Improved Orbital Constraints and Hα Photometric Monitoring of the Directly Imaged Protoplanet Analog HD 142527 B

Companions embedded in the cavities of transitional circumstellar disks have been observed to exhibit excess luminosity at Hα, an indication that they are actively accreting. We report 5 yr (2013–2018) of monitoring of the position and Hα excess luminosity of the embedded, accreting low-mass stellar companion HD 142527 B from the MagAO/VisAO instrument. We use pyklip, a Python implementation of the Karhunen–Loeve Image Processing algorithm, to detect the companion. Using pyklip forward modeling, we constrain the relative astrometry to 1–2 mas precision and achieve sufficient photometric precision (±0.2 mag, 3% error) to detect changes in the Hα contrast of the companion over time. In order to accurately determine the relative astrometry of the companion, we conduct an astrometric calibration of the MagAO/VisAO camera against 20 yr of Keck/NIRC2 images of the Trapezium cluster. We demonstrate agreement of our VisAO astrometry with other published positions for HD 142527 B, and use orbitize! to generate a posterior distribution of orbits fit to the relative astrometry of HD 142527 B. Our data suggest that the companion is close to periastron passage, on an orbit significantly misaligned with respect to both the wide circumbinary disk and the recently observed inner disk encircling HD 142527 A. We translate observed Hα contrasts for HD 142527 B into mass accretion rate estimates on the order of 4–9 × 10−10 M ⊙ yr−1. Photometric variation in the Hα excess of the companion suggests that the accretion rate onto the companion is variable. This work represents a significant step toward observing accretion-driven variability onto protoplanets, such as PDS 70 b&c.


INTRODUCTION
As the circumstellar environment surrounding young pre-main-sequence stars evolves, forming planets and binary companions disrupt and shape the circumstellar disk (e.g.Williams & Cieza 2011;Dong & Fung 2017).Embedded giant planets (protoplanets) and binary companions are expected to play dramatic roles in the formation of substructures such as cavities, rings, and spiral features within disks (e.g.Dodson-Robinson & Salyk 2011;Bae et al. 2018).In the past two decades, improvements in high-contrast imaging instrumentation and post-processing techniques have revealed these morphologically complex disks in striking detail (e.g.Tamura 2016; Avenhaus et al. 2018;Esposito et al. 2020;Garufi et al. 2020).Of particular interest are so-called "transition disks" (Dodson- Robinson & Salyk 2011;Espaillat et al. 2014), which host a wide central cavity depleted of dust.Gas has been observed to flow through these cavities (Casassus et al. 2013), fueling accretion onto the central star as well as onto the planetary (e.g.Sallum et al. 2015;Wagner et al. 2018;Haffert et al. 2019;Zhou et al. 2021) and stellar (Close et al. 2014a) mass companions that may be responsible for clearing them.This accretion is likely mediated by a circumsecondary (or circumplanetary) disk, observational evidence of which is accumulating at NIR (e.g.Lacour et al. 2016) and sub-mm (e.g.Benisty et al. 2021) wavelengths.
selection effect (Close 2020) or a reflection of differences in accretion physics at planetary masses (Aoyama et al. 2021;Marleau et al. 2022).
The inner thermally-emitting disk component was first inferred by SED modeling of NIR excess and unresolved sub-millimeter observations (Verhoeff et al. 2011;Boehler et al. 2017).Narrow "shadow" features have been observed along the outer disk cavity wall, suggesting that this inner disk may be inclined with respect to the outer disk (Marino et al. 2015).Bohn et al. (2021) provided the first resolved measurements of the inner disk using VLTI/GRAVITY observations, which they used to investigate the mutual alignments of the inner and outer disk components.They found that the K-band complex visibilities of their data were best fit by an inner disk model with an inclination of i d,inn = 23.76 ± 3.18 • and a longitude of ascending node of Ω d,inn = 15.44 ± 7.44 • .Refitting ALMA CO data to derive outer disk geometry, they found an outer disk inclination of i d,out = 38.21± 1.38 • and a longitude of ascending node of Ω d,out = 162.72.44 ± 1.38 • .From these measurements they inferred that the inner and outer disks of HD142527 are statistically significantly misaligned by 59 • .This difference in inclination is consistent with the inner disk generating the shadows observed in scattered light and is suggestive of dynamical disruption, namely a companion on an inclined orbit (Facchini et al. 2018).
In 2012, a companion candidate to HD 142527 A was detected with the VLT/NACO instrument (Lenzen et al. 2003;Rousset et al. 2003) using Sparse Aperture Masking (SAM) at 88 mas separation, with an estimated mass of 0.1 − 0.4 M (Biller et al. 2012).Subsequently, the Magellan Adaptive Optics (MagAO) team used the Visible Light Adaptive Optics (VisAO) instrument to confirm the companion by directly imaging HD 142527 B in Hα (∆mag = 6.33 ± 0.20 mag) and in 643nm continuum (∆mag = 7.50 ± 0.25 mag) with a combination of classical Angular Differential Imaging (cADI) and S-SDI (referred to hereafter as ASDI, Close et al. 2014a).The presence of Hα excess in the VisAO detection indicated that the companion was actively accreting at a rate of ∼ Ṁ = 5.9 × 10 −10 M yr −1 (Close et al. 2014a).
The companion was also imaged using the Gemini Planet Imager (GPI, Macintosh et al. 2014) in total intensity at Y-band with Angular Differential Imaging (ADI)/Principle Component Analysis (PCA) and polarized intensity light with Polarized Differential Imaging (PDI) and ADI/PCA (Rodigas et al. 2014).Interestingly, the polarized light detection was marginally spatially inconsistent with the total intensity source by ∼ 20 mas at 2σ confidence in their reduction, which they interpret either as scattered light from the disk around the companion, or a clump of dust separated from the companion2 .
The companion was subsequently characterized using VLT/NACO and Gemini/GPI SAM; it was confirmed to exhibit infrared excess indicative of a 1700K circumsecondary environment and was found to be significantly younger (1.0 ± 1Myr) than A (Lacour et al. 2016).Lacour et al. (2016) fit a mass of M B = 0.13 ± 0.03 M and a temperature of T B = 3000 ± 100K to the SED of the object.Using the SPHERE/SINFONI instrument (Eisenhauer et al. 2003), Christiaens et al. (2018) find T B = 3500 ± 100 K, a spectral type of M2.5, and therefore an age of 0.75±0.25Myrand mass of M B = 0.35 M with a recovered spectrum that is significantly brighter than that found in Lacour et al. (2016).They do not investigate this discrepancy3 .
The SPHERE IFS and IRDIS instruments were used to examine the companion via NIR direct imaging and SAM (Claudi et al. 2019).Their results suggest best fit temperatures in the range of 2600 − 2800K, closer to those fit in Lacour et al. (2016), and a spectral type M5-6.They also demonstrate variability in the differential flux of B with respect to the brightness of A on the order of half a magnitude between 1−1.6µm.While this could be due to the variability of the primary star, Claudi et al. ( 2019) quantified the variability of the primary and found it was insufficient to explain the variability in the continuum flux of HD 142527 B. They place dynamical constraints on the mass of HD 142527 B, finding M B = 0.26 +0.16  −0.14 M .Cugno et al. (2019) observed HD 142527 B using SPHERE/ZIMPOL in Hα with ADI+S-SDI in a similar fashion to Close et al. (2014a).They recovered the companion in three filters -a broad Hα filter, a narrow Hα filter, and a continuum filter.They confirmed Hα excess emission from the companion and estimated an accretion rate of ∼ Ṁ = 1 − 2 × 10 −10 M yr −1 , marginally lower than that derived by Close et al. (2014a).
Further study of HD 142527 B informs a number of open questions in the fields of star and planet formation.The extreme mass ratio (∼ 20 : 1) between the primary and secondary in this system, the fact that HD 142527 B is orbiting within a transition disk cavity, and its ongoing accretion make the system a higher mass analog to protoplanetary systems such as PDS 70 b&c, allowing for the refinement of techniques used to image these systems in visible light.
The HD 1425257 system itself is also an important probe of the processes of planet formation in binary systems.Improved orbital constraints can place limits on the mutual inclinations between the HD 142527 AB binary orbit and the inner and outer disk segments (Czekala et al. 2019).Recent VLTI/GRAVITY observations of the inner disk are especially important to consider, as it is likely that HD 142527 B is responsible for the inclination of the misinclined inner disk.Orbit fits to the HD 142527 binary allow mutual inclinations of all three components of the system (the inner disk, binary, and outer disk) to be determined and compared to hydrodynamical models.
The nature of the HD 142527 B companion itself is also broadly informative of star and planet formation processes, as its young age relative to HD 142527 A might indicate that it formed from the disk via disk instability.It could also be that HD 142527 B formed at the lower end of the IMF, was dynamically captured, and is disrupting planet formation around HD 142527 A. In either scenario, its motion necessarily drives the dynamical evolution of the disk (e.g.Aly et al. 2021).
Observations of the companion to date have only covered a 60−70 • orbital arc.The nature of HD 142527 B's orbit and its relationship to both the wide observed cavity in the circumbinary disk and the inner circumprimary disk is still an area of ongoing study that can be improved by further constraining the orbital elements, particularly the eccentricity and inclination of the binary orbit and the mutual inclinations of the various components.Until the orbit is very well characterized, it cannot be certain whether the companion can be held solely responsible for the massive cavity.
Additionally, relatively little is known about the density and dynamics of gas flow through transitional disk cavities or the reprocessing of material in circumsecondary accretion disks.Accretion rates and epochto-epoch variations in accretion rate can inform accretion processes for companions in these actively planetforming systems.
In this work, we build upon previous detections of Hα excess emission from HD 142527 B. We monitor the companion's astrometric motion in visible light and provide the most complete orbital solution to date, enabling a more precise comparison between the mutual inclinations of system components.We leverage additional epochs of observation at Hα to argue that the accretion onto the companion is likely variable on (at least) yearly timescales.

VisAO observations
We observed HD 142527 with the Magellan Adaptive Optics (Close et al. 2013;Morzinski et al. 2016) VisAO (Males et al. 2014) instrument in Hα SDI mode (Close et al. 2014b) on 7 nights between 2013 and 2018.Table 1 records general information about the VisAO observations of HD 142527.During observations, the telescope rotator was turned off, resulting in rotation of the FOV, which enables ADI.More field rotation allows for more reference images to be used in constructing the PSF model, improving its quality; this is especially important for tightly-separated companions like HD 142527 B. The total field rotation for each observation is noted in Table 1.HD 142527 A was dithered across the CCD throughout the observing night to mitigate the effects of near-focus dust spots in the images.MagAO uses a Natural Guide Star (NGS) to conduct wavefront corrections, and in all datasets the NGS used was the on-axis science target, HD 142527 A. The SDI observing mode splits the incoming light using a Wollaston beamsplitter, and results in simultaneous continuum (λ c = 656nm, ∆λ = 6.32nm) and Hα (λ c = 643nm, ∆λ = 6.20nm) images (512 × 1024 pixels in size) on the top and bottom of the CCD, respectively.We conducted a revised calibration of the absolute astrometric solution for the VisAO instrument, which is documented in detail in Appendix A. We determined an updated platescale and North Angle offset by tying observations of the θ 1 Ori B multiple system over 6 years to observations of the same system taken over 20 years with the precisely astrometrically calibrated Keck/NIRC2 instrument (Yelda et al. 2010;Service et al. 2016).The updated VisAO platescale is 7.95 ± 0.010 mas pix −1 and the updated North Angle offset is 0.497±0.192• counterclockwise.This new solution does not necessitate a revision of previously published results, as both values agree with previous calibrations within errorbars (Close et al. 2013;Males et al. 2014).However, the updated calibration improves the accuracy of VisAO astrometry, and validates the stability of the North Angle offset between instrument mountings, which occur every semester.

Preprocessing
VisAO data were reduced and preprocessed with the GAPlanetS data reduction pipleline described in detail in Follette et al. (2017) and Follette et al. (2022, in prep.).In short, images were dark subtracted, flat fielded (except for the 2013-04-11 dataset, for which no flatfields exist), split between Hα and continuum channels, and registered using a Fourier cross-correlation algorithm, which yields errors on the position of the central star of ∼ 0.1pix (Follette et al. 2022, in prep.).
In half of our observations (3 of 7, see Table 1), the central star was allowed to saturate in order to improve SNRs in the search for additional companions within the system.A stable instrumental ghost appears in VisAO images to the right of the natural guide star (NGS).We investigated the stability of the ghost as a photometric and astrometric calibrator for saturated data, and found it to be suitable for both purposes (see Appendix B).
Following alignment, images with cosmic rays within 50pix (≈ 0. 04) of the central star were rejected by hand.Images were then cropped to a 451 pixel (∼3.5) square surrounding the central star.
We then measured the "Hα scale factor" by conducting aperture photometry on the central star (or ghost, for saturated images) for each image in the sequence.We adopt the median of the ratios between the Hα flux and the continuum flux for each image in the sequence as the best estimate of the scale factor (F * ,Hα /F * ,Cont ), and the standard deviation as the uncertainty, and these values are recorded in Table 1.This scale factor is an estimate of the Hα excess of the primary star, and allows us to quantify the accretion variability of the host star itself and to correct for the effect of stellar accretion on measured contrasts for the companion (see Section 5).Note-The average seeing was determined by measurements taken from the DIMM, Magellan Baade, or both (in which case the two are averaged).For datasets where no external seeing information exists, the column is left blank.The stellar Hα/Cont.ratios are calculated by performing aperture photometry on every image in both wavelengths for each dataset.We extract photometry from the central star in the case of unsaturated datasets and from the instrumental ghost in the case of saturated data.We report the median ratio for each dataset in the table above.A ratio < 1 indicates a relatively quiescent phase in the accretion onto the primary.

PSF Subtraction
We conducted PSF subtraction using the python implementation of Karhunen-Loeve Image Processing, pyklip4 (Wang et al. 2015).The pyklip input parameters movement, annuli, and numbasis were chosen based on the optimization techniques described in Adams Redai et al. (2022, in prep) as developed for the the Giant Accreting Protoplanets Survey (Follette et al. 2022, in prep).In brief, the pyklip movement parameter is an exclusion parameter that removes images from the reference library in which a companion at a given separation from the star would have shifted (rotated with the sky) by fewer than the specified number of pixels relative to the image for which the PSF model is being constructed.The annuli parameter describes the size of annular zones that pyklip treats separately, where the width of the annulus is ∆r = (OWA − IWA)/annuli.Generically, the inner edge of each annulus is defined as r in = IWA + ∆r × n, but for HD 142527 B, the companion always lies within the inner annulus, so n=1 and r in = IW A, while r out = IW A + ∆r.Our images are non-coronagraphic, so do not have a hardwaredetermined inner working angle.However, we found that applying one in software generally improves our detections and we have adopted a fixed value of 1×FWHM here.The numbasis parameter controls the number of principle components, or KL modes, included in the constructed PSF.We chose to fix the maximum number of KL basis vectors used to construct the PSF at 100 and applied a pre-KLIP highpass filter of 1 × FWHM in all analyses to match the broader GAPlanetS Survey strategy.
We refer the reader to Adams Redai et al. (2022, in prep) for additional information on the optimization of pyklip parameters for GAPlanetS data.To briefly summarize the optimization method used here, we: 1. conduct a grid search in pyKLIP's movement, annuli, and numbasis (KL mode) parameters and generate KLIP images for each combination 2. compute six "image quality metrics" for each movement, annuli, and numbasis combination.These metrics are: peak SNR of the companion(s), average SNR of the companion(s), the "neighbor quality" of the previous two metrics (computed by smoothing their metric maps), false positive (>5σ) pixels, and contrast, normalized so that their best (highest for SNRs, lowest for contrast and false positive pixels) values are 1.
3. Sum or average across a desired choice of metrics, companion(s), and KL modes to select "optimal" values for each of the three KLIP parameters.
For this work, the target of the optimization for most datasets were "false planets" inserted into into the continuum images.In these datasets, we optimized on the sum of all 6 normalized image quality metrics averaged between 5 and 20 KL modes and across 4 to 8 false planets (as many as would fit with a radial spacing of 0.5×FWHM and an azimuthal spacing of 85 o between H , = 0.656 the IWA and control radius of each dataset) to select the optimal movement and annuli parameters.We then selected a single optimal numbasis value by maximizing the sum of the 6 normalized metrics for the optimal combination of annuli and movement.The optimal values for all three parameters are reported in Table 2.
Although the companion is recovered in all epochs using this parameter selection method, which optimizes for robust recovery of companions throughout the region between the IWA and control radius, it's not necessarily the best choice of parameters for the specific location of the companion.In the case of a robust known companion like HD 1425257 B, optimization can also be done on the location of the companion itself in Hα images, and we utilize this method to achieve higher SNR recoveries of the companion in the May 16, 2015 and April 27, 2018 datasets.For these epochs, we optimize on the sum of the peak and average SNR metrics for the companion, and select the annuli, movement, and numbasis parameters that maximize this sum.Since each epoch of VisAO data for HD 142527 B has a corresponding near-in-time NIR detection, we optimized the KLIP parameters on the known NIR location of the companion.Contrast curves and limits on additional companions in the HD 142527 cavity appear in the upcoming GAPlan-etS survey paper, Follette et al. (2022, in prep).

RESULTS
Using pyklip, we detect the companion in both filters in all epochs of observation except 2017-02-02, where the total field rotation was too small (16.1 • ) to achieve the necessary contrast to detect the very tightly separated companion5 .As expected, the observations with the greatest total field rotation (2015-05-15) and best atmospheric quality (2013-04-11) yielded the highest SNR recoveries of the companion.Figure 1 shows each recovery in Hα and continuum.We quantify the quality of each recovery using the forward modeling capabilities of pyKLIP.

Bayesian Forward Modeling
To determine the astrometry of the imaged companion, we implement KLIP forward modeling (Pueyo 2016) and conduct Bayesian Klip Astrometry (BKA, Wang et al. 2016).This technique involves the projection of a companion PSF estimate onto the basis vectors used to construct the primary KLIP PSF model to synthesize a forward-modeled PSF.This forward-model is fit to the post-KLIP data using an affine-invariant MCMC with emcee (Foreman- Mackey et al. 2013).This produces a posterior distribution of model fits that yields robust uncertainties on astrometry and photometry.We take the median of the posterior distribution of a given parameter as its measured value and the 16th and 84th percentiles as uncertainties on that median.The strength of KLIP forward modeling when compared to negative planet injection grid searches is the speed and precision of model fitting and convergence, which enables the application of a Bayesian analysis framework as well as the capabil- ity to model correlated noise within the images using a gaussian process.
For unsaturated data, we find that using a 2D Gaussian PSF with the FWHM of the median stellar PSF produces excellent forward model fits (see discussion in Appendix B).
For saturated datasets, we do not have a direct measure of the stellar FWHM.In Appendix B, we investigate the suitability of a range of forward models in extracting astrometry and photometry for the tight HD42527B companion in saturated datasets.In short, we find that the ghost is suitable as a photometric calibrator (see Appendix B), and determine that the ratio of the peak value of a Moffat fit to the ghost to the peak value of a Moffat fit to the unsaturated NGS PSF in unsaturated data is a stable quantity.We also find that the ghost is slightly out of focus, and that its FWHM is ∼ 7% wider than the stellar FWHM in unsaturated datasets.This result matches our expectations: as the ghost is produced by a reflection off of the MagAO 50/50 beamsplitter, it has a longer path than the 0th order source, and is therefore slightly out of focus.We assume that both the star-to-ghost brightness scaling and the FWHM discrepancy holds in the case where the NGS is saturated, and estimate the FWHM of our saturated datasets by taking 0.93×FWHM ghost .We find that the optimal forward model for saturated data is a Gaussian PSF whose FWHM was set to the 0.93 × FWHM ghost to match the expected FWHM of the image plane PSF.
In order to further quantify the strength of our detections, we utilize the PlanetEvidence class (Golomb et al. 2019) within pyklip to conduct a Bayesian model comparison and determine the SNR of HD 142527 B for a given detection.PlanetEvidence uses the nested sampling implementation pyMultiNest (Buchner et al. 2014;Feroz et al. 2009) to compare two models: H 0 , where the image contains only speckle noise, and H 1 , where the image contains a source at the position of HD 142527 B. PlanetEvidence returns marginal distributions of the parameters for the source and null cases, and calculates the SNR of the detection within the fitting region and the evidence values for H 0 and H 1 (Z 0 and Z 1 ).The log-ratio of these evidence values, log B 10 = log Z 1 /Z 0 , enables us to quantify the confidence with which one model can be favored over the other. 6This framework provides a more robust estimate of the quality of the detection, better capturing asymmetric speckle noise that can dominate at very close separations than SNR computed within an annulus.The SNR values reported in Table 4 are calculated from the residuals within the fitting region as described in Golomb et al. (2019).
For example, values for log B 10 > 5 are considered "strong" evidence against the null hypothesis.We achieved strong evidence (log B 10 > 10) in all cases, and unambiguous detections (log B 10 > 20) in eleven out of tweleve datasets.While the evidence is not strong for the detection in the continuum for the 2018-04-27 epoch, the search for the companion in this epoch is not a "blind" search.We can be reasonably certain that signal at this location is due to the real companion due to the near-in-time observations in the NIR (Claudi et al. 2019).Nevertheless, the continuum detection in 2018 should be approached with some skepticism.

Forward Model Results
Our Bayesian forward modeling returns posterior distributions of the x and y positions (in pixels) of the fit model, a flux scaling parameter α, and a length scale l of correlated noise within the image.We transform α into contrast by multiplying it by a constant 10 −2 (a pyklip input term setting the initial contrast of the forward model) and by the peak counts of the forward model (which in our case is 1 as the PSF model is normalized before it is passed to pyklip).
We calculate the astrometry given in Table 3 by applying the updated platescale and north angle offset (described in detail in Appendix A) to the separation and position angle measurements of the companion from the posterior distributions of forward model fits.We propagate 0.1pix uncertainties on the position of the host star from the image registration process, 1σ uncertainties on the position of the companion from the posterior distribution of forward model fits, and estimated uncertainties on our absolute astrometry (see Appendix A for details) into our calculations to obtain final uncertainty estimates on separation and PA measurements for the companion.
We determine the Hα contrast of the companion relative to the primary star reported in Table 3 as follows.First, we take the median and standard deviation on the forward model fit to the contrast (C B−A ) as the derived value and uncertainty, respectively.For the continuum filter, this value can be converted to a contrast in magnitudes via a simple magnitude transformation, namely ∆mag = −2.5 log C B−A .However, at Hα, the star itself is actively accreting, influencing the Hα contrast extracted from BKA.In order to measure the Hα contrast of the companion with respect to the stellar continuum, we multiply the BKA-derived contrast of the companion at Hα (F comp,Hα /F * ,Hα ) by the stellar Hα to continuum scale factor for the observations (F * ,Hα /F * ,Cont , determined as described in Section 3 and given in Table 1).This leaves us with the contrast of the companion at Hα relative to the stellar continuum (F comp,Hα /F * ,Cont ) and allows us to compare the brightness of the companion at Hα over time without being influenced by stellar Hα variability.We propagate errors on the BKA-derived contrast and scale factor through the logarithmic transformation to magnitude.
Table 3 records the results of our forward model fits to the astrometry and photometry of continuum and Hα images for each epoch.Figure 2 illustrates a representative model fit and residuals for the 2013-04-11 dataset, and galleries of forward model fits to other epochs can be found in Appendix C.

ANALYSIS & DISCUSSION
We calculate the separation and PA of the companion on each night of observation as the weighted mean of the x and y positions in the Hα and continuum filters, where the weights are the uncertainties in each filter (calculated as described in Section 4).The final uncertainty on the Hα and continuum averaged position is then the square root of the sum of the uncertainties on the individual measurements.Table 4 records the final VisAO astrometry together with previous astrometry from the literature, and these positions are plotted in Figure 3.

Orbit Fitting
By combining our derived astrometry with compiled results from the literature, we compute orbits fit to the motion of HD 142527 B with orbitize!(Blunt et al. 2020), an open-source python package that performs Bayesian orbit fitting for directly imaged companions7 .We use the parallel tempered (Vousden et al. 2016) affine-invariant (Foreman-Mackey et al. 2013) MCMC sampler in orbitize!, in order to determine the posterior probabilities for 8 orbital parameters: semi-major axis (a), eccentricity (e), inclination angle (i), argument of periastron of the companion's orbit (ω), longitude of ascending node (Ω), epoch of periastron passage (τ ), system parallax (π), and total mass of the binary system (M tot ).
We assume default orbitize!priors except on the total mass of the system and the system parallax.We set a Gaussian prior on the system parallax, N (µ = 6.356, σ = 0.047), following the measured Gaia eDR3 parallax (Gaia Collaboration et al. 2021).We adopt a Gaussian prior on the total system mass (M tot ) of N (µ = 2.3 M , σ = 0.3 M ) based on the results of Mendigutía et al. (2014) and Claudi et al. (2019).We refer the reader to Blunt et al. (2020) for information about orbitize!default priors, as well as more detailed descriptions of the variables involved in the orbital solution.Posterior distributions are computed by first initializing an MCMC with 50 walkers and 20 temperatures for a 100,000 step burn-in phase per walker; the walker samples are discarded before a total of 5,000,000 additional samples are recorded to approximate the posterior.
Table 5 records our estimates of the orbital elements for HD 142527 B. Figure 4 shows 100 randomly drawn orbits from our posterior distribution of orbit fits overplotted on the astrometry compiled in Table 4.The full corner plot visualizing the posterior distributions of all eight orbital parameters (Figure 14) can be found in Appendix D. Our additional VisAO astrometry provides marginally tighter constraints relative to previous orbital solutions.We note, however, that our MCMC approach is likely more robust under poorly-constrained posterior distributions than previous Least Squares Monte Carlo approaches (See discussion in §2.3.2 of Blunt et al. 2020).−0.72 agrees with their estimate within errorbars, but our refined Ω = 142.38 +5.51 −6.12 marginally disagrees; our solutions preferring a somewhat higher value.According to our orbital fits, periastron passage has likely already occurred for HD 142527 B.
HD 142527 B's orbit is interesting in the context of star and planet formation in that its eccentricity might indicate a stellar formation history.Directly imaged brown dwarf companions have been found to have a distribution of eccentricities that favor higher val- ues when compared to directly imaged planets (Bowler et al. 2020), suggesting that they form independent of the system and are captured, rather than arising from within the circumplanetary disk.HD 142527 may therefore be more analogous to objects like GQ Lup B (Wu et al. 2017;Stolker et al. 2021), a captured, accreting Very Low Mass object that is dynamically truncating/disrupting its circumprimary environment, than to accreting protoplanets.

Mutual Inclinations
In order to contextualize our results within the broader HD142527 system architecture, Figure 5 2020); these data were then interpolated to VisAO's platescale to facilitate comparison with VisAO observations of the companion (shown for all epochs in Figure 1).
The "mutual inclination" between the orbit of a binary star and its circumbinary disk is expected to inform the evolution of the circumbinary and circumstellar environments, and has wide-ranging implications for planet formation in binary systems (Czekala et al. 2019).While external binary companions may significantly truncate a circumstellar disk, perhaps suppressing planet occur-rence (Kraus et al. 2016), circumbinary disks around tighter binary systems result in circumbinary planets (e.g.Kostov et al. 2020Kostov et al. , 2021)).The HD 142527 system is interesting to consider in this context because it is one of a handful of systems with direct imaging measurements of both inner and outer disk components as well as the binary orbit.
The mutual inclination between the disk plane and binary orbit, θ, is defined as where θ is the angle between the angular momentum vector of the binary orbit and the midplane of the disk (Czekala et al. 2019).
We determine the posterior distribution of mutual inclinations between the binary orbit and both the inner and outer disk components (shown in Figure 15) by drawing random samples from the posterior distribution of i and Ω produced by orbitize!, and assuming a gaussian distribution for the disk orientation centered on the values most recently inferred from ALMA gas kinematic data for the outer disk (i disk,outer = 38.21± 1.38  clination's dependence on the bimodal Ω ).These values are not consistent with those reported previously (35 ± 5 • ) by Czekala et al. (2019), who adopted the Ω solution from Claudi et al. (2019).
Our best fit to the mutual inclination of the binary with respect to the inner disk is θ in− = 177.44+2.90 −2.80 • or 110.26 +1.82 −1.50 • .Taken together, our mutual inclination fits present a system where each component of the system is dramatically misinclined with respect to the others.
Modeling the interaction of the binary companion and the circumbinary disk, Price et al. (2018) demonstrated that the morphological features of the circumbinary disk (wide cavity, asymmetric dust horseshoe, and spiral arms) can be qualitatively reproduced solely through interaction with the binary companion under orbits representative of those found in Lacour et al. (2016).To best reproduce the observed morphology, Price et al. (2018) favor a narrower family of eccentric orbits (e = 0.6−0.7)with nearly perpendicular mutual inclinations with respect to the outer disk.
Our work lends some support to this hypothesis, as one of the equal, symmetrically-distributed peaks in θ (resulting from the degeneracy in Ω ) from our orbit     • suggestive of the near perpendicular configuration they describe.Our orbit fitting has not produced a distribution suggestive of such high eccentricities, however.

Photometry
Figure 6 plots the contrast of the companion relative to the stellar continuum over time, derived as described in Section 4. ∆mag measurements from Cugno et al. (2019) are overplotted to demonstrate consistency with our measurements.Between 2013-04-11 and 2015-05-15, the Hα contrast of the companion decreased by 0.68 mag.This is suggestive of a moderately variable accretion rate onto the companion on yearly timescales.Over the same baseline, the continuum did not vary within uncertainties.As there is not significant variability in the continuum contrast of the companion between epochs, this suggests that the extinction towards HD 142527 B is constant to ±0.2 mag.Accretion-driven variability in the stellar continuum flux at our cadence is similarly limited to at or below this value, though it likely occurs on shorter timescales (e.g.Stauffer et al. 2014).
The Close et al. (2014a) contrast measurements (∆mag Hα = 6.33 ± 0.20 mag, ∆mag cont = 7.50 ± 0.25 mag), derived from the same dataset, agree with our measurements within error bars, and we note our slightly higher continuum brightness estimate is likely due to improved PSF subtraction and forward modeling, which allows us to better quantify flux lost to this process.While taken in marginally different filters, the Cugno et al. (2019) contrast measurements (shown with square symbols in Figure 6) agree well with our results; their continuum contrast measurement falls along our median continuum contrast and their Hα contrast lies within the range of values we have measured, with similar uncertainties.
Our method for calculating the contrast of the companion with respect to the stellar continuum should should be insenstitive to stellar Hα variability, leaving only stellar continuum variability as a potential contaminant.The star has a known periodicity of ∼ 6 days, with a peak-to-valley amplitude of 0.09 mag in the Rband (Claudi et al. 2019), too small to account for the observed variation in the Hα channel.If stellar continuum variability were contaminating the observed Hα variation, we would expect to see it directly in our measurements of the companion's continuum contrast, and we do not observe such variation.We note that the companion's infrared continuum has been observed to vary on the order of 0.5mag (Claudi et al. 2019) but we do not observe a similar variability at visible wavelengths.
Our data are therefore suggestive of variability in the Hα emission of this directly imaged accreting companion, resulting in the observed time variability in the Hα excess.This variability has implications for future direct imaging protoplanet surveys.If accretion onto less massive companions is similarly variable, detection limits will need to be interpreted with some caution, as companions will be more quiescent at certain times.

Accretion Rate
We estimate the mass accretion rate onto HD 142527 B using Hα contrast measurements following a standard set of assumptions for accreting objects.First, we convert our contrast into a line luminosity measurement via where D is the distance to the system, Z pt = 2.339 × 10 −5 erg/cm 2 /s/µm is the Vega zeropoint of the Hα filter (Males et al. 2014), dλ = 0.006µm is the width of the Hα filter, and mag = 8.1 − A R is the R-band apparent magnitude of the central star (8.1, Cugno et al. 2019), corrected for extinction (A R = 0.05, Fairlamb et al. 2015;Cugno et al. 2019).
We convert line luminosity to an estimate of the total accretion luminosity of the star using the Rigliaco et al. (2012) empirical relationship between L Hα and L acc derived for T-Tauri stars, namely: where a = 1.25 ± 0.07 and b = 2.27 ± 0.23.From the accretion luminosity, we derive the mass accretion rate via the standard relation:  Table 6 records the line luminosities and mass accretion rates for each epoch calculated following these assumptions.The peak Hα excess, which occurs in the 2014-04-08 epoch, corresponds to a mass accretion rate estimate of 6 × 10 −10 M yr −1 .These mass accretion rate estimates differ from those calculated by Cugno et al. (2019, Ṁ = 1 − 2 × 10 −10 M yr −1 ), who assumed a smaller radius (0.9R given by Lacour et al. 2016).We assert that the larger radii inferred from the evolutionary model fit (Claudi et al. 2019) is more appropriate given the purported age of the system, and this yields a slight increase in mass accretion rate estimate.The values still agree within an order of magnitude.Our derived values are consistent with the value of 5.9 × 10 −10 M yr −1 reported by Close et al. (2014a); although they adopted a lower radius of 0.29R and a mass of 0.25 M , they did not conduct the same accounting for the Hα excess of the primary, which makes up the difference.Regardless of radius assumption, all derived accretion rates lie in a range consistent with observations of accretion rates onto low mass T-Tauri stars of similar masses and ages as HD 142527 B (e.g. 10 −9 − 10 −11 , Rigliaco et al. 2012).
Variability in mass accretion rates onto the primary stars of transitional disks has been observed previously.For example, the mass accretion rate of HD 142527 A changed by a factor of 7 over five years (Mendigutía et al. 2014).Variations in mass accretion rate on the order of factors of 2-10 on week-to-day long timescales have also been observed for accreting low mass stars more generally (e.g.Robinson & Espaillat 2019).We detect only marginally significant variability in the Hα contrast of HD 142527 B given our conservative uncertainties.To our knowledge, however, this is the first detection of accretion variability in a secondary companion within a transitional disk gap.
HD 142527 B is likely surrounded by a circumsecondary disk (Lacour et al. 2016) through which accreting material is processed.This circumsecondary disk is embedded within the cavity of the larger circumbinary disk, similar to the circumplanetary disk recently observed around PDS 70 c (Benisty et al. 2021).The non-detection results from Avenhaus et al. (2017) covered in §2 suggest that, if circumsecondary signal exists at the position of the companion, its contribution is much smaller than our photometric errors derived from Bayesian KLIP forward modeling.This motivates the need for a high-resolution ALMA search for circumsecondary material, as was recently done in the PDS 70 system (Benisty et al. 2021).
If the observed variability of HD 142527 B was due to the rotation of an accretion hotspot in and out of view, we would expect to see detectable continuum variability, similar to that observed in the NIR, but we do not.The observed variability could also be due to an accretion column rotating in and out of view.If that was the case, we might expect to observe large amplitude day-to-day variations between the 2015 observations, which we do not.The simplest explanation is that the accretion rate itself is variable over time, but more data are needed to verify this and to fully understand the degree and timescale of the variability of HD 142527 B.
Observations of accretion variability in planetary mass companions embedded within transition disk cavities have been difficult to date.The Hα lightcurve for PDS 70 b, for example, does not support large amplitude (> 30%) variability on month-long timescales (Zhou et al. 2021).Does this indicate that accretion onto objects orbiting within transition disk cavities is relatively stable when compared to accretion onto young stars?Recent modeling suggests that accretion onto a protoplanet from a circumplanetary disk may occur at a quasi-steady rate when averaged over week-long timescales, but should exhibit daily variability (Takasao et al. 2021).
The difficulty in obtaining accurate photometry of accretion emission onto planetary mass companions limits our ability to detect variations in their Hα excess.The next generation of instruments (e.g.MagAO-X, Males et al. 2020), which can achieve higher Strehl ratios in the visible, may be able to observe such variability for a wider range of embedded companions using similar methods.Our result could indicate that accretion onto companions within cavities is variable at a measurable level, at least for the highest mass companions.The caveat to this interpretation is that HD 142527 B is a stellar mass companion, whose accretion variability may be best understood in the domain of very-low-mass stellar accreting objects, rather than embedded planetary mass accretors.

CONCLUSIONS
In this paper, we present a five year monitoring campaign of the accreting companion HD 142527 B. We used the unique forward modeling capabilities of the Karhunen-Loeve Image Processing (KLIP) algorithm to achieve 1-2 mas precision on astrometric measurements taken over a 6 year time baseline, and validate an updated VisAO astrometric calibration solution (using the θ 1 Ori B2-B3 binary, see Appendix A) by demonstrating good agreement between VisAO observations of HD 142527 B and the literature.We combine literature astrometry and multi-epoch MagAO/VisAO observations to fit the orbit of HD 142527 B using orbitize!and derive a posterior distribution of orbital elements.We verify that the companion is on an inclined (i = 124.85± 4.56 • ), eccentric (e = 0.24 ± 0.15) orbit and is near periastron passage.We find that the HD142527 binary has a mutual inclination with respect to the outer disk of θ −out = 89.84+2.30   −1.65 • or 158.82 +2.76   −2.81   • (depending on the degenerate position of the ascending node, Ω).We also find a dramatic mutual inclination with respect to the newly directly detected inner disk, θ in− = 177.44+2.90 −2.80 • or 110.26 +1.82   −1.50 • .These newly derived inclinations could be used to guide hydrodynamical models of this system, in the context of disk warping, tearing, and precession in the presence of a disruptive companion.
While HD 142527 B may be too tightly separated at the time of writing to be detected without interferometric techniques, astrometry of the companion following periastron passage (∼ 2021) will be crucial in further constraining its orbit and therefore its mass.An improved mass estimate and updated characterization can yield, among other things, more precise accretion rate estimates.A single observation with the VLTI/GRAVITY interferometer, for instance, could track the companion's periastron passage, provide an improved mass constraint, and even place estimates (or upper limits) on the Br-γ emission from HD 142527 B. With an improved mass constraint, the companion could be better compared to evolutionary models, yielding an improved age determination and a better understanding of the companion's formation.GRAVITY observations could also detect or place additional constraints on the size of the circumsecondary disk around HD 142527 B, as was done for the PDS 70 planets (Wang et al. 2021).High resolution ALMA observations could directly detect the presence of a circumsecondary disk around B, provide constraining astrometry, and explore the inner disk surrounding A. In the future, fiber-fed spectroscopy of the companion with an instrument such as KPIC may allow for the measurement of the radial velocity, absorption, and accretion signatures of the companion at high spectral resolution, which could help constrain its currently uncertain age and spectral type (M2.5-M7).
Leveraging careful optimization of the pyKLIP algorithm, we achieved the most finely separated detection of a faint (∆mag > 6) directly imaged companion using a non-coronagraphic, non-interferometric instrument to date.We observe clear Hα excess in all epochs of observation, corresponding to mass accretion rates similar to those observed in young, isolated M-dwarfs.We observe tentative signs of variability in the Hα excess of the companion, suggestive of accretion variability.We estimate accretion rates for the HD 142527 B companion on the order of 4 − 9 × 10 −10 M yr −1 , assuming a radius based on evolutionary models.
Our results demonstrate that careful, long timebaseline observations from the current generation of high-contrast imaging instruments, combined with improvements in post-processing techniques, are able to place substantial constraints on both orbital motion and photometric variability, even for very tightly-separated directly imaged companions, provided a self-consistent data reduction and post-processing methodology.In the future, similar observations of systems such as PDS 70, LkCa 15, and AB Aur b (and other, newly discovered accreting protoplanets) with instruments such as MagAO-X will open new windows into the time variability of protoplanetary accretion and the process by which substellar companions form and evolve.

ACKNOWLEDGEMENTS
We sincerely thank the anonymous reviewer for their thoughtful, rigorous, and supportive review that contributed immensely to the improvement of this paper.Special thanks to Gabriel-Dominique Marleau and Yuhiko Aoyama for their fruitful discussions of accretion physics.We thank Sarah Blunt for her tremendous help with our orbit fitting, Jason Wang for his pyklip expertise, as well as Connor Robinson for his encouragement.Special thanks to David Sing for his encouragement (and for letting WOB use his server to run last minute orbits).WOB would like to thank Yevaud, Kalessin, and Morgoth, as well as Benjamina, Martin, Luke, Emmett, Jack, and Kate Balmer.
WOB and KBF acknowledge funding from NSF-AST-2009816.WOB thanks the LSSTC Data Science Fellowship Program, which is funded by LSSTC, NSF Cybertraining Grant #1829740, the Brinson Foundation, and the Moore Foundation; their participation in the program has benefited this work.KBF's work on this project was also supported by a NASA Sagan fellowship.LMC's work was supported by NASA Exoplanets Research Program (XRP) grants 80NSSC18K0441 and 80NSSC21K0397.KMM's work has been supported by the NASA XRP by cooperative agreement NNX16AD44G.
This paper includes data gathered with the 6.5 meter Magellan Telescopes located at Las Campanas Observatory, Chile.Some of the data presented herein were obtained at the W.M. Keck Observatory, which is operated as a scientific partnership among the California Institute of Technology, the University of California and the National Aeronautics and Space Administration.The Observatory was made possible by the generous financial support of the W.M. Keck Foundation.
This work has made use of data from the European Space Agency (ESA) mission Gaia (https://www.cosmos.esa.int/gaia),processed by the Gaia Data Processing and Analysis Consortium (DPAC, https://www.cosmos.esa.int/web/gaia/dpac/consortium).Funding for the DPAC has been provided by national institutions, in particular the institutions participating in the Gaia Multilateral Agreement.
WOB and KBF would like to acknowledge the land that the images used in this paper were observed from.Las Campanas Observatory, and the Magellan Clay Telescope, are built on Diaguita land.More on the Diaguita is available from the Museo Chileno de Arte Precolombino: http://precolombino.cl/en/culturasamericanas/pueblos-originarios-de-chile/diaguita/, and additional information from Diagutia activists can be found: https://upndsalta.blogspot.com.
The W. M. Keck Observatory, and the Keck II Telescope, are built on Native Hawaiian land.We are honored to be given the opportunity to conduct astronomy using data taken from this sacred place and would like to point to the informative Astrobites articles ("Mauna Kea and Modern Astronomy," https://astrobites.org/2018/11/09/mauna-kea-and-modern-astronomy/, "Maunakea, Western Astronomy, and Hawai'i," https://astrobites.org/2019/08/02/maunakea-western-astronomy-andhawaii/.We would like to encourage our colleagues to seek out additional information about the on-going protests against additional construction on the mountain and the historical precedent for these protests.
The authors would like to acknowledge the land they have conducted research from during the course of this investigation.WOB and KBF would like to acknowledge the Nonotuck land Amherst College occupies, the Nonotuck ancestors, their descendants, and the neighboring Indigenous nations: the Nipmuc and the Wampanoag to the East, the Mohegan and Pequot to the South, the Mohican to the West, and the Abenaki to the North.WOB would also like to acknowledge the Piscataway community, their elders and ancestors, as well as their future generations.WOB acknowledges that Johns Hopkins was founded upon the exclusions and erasures of many Indigenous peoples, including those on whose land this institution is located.errors on the platescale than previous calibrations, and smaller errors on the North Angle offset.It would appear that by measuring the platescale in only one filter with a limited number of observations, previous calibrations underestimated the platescale error, and that a longer baseline of observations allows improved precision in determination of the North Angle offset.
Figure 8 plots measurements of the VisAO platescale over time and across multiple filters.Figure 9 shows the orbital fit to the NIRC2 observations of θ 1 Ori B2-B3 and updated VisAO astrometry of the binary pair.
As a byproduct of this astrometric calibration, we have obtained astrometric measurements of the θ 1 Ori B2-B3 over nearly 20 years, as well as an updated, well-defined orbital solution for the binary.We record the astrometry from both NIRC2 and VisAO with other literature astrometry in Table 7.We record the orbital elements as drawn from the posterior distribution of fits to the NIRC2 astrometry in Table 8.

B. FORWARD MODEL CHOICE AND VISAO GHOST CALIBRATION
The VisAO CCD saturates at ∼ 16, 000 counts, and because of its small FOV, there are no other stars within the field that can be used as a PSF to forward model.There are no artificial satellite spots injected as astrometric or photometric calibrators, but there is an instrumental 'ghost' PSF that appears to the right of the natural guide star in each image (see Figure 10).We investigated the stability of the ghost and the scaling relationships between the ghost and central PSF using 10 unsaturated datasets taken as part of the GAPlanetS survey (Follette et al. 2017, andFollette et al. 2022, in prep.).
We find that peak of the ghost varies consistently with the peak of the central PSF.We adopt the empirical scaling relationships F c,Hα = 179.68± 4.59 × F g,Hα and F c,cont.= 196.31± 3.56 × F g,cont.However, we find that when fit with a Moffat distribution, the FWHM of the ghost is on average 7% larger than that of the central PSF.We therefore adopt the relationship FWHM c = 0.93 × FWHM ghost in determining the FWHM of saturated images.
This ghost calibration was conducted in part because we had initially used the instrumental ghost as a forward model PSF, and found it to be a poor fit.We then investigated the best choice of forward model for saturated data by comparing sum-of-squares residuals for the ghost itself, a Moffat distribution fit to the ghost, a Gaussian distribution fit to the ghost, and both distributions (Moffat and Gaussian) with reduced FWHM equal to 0.93 × FWHM ghost .For each PSF, we conducted the same procedure as in Section 4, forward modeling the PSF through KLIP and fitting it to the known companion, resulting in a posterior distribution of fits and a residual map.We found that for the majority of our images of HD 142527 B, Gaussian PSFs yielded the smallest residuals.We assume that, scaling by the above relationships, the counts under the Gaussian are equivalent to those under the unsaturated central PSF.This then enables us to conduct photometry using BKA.Males et al. (2014) using data taken in the Ys filter in 2014.Note the agreement between our measurement of the same Ys data (red diamond) and their value.We note that the Hα observations suffered from poor observing conditions and an unfavorable observing strategy for astrometry which resulted in very low signal-to-noise on the B2-B3 pair.No obvious trend in platescale with wavelength or time is present, and therefore we adopt the weighted average and standard deviation on the weighted average (represented by the blue dashed line and shaded region, respectively) as the updated platescale and platescale error for the instrument.

C. FORWARD MODEL FITS
Optimized post-KLIP images for each detection epoch, best fit BKA models, and the residuals between them are shown for HD 142527 B in the Hα and continuum filters in Figures 11 and 12, respectively.The marginal posterior distributions for the BKA fits, an example of which is shown in Figure 13, are distributed normally for all epochs.The most common correlation is a slight linear correlation between X and Y position (as seen in Figure 13).

D. ORBIT FIT POSTERIORS
This appendix details posterior distributions of orbital elements for our astrometric fits, computed using orbitize!. Figure 14 illustrates the posterior distribution of orbital elements for the HD 142527 AB binary.Figure 15 plots the mutual inclination parameters i , Ω from Figure 14, computed as described in Section 5, along with i d , Ω d drawn from gaussian distributions specified by the disk parameters fit by Bohn et al. (2021), and the resultant distribution of mutual inclinations θ. Figure 16 illustrates the posterior distribution for the orbit of the θ 1 Ori B2-B3 derived from Keck NIRC2 observtions, as described in Appendix A.
[mas]  M T [M ]   Figure 16.Posterior distribution of orbital elements fit to NIRC2 astrometry of θ 1 Ori B2-B3.The fit displays a posterior distribution typical of short-arc visual orbit fits, with the same bimodal ω and Ω noted previously, normally distributed π and Mtot, correlated semi-major axis, eccentricity, and inclinations that are otherwise relatively constrained.This orbit is so satisfyingly typical, I have petitioned for it act as the unofficial mascot of orbitize!but I have not yet been humored thusly.

Figure 1 .
Figure 1.Gallery of post-KLIP images showing the detections of HD 142527 B in Hα (top) and continuum (bottom).The colorbar normalized to the peak pixel value of the companion in each image.The cyan circle indicates the nearest in time position of HD 142527 B reported in previous literature(Lacour et al. 2016; Claudi et al. 2019).The cyan star indicates the position of HD 142527 A. The innermost pixels have been masked to r ∼ 1×FWHM for each dataset.

Figure 2 .
Figure 2. Representative pyklip forward model fits for HD 142527 B. Shown are the 2013-04-11 Hα (top row) and continuum (bottom row) detections.Based on the PlanetEvidence analysis, the evidence ratios log Z1/Z0 are 266.30and 75.89 in Hα and continuum, respectively.Both are considered extremely strong evidence in favor of the existence of the companion at this location.Our orbitize!orbit fitting yields well-converged unimodal distributions in all parameters except ω and Ω, which show bimodal distributions with peaks spaced 180 • apart.This a known degeneracy in visual orbits with a lack of RV constraints 8 .To avoid confusion, we report values of the first (0 − 180 • ) modes of ω and Ω in Table5, noting that solutions with values 180 • higher are equally likely.Previous orbital solutions to the motion of HD 142527 B have have shown it to be both inclined (i ∼ 125 ± 5• Lacour et al. 2016) and eccentric (e >0.2Lacour et al. 2016; Claudi et al. 2019).Our unimodal eccentricity distribution (e = 0.24 ± 0.15) agrees with the first family of eccentricities fit byClaudi et al. (2019) superimposes (a) 50 randomly drawn orbits from the posterior distribution of orbit fits, (b) a single epoch VisAO detection of the Hα point source, and (c) an H-band polarized intensity observation of the HD142527 outer disk.The image of the circumbinary disk was obtained by re-reducing GPI polarimetric data first published in Rodigas et al. (2014) using the updated GPI DRP described in De Rosa et al. (

Figure 3 .
Figure 3. Astrometric measurements of HD 142527 B relative to HD 142527 A (black star) from NACO, GPI, SPHERE,SINFONI, and VisAO (this work)  between 2012 and 2018.The crosses represent the reported uncertainties on each measurement.The companion experiences significant orbital motion (∆θ ≈ 65 • ) and decreases substantially in separation (∆ρ ≈ 30 mas) from 2016 to 2018.The VisAO values (yellow circles) agree well with astrometry from other instruments.Error bars on our measurements reflect the 1-2mas accuracy achievable using pyklip forward modeling.

Figure 4 .
Figure 4. 100 randomly drawn orbitize!orbits fit to the motion of HD 142527 B. Astrometry from Figure 3 is overplotted.Left: the orbits projected in RA and Dec, relative to HD 142527 A (black star).Right: the orbits in separation/position angle versus time.The new VisAO astrometry does not add new coverage of the orbital arc, but the 2018 recovery adds more weight to the near-in-time SPHERE NRM astrometry.

Figure 5 .
Figure 5.The orbit of HD 142527 B in context.The position of HD 142527 A is marked with a cyan star, and 250 randomly drawn orbits from fits to the astrometry of HD 142527 B are plotted with colors corresponding to their mutual inclination with the outer disk (θ −out).The outer image of the circumbinary disk is a Gemini/GPI scattered light polarized intensity H-band image reprocessed with an updated GPI pipeline and interpolated to VisAO's platescale.The inner image is the post-KLIP 2015/05/15 Hα detection from VisAO (also shown in Figure 1).Both images are normalized to their respective maximum pixel values before combination.

Figure 6 .
Figure6.The visible-light contrast of HD 142527 B with respect to the continuum of HD 142527 A over time.Brown and green diamonds mark our Hα and continuum contrast measurements, respectively.The grey shaded region represents the 1σ standard deviation of our measurements of continuum contrast, centered on the median.The red and blue squares mark the contrasts measured in nearly equivalent filters byCugno et al. (2019) using VLT/SPHERE/ZIMPOL.The amount of Hα excess appears to vary between many of the epochs, while the continuum contrast varies minimally and is consistent with uniformity within error bars.(Gullbring et al. 1998) assuming R in ∼ 5R , as inRigliaco et al. (2012).We adopt the dynamical mass of the companion (M B = 0.26M ) and best fit BHAC evolutionary model radius (R B = 1.2R ) fromClaudi et al.  (2019)  to compute our final accretion rate estimate from this equation.Table6records the line luminosities and mass accretion rates for each epoch calculated following these assumptions.The peak Hα excess, which occurs in the 2014-04-08 epoch, corresponds to a mass accretion rate estimate of 6 × 10 −10 M yr −1 .

Figure 9 .
Figure9.500 orbital fits to the NIRC2 astrometry of θ 1 Ori B2-B3 randomly drawn from the posterior distribution.Astrometry compiled in Table7is overplotted.Only the NIRC2 astrometry (orange triangles) was used to fit the orbit of the binary, and the corrected VisAO astrometry (red diamonds) falls along the fit orbits by design.

Figure 10
Figure10.A saturated image of HD 142527.The position of the instrumental ghost is marked with a black arrow.The Y axis is in pixels, and the X axis is labeled in milliarcseconds to illustrate the spatial extent of the 451 pixel crop and the position of the ghost.

Figure 11 .Figure 12 .
Figure 11.A gallery of BKA forward model best fits to HD 142527 B in the Hα filter, in chronological order from top to bottom.Data (left) is fit by BKA, yielding a best fit forward model (center), and their difference (right).

Figure 13 .
Figure13.A corner plot illustrating the posterior distribution of BKA forward model fits to the 2013-04-11 Hα data, which is representative of all other epochs.Note that all marginal parameters are normally distributed, and the only correlation is a slight linear correlation between X and Y position.

Figure 15 .
Figure15.Posterior distribution of fit (i , Ω ) and randomly sampled (i disk , Ω disk ) for both inner and outer disks yields a posterior distribution of θ, the mutual inclination angle between the binary orbit and the disk component.As inCzekala et al. (2019), the mutual inclination of the outer disk for this system is multimodal, but dramatically misaligned (θ 3 • ) no matter choice of Ω .Interestingly, one family of θ −out is nearly perpendicular, similar to the configuration described inPrice et al. (2018).