A Virgo Environmental Survey Tracing Ionised Gas Emission (VESTIGE).V. Properties of the ionised gas filament of M87

We have observed the giant elliptical galaxy M87 during the Virgo Environmental Survey Tracing Galaxy Evolution (VESTIGE), a blind narrow-band Halpha+[NII] imaging survey of the Virgo cluster carried out with MegaCam at the Canada French Hawaii Telescope (CFHT). The deep narrow-band image confirmed the presence of a filament of ionised gas extending up to ~ 3 kpc in the north-western direction and ~ 8 kpc to the south-east, with a couple of plumes of ionised gas, the weakest of which, at ~ 18 kpc from the nucleus, was previously unknown. The analysis of deep optical images taken from the NGVS survey confirms that this gas filament is associated with dust seen in absorption which is now detected up to ~ 2.4 kpc from the nucleus. We have also analysed the physical and kinematical properties of the ionised gas filament using deep IFU MUSE data covering the central 4.8 x 4.8 kpc^2 of the galaxy. The spectroscopic data confirms a perturbed kinematics of the ionised gas, with differences in velocity of ~ 700-800 km s^-1 on scales of<1 kpc. The analysis of 2D diagnostic diagrams and the observed relationship between the shock-sensitive [OI]/Halpha line ratio and the velocity dispersion of the gas suggest that the gas is shock-ionised.


Introduction
Giant ellipticals in the centre of rich clusters are among the most massive and luminous objects in the local Universe. They share with the cluster several properties such as a diffuse X-ray and stellar emission and a similar rest-frame velocity (e.g. Sarazin 1986). They are often used to trace the properties of the cluster dark matter halo in the study of the galaxy-halo connection (galaxy conformity), making them unique systems among all galaxy populations. Located in the deep of the potential well of a large dynamic structure, they have been formed through the accretion of baryonic matter in its different forms, from the diffuse intracluster medium through cooling flows (Cowie & Binney 1977, Fabian & Nulsen 1977, Fabian 1994, to the cannibalism of other galaxies orbiting within the cluster (Ostriker & Tremaine 1975. White 1976, Malumuth & Richstone 1984, Merritt 1985, Byrd & Valtonen 1990, and through multiple merging events that occurred during the formation of the cluster itself (e.g. De Lucia & Blaizot 2007, Boselli et al. 2014a).
Among these objects, the elliptical galaxy M87 (NGC 4486, 3C274, Virgo A) in the centre of Virgo is by far the most studied massive galaxy in the centre of a cluster, but also one of the most studied galaxies ever. At a distance of only 16.5 Mpc (Mei et al. 2007), M87 can be fully resolved at almost all frequencies, from the X-rays to the radio centimetric. At this distance, one arcsec corresponds to 80 pc, a small size compared to the optical extension of the galaxy which has an isophotal radius of 25 kpc (Cortese et al. 2012). This galaxy is a powerful radio source (Baade & Minkowski 1954) characterised by a prominent jet and two extended radio lobes (e.g. Hines et al. 1989, Owen et al. 2000. It also has a characteristic X-ray emission showing long filaments of hot gas extending up to 22 kpc to the east and 28 kpc to the south embedded in a diffuse halo (e.g. Young et al. 2002, Forman et al. 2007, Churazov et al. 2008. Similar filamentary structures have been observed also in narrow-band Hα imaging (Arp 1967, Ford & Butcher 1979, Baum et al. 1988, Heckman et al. 1989, Sparks et al. 1993, Gavazzi et al. 2000, in different UV lines with HST (Sparks et al. 2009, Anderson & Sunayaev 2018) and in the [CII]λ158 µm line with Herschel (Werner et al. 2013), indicating the multitemperature nature of the gas. In particular, the narrow-band Hα imaging data revealed a very complex filament extending to the south-east up to 8 kpc from the nucleus, with a few patchy regions, whose nature and origin are still not fully understood.
The Hα emission line traces the distribution of the ionised hydrogen with a typical temperature of 10 4 K (Osterbrock & Ferland 2006). For this reason this line has been often used to look for cooling flows in the centre of rich clusters of galaxies (e.g. Conselice et al. 2001). In spiral galaxies, where the gas is ionised by the young stellar populations, the Hα line is an excellent tracer of the recent star formation activity (Kennicutt 1998, Boselli et al. 2009). In early-type galaxies the gas can be ionised by a residual star formation activity (Kaviraj et al. 2007, Gavazzi et al. 2018 or by a low level ionisation due to hot, evolved (postasymptotic giant branch) stars (Gomes et al. 2016, Belfiore et al. 2016, the same stellar population responsible for the UV-upturn observed in the far ultraviolet (O'Connell 1999, Boselli et al. 2005. In M87 the gas can be also shock-excited by the central AGN or by the radio jet and counter-jet in the expanding lobes (Dopita & Sutherland 1995, Dopita et al. 1997, or heated by thermal conduction (Sparks et al. 2004;McDonald et al. 2010). The gas filament could be hot gas accreted from the intracluster medium cooling to T 10 4 K or cold gas stripped from a gasrich cluster galaxy after a gravitational interaction or lost during a recent ram pressure stripping event while crossing the halo of M87 (Sparks et al. 1993, Weil et al. 1997, Mayer et al. 2006. We have recently observed M87 as part of VESTIGE (A Virgo Environmental Survey Tracing Ionised Gas Emission), a CFHT large programme designed to make a blind deep narrowband Hα imaging survey of the whole Virgo cluster up to its virial radius (104 deg 2 , Boselli et al. 2018a). The core of the cluster, including M87 and its surrounding regions, has been mapped during the 2017A observing campaign. Thanks to a tuned observing strategy and a specific data reduction technique optimised to detect extended, low surface brightness features in the continuum-subtracted narrow-band images, the VESTIGE data overcome in terms of sensitivity and angular resolution those available in the literature, passing from a limiting surface brightness of Σ(Hα + [NII]) 10 −16 erg s −1 cm −2 arcsec −2 to Σ(Hα + [NII]) 5 × 10 −18 erg s −1 cm −2 arcsec −2 . The success of this original observing and data reduction strategy has been proven by the detection of extended tails of ionised gas in several Virgo spirals (Boselli et al. 2016, 2018a,b, Fossati et al. 2018. In this paper we present these new narrow-band imaging data obtained during the VESTIGE survey. We also present new photometric and spectroscopic data obtained at other frequencies useful for the study of the ionised gas filament. These include a deep spectroscopic IFU MUSE field in the central 1 × 1 of the galaxy taken as part of the first instrument verification run (Emsellem et al. 2014), GALEX UV from the GUViCS survey (Boselli et al. 2011), and optical images from the NGVS survey (Ferrarese et al. 2012). We then compare this new dataset with data at other frequencies available in the literature to have a complete view of the different phases of the interstellar medium (ISM) along the filament and in the surrounding intracluster medium (ICM). Since the properties of the multitemperature gas of M87 have been already studied in detail in the past, we refer the reader to these publications (Sparks et al. 1993(Sparks et al. , 2004Churazov et al. 2001Churazov et al. , 2008Young et al. 2002;Forman et al. 2007Forman et al. , 2017Werner et al. 2013;Simionescu et al. 2018). Here we limit our analysis to the novelties brought by our new dataset. The paper is structured as follows: the VESTIGE and MUSE observations and data reduction are presented in Sect. 2, along with the multifrequency data available in the literature. In Sect. 3 we analyse the new dataset through a comparative multifrequency analysis. Discussion and conclusions are given in Sects. 4 and 5. Consistently with the other VESTIGE publications, we assume a distance of 16.5 Mpc for M87 and the Virgo cluster (Gavazzi et al. 1999, Mei et al. 2007).

Observations and data reduction
2.1. Narrow-band imaging M87 and its surrounding regions have been mapped during the first observing semester of the VESTIGE project (2017A; Boselli et al. 2018a for details). The observations have been taken using MegaCam at the CFHT through the narrow-band filter MP9603 (λ c = 6591 Å; ∆λ = 106 Å) which includes at the redshift of M87 (vel = 1292 km s −1 , GOLDMine) the Hα line (λ = 6563 Å) and the two [NII] lines (λ =6548, 6583 Å), and through the r-band filter for the stellar continuum subtraction. MegaCam is composed of 40 2048×4096 pixels CCDs with a pixel size of 0.187 on the sky. The galaxy has been covered with 12 independent frames with a large dithering selected for the whole VESTIGE blind survey (15 arcmin in R.A., 20 arcmin in dec.). Each single exposure was of 600 sec in the narrow-band Hα filter and 60 sec in the broad-band r filter. The observations have been carried out in good seeing conditions, with a typical seeing of 0.90 and 0.83 arcsec in the narrow-band and broadband stacked frames, respectively.
The data have been reduced following the VESTIGE standard procedures presented in Boselli et al. (2018a) and Fossati et al. (in prep.). This consists in using Elixir-LSB (Ferrarese et al. 2012), a data reduction pipeline expressly designed to detect extended low surface brightness features associated with perturbed galaxies through an accurate determination and subtraction of any extended feature in the sky background.
The photometric calibration of the r-band frames has been derived following the standard MegaCam calibration procedures (Gwyn 2008), as extensively described in Boselli et al. (2018a). The calibration in the narrow-band filter has been done as described in Fossati et al. (in prep.). The typical uncertainty of VESTIGE in both bands is of 0.02-0.03 mag.
The subtraction of the stellar continuum for the determination of the image of the emitting gas is critical in early-type galaxies such as M87 which are characterised by a strong stellar emission in their centre, where the ionised gas is expected to be detected.
Given the wide band of the r filter (λ c = 6404 Å; ∆λ = 1480 Å) and the slight difference in the peak transmissivity of the two bands, the derivation of the stellar continuum in the narrow-band from the r-band depends on the spectral properties of the emitting source (Spector et al. 2012), thus on the colour that in earlytype galaxies is known to change radially from the core to the periphery (Boroson et al. 1983, Franx & Illingworth 1990, Suh et al. 2010, Roediger et al. 2011. As extensively described in Fossati et al. (in prep.), we have used several hundred thousands of unsaturated stars detected in the Hα narrow-band and in the r-and g-bands in the science frames to calibrate an empirical relation between the colour of the stars (expressed in AB magnitudes) and the normalisation factor: We applied this normalisation pixel by pixel on the stacked frame before the subtraction of the stellar continuum. The g − r colour map of the galaxy has been derived using the g-band frame taken with MegaCam during the NGVS survey (Ferrarese et al. 2012). To avoid the introduction of any extra noise in the sky regions, where there is no stellar continuum, this colourdependent normalisation is applied only whenever the signal in the r-and g-bands has a signal-to-noise S /N > 1. The continuum-subtracted image is then multiplied by the filter width (106 Å) to obtain the values of the line flux per pixel. The resulting Hα+[NII] stellar continuum subtracted image of M87 is shown in Fig. 1. The quality of this image is excellent: indeed, there are no extended dark regions often observed whenever the stellar continuum is over subtracted.

MUSE spectroscopy
M87 has been observed in 2014 with MUSE during the first instrument science verification run at the VLT. Details of the observations and data reduction are given in Emsellem et al. (2014) and Sarzi et al. (2018). MUSE provides a spectroscopic data cube on a rectangular 1 ×1 field with spaxels of 0.2 ×0.2 . The spectra cover the 4800-9000 Å range, with an instrumental resolution of ∼ 60 km s −1 at 5500 Å. Two 1800 sec exposures have been taken on the centre of the galaxy. Each exposure was followed by a 900 sec exposure in the outskirts of the galaxy for an accurate determination of the sky background. The data have been reduced using the v1.6 MUSE standard pipeline, as described in Sarzi et al. (2018). To subtract the stellar continuum emission and remove any possible contamination due to the underlying stellar absorption on the main Balmer lines we fitted the spectra using pPXF and GANDALF (Cappellari & Emsellem 2004;Sarzi et al. 2006). This code simultaneously models the stellar continuum and the emission lines in individual spaxels. The stellar continuum is modelled with a superposition of stellar templates convolved by the stellar line-of-sight velocity distribution, while the gas emission and kinematics are derived assuming a Gaussian line profile. Stellar templates are constructed using the MILES library (Vazdekis et al. 2010 Fossati et al. (2016). Groups of lines are fitted simultaneously using a combination of 1D Gaussian functions with fixed relative velocities. The noise is measured from the 'stat' data cube and renormalised on the line fluxes to take into account the correlated noise introduced by resampling and smoothing, as extensively described in Fossati et al. (2016). Figure 2 shows the Hα emission line map and its signal-to-noise map derived for a S /N>5.

UV
UV data can be used to identify the nature of the ionising source of the gas filaments observed in Fig. 1. Several GALEX exposures, available in both the FUV (λ c = 1539 Å) and NUV (λ c = 2316 Å) bands, have been collected during the GUViCS survey of the cluster (Boselli et al. 2011). We have combined the available frames with a sufficient signal to noise and obtained deep UV images of M87 with a typical integration time of 3278 sec in the FUV and 16267 sec in the NUV, and used them to create the FUV/NUV colour image. M87 has a diffuse light distribution in the UV, witnessing the presence of an evolved stellar population. The only evident structure in the image is the nucleus and the jet, which is present in both the FUV and NUV bands. The UV colour image does not show any clumpy structure which can be associated with a young stellar population. It is thus likely that, as in other massive early-type galaxies, the UV emission of M87 is dominated by evolved stellar populations (UV upturn;O'Connell 1999;Boselli et al. 2005Boselli et al. , 2014a.

Visible
Optical data are essential to see the distribution of dust seen in absorption. For M87 this is critical since in this object dust outside the nuclear region has never been detected in emission even in the most recent Spitzer (Perlman et al. 2007, Bendo et al. 2012, Ciesla et al. 2014 and Herschel (Ciesla et al. 2012, Cortese et al. 2014) observations 1 . M87 has been observed with HST during the ACS Virgo Cluster Survey (VCS) in the F475W and F850LP bands (∼ Sloan g and z) ) and in the F275W, F606W, and F814W bands by Bellini et al. (2015). The g − z colour image of the central ∼ 1 ×1 given in Ferrarese et al. (2006, their    The availability of deep and extended surveys targeting M87, allows us to investigate whether dust is also present at larger distances from the galaxy's center. To do so we used the g-band imaging gathered by the NGVS, that in the galaxy central regions reaches optimal seeing conditions, 0.62 , and a depth of g = 27.3 mag or µ g = 29.0 mag arcsec −2 . The continuum emission from the galaxy was retrieved using the IRAF tasks ISOFIT and C-MODEL (Ciambur et al. 2015) that allow for an accurate recovery of the galaxy surface brightness profile, hence of a two-dimensional model of the galaxy light. The resulting stellar continuum subtracted image of the inner 2 × 2 is shown in Fig.  3. Despite the poor image quality in the central 0.05 ×0.05 , affected by the presence of the AGN (see Bellini et al. 2015 for the detection of dust in this region), Fig. 3 clearly shows the presence of dust in absorption. This is detected along the filament located to the north of the jet and in several other patchy features close to the centre as already identified by previous analyses (Sparks et al. 1993, Ferrarese et al. 2006, Bellini et al. 2015. However, the high quality NGVS imaging also reveals a complicated network of dusty regions unknown so far. These are distributed around the galaxy in the western direction up to ∼2.4 kpc (0.5 ).
At its highest densities the most prominent dust feature (red contours in Fig. 3) absorbs the continuum light with an extinction factor A g = 0.010 ± 0.002 − 0.017 ± 0.002 mag. Such a value was estimated by comparing the NGVS g-band image with a dust-free model of the light distribution, the latter given by the ISOFIT two-dimensional map of the galaxy light. We note that, as consequence of the relatively small covering area of the dust, the ISOFIT model is a good approximation of the unextinguished light because the result of an azimuthal average. We also remark that this extinction factor perfectly agrees with the value obtained by Sparks et al. (1993). As in this work, assuming a standard gas-to-dust ratio d 100 and the relation: we can calculate the expected column density of the cold gas associated with the dust lane, N(H) 1.8 × 10 19 cm −2 . This number is consistent with the upper limit in the column density derived from observations of HI in absorption (van Gorkom et al. 1989, Dwarakanath et al. 1994 or from other absorption line measurements and from the shape of the X-rays spectrum in the nucleus (Sabra et al. 2003). Given the size of the filament, we estimate as in Sparks et al. (1993) that the associated total mass of cold gas is M gas,dust 2 × 10 7 M .

Narrow-band imaging
The continuum-subtracted narrow-band image of M87 shows prominent ionised gas filaments extending from the nucleus along an axis with a position angle of ∼ 55 deg measured clockwise from the north. Most of the extended emission is on the south-eastern side of the galaxy, where the emitting gas is detected up to ∼ 8 kpc from the nucleus, while only ∼ 3 kpc in the north-western direction.  Fig. 9), its total Hα luminosity is L(Hα) = 9.56 × 10 39 erg s −1 . As in Fossati et al. (2016) and Boselli et al. (2016) we can estimate the mean density and the total mass of the gas emitting in Hα along the filament using the relation: The dark regions outlined by green contours are dusty structures detected in absorption. The red contour corresponds to A g = 0.010± 0.002. The bright elongated region is the jet and the point sources globular clusters and foreground stars. (Osterbrock & Ferland 2006), where n e and n p are the number density of electrons and protons, α e f f Hα is the Hα effective recombination coefficient (α e f f Hα = 1.17 ×10 −13 ), V is the volume of the emitting region, f the volume filling factor, h the Planck's constant, and ν Hα the frequency of the Hα transition. Assuming a cylindrical distribution of the gas, with an orientation of 45 deg along the line of sight, and a filling factor f = 0.1, we estimate that the typical density of the gas is n e 0.3 cm −3 and its total mass M(ionised) f ilament 6.9 × 10 7 M . We stress, however, that this estimate, which is a factor of 10 larger than the mass derived by Sparks et al. (1993), is a very rough estimate given the very high uncertainty on the filling factor of the gas and on the volume of the filament.
The plume of ionised gas at ∼ 15 kpc to the east discovered by Gavazzi et al. (2000) is also evident given its high surface brightness (Σ(Hα+[NII]) 10 −16 erg s −1 cm −2 arcsec −2 , see Fig.  4). Figure 4 also shows a previously-unknown banana-shaped, low surface brightness feature (Σ(Hα + [NII]) = 5×10 −18 erg s −1 cm −2 arcsec −2 ) 3 kpc long at ∼ 3 kpc east from this plume (18 kpc from the galaxy nucleus). Figure 5 shows the velocity of the gas along the filament within the central 1 ×1 region. Thanks to the IFU mode of MUSE, the improvement in quality with respect to previous long slit data is astonishing (Heckman et al. 1989, Sparks et al. 1993, Werner et al. 2013. Figure 6 shows the map of the velocity dispersion of the gas. The gas does not follow any ordered motion. As already noticed by Sparks et al. (1993), the difference in velocity within the inner gas filament is very high, reaching 700-800 km s −1 on scales 1 kpc. On the contrary, the velocity dispersion is 100 km s −1 and is fairly uniform over the gas filament, with the only  Fig. 4. Zoom on the plume of ionised gas discovered by Gavazzi et al. (2000) (west), and on a new filament of ionised gas 3 kpc long (east). This zoomed region is indicated with a dashed green box in Fig. 1.

Gas kinematics
exception on the nucleus where it reaches 450 km s −1 because of the presence of an AGN. These velocity dispersions are generally larger than those measured within the tails of ionised gas observed in other perturbed cluster galaxies where these values are reached only within limited regions well outside the optical disc (Fumagalli et al. 2014, Consolandi et al. 2017, Poggianti et al. 2017, Bellhouse et al. 2017. For comparison, within the same MUSE field the velocity dispersion of the stellar component goes from 370 km s −1 in the central 2 arcsec (160 pc) to 260 km s −1 at ∼ 1.6 kpc, whereas the velocity rotation is always 10 km s −1 (Emsellem et al. 2014). The steep gradient in the velocity field observed in the regions at (-1.4; 0.0) kpc from the nucleus (Fig. 5, left panel), which is also characterised by a high velocity dispersion (Fig. 6, left panel) (and a high uncertainty on both the rotational velocity and the velocity dispersion, Figs. 5 and 6, right panels) suggests the presence of two physically and kinematically distinct components in the ionised gas. Cold gas has been detected in the bright ionised gas region at the southeast edge of the counter-jet radio lobe at (-2.2; -1.5) kpc from the nucleus through the emission of the 12 CO(2-1) (Simonescu et al. 2018) and [CII] (Werner et al. 2013) lines. Their recessional velocity is consistent within the errors. Sparks et al. (1993), using geometrical arguments based on the presence of a dust lane absorbing the stellar emission, concluded that the associated gas filament in the western direction is in the foreground, and thus it is flowing out from the nucleus. Using similar arguments based on the relative position of the ionised gas filament and of the radio lobe, combined with the lack of any dust component, they also concluded that the southeastern gas filament is in the background of the galaxy.

Emission line properties
Several gas emission lines in the MUSE spectral range (4800-9000 Å) can be used to derive the physical properties of the gas through diagnostic diagrams. The Hα/Hβ Balmer decrement is generally used to derive the dust attenuation within the ionised gas. The Balmer decrement within the MUSE field is shown in Fig. 7. The Hα/Hβ ratio is fairly constant along the filament, with values close to Hα/Hβ 2.8-3.5. These values are close to those expected whenever the gas is photoionised by the emission of young and massive stars and the dust content is low (E(B − V) 0.08). Similar values, however, are also expected whenever the gas is collisionally-heated by cosmic rays accelerated by magnetohydrodynamic waves in the filament (Hα/Hβ 3.7-4.4; Ferland et al. 2009). The comparison of Fig. 7 and Fig. 3 gives a consistent picture where dust attenuation is at the origin of the observed Balmer decrement.
The [SII]λ6716Å/[SII]λ6731Å, on the other hand, is tightly connected to the density of the gas (Osterbrock & Ferland 1989). Figure 8 shows that, within the inner 1 ×1 region of M87 mapped by MUSE the sulphur doublet ratio smoothly increases from [SII]λ6716Å/[SII]λ6731Å 0.9 (n e 6 × 10 2 cm −3 assuming a temperature of 10 4 K and using the recent calibration of Proxauf et al. 2014) close to the nucleus to [SII]λ6716Å/[SII]λ6731Å 1.35 (n e 80 cm −3 ) at the edges of the field 2 . These low densities are similar to those encountered in other tails of ionised gas in perturbed cluster galaxies (e.g. Poggianti et al. 2017). We notice that a large [SII]λ6716Å/[SII]λ6731Å≥ 1.4 ratio has been also measured by Gavazzi et al. (2000) in the plume of ionised gas at ∼ 15 kpc to the east.
Unfortunately the high noise induced by the bright stellar continuum in these inner regions prevents the detection of the weak [NII] line at λ = 5755 Å, we are thus not able to estimate the variation of the gas temperature along the filament using the [NII]λ6548 + [NII]λ6583Å/[NII]λ5755Å line ratio (Osterbrock & Ferland 2006). The detection of the [CIV] line at λ 1549 Å by Sparks et al. (2012) in the ionised gas, however, suggests that the gas might reach temperatures of T = 10 5 K. At this temperature, the iron lines [FeXIV] at λ 5303 Å and [FeX] at λ 6374 Å should also be present (Heckman et al. 1989). As explained above, however, the detection of these weak emission lines is hampered by the bright stellar continuum. Figure 9 shows the map of different line ratios within the MUSE field, including the shock-sensitive [OI]λ6300Å/Hα, [SII]λ6716,6731Å/Hα and [NII]λ6548,6583Å/Hα line ratios (Rich et al. 2011). Figure 9 shows high values for all the shocksensitive tracers, both close to the AGN or along the filament, suggesting that within these regions the gas is mainly ionised by shocks. For comparison, lower values ([OI]λ6300Å/Hα =0.18; [SII]λ6716,6731Å/Hα=0.53) have been measured in the plume of ionised gas discovered by Gavazzi et al. (2000) at 15 kpc east from the galaxy nucleus. Figure 10 shows three typical line diagnostic diagrams generally referred to as BPT diagrams (Baldwin et al. 1981). Thanks to the IFU mode of MUSE and to its excellent sensitivity, the difference with respect to published BPT diagrams all based on long-slit spectroscopy (Heckman et al. 1989, Crawford et al. 1999, is stunning. The three BPT diagrams consistently indicate that the gas filament is not photoionised by massive young stars, in agreement with the lack of young stellar associations in the UV images of the galaxy. All the points, indeed, are located to the upper right of the typical curves delimiting the position of HII regions from AGN. The handful of dots in the photoionisation regime in the upper and lower panels of Fig. 10 are located along the jet, thus their line ratio estimates is highly uncertain.
To identify the possible ionising source of the gas within the filament we plot in Fig. 11 the relationship between the velocity dispersion of the gas σ and the three different shocksensitive line ratios [NII]λ6584/Hα, [SII]λ6716,6731Å/Hα, and [OI]λ6300Å/Hα. Points are colour-coded according to their distance from the galaxy centre. A strong relationship between the two variables has been interpreted in the literature as a further indication that the gas is excited by shocks (Rich et

Comparison with other bands
The spectacular images at hand allow us to make a detailed comparison between the distribution of the ionised gas emission and the other components of the interstellar medium of the galaxy. Figure 12 shows with a sensitivity and resolution never reached so far (Sparks et al. 1993, Werner et al. 2013) how the hot gas traced by the X-rays emission from Chandra compares with the ionised gas emission. As noticed in previous works (Sparks et al. 1993, Werner et al. 2013, at large scale the main ionised gas filaments are not associated with the hot gas distribution which has a structured shape only in the 0.5-1.0 and 1.0-3.5 keV bands, while at smaller scales the coincidence between well defined features is present (e.g. Sparks et al. 2004). We remark that the plume of ionised gas first detected by Gavazzi et al. (2000) at 15 kpc to the east, and the extended low surface brightness filament 3 kpc to the east of this plume, at 18 kpc from the nucleus, first detected in this work, correspond to sharp edges in the X emission in the 0.5-1.0 and 1.0-3.5 keV bands. Figure 13 shows the comparison between the ionised gas distribution traced by the Hα+[NII] and the distribution of the dust component as derived by the NGVS image (see Sect. 3.3). As already noticed by Sparks et al. (1993), there is a quite clear association between the north-west filament located just on the northern side of the jet and a dust lane. The extremely deep NGVS image allowed us to detect several other dust features previously unknown. Clearly most of them are also associated with prominent ionised gas filaments, as for instance is the structure at 14 arcsec ( 1 kpc) to the north of the nucleus called "bar structure" in Sparks et al. (2004), although dust and ionised gas are here distributed in a different way, the former on a circular loop, the latter along an elongated structure. Dust is also present on the "patch" at 10 arcsec ( 800 pc) south to the nucleus, and in a bright region at 10 arcsec ( 800 pc) to the west. There is also a clear association in the four ionised gas filaments to the north-west, west, and south-west of the nucleus, at roughly 32-to-24 arcsec (2.6-1.9 kpc). Because the surface brightness of the galaxy rapidly decreases outward, we are not able to detect dust in absorption at distances larger than 30-35 arcsec (2.4-2.8 kpc), we thus do not know whether the long south-eastern filament is associated with any dusty feature.  We can also compare the distribution of the ionised gas to that of the cold gas component derived from the 12 CO(2-1) line by Simionescu et al. (2018) using ALMA data and from the [CII] line by Werner et al (2013). Since the gas has been detected only  on a very limited region at the edge of the counter-jet in the south-eastern direction, we show the comparison on a zoomed region (Fig. 14). The cold gas, which has a mass of M(H 2 ) 5 × 10 5 M , is located on a peak of ionised gas emission. We do not see, however, any associated compact source in both the broadand narrow-band images, indicating the lack of any star forming system. Figure 15 compares the ionised gas emission to the radio continuum emission at 115 GHz. As already noticed in previous works, the ionised gas emission is located northern to the promi-nent jet, and avoids the counter-jet lobe in the south-eastern direction. As a final image, we show in Fig. 16 the pseudo-colour image of M87 obtained by combining Chandra 1.0-3.5 keV, Hα+ [NII] and radio continuum at 90 cm frames of the galaxy. This figure allows us to appreciate how the emission at these different wavelengths correlates at large scale. Again we see how Distance to Kauffmann et al. (2003, Top) and Kewley et al. (2001Kewley et al. ( , 2006Center and Bottom) (Kauffmann et al. 2003;Kewley et al. 2001, Kewley et al. 2006). Data are colour coded according to their minimum distance from these curves. The black and gray crosses indicate the typical error on the data for lines with S/N 15 and S/N 5, respectively. The gray contours show the distribution of a random sample of nuclear spectra of SDSS galaxies in the redshift range 0.01 -0.1 and stellar masses 10 9 ≤ M star ≤ 10 11 M . The magenta star shows the position in the diagram of the plume of ionised gas observed at 15 kpc to the east by Gavazzi et al. (2000). . Each point is colour-coded according to its distance from the galaxy centre, as indicated in the right lower panel.
the Hα emission avoids the main radio structures in the inner regions (Sparks et al. 1993), while the ionised gas plume and filaments at 15 and 18 kpc east from the nucleus are located within the radio lobe (Gavazzi et al. 2000).

Discussion
The detailed discussion of the properties of the different gas phases of M87 has been the topic of several publications (e.g. Sparks et al. 1993, Forman et al. 2007, Churazov et al. 2008, 2013, Simionescu et al. 2018. We thus refer the reader to these works. Here we just focus on the main novelties provided by our data, mainly due to their higher X-rays 0.5-1.0 keV ⇡ 5 kpc X-rays 1.0-3.5 keV X-rays 3.5-7.5 keV sensitivity and angular resolution with respect to those gathered in previous works.

The source of ionisation
The exquisite quality of the MUSE data allowed us to make a 2D BPT diagram along the ionised gas filament and compare it to the prediction of the photoionisation models of Kewley et al. (2001). Figure 10 (upper panels) clearly shows that, in the case of photoionisation, the [NII]/Hα line ratio is expected to increase with increasing metallicity, but it never reaches the values observed in the gas filaments of M87. We recall that in Virgo the metallicity of the ICM derived from ASCA is only 0.39-0.55 Z (Matsumoto et al. 2000;Fukazawa et al. 2004). Photoionisation due to young massive stars does not seem to be at play even in the prominent ionised gas region at the south-eastern edge of the counter-jet, where molecular gas has been detected (Fig. 14). Indeed, we do not observe any compact stellar counterpart in image. The molecular gas has been detected at ∼ 40" south-east from the nucleus of M87. The Hα filament is located at the south-east edge of the cavity formed by the counter-jet.
both the UV and optical images, we thus deduce that there is no star formation associated with this molecular clouds. We recall that Simoniescu et al. (2018), using dynamical arguments, concluded that because of its velocity dispersion of 27 km s −1 , the molecular gas is here stable, and it is thus highly unlikely that it can collapse to form new stars. Positive feedback does not seem to be present in this object. This result is consistent with what found by Johnstone et al. (2007) after comparing the predictions of photoionisation models with the optical and infrared line properties of cooling-flow galaxies in the core of clusters. Other sources of ionisation have been proposed, such as the pure collisional heating due to cosmic rays accelerated by magnetohydrodynamic waves present along the filament (Ferland et al. 2009, McDonald et al. 2012, dust scattering (Mattila et al. 2007, Seon & Witt 2012, thermal conduction from the hot phase (Sparks et al. 2004), or ionisation due to the central AGN. The deep optical image shown in Fig. 3 (Fig. 10), while it is probably negligible in the outer filament (McDonald et al. 2012). Our analysis also suggests that the gas might be ionised by shocks (Fig. 11). The distribution of the ionised gas at the edges of the radio jet or of the counter-jet might indicate that the shocks can be produced by the energy injected by the AGN in the ISM. The contribution of shocks to the ionisation of the prominent gas filaments of M87 seems more important than in other central galaxies in cooling flow clusters given its higher [OIII]/Hα line ratio (McDonald et al. 2012).

The origin of the filament
The kinematical properties of the gas (Fig. 5) compared with those recently derived from the planetary nebulae (Longobardi et al. 2015(Longobardi et al. , 2018 and the globular clusters (Romanowsky et al. 2012) in the outer halo, from the stellar kinematics in the inner regions (Emsellem et al. 2014, Arnold et al. 2014, or from the observed substructures in the colour distribution of the globular clusters (Powalka et al. 2018), seem to suggest that the ionised gas filament is the remnant of a gas-rich galaxy which has been recently accreted by M87, as first proposed by Sparks et al. (1993), or has crossed the halo loosing most of its gas reservoir after a ram pressure stripping event (Mayer et al. 2006). The rough estimate of the total mass of the ionised gas in the filament (M f ilament 6.9 × 10 7 M ) is comparable to that of the gas that would have been associated with the observed dusty filaments if this dust had been accreted from an external object (M gas,dust 2 × 10 7 M ). This gas mass, however, should be considered as a lower limit to the total gas mass lost by the star-forming galaxy because dust might have been destroyed by sputtering in the harsh intertsellar radiation field of M87 (Goudfrooij & de Jong 1995, Temi et al. 2007). The kinematical properties of ∼ 300 planetary nebulae in the outer halo and a detailed colour analysis of the galaxy led Longobardi et al. (2015Longobardi et al. ( , 2018 to conclude that M87 has recently ( 1 Gyr) accreted a massive (M star 6 × 10 9 M ) star forming system. The comparison between the distribution of ∼ 500 globular clusters in the phase diagram with tuned simulations suggests that the galaxy has been continuously accreting smaller systems for ∼ 1 Gyr (Romanowsky et al. 2012). Furthermore, M87 has also a kinematically distinct component in the inner ∼ 1.2 kpc radius which might have been formed after a gravitational perturbation (Emsellem et al. 2014). Accretion of satellites is also suggested by the low surface brightness fine structures in the deep optical image of M87 (Mihos et al. 2017). The complex kinematics along the filament rules out a simple scenario where the gas has been lost while the star forming progenitor was spiraling around M87, as observed in the central galaxy of the cluster A2052 (Balmaverde et al. 2018), and get virialised on short timescales as expected in massive objects. Indeed, the energy ejected by the AGN can mix, excite, and accelerate the gas which is conserving its angular momentum while infalling within the potential well of the galaxy, producing the peculiar velocity field shown in Fig. 5. Complex structures in the velocity field of the ionised gas filaments have been observed in other radio galaxies such as NGC 1275 in Perseus (Gendron-Marsolais et al. 2018), where star formation is also present (Canning et al. 2014).
An alternative interpretation is that, as in other central cluster galaxies, the ionised gas filament might be produced by the diffuse hot gas halo which, for local instabilities, cools out to form magnetically supported narrow filaments. The velocity map shown in Fig. 5 reveals generally negative velocities for the gas adjacent to the jet and positive velocities on the other side. This is in general consistent with the scenario that at least some ionised emitting gas interacts with jets and the counter-jet. Indeed, the ionised gas filaments of M87 avoids the radio bubbles (Figs. 15 and 16) as often observed in cooling-flow galaxies (Russell et al. 2017). Jets can trigger buoyant bubbles filled by radio plasma, rising in the X-ray atmosphere of M87 (Churazov et al. 2001). Such kind of rising bubbles can have velocities of 60% -70% of local sound speed, or 300 -470 km/s for kT = 1 -1.8 keV . The ionised gas filaments at the south-east align with the current jet / counter jet, and might have been formed out of thermal instabilities in uplift gas from the central region by radio bubbles (e.g., Churazov et al. 2001;Forman et al. 2007). The ionised gas inside the radio contours in the western direction might thus be at the edges of the radio bubble (projection effects).

Conclusions
The analysis of deep narrow-band Hα+ [NII] images of the elliptical radio galaxy M87 in the centre of the Virgo cluster confirmed the presence of an ionised gas filament extending up to 3 kpc in the north-western and 8 kpc in the south-eastern direction. It also allowed the detection of a low surface brightness plume of ionised gas at 18 kpc to the east of the nucleus, located at the edge of the radio continuum counter-jet lobe and corresponding to a sharpe edge in the 0.5-1.0 keV X-rays emission. The kinematical properties of this gas filament derived using new MUSE IFU data revealed the presence of a very perturbed velocity field, with differences in velocity as high as ∼ 800 km s −1 on scales 1 kpc, probably due to different gas components located along the line of sight. The position of the spaxels along several diagnostic diagrams, and the observed relationship between the shock-sensitive [OI]/Hα line ratio and the velocity dispersion of the gas, consistently suggest that the gas is shock-ionised probably by the expanding bubbles of the rotating radio jet. The mass of the ionised gas in the filament, or that of the cold gas expected to be associated with the dust seen in absorption (both a few 10 7 M ) might suggest that the filament is a remnant of gas coming from a galaxy recently captured by M87. The ionised gas filament, however, could also be gas cooled out from the hot gas halo along magnetically supported narrow filaments.