High-contrast H$\alpha$ imaging with Subaru/SCExAO+VAMPIRES

We present current status of H$\alpha$ high-contrast imaging observations with Subaru/SCExAO+VAMPIRES. Our adaptive optics correction at optical wavelengths in combination with (double) spectral differential imaging (SDI) and angular differential imaging (ADI) was capable of resolving a ring feature around omi Cet and detect the H$\alpha$ counterpart of jet around RY Tau. We tested the post-processing by changing the order of ADI and SDI and both of the contrast limits achieved $\sim10^{-3}-5\times10^{-4}$ at $0.3^{\prime\prime}$, which is comparable to other H$\alpha$ high-contrast imaging instruments in the southern hemisphere such as VLT/SPHERE, VLT/MUSE, and MagAO. Current wavefront sensing and adaptive optics correction at optical wavelengths empirically depends on airmass and Subaru/VAMPIRES provides great opportunities for H$\alpha$ high-contrast imaging for northern hemisphere targets.


Introduction
Observing hydrogen lines provides fruitful opportunities to investigate mass accretion, shocks, jets, atmospheres, and other astrophysical phenomena. A variety of observations have been implemented but previous observations basically targeted an isolated object or multi-objects that can be spatially resolved under seeing-limited conditions. An improvement in instrumentation has enabled better angular resolution than set by the seeing limit -adaptive optics (AO) 1 systems can make real-time corrections to wavefront distortions by the Earth's atmosphere using a guide star and delivers a point spread function (PSF) close to the diffraction-limited one. In particular, AO is very important for high-contrast imaging of exoplanets or protoplanetary disks by removing stellar halo and instrumental speckles that bury such faint signals in the stellar halo. However, the AO correction at optical wavelengths was difficult in the early years of the AO instruments because the Fried parameter (∝ λ 6/5 ) 2 is so small at optical wavelengths and hence a large number of actuators are required across the telescope aperture as well as fast temporal operation for wavefront sensing.
Therefore, classical AO instruments were mostly applied near-infrared (NIR) wavelengths (e.g. JHKLM band), which benefit exoplanetary science because low-mass objects have less contrast with respect to their host stars in the NIR and thus high-contrast imaging at these wavelengths provides the best sensitivity.
Recently, further improvements in instrumentation have made it possible to operate AO correction at optical wavelengths (e.g. VLT/MUSE, 3 VLT/SPHERE, 4 MagAO, 5 and Subaru/SCExAO 6 ) and have kicked off a new era of high-contrast imaging at Hα (λ =656.28 nm). One of the most important subjects of Hα observation with AO is active mass accretion onto protoplanets (e.g. PDS 70 bc 7,8 ). Recent disk observations with ALMA or NIR polarimetric observations have shown a variety of asymmetric features within 1 that may be related to planet formation (e.g. gap, ring, or spiral 9, 10 ), but the number of confirmed protoplanets is still smaller than the number of predictions of potential protoplanets. Therefore, the planet formation mechanism is still controversial and exploring for planets at such inner regions while avoiding the stellar halo with high-performance AO correction is important. The performance at optical wavelengths with current instruments highly depends on airmass. For example, the performance of MUSE/NFM (Narrow-Field-Mode) empirically decreases with airmass ≥ 1.6 1 . We also checked some archival MUSE data 2 and compared  14 2 2MASS 15 airmass and full width at half maximum (FWHM) in the Hα channel (see Table 1), which confirms the degradation of AO correction at high airmass. Poor AO correction leads to ineffective post-processing to remove the stellar halo and to achieve high contrast. As VLT and MagAO are located in the southern hemisphere, their observations of northern targets, e.g. the Taurus star forming region, may not have sufficient sensitivity to detect Hα in the vicinity of a star.
Here we have installed a new observing mode of The Visible Aperture Masking Polarimetric Imager for Resolved Exoplanetary Structures (VAMPIRES) at Subaru Telescope, 11 which is the only instrument capable of Hα imaging fed by AO in the northern hemisphere at the moment.
In this paper we present the current performance of high-contrast Hα observations with VAM-PIRES. The effective bandwidth of Hα is narrower than the widths of broad-band filters used in the optical-NIR astronomy 12,13 and thus investigating such emissions or absorption's requires a specified narrow-band filter or spectroscopic instrument with R 1000. VAMPIRES adopts a narrow-band filter and details of the specifications are described in Section 2. Section 3 presents our engineering observations and the results. Finally we summarize our work and briefly mention future prospects with AO upgrades in Section 4.

SCExAO+VAMPIRES Specifications
The VAMPIRES instrument 11  VAMPIRES is capable of performing diffraction limited, polarimetric imaging at visible wavelengths (600 to 800 nm). It uses two electron-multiplying CCD (EMCCD) cameras which in normal operation each record an orthogonal polarisation, which in combination with a ferroelectric liquid crystal modulator and half-waveplate allow precise polarimetric differential imaging (PDI).
The high speed of the cameras (up to ∼1000 frames/sec depending on subwindow size) allows lucky imaging techniques to be used to enhance resolution, and also include non-redundant masks for super-diffraction limited imaging.
With the upgrade presented here, the polarisation-splitting optics can be automatically interchanged for wavelength-splitting optics, recording simultaneous images in a narrow band centred at Hα (λ cen =656.3 nm, ∆λ = 1.0 nm) in one camera and an adjacent continuum bandpass to allow non-common path errors to be mitigated by a 'double-difference' approach during data reduction (see Figure 1 and the instrument papers for details 6,11 ). The splitting of the light before encountering the filters is performed via a non-polarizing beamsplitter, to minimise mixing of polarization effects with spectral differential imaging (SDI; see Section 3) 18 signal.

Target Selection
For science verification we target omicron Ceti (omi Cet) and RY Tau 22 While detectable in spectroscopy, this is very difficult to resolve spatially due to the expected low spatial separation (several 10s of milliarcseconds) between the star and shock region.
We also use SAO 105500 (γ Sge) data for a comparison of the post-processing results with those of omi Cet and RY Tau. Previous spectroscopic observation proves that this single star does not present any Hα emission/absorption features 23,24 and is good for the comparison of the results.

Observation
SCExAO is operated under angular differential imaging (ADI) mode 25 by fixing the pupil, with which we can utilize a combination of SDI and the ADI technique to detect faint objects around the target star. VAMPIRES uses a beam-switcher where both filters can be converted frequently and hence the spectral content of the beams is switched between the two detectors. In order to reduce further bias, which arises from non-common path aberrations and systematic differences between the detectors, and to achieve better sensitivity we utilize double differential imaging (DDI) 26 Table 3 summarizes the Strehl ratios of our data. We used the beginning part of these data sets for measuring them. To estimate the Strehl ratio we 1) computed the ratio between the total flux inside the core of the PSF (2.44λ/D in diameter) and the total flux of the PSF inside a circle of 85λ/D (to reject the diffraction pattern created by the quilting mode of the deformable mirror) and 2) computed the same ratio calculated for a simulated perfect PSF using the shape and orientation of the pupil. The Strehl ratio is then the flux ratio of the on-sky image divided by the flux ratio of the simulated image. The Strehl ratios measured from the single exposures indicate the performance of the 'fast-AO correction' with SCExAO+VAMPIRES. For comparison we also show the Strehl ratios with the long exposure (simply cube-combined PSF without image registration). We note that at the data reduction stage we do image registration of each slice and then conduct the post-processing technique (see Section 3.3), the performance of which is indicated in the last column of Table 3. For the SAO 105500 and RY Tau data we see the same characteristics in that the top 5% PSF has the highest Strehl ratio. The single exposure of RY Tau has lower SNR because this target is faint in the optical wavelength, which may affect the background evaluation and the Strehl ratio measurement. For the omi Cet data the long-exposure PSF has the highest Strehl ratio but this is likely related to measurement errors of the Strehl ratio and the background: we combined 3201 slices to make the long-exposure image (and ∼2880 slices to make the combined image) and the background values are much better estimated than the single exposure images. Figure 3 illustrates difference between State 1 (cam1: continuum, cam2: Hα) and State 2 (cam1: Hα, cam2: continuum) so that we can conduct the DDI technique to reduce effects of  4 These targets are variable at optical wavelength and we adopt photometric data provided by AAVSO on the nearest date to our observations. 5 20 msec/slice × 3201 slices/cube × 7 cubes + 20 msec/slice × 791 slices/cube × 1 cube. 6 20 msec/slice × 3201 slices/cube × 7 cubes. 7 ∼ 1.65 × 1.65 . 8 1 sec/slice × 101 slices/cube × 24 cubes + 1 sec/slice × 73 slices/cube × 1 cube.  1 We selected the single exposures whose PSFs have top 5%, 30%, and 50% peaks among all the single exposures using the fitted PSF information and then estimated the Strehls from these selected PSFs. This value indicates the performance of the short-exposure AO correction. 2 We combined a data cube into an image without image registration. This value indicates the performance of the long-exposure AO correction (exposure time corresponds to the product of the single exposure time and the number of slices in the data cube, see also Table 2). 3 We selected a set of slices in a data cube (top 90% peaks), and then shifted all the images to align the center of the PSF, and finally combined them to make the combined image. This data set is basically used for the post-processing in this study. 4 Equivalent integration time.
the non-common path aberrations, which is explained in Section 3.3.1. The VAMPIRES output constitutes a data cube (x, y, and time) and exposure time/cube format information is summarized in Table 2. Note that we replaced a Hα narrow-band filter before the RY Tau observation and we observed SAO 105500 and omi Cet with a different Hα filter (λ = 656.4 nm, ∆λ = 2.0 nm).

Data Reduction
As SCExAO+VAMPIRES enables fast-AO correction, we can obtain images with very short exposures. Then we can select 'good' PSFs among all the data set like lucky imaging. After dark subtraction we read all continuum slices, which do not basically include any asymmetric features related to Hα from the central star, to fit PSFs for good-data selection and image registration. We used a criterion of fitted peak and selected 90% good data sets that would then be reduced by post-processing. The typical FWHMs of the selected data sets were 45 mas (7 pix) and 55 mas (8.5 pix) for omi Cet and RY Tau, respectively. Hereafter, we show several methods of reducing the data by combining the ADI and SDI techniques to look for differences in the outputs because, for instance, a previous VLT/SINFONI observation 28 suggested a specific order to utilize ADI and SDI reduction techniques may change the overall sensitivity.

SDI and SDI+ADI
The advantage of the SDI reduction is that in principle we can subtract continuum components as a reference PSF from the Hα image that includes both Hα and continuum information. AO correction works with almost the same efficiency in both filter bands and enables us to simply subtract the continuum image from the Hα image after correcting throughputs between cameras and filter transmission functions. The coefficients used for the throughput correction are assessed by comparing photometric results of instrumental laser PSFs.  Figure 3. Applying the DDI technique to the SDI reduction can further suppress the effects of the bias on the final-reduced image. After PSF fitting of continuum slices using the Moffat function 29 we repeat producing a SDI-reduced slice by subtracting a good-continuum slice in the VAMPIRES data cube from a corresponding Hα slice, which can attenuate the influence of the atmospheric turbulence at each short exposure. Here we used the fitted peak of each PSF as a criterion for judging good PSFs.

SDI reduction
Next we made a combined SDI-reduced image per data cube. As SCExAO is operated in ADI mode we derotate images by differences of parallactic angle and then combine this data set into a SDIreduced image. At each state we conduct SDI reduction and finally obtain double-SDI-reduced image using two SDI-reduced images. We note that in this study we mainly intend to reduce the continuum component from the Hα image and that we do not scale the reference (continuum) image by the difference of wavelength from Hα to attenuate the speckle noise.

SDI+ADI reduction
By applying the ADI technique to the residual of the SDI result we can further suppress the speckle noise. The data set of combined SDI-reduced image per data cube can also be applied to ADI reduction (see Section 3.3.2), which leads to two SDI+ADI images in State 1 and 2. We then conduct double-differential imaging to obtain the final double-SDI+ADI image.

ADI+SDI
ADI makes a likely reference PSF that includes the starlight and instrumental speckles by rotating the FoV, then subtracting it from the raw images, and finally derotating and combining the subtracted images. This post-processing technique has been widely used for high contrast imaging, and has been used to detect faint companions and/or other asymmetric features within a few arcseconds. We made a set of images by combining selected good slices among one data cube, which is then input into ADI reduction algorithms. In this paper we utilized Karhunen-Loéve Image Projection algorithms (KLIP) 30 with pyKLIP algorithms 31 to produce the most likely reference PSF from the set of combined SDI-reduced images, where we adopted optimization and subtraction area as large as the whole VAMPIRES FoV.
After PSF fitting of continuum slices we made both combined continuum and Hα images per data cube. Then we applied pyKLIP to 4 data sets (registered images of continuum/Hα in State 1/2). Each reduced image was then used for further double-SDI reduction (see Figure 3) and we finally got the ADI+double-SDI image. We note that the ADI+SDI reduction substantially ignores the advantage of simultaneous acquisition of Hα and continuum at each exposure.

Results
We present a variety of results that were reduced via SDI and ADI reduction techniques. We note that in this study we do not analyze Hα intensities of each detection and discuss mechanisms of (possible) Hα emissions from omi Cet, omi Cet B, and RY Tau. Figure 4a shows a single exposure raw Hα image of omi Cet. We note that omi Cet B was detected in both combined Hα and continuum images from one data cube (see Figures 4b and 4c). Figures   4d and 4e compare the SDI-reduced images of State 1 and 2 respectively. Figure 4f shows the double-SDI result and Figure 5a plots an azimuthally-averaged radial profile of the double-SDI result. There is a ring-like feature at a separation of ∼ 0.03 . We also checked encircled energy of both combined Hα and continuum images and Hα profile is slightly brighter than the continuum profile at ∼ 0.03 (see Figure 5b) , though the difference is marginal compared with the Hα and continuum observations of η Carinae. 32 This feature represents either of the expected astrophysical feature -limb brightening of the shock feature at Hα -or the difference of PSFs at wavelengths between the Hα and the continuum filters. To further test whether our data reduction can be affected by an artifact (e.g. different AO correction), we apply the double-SDI reduction to the SAO 105500 data (see Section 3.4.3 for the result). A scientific analysis of these resolved observations of the Hα shock region will be presented in a forthcoming paper. Figures 6a and 6b compare the outputs of combining ADI and double-SDI reduction tech-niques. For ADI reduction we adopted KL=3 to show our outputs because we had a small number of data cubes and the larger KL number does not work properly. The smaller number of KL modes subtracts the starlight less efficiently and thus leaves a lot of residuals. Both of the reduction approaches resulted in almost the same outputs and detected omi Cet B with signal-to-noise ratios (SNRs) > 5000 but with the ADI+double-SDI reduction (Figure 6b) there still remains some stellar halo at the inner working angle. The noise was estimated by calculating the standard deviation within an annular region at different separations, which is used for signal-to-noise (SN) maps and detection limits. and that SDI reduction does not lose the signal of Hα. Therefore we injected fake sources in the Hα images and calculated the throughputs of ADI reduction. In case a companion is detected in both continuum and Hα filters such as omi Cet B it is better to use photometric results from both combined Hα and continuum images to accurately obtain Hα intensity.         'Double-SDI+ADI' and 'ADI+double-SDI' correspond to running ADI reduction after double-SDI reduction (Figure 6a) and the reverse order (Figure 6b), respectively.

RY Tau
Both of the reduction ways detected the jet with SNR∼4-5 but its SNR is slightly higher in the SDI+ADI result (Figure 9c). Figures 9b, 9d, and 9f present cropped SN maps of Figures 9a, 9c, and 9e, respectively. Indeed a simple ADI reduction could detect the same feature to some extent as seen in the SDI+ADI image, but there is another feature that corresponds to speckles because of insufficient AO correction at optical wavelengths.
The most significant feature in our post-processed image extends ∼ 0.3 , which is consistent with the SPHERE/ZIMPOL Hα observation. 19 The SPHERE observation also reported a fainter and more extended Hα region extending ∼ 1 and our SDI+ADI-reduced image marginally confirms the inner part (ρ ∼ 0.3 − 0.6 ) of this feature with SNR∼3.

SAO 105500
For comparing the post-processing results of the other two targets, we reduced the SAO 105500 data. We note that because of small field rotation we could not conduct the ADI reduction and we show the SDI results in this section. Figures 11a, 11b, 11c, and 11d show the single Hα image and    (Figures 4d, 4e). The residual at the center probably reflects the difference of its spectrum at wavelengths of the Hα-filter and the continuum-filter. Figure 11e shows an azimuthally-averaged radial profile of the surface brightness and there is no ring-like feature as seen in the case of omi Cet (Figure 4f). Figure 12 compares difference of the encircled energies between Hα and continuum PSFs. The slight rise in the SAO 105500 profile by ∼ 0.4% at 0.3 is likely caused by the difference of AO correction at Hα and continuum wavelengths. The omi Cet profile has much greater difference by ∼ 5% at 0.3 . Considering that AO correction to SAO 105500 works as effectively as to omi Cet, the difference profile at omi Cet may not only be affected by the difference of AO correction. We need to model the expected Hα shock around omi Cet to discuss in detail as mentioned in Section 3.4.1.
We also show the 5σ detection limits of the SAO 105500 data in Figure 14. The humps at ∼ 0.8 are affected by diffraction patterns arising from the bright central star. These detection limits reach similar contrast levels to the omi Cet case because of almost the same AO efficiency.

Summary and Future Prospects
We present high-contrast Hα observations with Subaru/SCExAO+VAMPIRES. VAMPIRES adopts the beam-switching system where we can efficiently conduct SDI reduction by utilizing DDI technique. We also show the combination of ADI reduction with SDI with our engineering data sets of omi Cet and RY Tau. Our double-SDI reduction could resolve a ring feature around omi Cet (corresponding to the expected shock arising from this star's pulsations) as well as detect omi Cet B in both continuum and Hα filter, and SDI+ADI reduction could resolve a jet around RY Tau. We achieved ∼ 10 −4 at 0.5 for omi Cet and ∼ 5×10 −4 at 0.5 for RY Tau, respectively. Our detection limits in apparent flux are comparable to other high-contrast Hα imaging with SPHERE/ZIMPOL, MUSE, and MagAO located in the southern hemisphere. Our instrument will provide great opportunities of implementing high-contrast Hα explorations of northern targets.
In the next few years, AO188 will benefit from two major upgrades that will benefit SCExAO and VAMPIRES. AO188's 188-element deformable mirror will be replaced with a 64x64-element deformable mirror, which will improve significantly the wavefront correction at all wavelengths.
The second upgrade is the addition of a NIR PyWFS inside AO188, which will be used to probe redder targets such as TTS. 35 SCExAO is also being upgraded to perform PDI at NIR wavelengths, using a fast detector similarly to VAMPIRES, and also by doing spectro-polarimetry using CHARIS. A differential imaging mode is also envisioned to include a SDI mode at the Paβ (1.    µm) wavelength. In case of planet formation, for instance, theoretical models suggest that hydrogen emissions depend on several parameters such as number density of hydrogen, pre-shock velocity, extinction of the source, and filling factor of the emission 36 and obtaining only Hα cannot solve degeneracy between these parameters. Therefore Paβ will help to solve the degeneracies and promote detailed discussions of accretion mechanisms. Currently a limited number of instruments enable high-contrast imaging explorations of Paβ (e.g. Keck/OSIRIS 37 ) and SCExAO will be more useful instruments for hydrogen emission observations. The authors wish to acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain.

Appendix A: Filter Information
For high-contrast Hα imaging with VAMPIRES we designed the bandpass of the filter to be able to transmit sufficient signal in light of various broadening effects. For example, the maximum accretion velocity of a Jupiter mass planet is ∼ 60 km/s. To account for an object larger than a Jovian planet, we adopted a velocity of ±100 km/s. In addition, the RV motion of a planet around such a star is 10 km/s and the RV motion of the star from the observer is 50 km/s. Therefore, we set a budget of 200 km/s, which corresponds to a 0.8 nm wavelength range at 600 nm, which should cover all the astrophysical features one may like to study.
We found an off-the-shelf filter from Chroma that had specifications very close to our requirements (3 nm bandwidth; see also the manuscript webpage at Chroma) and requested it be modified to have a FWHM bandwidth of 2 nm instead. This is still larger than the 0.8 nm needed and was chosen because of the uncertainty the manufacturer offered in the center wavelength of the filter and our ability to align it with respect to the beam. This filter was purchased and installed. width of Hα is ∼100-120 km/s. 8,38 We found that detecting such a signal with the 2 nm-bandwidth filter will require higher contrast, enough to detect the accreting planet. Therefore we decided to replace the previous Hα filter with a new filter (1 nm-bandwidth) in January 2020. This bandwidth is the minimum acceptable value for the Hα imaging with our assumptions. Out of band, the optical density (OD) of the new Hα filter is larger than 4 (OD4 means that the out-of-band suppression is 0.01%) across 400-1200 nm, which covers the entire bandpass of the EMCCD. And similarly the suppression of the old Hα and continuum filters was better than OD5 (the out-of-band suppression is 0.001%) from 400-1000 nm. The detailed information for the new Hα filter is shown at Alluxa.