burg, R. F. J., McGee, S. L., et al.2014, ApJ, 796, 65Noble, A. G., Webb, T. M. A., Muzzin, A., et al. 2013,ApJ, 768, 118Owers, M. S., Hudson, M. J., Oman, K. A., et al. 2019,ApJ, 873, 52Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011,Journal of Machine Learning Research, 12, 2825Peng, Y., Maiolino, R., & Cochrane, R. 2015, Nature, 521,192Peng, Y.-j., Lilly, S. J., Kovaˇc, K., et al. 2010, ApJ, 721, 193Pintos-Castro, I., Yee, H. K. C., Muzzin, A., Old, L., &Wilson, G. 2019, ApJ, 876, 40Poggianti, B. M., von der Linden, A., De Lucia, G., et al.2006, ApJ, 642, 188Poggianti, B. M., Moretti, A., Gullieuszik, M., et al. 2017a,ApJ, 844, 48Poggianti, B. M., Jaff´e, Y. L., Moretti, A., et al. 2017b,Nature, 548, 304Postman, M., Franx, M., Cross, N. J. G., et al. 2005, ApJ,623, 721Ramos-Mart´ınez, M., G´omez, G. C., & P´erez-Villegas, ´A.2018, MNRAS, 476, 3781Rhee, J., Smith, R., Choi, H., et al. 2017, ApJ, 843, 128 Ricarte, A., Tremmel, M., Natarajan, P., & Quinn, T. 2020,ApJL, 895, L8Roediger, E., Br¨uggen, M., Simionescu, A., et al. 2011,MNRAS, 413, 2057Russell, H. R., McNamara, B. R., Fabian, A. C., et al.2019, MNRAS, 490, 3025S´anchez, S. F., Kennicutt, R. C., Gil de Paz, A., et al.2012, A&A, 538, A8Schaefer, A. L., Croom, S. M., Allen, J. T., et al. 2017,MNRAS, 464, 121Schaefer, A. L., Tremonti, C., Pace, Z., et al. 2019, ApJ,884, 156Smith, R. J., Lucey, J. R., Hammer, D., et al. 2010,MNRAS, 408, 1417Sobral, D., Best, P. N., Smail, I., et al. 2011, MNRAS, 411,675Vijayaraghavan, R., & Ricker, P. M. 2013, MNRAS, 435,2713Vulcani, B., Treu, T., Nipoti, C., et al. 2017, ApJ, 837, 126Vulcani, B., Poggianti, B. M., Gullieuszik, M., et al. 2018,ApJL, 866, L25Wang, J., Xu, W., Lee, B., et al. 2020, ApJ, 903, 103Wegner, G. A. 2011, MNRAS, 413, 1333Wegner, G. A., Chu, D. S., & Hwang, H. S. 2015, MNRAS,447, 1126Wegner, G. A., Umetsu, K., Molnar, S. M., et al. 2017,ApJ, 844, 67Weinmann, S. M., Kauffmann, G., van den Bosch, F. C.,et al. 2009, MNRAS, 394, 1213Yee, H. K. C., Ellingson, E., Abraham, R. G., et al. 1996,ApJS, 102, 289Yoon, H., Chung, A., Smith, R., & Jaff´e, Y. L. 2017, ApJ,838, 81Yoshida, M., Yagi, M., Komiyama, Y., et al. 2008, ApJ,688, 918Yun, K., Pillepich, A., Zinger, E., et al. 2019, MNRAS, 483,1042 Liu et al.
APPENDIX A. REDUCTION OF INTERFEROMETRICFRINGESBelow we describe the low-pass filtering (LPF) tech-nique used to remove fringes on the SITELLE datacube.A source detection with an S/N threshold of 3 is firstrun on the stack image using the python photometricpackage photutils to pick out possible sources. Thesesources are masked before the LPF procedure to avoidthem being over-subtracted. Next, LPF is performed oneach channel by convolving the image with an ellipticalgaussian kernel elongated in the x-axis to construct alow-frequency components image and then subtractingit from the original image. The method is based on thefact that the fringes are high-frequency components thatare relatively continuous along the x-axis but may radi-cally change its brightness in the y-axis. In practice, weadopt a kernel size with FWHM of 28 pix in the x-axisand 7 pix in the y-axis. The kernel size is determinedby experimenting with different sizes to reach a trade-offbetween reducing the high-frequency fringes and avoid-ing over-subtraction of faint candidates that are possiblymissed in the initial detection.After iterating through all the channels, a source de-tection is run again on the low-pass-filtered datacube,again with all of the detected sources masked. Theconvolve-subtract-detect-mask process is repeated untilthe change in the number of detected sources is smallerthan 5% in order to capture as many faint sources as pos-sible. The efficacy of the LPF is clear by comparing theupper two panels of Figure 14 showing the same cutoutof a channel that suffered from fringes before and af-ter LPF, where contamination from fringes occurring inthe left panel has been mitigated as shown in the rightpanel. Two sample spectra extracted in the A2390Efield, which are ELGs identified using procedures in Sec-tion 3.6, are shown in Figure 15 to demonstrate the effectof the LPF on the extracted spectra. The LPF process-ing can greatly reduce additional noise caused by fringesin channels between 7980 ˚A and 8030 ˚A where strongsky-lines are located, thus facilitating the detection andidentification of faint emission-line candidates.Because of the finite kernel size used, in some chan-nels there exist fringe residuals whose spatial variationis smaller than the kernel. This can be observed inthe upper-right panel of Figure 14. To further cleanthe residuals and facilitate candidate detection, we ap-ply a moving average processing to construct a new de-tection datacube: the low-pass-filtered datacube is con- ◦ One Channel(Before LPF) One Channel(After LPF) h m s s s s s ◦ Moving Averaged(Before LPF) h m s s s s s Moving Averaged(After LPF)
RA (J2000) D ec ( J ) Figure 14.
Improvement by post-processing the spectraldatacube in the presence of interferometric fringes. Upperleft: image of a single channel showing fringes after subtract-ing the large-scale background. The brightness of the fringesdoes not follow a regular pattern. Upper right: image of thesame channel processed with the LPF process. Fringes havebeen mitigated, although some residuals remain in the im-age. Lower left: image of the same channel but each pixel ismanipulated by a moving average using a 3x3x5 box. Lowerright: image of the same channel applying both LPF andmoving average processing. The background residuals arelargely suppressed. For direct visual comparison, the fourpanels share the same contrast. The image contrast is inarcsinh stretch to visually augment the small difference inthe background. volved with a 3x3x5 averaging kernel (2D spatial + 1Dspectral), i.e., for a single spaxel its value is taking themean of contiguous pixels in 2D and in the adjacentfour channels. This takes advantage of the feature ofthe fringe pattern that it moves across the field as thescanning/wavelength proceeds/increases. As a result,residuals in nearby channels are expected to be moreor less cancel out. The moving average processing alsoreduces the sky noise. Note this new datacube is onlyused for source detection, not for centroid measurementin Section 4.1.
ITELLE Cluster ELGs Wavelength (˚A) − − − F l u x ( − e r g / c m / s / ˚ A ) Before LPFAfter LPF
Wavelength (˚A) − . − . − . − . . . . . . F l u x ( − e r g / c m / s / ˚ A ) Before LPFAfter LPF
Figure 15.
Sample spectra of two ELGs demonstrat-ing the effectiveness of LPF in reducing additional noisecaused by fringes (gray: before LPF; blue: after LPF)for channels around strong sky-lines (gray band). Upper:[N ii ] λλ ∼ . iii ] λλ ∼ . UNCERTAINTIES IN CENTROID ANALYSISB.1.
Uncertainty Propagation from Centroids toAngles and Offsets
The uncertainties of flux-weighted centroids (¯ x, ¯ y ) aregiven by error propagation through σ x = (cid:88) i σ I i (cid:18) ∂ ¯ x∂I i (cid:19) = (cid:88) i σ I i (cid:34) ∂∂I i (cid:32) (cid:80) j x j I j (cid:80) j I j (cid:33)(cid:35) = (cid:88) i σ I i ( (cid:80) j I j ) (cid:34) − (cid:80) j x j I j (cid:80) j I j ∂ ( (cid:80) j I j ) ∂I i + ∂ ( (cid:80) j x j I j ) ∂I i (cid:35) = (cid:80) i σ I i · ( x i − ¯ x ) ( (cid:80) j I j ) , (B1) σ y = (cid:80) i σ I i · ( y i − ¯ y ) ( (cid:80) j I j ) . (B2)Let ∆¯ x = ¯ x E − ¯ x C and ∆¯ y = ¯ y E − ¯ y C , the uncertain-ties in the offset and difference angle are given by errorpropagation through: σ ∆ d = σ (cid:104) (cid:2) (∆¯ x ) + (∆¯ y ) (cid:3) (cid:105) = (cid:34) ∆¯ x · σ ∆¯ x (cid:112) (∆¯ x ) + (∆¯ y ) (cid:35) + (cid:34) ∆¯ y · σ ∆¯ y (cid:112) (∆¯ x ) + (∆¯ y ) (cid:35) = (cid:113) ((∆¯ x · σ ∆¯ x ) + (∆¯ y · σ ∆¯ y ) ) / ∆ d , (B3) and σ θ d = σ (cid:104) atan(∆¯ y/ ∆¯ x ) (cid:105) = σ (cid:104) ∆¯ y/ ∆¯ x (cid:105) (∆¯ y/ ∆¯ x ) + 1= (∆¯ y/ ∆¯ x ) · (cid:112) ( σ ∆¯ y / ∆¯ y ) + ( σ ∆¯ x / ∆¯ x ) (∆¯ y/ ∆¯ x ) + 1= ( (cid:113) (∆¯ x · σ ∆¯ y ) + (∆¯ y · σ ∆¯ x ) ) / ∆ d , (B4)where σ ∆¯ x = (cid:113) σ x E + σ x C and σ ∆¯ y = (cid:113) σ y E + σ y C .B.2. An Empirical Assessment for Seeing-limitedCentroid Measurements
The measured centroid offsets in emission and contin-uum are in general small relative to the seeing and there-fore suffer from smearing effects. However, it should benoted that the centroid of an object can be measured toa considerably higher precision than the seeing, depend-ing on the S/N of the object. We perform several exper-iments with a control sample to demonstrate the robust-ness and effectiveness of the centroid measurements.The control sample is constructed using unsatu-rated stars cross-matched with PAN-STARRS (Cham-bers et al. 2016) and is matched with the ELG samplein terms of S/N as follows: we randomly draw a starfrom the crossmatch and choose a medium wide win-dow in its spectrum with channels in it to represent thepseudo-emission. The rest of the channels are used asthe continuum. Edges and channels in the presence ofstrong sky-lines are excluded. The pseudo-emission andcontinuum images are constructed from these channelswith which centroids are measured in the same approachas Section 4.1 except that the continuum is not sub-tracted from the emission. This process is repeated for500 times field by field to build a parent sample for eachfield. We then resample the measurements by their pho-tometric S/N for N = 100 times according to the dis-tribution of S/N of the ELGs on the emission image.The sample size N is chosen to match the average sam-ple size of ELGs detected in each field for the statisticaltests below. We construct 10 different control samplesto account for sample variation.With the control samples, we test whether there is sys-tematic bias in the centroid measurement and quantifythe uncertainties from random noise. Because the mea-surements are based on stars, the centroids should haveno intrinsic offset unless there is any systematic bias.We first perform a Hotelling’s T test on the locationof the distribution of centroid difference x = (∆ x, ∆ y )with the null hypothesis H : x = x = (0 , t Liu et al. d (pix ) D e n s i t y ( , ) Figure 16.
Distributions of (∆ d ) measured from 10 controlsamples made up of stars in the A2390C field. Red curvesshow the best fits to χ distributions. Distributions of fittedcenters, standard deviations, and fractions of measurementswithin the adopted threshold are displayed in small panels. statistic is given by t = n ( x − x ) (cid:48) ˆ S − ( x − x ) ∼ χ p (B5)with x representing the sample mean, ˆ S = N − (cid:80) Ni =1 ( x i − x ) ( x i − x ) (cid:48) to be the sample covarianceand p = 2 to be the degree of freedom. The last condi-tion in the equation above satisfies assuming the largesample approximation. If t is larger than the criti-cal value, the upper 1- α quantile χ p ( α ), H is then re-jected at confidence level of α . With the computed t in each field, we cannot reject H at confidence levelof α = 0 .
01 in 9/8/9/9 out of 10 control samples forA2390C/A2390E/A2390W/A2465C, that is, no signifi-cant systematic bias is found for the centroid measure-ments.The other test is motivated by the fact that (∆ d ) =(∆ x ) + (∆ y ) ∼ χ p =2 under normality and indepen-dence assumptions for ∆ x and ∆ y . We then fit a χ ( µ , σ ) distribution for (∆ d ) in each field with thecenter µ and standard deviation σ indicating the over-all bias and degree of deviation. Figure 16 displaysthe distributions of (∆ d ) and their fits (red line) outof the 10 control samples for the A2390C field as anexample. µ are close to 0, which further proves thesmall systematic bias, and σ are at least two timessmaller than the adopted threshold. The contamina-tion level can also be revealed by the fraction of mea-surements of offset below the threshold (black dashedline), f , where on average ∼
95% of measurements fallwithin the threshold. The average f is 93%/90%/93%for A2390E/A2390W/A2465C. . . . . . . . . . . . Centroid Offset ∆ d m [kpc] N u m b e r o f E L G A2390 MorphA2390 LW . . . . . . . . . . . Centroid Offset ∆ d m [kpc] A2465 MorphA2465 LW
Figure 17.
Distribution of morphological centroid offset∆ d m between the emission and the stellar continuum forELGs in A2390 (combining three fields, left) and A2465C(right). The distributions of ∆ d from light-weighted cen-troids are overplotted as green outlines. The 1kpc thresholdused is marked as the black dashed line in each panel. Diffence Angle θ d,m N u m b e r o f E L G A2390
MorphLW
Diffence Angle θ d,m A2465
MorphLW
Figure 18.
Histogram of difference angle measured frommorphological centroids ( θ d,m ) for cluster member ELGs inA2390 fields (left) and A2465C (right) with ∆ d m > θ d fromlight-weighted centroids are overplotted as green outlines. In summary, we conclude that although our centroidmeasurements are seeing-limited, the systematic biasand random uncertainties in the process of measurementare small and do not impact our conclusions. C. RESULTS USING MORPHOLOGICALCENTROIDSIn the main text, the emission and continuum cen-troids are measured in a light-weighted procedure. Al-ternatively, we can measure the morphological centroidsof the output segmentation of emission and continuumby simply giving equal weights to all the pixels withinthe border:(¯ x m , ¯ y m ) = (cid:32) N (cid:88) i x i , N (cid:88) i y i (cid:33) , (C6)where N is the number of summed pixels. This is moti-vated by attempting to give higher weights to the weakerbut more spatially distorted ionized gas, e.g. gas tails ITELLE Cluster ELGs σ x m = σ y m = N asan empirical error. The propagation to ∆ d m and θ d,m isthe same as in Appendix B.1. Because the morphologi-cal centroids are sensitive to the way how segmentationis performed and they do not have well-defined errors,our preference is to use the light-weighted centroids forour analysis; however, the morphological centroids pro-duce essentially the same conclusions.We can measure the difference vector d using the mor-phological centroids, and accordingly measure the cen-troid offset ∆ d m and difference angle θ d,m in the samemanner as in Section 4.1. We show the distributionsof ∆ d m in Figure 17. The distributions of centroid off-sets are similar but slightly more skewed to larger valuesin both clusters. This is as expected as morphologicalcentroids put more stress on the faint outskirts.Another selected sample is generated following thesame criteria in Section 4.1 except for requiring ∆ d m > σ ∆ d m , with 49 ELGs obtained for A2390 and 28 ELGsobtained for A2465. The distributions of θ d,m are shownin Figure 18. Similar to the ones in Figure 9, they alsovisually deviated strongly from a uniform distribution.The difference lies in that the major peak being shiftedto a higher value (30 ◦ − ◦ ) in A2390, which can be in-ferred as a result of more influence from the tangentialcomponent of the ram pressure. Furthermore, the minorpeak toward 180 ◦ in A2465 is no longer significant. TheK-S test on A2465 rejects the uniformity at 1% confi-dence level ( p = 0 . θ d,m in A2390 ( p = 0 ..