GHαFaS: Galaxy Hα Fabry-Perot System for the William Herschel Telescope

The internal kinematics of H II regions are of interest because of the light shed on the evolutionary properties of the massive stars that ionize them and on their interaction with the surrounding ISM. A key issue here is that the velocity dispersion measured in Hα of the majority of H II regions observed in external galaxies, which are necessarily selected as the most luminous regions, is found to be supersonic. A basic question is what is the source of input energy to maintain these line widths, and the linked possibility that the line widths could be used as a luminosity index and, hence, as a distance-measuring standard candle. A long-term debate was opened by Terlevich & Melnick (1981), who suggested that the line widths vary as the fourth power of the luminosity and that this reflects virial equilibrium within the regions. This was addressed in some detail by Chu & Kennicutt (1994a) in a study of 30 Doradus, and by Chu & Kennicut (1994b), who found, for a large sample of bright H II regions, that their line profiles are broadened by turbulence as well as wind-induced line splittings; i.e., that, in general, virialization cannot be assumed. These studies are merely pointers to the wealth of physical inference that will be available using GHαFaS to investigate how metals are distributed around galaxies by the outflows from H II regions and by similar flows around galactic centres, and how the ionization structure of H II regions relates to their kinematic structure.

In the last two decades, the extraordinary variety of morphologies of planetary nebulae (PNe) has been widely recognized (e.g., Balick & Frank 2002). Their study is crucial in order to fully understand mass loss, the physical process that governs the latest evolution of low mass stars and plays a basic role in the chemical enrichment of galaxies. A precise determination of the 6D geometry and kinematics of PNe requires 2D spectroscopy, at a resolution R≥30,000 for "normal" PNe (and lower for the highly collimated, high-velocity components) and in the brightest nebular emission lines ([O III], [Hα], and [N II]). GHαFaS can provide an efficient and accurate way to observe all the different macro and microstructures present in the nebulae, which include—often in the same object—multiple spherical shells, bipolar or multipolar lobes, point-symmetrical features, symmetric knots, filaments, and jets. Comparison of the images and radial velocity fields with sophisticated radiative-hydrodynamical simulations (e.g., García-Segura et al. 1999) can then reveal the physical processes at work in shaping PNe. By obtaining fully monochromatic images of the nebulae in different emission lines, we can determine the density, temperature, and ionization degree, throughout the objects. This is also essential for a proper 3D modeling of the nebulae and a correct determination of their physico-chemical properties (cf. Gonçalves et al. 2006). Ultimately, the most important questions to be answered are the following ones: Why do stars loose their spherical symmetry at the tip of the asymptotic giant branch or immediately after? Can single stars do it, or is the majority of the observed PNe the result of binary evolution (e.g., Moe & de Marco 2006)?

The evolution of structures in galaxies is given by the interactions with neighboring galaxies, gravitational perturbations, and/or winds from star-forming regions. The IAC, LAE, and LAM groups have developed a set of robust data analysis tools that were used successfully on previous Fabry-Perot observations covering out to 12 kpc in the nearby grand-design Sc spiral galaxy, M 74. These tools are optimized to investigate the large-scale dynamics as well as feedback from individual H II regions into their surrounding interstellar medium in order to study the evolution of structures in galaxies using two-dimensional kinematical observations (Zurita et al. 2001; Hernandez et al. 2005a, 2005b; Fathi et al. 2005, 2007a, 2007b). After the careful examination of the resonance structures by measuring the pattern speeds of the bars in M 100 (Hernandez et al. 2005b), NGC 1068 (Emsellem et al. 2006), and NGC 6946 (Fathi et al. 2007b), there is now a solid framework for kinematically studying the resonant interaction in disks of galaxies. The combination of these studies with the emerging theoretical foundation (Maciejewski 2006; Maciejewski & Athanassoula 2007; Athanassoula 2005) has opened the possibility of adding the global kinematical information to the local effects at or near the Lindblad and ultraharmonic resonance radii, and compare the strength and degree of symmetry of the density waves. Such quantification is important to study to what degree bars infer dynamical mixing of the gas and stars in their parent disks.

Due to the mass concentration toward the center of the gravitational potential in galaxies, their nuclear region are suitable environments for the formation of supermassive black holes (SMBHs), which can become active by infall of interstellar gas. Recent works (Tremaine et al. 2002 and references therein) show, in addition, that the mass of the SMBH is proportional to that of the bulge of its host galaxy. This relation suggests a connection between the large-scale properties of galaxies and the nuclear properties (Haehnelt & Kauffmann 2000). The process responsible for such a relation has to transfer matter and angular momentum within galaxies. Pioneer works have proposed that bars can trigger the mass flows to help grow and feed SMBHs in galaxies (e.g., Shlosman et al. 1989), but observations have not confirmed this. Thus, measuring the velocity at which gas is falling from the outer parts of a galaxy toward the center is imperative for constraining the links between large-scale properties and activity in the nucleus of galaxies. The dynamical analysis methods developed in Fathi et al. (2005, 2006) have shown to be useful for quantifying the kinematical effects of gravitational perturbations due to bars and spiral arms. Applying these methods to high-spatial-resolution data for a large number of objects, delivered by GHαFaS, will provide necessary details required for linking the kiloparsec-scale kinematics with the parsec-scale processes.

The large size and homogeneity of surveys of local galaxies (SDSS, 2dF, 2MASS) have yielded extraordinary advances in our understanding of galaxy structure and evolution. One of the most important realizations of galaxy formation studies is the recent detection from SDSS and 2dF of a robust bimodality in the distribution of galaxy properties, with a characteristic transition scale at stellar mass ∼3 × 10¹⁰ M_⊙, corresponding to a halo (virial) circular velocity V ∼ 120 km s^-1 (Kauffmann et al. 2003; Dekel & Birnboim 2006). However, large surveys like the SDSS lack measurements of the dynamical mass, because it must include coverage to luminosities below the fundamental galaxy transition scale at V ∼ 120 km s^-1. Central to such a study, the number of galaxies per unit volume as a function of halo circular velocity—the circular velocity function, N(V) —is a robust prediction of cosmological models that has never been directly tested (Klypin et al. 1999; Newman & Davis 2002). The Virgo cluster is an ideal laboratory for analyzing, in depth, these processes using GHαFaS observations.

In studying high-redshift galaxies, it is necessary to disentangle distance effects from evolution ones (e.g., Förster Schreiber et al. 2006). Even at low redshifts, controversy may exist on the nature and on the history of merging, interacting (including compact groups [CGs] of galaxies) or even blue compact galaxies that are suspected of constituting a large fraction of the primordial building blocks leading to the present-day galaxies (e.g., Kunth & Östlin 2000; Amram et al. 2004, 2007). In compact groups, high spatial and spectral resolution data from GHαFaS allow observation that the broader Hα profiles are located in the overlapping regions between the galaxies within a group. At higher redshift such a broadening would be interpreted as an indicator of rotating disk, and this system could have been catalogued like a rotator instead of a merger (Flores et al. 2006). On the other hand, in cases when the spatial resolution increases, it becomes possible to address the problem of the shape of the inner density profile in spirals (CORE vs CUSP controversy), which remains one of the five main further challenges to the Λ CDM theory proposed by Singh (2006).

The lack of resolution induce series of biases: (i) the size of the galaxy is artificially enlarged but, in the same time, the size of the galaxy is diminished due to flux limitations; (ii) the inclination is lowered when the spatial resolution decreases; (iii) the velocity gradient is lowered along the major axis while, in the same time, the velocity gradient is enlarged toward the minor axis. The consequence is that there is often no indication that the maximum velocity of the rotation curve is reached, leading to uncertainties in the TF relation determination. In addition, the (fine) structures within the galaxies (bars, rings, spiral arms, bubbles, etc.) are attenuated or erased and the determination of the other kinematical parameters (position angle, center, inclination) becomes highly uncertain.

The Nasmyth focii of the WHT offer two optical modes: one with an optical derotator (in the UV/optical) provides a FOV of 2.5^' × 2.5^' with a throughput of 75%, while the other, without derotator, gives access to a FOV of 5^' × 5^'' with no optical throughput degradation. In the design of the optical system, two cases—one with derotator and one without—were studied. Taking into account the numerous scientific cases exposed above, we opted to keep the FOV (5^' × 5^') as large a possible and decided not to use the optical derotator, which implied that a post observation alignment of the data had to be developed (see § 4.1).

Optical calculations and optimization were done using ZEMAX. Table 2 presents the detailed optical prescription. The optical prescription for GHαFaS is beginning at surface numbered 9. Before the ninth surface, the prescription is the one the WHT for the nasmyth focus. The focal reducer stands on the optical table of the GHRIL platform at the second Nasmyth focus of the WHT. It has a total length of 770 mm (distance from the field lens to the detector plane). The length between the focal plane and the field lens is 50 mm and between the field lens and the collimator lens 345 mm. The total free space of the collimated beam between the collimator lens and the first meniscus of the objective is 200 mm. In the collimated beam the pupil has a diameter of 40 mm (angles are less than 5° on the pupil) in order to use etalon with an aperture diameter greater than 45 mm and avoid any vignetting (see Fig. 1).

Zoom In Zoom Out Reset image size

Globally, as GHαFaS is scanning small wavelength ranges (typically the FWHM of a filter is 15 Å), it should be considered as an achromatic system. The optical calculations, design, and manufacturing were relatively simple. This configuration allowed us to use 3 commercial lenses while two custom lenses had to be fabricated by BMV Optical Technologies ¹¹. The optical performances are shown in Table 1 and the spot diagram is given in Figure 2.

Zoom In Zoom Out Reset image size

From left to right, the optical beam will meet (a) the focal plane, where a mask has been put to avoid any diffuse light, (b) the field lens (surface 10 in the optical prescription in Table 2), (c) the collimator lens (surface 12), (d) a filter in the filter wheel, (e) the Fabry-Perot etalon in the pupil, (f) the first objective lens (surface 20), (g) the second objective lens (surface 23), (h) the last objective lens (surface 25), to be, finally, focused on the detector. Perpendicular to the optical axis, in the plane of the optical table, between the field lens and the collimator, the calibration unit (replication of the focal plane of the telescope) is founded. The calibration unit is in place when the calibration mirror is in the optical axis.

Finally, Table 3 presents the available Fabry-Perot etalons showing the ability of GHαFaS to reach high resolution by simply changing the etalon, an operation that can be done in less than 2 minutes.

The spot diagram (Fig. 2) shows the optical performances of the instrument. It clearly stresses the size of the optical spot on the detector in various positions (conjugated of various positions on the sky). It is given for only a wavelength (6560 Å) as observations are typically done in Hα mode. The spot size is sampled by 1 pixel in the center and in almost the detector, whereas 2 pixel are needed to have 50% of the encircled energy on the edges of the detector. The instrument is able to well sample the seeing (typically 0.6^'' pixel^-1 to 0.8^'' pixel^-1) with 2 pixels.

The mechanical design was performed at the Université de Montréal. All parts are made of aluminum. The lens' cylinders were painted inside to avoid any parasite reflection and included different masks. Lenses have been tested with a Zygo interferometer to check their quality. The optical and mechanical integration were done at Montreal, whereas the IPCS–Focal reducer integration was done at Marseille. Figure  3 presents a top view and a cut along the optical axis parallel to the optical breadboard of the instrument on the optical table of the GHRIL.

Zoom In Zoom Out Reset image size

Finally, the calibration arm, with its mirror-controlled arm, has a neon lamp for calibration purpose. The Ne [6598] emission is used to calibrate the phase shifts of the Fabry-Perot etalon. This lamp will be replaced, at a later stage, by a Photon etc Tunable Source (see http://www.photonetc.com), which is able to provide any calibration wavelength.

GHαFaS is fully automated and controlled via ethernet modules. Table 4 presents a summary of all automated items in their operation modes. The filter wheel, the shutter, and the calibration arms have their own ethernet modules designed by J-L. Gach and P. Ballard from LAM. The IPCS has its own dedicated link and network to avoid a mixing of the information between ethernet modules and data. The etalon is controlled via a RS-232 link with its CS100 controller.

3.2.1. IPCS

The IPCS is composed of a photocathode placed in front of a commercial CCD, which amplifies the photons (from a factor 10⁵ to 10⁷; see Gach et al. 2002; Hernandez et al. 2003 and references therein for a complete description and history of such devices). It provides a system where the detection of a photon is around 10⁶ times above the readout noise of the commercial CCD. Therefore, the IPCS has essentially no readout noise.

The photocathode in the WHT camera is a GaAS type HAMAMATSU V7090-61¹² with a quantum efficiency of ≲26% (this value includes the loss due to the MCP stack). The spectral range is from 400 to 850 nm and is very flat over these wavelengths. It uses two microchannel plates to provide an electron amplification (after the photoelectic effect that converts photons to electrons) of 10⁷ (see Fig. 4). Each channel is 6 μm wide.

Zoom In Zoom Out Reset image size

Finally, the electrons are projected onto a P-43 phosphor screen via a bundle of optical fibers and then is imaged on the focal plane of a DALSA fast readout commercial CCD (Panthera 1M60) using two objectives (from LINOS technology). Data are then transferred to an acquisition computer using the "camera link" cable via an optical hub and to a dedicated ethernet link at 1 Gbits s^-1 (Balard et al. 2006).

3.2.2. Cooling System: The PELTIER Effect Simplified

The readout noise affects CCDs at low light levels when very faint flux objects are observed. Classical CCDs reach nowadays high quantum efficiency (up to 95%), and very low readout noise (down to 2–3 electrons and may be 1 in a near future, Gach et al. 2003). With the IPCS, observations of faint flux objects are split into several short ones (in particular, this method also helps reduce pollution by cosmic rays). Thus, the maximal signal-to-noise ratio (S/N) per pixel achievable by an "ideal" CCD is described by

where N is the number of photons collected per pixel during the exposure time, σ the readout noise of the CCD in electrons, and T the thermal noise in electrons per pixel. Although the T term is close to zero in more recent CCD or IPCS devices and can be neglected, additional limitations constrain the usage of CCDs. n is the number of exposures. The S/N decreases dramatically when N is small and n is large. This is the case in multiplex instruments (scanning instruments). For an IPCS, the S/N is described by the following formula:

As IPCS, in fact, do not have readout noise, they are also less affected by cosmic rays because one event is seen as one photon only—a decisive advantage with respect to CCDs, where long exposures are required in the case of faint objects. Although the first generation of IPCS had a poorer quantum efficiency with respect to CCDs and were affected by image distortion, they were still competitive in multiplex instruments or in speckle interferometry because of the absence of readout noise.

Multiplex instruments collect a large number of images. For example, the scanning Fabry-Perot FaNTOmM uses up to 64 channels to reconstruct the interferometric map of an object emission line, in order to determine its velocity field, monochromatic image, and velocity dispersion maps (see Hernandez et et al., 2003). The global S/N of a multiplex observation (S/N_m) is the quadratic sum of the S/Ns of each channel, since there is no noise correlation between channels. Neglecting thermal noise, the S/N_m is given by

where N is the number of photons expected during the whole exposure, n the number of channels, and σ the readout noise of the CCD.

During multiplex observations, the emission usually appears only in a few channels, which could be different for each pixel according to the Doppler shift. Experience with CIGALE and FaNTOmM showed that, emission lines appear most of the time in only 25% of the channels. This lowers the above "ideal" S/N_m. The "worst" case is when the line is detected only in 2 channels. This does not affect the comparison between CCD and IPCS usage, but gives a more precise idea of the kind of objects that could be observed with such instruments. In this "worst" case the S/N_m is given by

Because of the very small readout time (10 ms with the WHT camera and 25 ms with the FaNTOmM one), in the case of an IPCS, it is possible to observe each channel several times during the observation and then average the sky variations (because the first and last images may not be comparable in terms of seeing and transparency when observing with a ground-based telescope). Typically, each channel is observed 5 to 15 s, and when the last channel has been integrated, the first is observed again. Each set of n channels is called a cycle, the duration of which is typically 5 to 15 minutes (depending on the exposure time per channel and the number of channels). The cycle exposure time must be under the OH timescale variations (typically 15  minutes; see Ramsay et al. 1992). Individual cycles are then summed to obtain the S/N required without any loss of events. Many short cycles are then preferred to a single cycle with sky variation. The overlay time of the data acquisition system is under 10 ms, so an exposure time of, at least, 1 s per channel, e.g. 48 s per cycle, is the shortest optimal exposure time. The experience of the LAE and LAM teams lead channel exposure time to 5 s, so a cycle time of around 5 minutes. The total observation consists of several cycles. S/N_m can be rewritten as:

where N is the number of photons expected during the whole exposure, n the number of channels, m the number of cycles, and σ the readout noise of the CCD. Meanwhile, the IPCS S/N_m obtained is independent of the number of cycles and the atmosphere-free S/N is

These two equations are used to generate Figure 5. The faint detection limit is for a S/N of 3. It can be seen on this figure that an IPCS can reach higher S/N than a classical CCD for very faint fluxes.

Zoom In Zoom Out Reset image size

The measured QE of the whole system (including MCP losses) was measured to be 26% at Hα. The overall system presents the same characteristics as the IPCS of FaNTOmM (Gach et al. 2002; Hernandez et al. 2003). The system is limited by the process itself. The Dalsa camera can be read with two speeds depending on the binning mode (1024 × 1024 pixels or 512 × 512 pixels) chosen. When two events occur within the same frame (16.6 ms or 10 ms) at the same location, only one event is counted instead of two. The mean number M of missed photons can be evaluated assuming a Poissonian process for the photon emission by the following equation:

where λ = N/frame per second is the mean number of photons expected during the resolved period of the detector and N the total number of photons.

Figure 6 shows the percentage of nonlinearity of an IPCS detector at different possible frame rates. This shows that the system loses its linearity when the flux increases. Therefore this detector is used exclusively for faint fluxes. Figure 6 shows also that this effect is less important for higher frame rates. The blue lines (dot or continuous) represent the FaNTOmM GaAs system, and the red one is the improved GHαFaS camera.

Zoom In Zoom Out Reset image size

As GHαFaS is mounted on the optical table of the Nasmyth focus, our observations are affected by the field rotation. The optical derotator provided by the Isaac Newton Group (ING) of Telescopes has the disadvantage of covering a field of only 2.5^' × 2.5^', which is too small for the scientific programs envisaged for GHαFaS. Our observation strategy allows us to store the image of the individual channel separately, which is ideal for derotating the observed cube a posteriori (for examples of derotation, see Figure 7). Following the information provided on the ING web pages, the field rotation is ≈0.29^'', i.e., negligible, during the exposure of each individual cycle (∼4 minutes).

Zoom In Zoom Out Reset image size

Most of the data reduction steps are carried out using an IDL-based data reduction package developed by Daigle et al. (2006b). However, because we do not use the optical derotator, the data are affected by the rotation. In order to correct for this effect, we combine Daigle's reduction package with a number of available packages, KARMA¹⁴, SWARP¹⁵ and IMWCS¹⁶. The complete steps of the GHαFaS data reduction are

• 1.  
  We create an initial and rough estimate of the data quality by reducing the observed data cube. This step allows also a good phase calibration, which we use to reduce the individual cycles.
• 2.  
  We reduce each individual observed cycle.
• 3.  
  We create a cube image for each reduced cycle. Investigating each of these images, we have found that for almost all the observed galaxies, the effect of the sky rotation is marginal within the 4 minutes exposure to scan one cycle. This is also confirmed by the ING Web pages.
• 4.  
  The collapsed cube image of each individual cycle is then used to find the three brightest point sources. The IMWCS and KARMA tools are then used to include the appropriate keywords in the header of each image.
• 5.  
  We found that a simple rotational transformation does not provide the adequate accuracy, hence we use the SWARP package to match all cycles regardless of the complexity of the transformation required.
• 6.  
  We apply the same transformation to each individual channel within each cycle, and rebuild a corrected version of the data cube.
• 7.  
  Finally, we reduce the corrected data cube following all the standard reduction procedures presented in Daigle et al. (2006b).

Note that this process is made possible only because our detector has a high time resolution and has no readout noise. This reduction process does not add any excess noise to the observational data and keeps the very high overall sensitivity of the instrument.

GHαFaS data reduction involves the standard Fabry-Perot data reduction steps: cube construction, phase calibration, sky subtraction, and kinematics derivation, which is done using the software package of Daigle et al. (2006b) and the GIPSY package (Vogelaar & Terlouw 2001).

By construction, GHαFaS covers a reasonably large field in order to observe nearby extended objects and provide simultaneous sky coverage to allow for reliable sky subtraction. However, some galaxies extend beyond the full GHαFaS field. In order to sample the sky variations properly during the long observation of the object, the telescope is pointed to a blank field away from the galaxy once every three cycles. The sky cubes are thereafter treated in the same way as the galaxy cube. The reduced sky cube is then rescaled to the galaxy count level and subtracted from the galaxy cube. In order to examine the accuracy of the sky subtraction procedure, the sky cube is used in four different ways before subtraction from the galaxy. All the sky spectra are then rearranged to build one spectrum representative of the entire sky cube, but also fitted with first, third, and fourth order polynomials to each channel image of the sky cube. Subtracting each sky representation separately, we end up with four different reduced galaxy cubes.

The free spectra range (FSR) determines the depth of the data cube and is a function of the observed wavelength (λ) and the interference order (p) at this wavelength: FSR = λ/p. The number of channels is linked to the finesse of the Airy function of the etalon (in fact, the number of channel respects the Nyquist criteria: n = 2.2 × the effective Finesse) and has no relation to the FSR. To study normal galaxies, a FSR of 0.8 nm (360 kms at Hα) is used. If the velocity range of a galaxy (with its inclination factor) is larger than the FSR (which is not frequent), one will find in order −1 or +1 the rest of the velocity field (filters, used in front of the etalon, typically block all etalon order except p + 1, p, p - 1). So a jump in the velocity field is generally found and easily corrected. This could happen with active galactic nuclei of galaxy. In that particular case, another etalon is used to increase the FSR. Table 2 presents the FSR for different etalons.

Each reduced galaxy cube is treated independently to derive the Hα kinematics by means of deriving the moment maps with consistency checks by applying single Gaussian profiles to the spectra (e.g., Fathi et al. 2007a). Analyzing the kinematical maps, it was found that the sky subtraction does not significantly change the derived kinematics. In order to gain more field coverage, a two-dimensional Voronoi tessellation method (Cappellari & Copin 2003; Daigle et al 2006b) was applied on the faint regions of the observed field. Finally, the maps were cleaned by removing all values that are derived from spectra with amplitude-over-noise (A/N) less than 10 (the noise being the standard deviation of the parts of the spectra outside the Hα emission line). Although this seems a strict criterion, it was chosen to continue the analysis of the velocity field based on spectra that are reliable not only for deriving reliable velocities, but also velocity dispersion. GHαFaS instrumental dispersion has been measured using the neon lamp and was found to be σ_inst = FWHM/2.35 = 2.5 km s^-1. It was then subtracted quadratically from the Hα velocity dispersion, and the derived velocities were corrected for the Heliocentric velocity drift.

To give an idea of the efficiency of the pipeline reduction, two panels of results are given, representing the scientific drivers of the instrument. The final kinematical maps are presented in Figure 8 for four objects while Hα monochromatic images are represented in Figure 9. Finally,monochromatic image in the [N II] line and the first order of the PNe M1-75 is presented in Figure 10. It is not the goal of this paper to comment, analyze, or derive kinematical parameters. The final, complete, and detailed data reduction and analysis will be available in forthcoming papers.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Finally, the status of the data reduction of the first run (17 objects) over 6 nights are presented in Table 5.