Draft version March 20, 2024

Typeset using LATEX preprint style in AASTeX631

Unveiling MOA-2007-BLG-192: An M Dwarf Hosting a Likely Super-Earth

Sean K. Terry ,1, 2 Jean-Philippe Beaulieu ,3, 4 David P. Bennett ,1, 2 Euan Hamdorf,4
Aparna Bhattacharya,1, 2 Viveka Chaudhry,5 Andrew A. Cole ,4 Naoki Koshimoto ,6
Jay Anderson ,7 Etienne Bachelet,8 Joshua W. Blackman ,9 Ian A. Bond ,10
Jessica R. Lu ,11 Jean Baptiste Marquette ,12, 3 Clément Ranc ,3
arXiv:2403.12118v1 [astro-ph.EP] 18 Mar 2024

Natalia E. Rektsini ,4, 3 Kailash Sahu ,7 and Aikaterini Vandorou 1, 2

1
Department of Astronomy, University of Maryland, College Park, MD 20742, USA
2
Code 667, NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA
3
Sorbonne Université, CNRS, Institut d’Astrophysique de Paris, IAP, F-75014 Paris, France
4
School of Natural Sciences, University of Tasmania, Private Bag 37 Hobart, Tasmania, 70001, Australia
5
Sidwell Friends School, Washington, DC 20016, USA
6
Department of Earth and Space Science, Graduate School of Science, Osaka University, Toyonaka, 560-0043, Japan
7
Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
8
IPAC, Caltech, Pasadena, CA 91125, USA
9
Physikalisches Institut, Universität Bern, Gessellschaftsstrasse 6, CH-3012 Bern, Switzerland
10
School of Mathematical and Computational Sciences, Massey University, Auckland 0745, New Zealand
11
Department of Astronomy, University of California Berkeley, Berkeley, CA 94701, USA
12
Laboratoire d’Astrophysique de Bordeaux, CNRS, B18N, allée Geoffroy Saint-Hilaire, Pessac, France

ABSTRACT
We present an analysis of high angular resolution images of the microlensing target
MOA-2007-BLG-192 using Keck adaptive optics and the Hubble Space Telescope. The
planetary host star is robustly detected as it separates from the background source star
in nearly all of the Keck and Hubble data. The amplitude and direction of the lens-
source separation allows us to break a degeneracy related to the microlensing parallax
and source radius crossing time. Thus, we are able to reduce the number of possible
solutions by a factor of ∼2, demonstrating the power of high angular resolution follow-
up imaging for events with sparse light curve coverage. Following (Bennett et al. 2023),
we apply constraints from the high resolution imaging on the light curve modeling to
find host star and planet masses of Mhost = 0.28 ± 0.04M⊙ and mp = 12.49+65.47 −8.03 M⊕
at a distance from Earth of DL = 2.16 ± 0.30 kpc. This work illustrates the necessity
for the Nancy Grace Roman Galactic Exoplanet Survey (RGES ) to use its own high
resolution imaging to inform light curve modeling for microlensing planets that the
mission discovers.

Keywords: gravitational lensing: micro, planetary systems

Corresponding author: S. K. Terry

[email protected]
2 Terry et al.

1. INTRODUCTION
Since the early 1990’s, surveys of the galactic bulge have searched for variations in the brightness
of background stars (sources) induced by the gravitational field of foreground objects (lenses). The
number of lensing events detected has dramatically increased from a few dozen per year in the 1990’s
(Udalski et al. 1994; Alcock et al. 1996) to thousands per year currently. At present there are three
primary ground-based surveys that contribute to these lensing event detections: OGLE (Udalski
et al. 2015), MOA (Bond et al. 2001), and KMTNet (Kim et al. 2016). NASA’s Nancy Grace Roman
Space Telescope (Roman) is scheduled to launch in the next several years and will conduct the Roman
Galactic Bulge Time Domain Survey (GBTDS ). As part of this bulge survey, the Roman Galactic
Exoplanet Survey (RGES ) will be the first dedicated space-based gravitational microlensing survey
and is expected to detect over 30,000 microlensing events and over 1400 bound exoplanets during its
five-year survey (Penny et al. 2019). This mission will complement previous large statistical studies
of transiting planets from missions like Kepler /TESS and radial velocity planets from many ground-
based RV surveys. The GBTDS is also expected to discover free-floating planets that do not orbit
any host star (Johnson et al. (2020); Koshimoto et al. (2023); Sumi et al. (2023), Johnson et al. in
prep).
As of this writing, microlensing has detected ∼200 planets at distances up to the Galactic Bulge.
As for most transient phenomena, one limitation of this method for fully characterizing microlensing
systems is the cadence at which the photometric data is obtained by the dedicated ground-based
surveys. An effective way to increase the sampling for a given microlensing event is to issue a public
alert so that observatories around the world can observe ongoing events as a ‘follow-up’ network of
telescopes. MOA-2007-BLG-192 was the first planetary microlensing event detected without follow-
up observations from other observatories. The initial analysis reported a low-mass planet orbiting a
very-low mass host star or brown dwarf (Bennett et al. 2008). Due to the lack of follow-up network
data for this microlensing event, there are significant gaps in the photometric light curve coverage,
which leads to uncertainties in the derived lens system parameters. There are also additional degen-
eracies in the interpretation of this lens system that arise from the possible planet-star separations
and microlensing parallax. The details of these degeneracies are discussed further in Section 2.1.
One way to mitigate some of these degeneracies is by resolving the source and lens independently
with high angular resolution imaging (i.e. Hubble Space Telescope (HST), Keck AO, Subaru AO)
several years after peak magnification (Bennett et al. 2006, 2007). This high angular resolution
imaging can enable measurements of the lens-source separation, relative proper motion, and lens flux
which can then be used with mass-luminosity relations (Henry & McCarthy 1993; Henry et al. 1999;
Delfosse et al. 2000) to calculate a direct mass for the host.
This current analysis is part of the NASA Keck Key Strategic Mission Support (KSMS) program,
“Development of the WFIRST Exoplanet Mass Measurement Method” (Bennett 2018), which is
a pathfinder project for the Nancy Grace Roman Space Telescope (formerly known as WFIRST )
(Spergel et al. 2015). For several years now, the KSMS program has measured the masses of many
microlensing host stars and their companions (Bhattacharya et al. 2018; Vandorou et al. 2020; Ben-
nett et al. 2020; Blackman et al. 2021; Terry et al. 2021, 2022), all of which are included in one
of the most complete statistical studies of the microlensing exoplanet mass ratio function (Suzuki
et al. 2016, 2018). This statistical study shows a break and likely peak in the mass-ratio function for
wide-orbit planets at about a Neptune mass which is at odds with the runaway gas accretion scenario
 MOA-2007-BLG-192 3

of the leading core accretion theory of planet formation (Lissauer 1993; Pollack et al. 1996), which
predicts a planet desert at sub-Saturn masses (Ida & Lin 2004) for gas giants at wide orbits.
This paper is organized as follows: In Section 2 we present the light curve re-analysis of MOA-
2007-BLG-192 and explain the challenges in the modeling posed by lack of photometric coverage and
degeneracies. In Section 3, we describe the high angular resolution HST and Keck adaptive optics
(AO) observations and analysis. Section 4 details our direct measurement of the lens system flux
and lens-source separation in the Keck and HST data which allows us to reduce the total number of
solutions. Section 5 describes the newly derived lens system properties from the light curve modeling
that incorporates the high resolution imaging results. Finally, we discuss the overall results and
conclude the paper in Section 6.
2. PRIOR STUDIES OF THE MICROLENSING EVENT MOA-2007-BLG-192
2.1. Fitting the microlensing light curve
MOA-2007-BLG-192 (hereafter MB07192), located at RA (J2000) = 18:08:03.80, DEC (J2000) =
−27:09:00.27 and Galactic coordinates (l, b = (4.03, −3.39)) was first alerted by MOA on May 24,
2007. Due to the faintness of the source and poor weather at the MOA telescope, the event was not
alerted until the day that the planetary deviation was observed in the light curve.
Figure 1 shows the observed light curve with OGLE (blue) and MOA (red) data as well as the best-
fit planetary model (2L1S) from our re-analysis of the light curve modeling. The original light curve
analysis for this event was presented by Bennett et al. (2008) (hereafter B08). The only photometric
monitoring of the target during magnification was conducted in OGLE-I and MOA-R bands. Due
to the faintness of the source, there is no direct V -band measurement of the target from OGLE or
MOA. In order to get a source color estimate, earlier studies used the photometric measurements
from these two data sets and converted to (V-I) color following Gould et al. (2010). As apparent in
Figure 1, there are significant gaps in the photometric coverage for this event. Due to this incomplete
coverage, there are multiple binary lens solutions, with similar mass-ratio that can equally explain
the deviations in the light curve due to a binary lens system.
This lack of coverage also resulted in large uncertainties in the measurement of the angular source
size and a poorly determined angular Einstein radius, θE . However, these various solutions all gave
a low mass planetary system with a mass ratio of q ∼ 2 × 10−4 , and with quite large errors on
the reported q’s. Using the constraints from microlensing parallax and the source star size, B08
concluded that the system was composed of a 0.06+0.028 +4.9
−0.021 M⊙ object orbited by a 3.3−1.6 M⊕ super-
Earth. We note at the time of the B08 publication the MOA team was unaware of systematics in
their photometry due to chromatic differential refraction effects (Bennett et al. 2012a). This led
to an erroneous measurement of microlensing parallax (πE ) reported in their study. Further, the
caustic-crossing models presented in B08 contributed to relatively small error bars on the derived
planet mass (see Figure 5 in B08). These caustic-crossing models have now been largely ruled out
by this study, therefore the planet mass error bars have increased (e.g. Section 5).

2.2. Constraining the lensing system with adaptive optics

Kubas et al. (2012) (hereafter K12) obtained two epochs with NACO AO imaging on the Very Large
Telescope (VLT) shortly after the peak of the microlensing event when the target was still magnified
by a factor of 1.23, as well as 18 months later at baseline. They observed in three bands, J, H, KS ,
4 Terry et al.

this was the first microlensing event for which a fairly large AO data set had been obtained. The
AO data was reduced with the Eclipse package (Devillard 1997) and the authors performed PSF
photometry using the Starfinder tool (Diolaiti et al. 2000). The absolute calibration was performed
by a two-stage process using 2MASS and data collected by the IRSF telescope in South Africa.
Knowing the source flux from the microlensing fit, they detected excess flux in all three near-IR
bands. Assuming that all the excess flux comes from the lens, they obtained new constraints on the
lensing system. Combining the results of the two epochs, they derived that the lens has the following
magnitudes: JL = 20.73 ± 0.32, HL = 19.94 ± 0.35, KL = 19.16 ± 0.20. Using these constraints,
and the (erroneous) microlensing parallax fit by B08, they concluded that the lensing system is a
0.084+0.015 +100 +5.2 +0.51
−0.012 M⊙ M dwarf at a distance of 660−70 pc orbited by a 3.2−1.8 M⊕ super-Earth at 0.66−0.22
AU.

2.3. Why revisiting this system?

MB07192 is an important event from the Suzuki et al. (2016) sample of cold planets. Its mass ratio
is in the region where a change of slope has been observed in the mass-ratio function. Also MOA have
recently improved their photometry methods, so we have re-reduced the MOA photometry following
Bond et al. (2017). This re-reduction includes corrections for systematic errors due to chromatic
differential refraction (Bennett et al. 2012a). This has direct consequences on the microlensing model
compared to the initial studies, which affects the fitting parameters like microlensing parallax, finite
size of the source star, and other higher order effects. Additionally, over the years we have refined
our procedures to process, analyze, and calibrate AO data as well as update extinction correction
calculations. We will therefore adopt our standard method described by (Beaulieu et al. 2018) and
re-analyze the KS NACO data.
Finally, we have obtained recent Keck-NIRC2 and HST observations in 2018 and 2023, which should
give us the opportunity to independently resolve the source and lens and measure the magnitude and
direction of their relative proper motion.

Table 1. New HST and Keck observations in this work

Epoch (UT) Instrument PA Filter Texp Nexp ∆t Reference

(yyyy-mm-dd) (deg) (sec) (yr)
2012-03-30 WFC3−UVIS 131.8 F606W (V) 1760 8 4.85 (a)
F814W (I) 1640 8
2014-03-30 WFC3−UVIS 131.8 F606W (V) 1760 8 6.85 (b)
F814W (I) 1640 8
2018-08-06∗ NIRC2 0.0 KS 900 15 11.20 (c)
2023-08-06 WFC3−UVIS 309.5 F814W (I) 600 2 16.20 (d)
Note: ∆t gives the amount of time (in years) since the peak of the microlensing event.
∗
The 2018 epoch is from Keck, all other epochs are from HST.
References. (a) Bennett et al. (2012b), (b) Bennett et al. (2014), (c) Bennett (2018), (d) Sahu et al. (2023)
 MOA-2007-BLG-192 5

400 MOA R
OGLE I
magnification 300

200

100

0
4242 4244 4246 4248 4250

400 2L1S
1L1S
magnification

300

200

100
residual

10
0
10
4245 4246
HJD 2450000

Figure 1. Best fit light curve with constraints from the high resolution follow-up data as described in
section 3. The 2L1S model shown is from the second column of Table 6 with u0 < 0 and s < 1.

3. HIGH ANGULAR RESOLUTION FOLLOW-UP WITH HST AND KECK

3.1. Preparing the absolute calibration data set
We use our own re-reduction of data from the VVV survey (Minniti et al. 2010) obtained with
the 4m VISTA telescope at Paranal (see Beaulieu et al. (2018)). We cross identified these JHKS
catalogues with the VI OGLE-III map (Udalski et al. 2015). We then obtain an OGLE-VVV cata-
logue of 8500 objects with V IJHKS measurements, covering the footprint of the HST and KECK
observations. We subsequently used this catalog to calibrate HST and KECK data, and we also
revisit the VLT/NACO data.
3.2. Keck NIRC2
6 Terry et al.
N Keck K
E

Star 1

Star 2

0.35"

1-star residual 2-star residual

0 4 16 28 40 52 64 76 88
Counts
Figure 2. Top Left: The co-added sum of 15 Keck NIRC2 narrow camera images, each with an exposure
time of 60 seconds. The target is indicated with a square outline. Top Right: Zoomed image of the MB07192
blended source and lens stars. The magnitude of the separation in this epoch is 29.3 ± 1.1 mas. Bottom
Left: The residual image from a single-star PSF fit with DAOPHOT. A clear signal is seen due to the
blended stellar profiles. Bottom Right: The residual image for a simultaneous two-star PSF fit, showing a
significantly improved subtraction. The color bar represents the pixel intensity (or counts) in the bottom
panel residual images.

The target MB07192 was observed with the NIRC2 instrument on Keck-II in the Kshort band
(λc = 2.146µm, hereafter Ks) on August 5 and 6, 2018. The two nights of data were combined using
the KAI reduction pipeline (Lu 2022). The pipeline registers the images together, applies flat field
correction, dark subtraction, as well as bad pixel and cosmic ray masking before producing the final
combined image that we analyze. The data from both nights are of similar quality, with an average
point spread function (PSF) full width at half maximum (FWHM) of 66.2 mas for the August 5 data,
and 67.5 mas for the August 6 data.
For the 2018 Ks band observations, both the NIRC2 wide and narrow cameras were used. The
pixel scales for the wide and narrow cameras are 39.69 mas/pixel and 9.94 mas/pixel, respectively.
All of the images were taken using the Keck-II laser guide star adaptive optics (LGSAO) system. For
the narrow data, we combined 15 flat-field frames, six dark frames, and 15 sky frames for calibrating
 MOA-2007-BLG-192 7
Table 2. HST, VLT NACO, and Keck single-star PSF photometry

Data set V I J H KS
HST 2012 23.88 ± 0.02 20.93 ± 0.01 19.04 ± 0.01 18.28 ± 0.01
HST 2014 23.83 ± 0.02 20.90 ± 0.01
K12 NACO ep.1 19.21 ± 0.04 18.28 ± 0.04 17.95 ± 0.04
K12 NACO ep.2 19.32 ± 0.07 18.55 ± 0.11 17.99 ± 0.04
NACO ep.1 17.80 ± 0.05
NACO ep.2 17.92 ± 0.05
KECK 2018 17.88 ± 0.05
HST 2023 21.01 ± 0.04
Note: We provide the magnitudes measured at the source position for MB07192. We recall the measured magnitudes
from K12 for the two epochs. We underline that the time of the first epoch, the source was still amplified by ∼ 0.15
mag. We re-analyzed the NACO KS images and calibrated against VVV for the two epochs. Finally, we provide our
flux calibration in KS with Keck-NIRC2.

the science frames. A total of 15 Ks band science frames with an integration time of 60 seconds
per frame were reduced using KAI which corrects instrumental aberrations and geometric distortion
(Ghez et al. 2008; Lu et al. 2008; Yelda et al. 2010; Service et al. 2016).
Because of the potentially significant effects of a spatially varying PSF in ground-based AO imaging
(Terry et al. 2023), we made a careful selection of bright and isolated reference stars that were used
to build the empirical PSF model. Each of the eight selected PSF reference stars has magnitude
−0.7 < m < 0.7 and separations −4′′ < r < 4′′ from the target. The resulting PSF model has a
FWHM in the x and y directions of 6.8 pixels and 6.5 pixels, respectively.
Further, a co-add of 4 wide camera images were used for photometric calibration using the cat-
alogue prepared in section 3.1. The wide camera images were flat-fielded, dark current corrected,
and stacked using the SWarp software (Bertin 2010). We performed astrometry and photometry
on the co-added wide camera image using SExtractor (Bertin & Arnouts 1996), and subsequently
calibrated the narrow camera images to the wide camera image by matching 80 bright isolated stars
in the frames. The uncertainty resulting from this procedure is 0.05 magnitudes.

3.3. Resolving the Source and Lens in Keck/NIRC 2

Given the lens detections from HST 2012 and 2014 data, the lens and source stars have a predicted
separation of 0.65×FWHM in 2018, we expect the stars to be partially resolved, so it is necessary to
use a PSF fitting routine to measure both targets separately. Following the methods of Bhattacharya
et al. (2018) and Terry et al. (2021), we use a modified version of the DAOPHOT-II package (Stetson
1987), that we call DAOPHOT MCMC, to run Markov Chain Monte Carlo (MCMC) sampling on
the pixel grid encompassing the blended targets. Further details of the MCMC routine are given in
Terry et al. (2021) and Terry et al. (2022).
8 Terry et al.

The stellar profile does not appear to be significantly extended in the NIRC2 data, as seen in
the top-right panel of Figure 2. However, using DAOPHOT MCMC to fit a single-star PSF to the
target produces the residual seen in the lower-left panel of Figure 2, which shows a strong signal
due to extended flux from the blended star (presumed lens). Re-running the routine in the two-star
fitting mode (e.g. simultaneously fit two PSF models) produces a significantly better fit as expected,
with a χ2 improvement of ∆χ2 ∼ 784. The two-star residual is nearly featureless, as can be seen
in the lower-right panel of Figure 2. Table 4 shows the calibrated magnitudes for the two stars of
K1 = 18.94 ± 0.10 and K2 = 18.39 ± 0.09.
The final error bars on the Keck photometry and astrometry that we report in Tables 4 and 5 are
determined with a combination of MCMC and jackknife errors. The jackknife method (Quenouille
1949, 1956; Tierney & Mira 1999) allows us to determine uncertainties due to PSF variations between
individual Keck images. From the total of 15 Keck images, we construct N = 14 co-added jackknife
images, with each combined image containing all but one successive image in each iteration. This
method is also sometimes called the “drop-one” or “leave-one-out” method. The jackknife errors are
then calculated via the equation:
r
N −1X
σx = (xi − x̄)2 , (1)
N
where xi is a given value for the ith jackknife image, and x̄ is the mean value for the jackknife images.
See Bhattacharya et al. (2021) and Terry et al. (2022) for further details on the jackknife method.
From the dual-star PSF fitting in Keck, we find a difference in K-band magnitude between the
two blended stars of KS1 − KS2 = −0.55 ± 0.13. Since the two stars are similar enough magnitude
in K, at this point we simply apply arbitrary labels of ‘star 1’ and ‘star 2’ to the two stars in Keck.
However, our subsequent analysis of the HST data will allow us to confidently determine which star
is the source and which is the lens (Section 3.7).
3.4. HST WFC3/UVIS: 2012-2014-2023 Data
The target MB07192 was observed a total of three times with the WFC3/UVIS camera on the
Hubble Space Telescope (HST ). The first observation took place on 23 February 2012 in the F555W,
F814W, F125W and F160W filters. A second epoch of observations were obtained on 30 March
2014 with the same four filters, and finally a third epoch was obtained on 06 August 2023 with
just two exposures in the F814W filter. The datasets are from proposals GO-12541, GO-13417
(PI: Bennett), and GO-16716 (PI: Sahu), and were obtained from the Mikulski Archive for Space
Telescopes (MAST). We flat-fielded, stacked, corrected for distortions and performed PSF photometry
with the program DOLPHOT (Dolphin 2000). Because of the disparate sensitivities, the visible
images obtained with the UVIS module (F555W and F814W) and the near-IR images obtained with
the IR module (F125W and F160W) were reduced separately.
The drizzled, stacked frames with the astrometric solutions from STScI were used as the reference
image for source finding. We used Dolphot to correct for pixel area distortions, remove cosmic rays,
and perform PSF fitting photometry of the individual frames. Because of the crowded nature of the
field, the sky background was determined iteratively and many artifacts due to bright stars were
rejected. Dolphot uses a library of reference PSFs for each filter and applies corrections to these dur-
ing photometry to minimize the residuals when the photometered stars are subtracted. The PSF-fit
magnitudes are aperture corrected to a standard circular aperture of radius 0.′′ 5 and matched across
 MOA-2007-BLG-192 9

all filters. In order to eliminate marginal detections and stars badly impacted by bright neighbors,
we rejected stars with signal-to-noise ratio <5 and crowding parameter >0.75 as well as any objects
flagged by the software as too sharp or too extended to be stellar. The output magnitudes are given
in the STScI Vegamag system (m555, m814, m125, m160). Note that we used only main sequence
stars for calibration, and ignore color terms between VVV and the STSCI Vegamag system.

3.5. HST Multiple Star PSF Fitting

In addition to the photometry obtained using Dolphot, we performed multi-star PSF fitting on
the target in all three of the HST epochs. Since the 2012 and 2014 epochs are approximately 6.4
and 4.4 years before the Keck observations, we expect the separation between the source and lens
star to be 0.619× and 0.734× smaller in these HST images compared to Keck. The 2023 HST
images were taken approximately 5.0 years after the Keck observations, so we expect the lens and
source separation to be 1.445× larger in this epoch than the Keck images. Because each HST
observation is separated by at least several years and some epochs were taken at different position
angles (PA), we performed coordinate transformations between the Keck observation and each of the
HST observations independently. We do this by cross-matching ∼two dozen isolated and bright (but
not saturated) stars in each dataset, and then calculate the linear (i.e. first order) transformation
between the pixel positions in the HST and Keck catalogs. The transformations are listed as follows:

2012: xhst = −0.170xkeck + 0.186ykeck + 804.011

yhst = −0.184xkeck − 0.169ykeck + 1175.664

2014: xhst = −0.169xkeck + 0.186ykeck + 802.920

yhst = −0.185xkeck − 0.169ykeck + 1175.561

2023: xhst = 0.164xkeck − 0.188ykeck + 283.268

yhst = 0.187xkeck + 0.166ykeck + 225.342

The average RMS scatter for these relations is σx ∼ 0.25 and σy ∼ 0.20 HST/UVIS pixels for the
same 16 stars used in each transformation. Given the varying baseline between the earliest and latest
HST epochs and the 2018 Keck epoch, this scatter of ∼13 mas can be at least mostly explained by
an average proper motion of ∼2.5 mas/yr in each direction. We note the 2012 and 2014 data were
taken with the larger sub-array chip, UVIS2-2K2C-SUB, while the recent 2023 data were taken with
the smaller chip, UVIS2-C1K1C-SUB. Using the smaller sub-array chip allows for CTE losses to be
minimized as the detector has measurably degraded since the 2012 and 2014 epochs were obtained.
This HST analysis was performed using a modified version of the codes developed in Bennett
et al. (2015) and Bhattacharya et al. (2018), which analyzes the original individual images with no
resampling. This avoids any loss in resolution that can occur when dithered, undersampled images
are combined. The top-left panel of Figure 3 shows the target and surrounding HST stars from
the combined I-band image in 2014. A zoom on the target is shown in the top-right panel, which
also shows an unrelated star to the North of MB07192. The lower panels of Figure 3 show the
10 Terry et al.

residual images after fitting a single PSF model and simultaneously fitting two PSF models to the
blended stars. The single-star residual shows the typical signal that we would expect for two highly
blended stars, the direction and amplitude of the measured separation here is consistent with the
2018 detection in Keck (Table 5). The V -band detection is at a lower confidence than the I-band
detection (∼2σ vs. > 5σ), this leads to a larger error on the measured V -band lens magnitude (Table
4) and significantly larger error on the measured lens-source separations in HST V -band (see Table
5).
Given the strong detection in the Keck data, we impose separation constraints when analyzing the
earlier HST epochs, particularly the 2012 epoch in V -band, where the lens detection is most marginal.
We convert the Keck relative proper motion value (µrel,H = 2.63 ± 0.13 mas/yr) to constraints on the
position of the lens and source in the 2012 HST images, while taking into account the 4.8520 years
between the microlensing event peak and the 2012 Hubble observations. We note that in all of our
HST PSF fitting procedures we include the unrelated faint nearby neighbor as a third star to avoid
any interference of its PSF with our measurement of lens-source separation. Between the 2012 and
2023 epochs, the unrelated neighbor star moves ∼1 HST pixel closer to MB07192.
For all three HST epochs (2012, 2014, 2023), the F814W fits converge to a consistent solution with
‘star 1’, to the North as the slightly brighter star (∆mF814W ∼ 0.1). For the two epochs of F555W
data (2012, 2014), the PSF fit converged to a unique solution in the 2014 data without requiring any
separation constraint but the 2012 fit required a separation constraint to be imposed, as mentioned
previously. In all HST F814W fits, ‘star 2’, to the South is slightly fainter than ‘star 1’. Our reduction
and fitting code places the star coordinates from both filters into the same reference system, so all
stars have positions that are consistent between both passbands. The best-fit magnitudes (calibrated
to OGLE V and I) from the 2014 HST epoch are given in Table 4, and the best-fit positions in all
epochs and filters are given in Table 5.
The HST data were calibrated to the OGLE-III catalog (Szymański et al. 2011) using eight rela-
tively bright isolated OGLE-III stars that were matched to HST stars. The same eight stars were
used in each epoch. For the best quality HST data in both filters (i.e. 2014 epoch), the calibrations
yielded I1 = 21.56 ± 0.15, V1 = 24.93 ± 0.32, I2 = 21.68 ± 0.16, and V2 = 24.25 ± 0.18. The magnitude
of both lens and source stars combined is measured to significantly higher precision, I12 = 20.87±0.02
and V12 = 23.79 ± 0.04. This combined magnitude allows us to place a stronger constraint when
re-evaluating the light curve photometry. During our PSF fitting, the two blended stars can trade
flux back and forth which results in larger errors on the individual stars’ magnitude.

3.6. The Extinction Towards the Source star

The OGLE extinction calculator1 is a standard way to estimate the extinction for a galactic bulge
field, and it has been commonly used for many years. The calculator is derived from the reddening
and extinction study of Nataf et al. (2013). For MB07192, the calculator gives E(V −I) = 1.10±0.127
and an extinction AI = 1.3. These standard extinction maps have recently been superseded by the
Surot et al. (2020) analysis of the VVV survey and give E(J − Ks) = 0.329 ± 0.018 at the location
of the target. We then follow Nataf et al. (2013) in adopting E(J − KS )/E(V − I) = 0.3433, and

1
https://ogle.astrouw.edu.pl/cgi-ogle/getext.py
 MOA-2007-BLG-192 11
Table 3. Extinction estimates towards the source

Ext. map E(V − I) E(J − Ks ) AV AI AJ AH AK S

B08 1.12 ± 0.09 2.73 ± 0.13 1.61 ± 0.10
K12 0.43 ± 0.14 0.72 ± 0.10 0.46 ± 0.10 0.29 ± 0.10
ext. calc 1.10 ± 0.13 2.43 ± 0.16 1.33 ± 0.1
this study 1.10 ± 0.06 0.33 ± 0.02 2.45 ± 0.15 1.35 ± 0.07 0.44 ± 0.02 0.24 ± 0.01 0.11 ± 0.01
Note. We summarize here the different estimates for the extinction towards the source, in the initial study (B08),
the follow up work with NACO data (K12), and this study. Extinction values are derived from a combination of the
methods described in Nishiyama et al. (2009), Bennett et al. (2010), Nataf et al. (2013), and Surot et al. (2020) (see
section 3.6).

N HST I
E

Star 1

Star 2

0.35"

1-star residual 2-star residual

0 4 9 18 28 37 46 56 66
Counts
Figure 3. Similar to Figure 2, but for the 2014 HST data (eight exposures). The zoomed inset and residual
image panels show 100 × 100 super-sampled pixels where the observed dither offsets are accurate to 0.01
pixels. The color bar represents the pixel intensity (counts) seen in the top-right and lower-left/right panels.
12 Terry et al.

AI = 0.7465E(V − I) + 1.37E(JKS ), with which we derive the extinctions. Following Nishiyama

et al. (2009), we obtain the extinctions summarized in Table 3 along with prior estimates from the
literature. For our subsequent analysis, we adopt the numbers from the last row of Table 3 (i.e. this
work).

3.7. Identifying the Source and Lens Stars

With the HST V and I-band measurements described in Section 3.5, we can now attempt to
determine which star is the source and which is the lens. As mentioned previously, since the original
discovery paper of Bennett (2008), the MOA group has begun detrending its photometry to remove
systematic errors caused by differential atmospheric refraction (Bennett & Rhie 2002; Bond et al.
2017). Following Bond et al. (2017), we correct the MOA photometric data and perform re-modeling
of the MOA + OGLE photometry. This re-analysis yields an estimate of the source star I-band
magnitude of IS = 21.8 ± 0.05 with a color of VS − IS = 2.7 ± 0.2. This source I-band magnitude is
within 1σ of the HST I-band magnitude for ‘star 2’, and just over 1σ fainter than the HST I-band
magnitude for ‘star 1’. Additionally, this estimated source color is a closer match to the measured
HST V − I color of ‘star 2’ (2.57 ± 0.24) as can be seen in Figure 4. These results support the
identification of ‘star 2’ as the true source star. However, since the ground-based V -band estimate
of the source comes from a relatively weak relationship (OGLE I − MOA R), we conduct a further
verification of the source and lens using their relative proper motions as measured in HST and Keck.
We calculate the 2D prior probability distribution of the lens-source relative proper motion using
the Koshimoto et al. (2021) Galactic model to determine which stars are the preferred lens and source.
Figure 5 shows this proper motion distribution for MB07192, with two locations for the possible lens
(North and South lens). We calculate these µrel priors from the distribution of single lens stars that
√
reproduces the Einstein radius crossing time that accounts for the host star mass, i.e., tE / 1 + q.
The results show that there is a preference for the North star to be the lens star considering the stellar
distribution along this sight-line. The relative probability is PN /PS = 25.88/12.43 = 2.08; this means
the North star is > 2× more likely to be the lens than the South star. So, using the locations of ‘star
1’ and ‘star 2’ on the CMD (Figure 4) and the relative proper motion prior probability distribution
(Figure 5), we identify star 2 to be the source star and star 1 to be the lens star which hosts the
planet. We therefore label the source and lens on the CMD in Figure 4 as well as the stars in Table
4.

Table 4. HST and Keck multi-star PSF photometry

Star V Mag I Mag K Mag

Star 1 (Lens) 24.93 ± 0.32 21.56 ± 0.15 18.39 ± 0.09
Star 2 (source) 24.25 ± 0.18 21.68 ± 0.16 18.94 ± 0.10
Lens + Source 23.79 ± 0.04 20.87 ± 0.02 17.88 ± 0.05
Note. V and I magnitudes are calibrated to the OGLE-3 system and K magnitudes are calibrated to the 2MASS
system as described in section 3.
 MOA-2007-BLG-192 13

Figure 4. The observed color-magnitude diagram (CMD) for the MB07192 field. The OGLE-3 stars within
90 arcseconds of MB07192 are shown in black, with the HST CMD of all detected sources from the 2014
epoch shown in green. The red point indicates the location of the red clump centroid, and the purple and
orange points show the source and lens colors and magnitudes from the 2014 HST observations.

4. LENS-SOURCE RELATIVE PROPER MOTION

The Keck (2018) and HST (2012, 2014, 2023) follow up observations were taken between 4.85 and
16.20 years after the peak magnification which occurred in May 2007. The motion of the source
and lens on the sky is the primary cause for their apparent separation, however there is also a small
component that can be attributed to the orbital motion of Earth. As this effect is of order ≤ 0.10
mas for a lens at a distance of DL ≥ 2 kpc, we are safe to ignore this contribution in our analysis as
it is much smaller than the error bars on the stellar position measurements. The mean lens-source
relative proper motion is measured to be µrel,H = (µrel,H,E , µrel,H,N ) = (0.653 ± 0.291, 2.790 ± 0.274)
mas yr−1 , where ‘H’ indications that these measurements were made in the heliocentric reference
frame, and the ‘E’ and ‘N’ subscripts represent the East and North on-sky directions respectively.
14 Terry et al.

100
3 North lens

Relative probability
(mas/yr)

0 10 1
hel, N

South lens
3
3 0 3 10 2

hel, E (mas/yr)
Figure 5. The probability distribution for the north and east components of lens-source relative proper
motion (µrel ) using the Galactic model from Koshimoto et al. (2021) and genulens (Koshimoto & Ranc
2021). The possible lens positions (North and South) are plotted in black and given by the relative motion
of the two stars detected in the HST and Keck data. Importantly, this distribution uses tE values that are
close to the measured tE value from the light curve modeling (tE ∼ 99.5 days). This implies that the North
star is > 2× more likely to be the lens than the South star.

Our light curve modeling is performed in the geocentric reference frame that moves with the Earth
at the time of the event peak. Thus, we must convert between the geocentric and heliocentric frames
by using the relation given by Dong et al. (2009):
ν⊕ πrel
µrel,H = µrel,G + , (2)
AU
where ν⊕ is Earth’s projected velocity relative to the Sun at the time of peak magnification. For
MB07192 this value is ν⊕E,N = (25.772, 1.237) km/sec = (5.433, 0.261) AU yr−1 at HJD′ = 4245.45.
With this information and the relative parallax relation πrel ≡ AU (1/DL − 1/DS ), we can express
equation 2 in a more convenient form:

µrel,G = µrel,H − (5.433, 0.261) × (1/DL − 1/DS ) mas/yr, (3)

where DL and DS are the lens and source distance, respectively, given in kpc. We have directly
measured µrel,H from the HST and Keck data, so this gives us the relative proper motion in the
geocentric frame of µrel,G = 3.10 ± 0.19 mas/yr. As a reminder, the lens and source distance we
use in Equation 2 are inferred by the best-fit light curve results which include constraints from the
high-resolution imaging.
 MOA-2007-BLG-192 15

Figure 6. Left: The MCMC distribution for πE from the light curve modeling without any constraint from
the high-resolution imaging. Right: The MCMC distribution for πE from the light curve modeling with
the inclusion of high-resolution imaging constraints. The color bar represents the χ2 differences from the
best-fit light curve model: red represents ∆χ2 < 1, orange represents ∆χ2 < 4, green represents ∆χ2 < 16,
and purple represents ∆χ2 < 25. The two components of the relative proper motion that were measured by
HST and Keck allow the north component, πE,N to be tightly constrained.

5. LENS SYSTEM PROPERTIES

As has been shown in prior work (Bhattacharya et al. 2018; Bennett et al. 2020; Terry et al. 2021;
Rektsini et al. 2024), we find it particularly useful to apply constraints from the high resolution follow-
up observations to the light curve models (we deem this “image-constrained modeling”). This can
help prevent the light curve modeling from exploring areas in the parameter space that are excluded
by the high resolution follow-up observations. We refer the reader to Bennett et al. (2023) for a full
description of the methodology for applying these constraints to the modeling and an exhaustive list
of the light curve + high-res imaging parameters that are important for obtaining full solutions for
planetary lens systems in this context.
We use a modified version of the light curve modeling code eesunhong (Bennett & Rhie 1996;
Bennett 2010) to incorporate constraints on the brightness and separation of the lens and source
stars from the high resolution imaging via HST and Keck (Bennett et al. 2023). Ideally, we want
16 Terry et al.
Table 5. Measured Lens-Source Separations from HST and Keck

Separation (mas)
Year East North Total
2012 (HST V ) 2.28 ± 4.60 15.58 ± 4.96 15.75 ± 6.78
(HST I) 1.01 ± 1.39 18.17 ± 1.71 18.20 ± 2.23
2014 (HST V ) 9.84 ± 4.74 22.17 ± 4.01 24.26 ± 6.22
(HST I) 3.12 ± 1.23 21.62 ± 1.03 21.84 ± 1.63
2018 (Keck K) −0.34 ± 1.03 29.37 ± 1.01 29.38 ± 1.46
2023 (HST I) −1.97 ± 1.49 43.13 ± 1.68 43.17 ± 2.26
µrel,H,E (mas/yr) µrel,H,N (mas/yr) µrel,H (mas/yr)
weighted mean 0.63 ± 0.29 2.76 ± 0.27 2.83 ± 0.37

to use a mass-distance relation coupled with empirical mass-luminosity relations to infer the mass
and distance of the host star. In order to do this, we need to know the distance to the source star,
DS . Thus we are required to include the source distance as a fitting parameter in the re-modeling of
the light curve with imaging constraints. We include a weighting from the Koshimoto et al. (2021)
Galactic model as a prior for DS , and we also use the same Galactic model to obtain a prior on the
lens distance for a given value of DS . This prior is not used directly in the light curve modeling, but
instead is used to weight the entries in a sum of Markov chain values.
The angular Einstein radius, θE , and the microlensing parallax vector, πE , give relations that
connect the lens system mass to the source and lens distances, DS and DL (B08, Gaudi (2012)). The
relations are given by:

c2 2 DS DL
ML = θ , (4)
4G E DS − DL
and

c2 AU DS − DL
ML = , (5)
4G πE 2 DS DL
where ML is the lens mass, G and c are the gravitational constant and speed of light. DL and DS
are the distance to the lens and source, respectively. As mentioned previously, the measurement of
µrel,H from the high resolution imaging allows us to measure µrel,G to high precision, which ultimately
lets us determine θE . Additionally, the two components of the µrel measurement enables a much
tighter constraint on the possible values of πE,N . The north direction in particular is usually only
weakly constrained because it is typically perpendicular to the orbital acceleration of the observer
for microlensing events towards the Galactic bulge. The geocentric relative proper motion and the
microlensing parallax are related by:

πrel µrel,G
πE = , (6)
tE |µrel,G |2
 MOA-2007-BLG-192 17
Table 6. Best Fit Model Parameters with µrel and Magnitude Constraints

u0 < 0 u0 > 0
Parameter s<1 s>1 s<1 s>1 MCMC Averages
tE (days) 99.469 98.722 100.111 99.262 99.577 ± 3.919
t0 (HJD′ ) 4245.446 4245.448 4245.431 4245.436 4245.440 ± 0.0070
u0 −0.0027 −0.0029 0.0035 0.0004 −0.0027 ± 0.0012
(u0 > 0) 0.00195 ± 0.00155
s 0.9102 1.0311 0.8780 1.1441 0.8728 ± 0.0667
(s > 1) 1.0951 ± 0.0938
α (rad) 2.1061 1.9288 4.5473 3.2862 2.4364 ± 0.5075
(u0 > 0) 3.9167 ± 0.6305
log(q) −3.9751 −3.9975 −3.6017 −3.7937 −3.8690 ± 0.5253
t∗ (days) 0.0562 0.0539 0.0567 0.0547 0.0551 ± 0.0044
πE,N 0.3161 0.3152 0.3119 0.3133 0.3154 ± 0.0218
πE,E −0.2364 −0.2308 −0.2338 −0.2300 −0.2359 ± 0.0474
Ds (kpc) 7.8423 7.1562 6.9687 7.1156 7.049 ± 1.163
Fit χ2 4760.94 4760.97 4761.45 4761.53

so with the measurement of πE,E and µrel,H , we can use equations 2 and 6 to solve for πE,N . This
tight constraint on the north component of the microlensing parallax can be seen in Figure 6, where
the left panel shows the distribution in πE,N largely unconstrained. When the constraint from the
high resolution measurement of µrel is applied, the distribution collapses to a relatively small region
centered on πE,N ∼ 0.3.

Additionally, since we have a direct measurement of lens flux in the V , I, and K-bands, we utilize
the Delfosse et al. (2000) empirical mass-luminosity relations in each of these passbands as described
by Bennett et al. (2018). We consider the foreground extinction in each passband (i.e. Table 3) and
generate the relations in conjunction with the mass-distance relations given by equations 4 and 5.
Figure 7 shows the measured mass and distance of the MB07192 lens. The blue (HST V ), green (HST
I ), and red (Keck K ) curves represent the mass-distance relations obtained from the empirical mass-
luminosity relations with lens flux measurements given in Table 4. The dashed lines represent the 1σ
error from the Keck and HST measurements. Further, the mass-distance relation obtained from the
measurement of θE (i.e. equation 4) is shown in solid brown. Considering only these two relations
(empirical mass-luminosity and θE ), there is overlap for a significant amount of mass and distance
space. This is sometimes referred to as the “continuous degeneracy” (Gould 2022). Fortunately this
degeneracy is broken when we include the constraint from the microlensing parallax measurement,
πE , shown as the solid teal curve in Figure 7.
Table 6 shows the results of the four degenerate light curve models and the Markov chain average
for all four models. Although we are able to successfully reduce the number of possible solutions
18 Terry et al.

Figure 7. The mass-distance relation for MB07192L with constraints from the lens flux measurement in blue
(HST V ), green (HST I ), and red (Keck K ). Dashed lines show the 1σ error bars for each passband. The
solid teal region shows the mass-distance relation calculated using the microlensing parallax measurement
(πE ), and the solid brown region shows the mass-distance relation calculated using the angular Einstein
radius measurement (θE ).

presented in K12 by a factor of two, the close/wide and u0 degeneracies still remain. Further, the
host star mass is very precisely measured now, however the best-fit mass ratio, q, remains largely
uncertain because of poor sampling of the light curve. This also results in a large error in the inferred
planet mass (see Table 7).
The MB07192 lens system is located at a distance of ∼2.2 kpc and has a log mass ratio of
log10 (q) = −3.87 ± 0.53. The host star is directly detected in several high resolution imaging
passbands, enabling us to precisely measure its mass to be Mhost = 0.28 ± 0.04M⊙ with a less-
precisely measured mass of the planet to be mplanet = 12.49+65.47
−8.03 M⊕ . These masses are consistent
with a planet with mass between a super-Earth and sub-Saturn orbiting an M dwarf star near the
bottom of the main sequence. We note the most likely mass for the planet is in the super-Earth
regime (∼3 − 12M⊕ ), as given by the top-left panel in Figure 8. The best-fit solution gives a 2D
projected separation of a⊥ = 2.02 ± 0.44 AU. These physical parameters are calculated from the
best-fit solution which takes a combined weighting of several models along with a Galactic model
prior based on Koshimoto et al. (2021). Table 7 gives the derived lens system physical parame-
ters along with their 2σ ranges. Lastly, Figure 8 shows the final posterior probability distributions
 MOA-2007-BLG-192 19

for the planetary companion mass, host star mass, 2D projected separation, and lens system distance.

Table 7. Lens System Properties with Lens Flux Constraints

Parameter Units Values & RMS 2-σ range

Angular Einstein Radius (θE ) mas 0.854 ± 0.043 0.775 − 0.947
Geocentric lens-source relative proper motion (µrel,G ) mas/yr 3.14 ± 0.15 2.84 − 3.44
Host Mass (Mhost ) M⊙ 0.28 ± 0.04 0.23 − 0.37
Planet Mass (mp ) M⊕ 12.49+65.47
−8.03 2.75 − 105.06
2D Separation (a⊥ ) AU 2.02 ± 0.44 1.26 − 2.86
3D Separation (a3d ) AU 2.44+1.39
−0.68 1.38 − 9.65
Lens Distance (DL ) kpc 2.16 ± 0.30 1.75 − 2.76
Source Distance (DS ) kpc 7.05 ± 1.17 4.83 − 9.38

6. DISCUSSION & CONCLUSION

Our high resolution follow up observations of the microlensing target MB07192 have allowed us
to make a direct measurement of the lens system flux in multiple passbands (V , I, K) as well as a
precise determination of the amplitude and direction of the lens-source relative proper motion µrel .
We perform simultaneous multiple-star PSF fitting to obtain best-fit positions and fluxes for both
stars across two independent platforms (HST and Keck). The lens flux measurements we make
enable us to use mass-luminosity relations and new constraints on higher order light curve effects
(πE , θE ) to measure a precise mass and distance for the lens system.
Further, we demonstrate the importance of applying constraints from high resolution follow up
imaging on the microlensing light curve modeling. Particularly, the microlensing parallax effect, which
is present in all microlensing events observed from a heliocentric reference frame, is tightly constrained
when the direction of µrel can be measured through high resolution imaging. This measurement is
critically important for several reasons; poor light curve sampling (i.e. for MB07192) can result in
a lack of a microlensing parallax signal from the light curve alone, even for long timescale events.
Second, the mass-distance relation that results from a direct measurement of πE (via lens-source
separation) allows for the “continuous degeneracy” to be completely broken.
Although the host mass and lens system distance have now been precisely measured as a result
of the direct detection in HST and Keck, the sampling of the light curve photometry during the
microlensing event remains poor. This means that the large uncertainty in the mass ratio parameter,
(q in Table 6), results in a large error in the inferred mass of the planetary companion (mp in Table
7). As previously mentioned in Section 2, the large uncertainty in the planetary companion mass
comes from a combination of factors; the event is located in a MOA field with a relatively low
cadence which leads to poor sampling of the light curve, and the planetary signal was not detected
in real time. It was several days after the photometric peak that the anomaly in the light curve was
20 Terry et al.

Figure 8. The posterior probability distributions for the lens system physical parameters: planetary com-
panion mass (upper-left), host mass (upper-right), 2D projected separation (lower-left), and lens distance
(lower-right). The vertical black line shows the median of the probability distribution for each parameter.
The central 68% of the distributions are shown in dark blue with the remaining central 95% shown in light
blue.

alerted.
In conclusion, the distance to the MB07192 lens system is ∼3× larger than previously reported,
now at a distance of approximately 2 kpc. Both the mass of the host star and planetary companion
are also 2 − 5× larger than previously reported, which now extends the possible mass range for
the planet to a possible sub-Saturn class planet. However, as the top-left panel of Figure 8 shows,
the most likely mass for the planetary companion remains in the super-Earth range (∼3 − 12M⊕ ).
The previous studies reported a smaller planet mass and also underestimated the error bars on the
planet mass for several reasons. All of the B08 models report a too-large microlensing parallax value,
which led to a smaller derived planet mass than the median value of the planet mass that we report
here. Further, the B08 and K12 caustic-crossing models contributed significant weighting to the
combined results which gave much smaller error bars on the derived planet mass. Our new results
have ruled out the caustic crossing models, which now gives larger error bars on the derived planet
mass, particularly the upper 1σ error.
The results of this work have several implications for the upcoming RGES. If Roman is expected
 MOA-2007-BLG-192 21

to employ lens flux measurement methods similar to those described in this work, then a careful
selection of secondary observing filters must be made to avoid or minimize instances of the “con-
tinuous degeneracy”. For example, the mass-luminosity relation given by a bluer Roman passband
would have a smaller overlap with the mass-distance relation given by θE compared to other redder
Roman filters. This effect is more severe for nearby M dwarf lenses (i.e. Figure 7). Also, for
Roman detected events with very faint sources or very short Einstein timescales that don’t have a
measurable microlensing parallax signal, a successful lens-source flux measurement by Roman itself
will be important for breaking possible degeneracies like those discussed in this work.

This paper is based in part on observations made with the NASA/ESA Hubble Space Tele-
scope, which is operated by the Association of Universities for Research in Astronomy, Inc., under
NASA contract NAS 5-26555. The Keck Telescope observations and data analysis were supported
by a NASA Keck PI Data Award, 80NSSC18K0793, administered by the NASA Exoplanet Science
Institute. Data presented herein were obtained at the W. M. Keck Observatory from telescope
time allocated to the National Aeronautics and Space Administration through the agency’s scien-
tific partnership with the California Institute of Technology and the University of California. The
Observatory was made possible by the generous financial support of the W. M. Keck Foundation.
The authors wish to recognize and acknowledge the very significant cultural role and reverence
that the summit of Maunakea has always had within the indigenous Hawaiian community. We are
most fortunate to have the opportunity to conduct observations from this mountain. The material
presented here is also based upon work supported by NASA under award number 80GSFC21M0002.
This work was supported by the University of Tasmania through the UTAS Foundation and the
endowed Warren Chair in Astronomy and the ANR COLD-WORLDS (ANR-18-CE31-0002). This
research was also supported in part by the Australian Government through the Australian Research
Council Discovery Program (project number 200101909) grant awarded to Cole and Beaulieu. Some
of this research has made use of the NASA Exoplanet Archive, which is operated by the California
Institute of Technology, under the Exoplanet Exploration Program.

Software: DAOPHOT-II (Stetson 1987), daophot mcmc (Terry et al. 2021), eesunhong (Bennett
& Rhie 1996), genulens (Koshimoto & Ranc 2021), hst1pass (Anderson 2022), KAI (Lu 2022),
Matplotlib (Hunter 2007), Numpy (Oliphant 2006).

REFERENCES
Alcock, C., Allsman, R., Axelrod, T., et al. 1996, Bennett, D., Bond, I., Udalski, A., et al. 2008,
Astrophysical Journal v. 461, p. 84, 461, 84 The Astrophysical Journal, 684, 663
Anderson, J. 2022, Instrument Science Report Bennett, D., Sumi, T., Bond, I., et al. 2012a, The
WFC3, 5, 55 Astrophysical Journal, 757, 119
Beaulieu, J. P., Batista, V., Bennett, D. P., et al. Bennett, D. P. 2008, in Exoplanets (Springer),
2018, AJ, 155, 78, 47–88
doi: 10.3847/1538-3881/aaa293 —. 2010, ApJ, 716, 1408
Bennett, D. 2018, Development of the WFIRST Bennett, D. P., Anderson, J., Bond, I. A., Udalski,
Exoplanet Mass Measurement Method, A., & Gould, A. 2006, ApJL, 647, L171,
10.26135/KOA4, 2018B, N139 doi: 10.1086/507585
22 Terry et al.

Bennett, D. P., Anderson, J., & Gaudi, B. S. 2007, Gaudi, B. S. 2012, ARA&A, 50, 411,
ApJ, 660, 781, doi: 10.1086/513013 doi: 10.1146/annurev-astro-081811-125518
Bennett, D. P., & Rhie, S. H. 1996, ApJ, 472, 660, Ghez, A. M., Salim, S., Weinberg, N., et al. 2008,
doi: 10.1086/178096 ApJ, 689, 1044
—. 2002, ApJ, 574, 985, doi: 10.1086/340977 Gould, A. 2022, arXiv preprint arXiv:2209.12501
Bennett, D. P., Rhie, S. H., Nikolaev, S., et al. Gould, A., Dong, S., Bennett, D. P., et al. 2010,
2010, ApJ, 713, 837, ApJ, 710, 1800,
doi: 10.1088/0004-637X/713/2/837 doi: 10.1088/0004-637X/710/2/1800
Bennett, D. P., et al. 2012b, HST Proposal Henry, T. J., Franz, O. G., Wasserman, L. H.,
GO-12541 et al. 1999, ApJ, 512, 864, doi: 10.1086/306793
—. 2014, HST Proposal GO-13417 Henry, T. J., & McCarthy, Donald W., J. 1993,
Bennett, D. P., Bhattacharya, A., Anderson, J., AJ, 106, 773, doi: 10.1086/116685
et al. 2015, ApJ, 808, 169, Hunter, J. D. 2007, CSE, 9, 90
doi: 10.1088/0004-637X/808/2/169 Ida, S., & Lin, D. N. C. 2004, ApJ, 604, 388,
Bennett, D. P., Udalski, A., Bond, I. A., et al. doi: 10.1086/381724
2018, The Astronomical Journal, 156, 113 Johnson, S. A., Penny, M., Gaudi, B. S., et al.
Bennett, D. P., Bhattacharya, A., Beaulieu, J.-P., 2020, ApJ, 160, 123
et al. 2020, AJ, 159, 68, Kim, S.-L., Lee, C.-U., Park, B.-G., et al. 2016,
doi: 10.3847/1538-3881/ab6212 Journal of The Korean Astronomical Society,
Bennett, D. P., Bhattacharya, A., Beaulieu, J.-P., vol. 49, issue 1, pp. 37-44, 49, 37
et al. 2023, arXiv preprint arXiv:2311.00627
Koshimoto, N., Baba, J., & Bennett, D. P. 2021,
Bertin, E. 2010, Astrophysics Source Code Library
The Astrophysical Journal, 917, 78
Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393,
Koshimoto, N., & Ranc, C. 2021,
doi: 10.1051/aas:1996164
doi: 10.5281/zenodo.4898012
Bhattacharya, A., Beaulieu, J. P., Bennett, D. P.,
Koshimoto, N., Sumi, T., Bennett, D. P., et al.
et al. 2018, AJ, 156, 289,
2023, AJ, 166, 107,
doi: 10.3847/1538-3881/aaed46
doi: 10.3847/1538-3881/ace689
Bhattacharya, A., Bennett, D. P., Beaulieu, J. P.,
Kubas, D., Beaulieu, J., Bennett, D., et al. 2012,
et al. 2021, AJ, 162, 60
Astronomy & Astrophysics, 540, A78
Blackman, J., Beaulieu, J., Bennett, D., et al.
2021, Nature, 598, 272 Lissauer, J. J. 1993, ARA&A, 31, 129,
Bond, I. A., Abe, F., Dodd, R. J., et al. 2001, doi: 10.1146/annurev.aa.31.090193.001021
MNRAS, 327, 868, Lu, J. 2022, doi: 10.5281/zenodo.6522913
doi: 10.1046/j.1365-8711.2001.04776.x Lu, J., Ghez, A., Hornstein, S. D., et al. 2008,
Bond, I. A., Bennett, D. P., Sumi, T., et al. 2017, ApJ, 690, 1463
MNRAS, 469, 2434, doi: 10.1093/mnras/stx1049 Minniti, D., Lucas, P. W., Emerson, J. P., et al.
Boyajian, T. S., van Belle, G., & von Braun, K. 2010, NewA, 15, 433,
2014, AJ, 147, 47, doi: 10.1016/j.newast.2009.12.002
doi: 10.1088/0004-6256/147/3/47 Nataf, D. M., Gould, A., Fouqué, P., et al. 2013,
Delfosse, X., Forveille, T., Ségransan, D., et al. ApJ, 769, 88, doi: 10.1088/0004-637X/769/2/88
2000, A&A, 364, 217. Nishiyama, S., Tamura, M., Hatano, H., et al.
https://arxiv.org/abs/astro-ph/0010586 2009, ApJ, 696, 1407,
Devillard, N. 1997, The Messenger, vol. 87, p. doi: 10.1088/0004-637X/696/2/1407
19-20, 87, 19 Oliphant, T. E. 2006, A guide to NumPy, Vol. 1
Diolaiti, E., Bendinelli, O., Bonaccini, D., et al. (Trelgol Publishing USA)
2000, A&AS, 147, 335, doi: 10.1051/aas:2000305 Penny, M. T., Gaudi, B. S., Kerins, E., et al. 2019,
Dolphin, A. E. 2000, PASP, 112, 1383, ApJS, 241, 3, doi: 10.3847/1538-4365/aafb69
doi: 10.1086/316630 Pollack, J. B., Hubickyj, O., Bodenheimer, P.,
Dong, S., Gould, A., Udalski, A., et al. 2009, ApJ, et al. 1996, Icarus, 124, 62,
695, 970, doi: 10.1088/0004-637X/695/2/970 doi: 10.1006/icar.1996.0190
 MOA-2007-BLG-192 23

Quenouille, M. H. 1949, The Annals of Suzuki, D., Bennett, D. P., Ida, S., et al. 2018,
Mathematical Statistics, 20, 355–375 ApJL, 869, L34, doi: 10.3847/2041-8213/aaf577
—. 1956, Biometrika, 43, 353–360 Szymański, M. K., Udalski, A., Soszyński, I., et al.
Rektsini, N. E., Batista, V., Ranc, C., et al. 2024, 2011, AcA, 61, 83,
arXiv preprint arXiv:2401.17549 doi: 10.48550/arXiv.1107.4008
Terry, S. K., Bhattacharya, A., Bennett, D. P.,
Sahu, K., et al. 2023, HST Proposal GO-16716
et al. 2021, AJ, 161, 54
Service, M., Lu, J. R., Campbell, R., et al. 2016, Terry, S. K., Bennett, D. P., Bhattacharya, A.,
PASP, 128, 095004, et al. 2022, The Astronomical Journal, 164, 217
doi: 10.1088/1538-3873/128/967/095004 Terry, S. K., Lu, J. R., Turri, P., et al. 2023,
Spergel, D., Gehrels, N., Baltay, C., et al. 2015, Journal of Astronomical Telescopes,
arXiv e-prints, arXiv:1503.03757. Instruments, and Systems, 9, 018003
https://arxiv.org/abs/1503.03757 Tierney, L., & Mira, A. 1999, Stat Med, 18, 2507
Stetson, P. B. 1987, PASP, 99, 191, Udalski, A., Szymanski, M., Kaluzny, J., et al.
doi: 10.1086/131977 1994, Astrophysical Journal, Part 2-Letters
Sumi, T., Koshimoto, N., Bennett, D. P., et al. (ISSN 0004-637X), vol. 426, no. 2, p. 69-72, 426,
2023, AJ, 166, 108, 69
doi: 10.3847/1538-3881/ace688 Udalski, A., Szymański, M., & Szymański, G.
2015, arXiv preprint arXiv:1504.05966
Surot, F., Valenti, E., Gonzalez, O. A., et al. 2020,
Vandorou, A., Bennett, D. P., Beaulieu, J.-P.,
A&A, 644, A140,
et al. 2020, AJ, 160, 121,
doi: 10.1051/0004-6361/202038346 doi: 10.3847/1538-3881/aba2d3
Suzuki, D., Bennett, D. P., Sumi, T., et al. 2016, Yelda, S., Lu, J. R., Ghez, A. M., et al. 2010, ApJ,
ApJ, 833, 145, 725, 331, doi: 10.1088/0004-637X/725/1/331
doi: 10.3847/1538-4357/833/2/145
24 Terry et al.

APPENDIX

2023 HST SNAPSHOT IMAGES

We show the four-panel figure created from the two exposures taken during the August 2023 Snapshot
Program (Sahu et al. 2023). The stacked frame has noticeably larger Poisson noise than the previous
HST epochs which have 4× more exposures. The longer time baseline between the peak of the
microlensing event and the 2023 HST data helps to mitigate the lack of exposures, as the larger
separation between source and lens can be clearly detected in this epoch.
N HST I
E

Star 1

Star 2

0.35"

1-star residual 2-star residual

0 4 9 18 28 37 46 56 66
Counts

Figure 9. Top Left: The 2023 HST I band stack image created from two individual exposures from the
Snapshot Program. The target is indicated with a yellow box. Top Right: Zoomed image of the target, with
the two points indicating the best-fit positions for the two stars from the multi-star PSF fitting. We note
the unrelated neighbor star has moved closer to the target(s) by ∼1 pixel between the 2012 and 2023 HST
data. Bottom Left: The residual image from a single-star PSF fit, showing a strong signal of the blended
lens(source). Bottom Right: Residual image for the simultaneous two-star PSF fit, showing a smoother
subtraction with Poisson noise remaining as well as systematics due to the less-characterized PSF model.
The color bar represents the pixel intensity (counts) seen in the top-right and lower-left/right panels.

This 2023 data, in conjunction with the previous HST epochs, largely confirm that the two stars
we detect are the true source and lens separating from each other with their expected relative proper
motions. This multi-epoch tracking rules out the scenarios in which we are detecting an unrelated
blend or a bound stellar companion to either the source or lens.
 MOA-2007-BLG-192 25

FULL LIGHT CURVE MODELING COMPARISON

We show in Figure 10 the comparison of the fitting parameters between the light curve modeling
from photometry only and from photometry plus HST /Keck AO imaging. In Sections 3 and 5
we explained one of the strongest high-res imaging constraints is that of the microlensing parallax
vectors, particularly the North component, πE,N . Additionally, the tighter constraint on the source
radius crossing time, t∗ , comes primarily from the µrel,H measurement derived from the Keck data
via the following equation:

θ∗
t∗ = (1)
µrel
where θ∗ is the angular size of the source star, which we estimate using surface brightness relations
from Boyajian et al. (2014), considering only stars spanning the range in colors that are relevant for
microlensing targets. And µrel in equation 1 comes from the best-fit lens-source separation measure-
ment in Keck.
The caustic crossing models given in the prior studies (B08, K12) are ruled out by this Keck
data, and the models with a close approach to a caustic cusp do not strongly constrain t∗ very well.
Ultimately, we can further reduce the total number of possible solutions from K12 (8 solutions) by a
factor of 2, which leaves a four-fold degeneracy remaining (i.e. s → 1/s and u0 < 0, u0 > 0).
26 Terry et al.
tE [days] = 99.4+3.96
3.84

+4.2454e3 t0 [HJD ] = 4.25e + 03+0.00392

0.00421
1.8 2.0 2.2 20..472 0.80 0.88 0.96 1.04 2.55 2.70 2.85 3.00 0.03 0.04 0.05 0.06 0.07
t0 [HJD ]

1e 3 u0 = 0.00271+0.000151
0.000262
u0

s = 0.888+0.0247
0.0478
s

[rad] = 2.04+0.0582
0.135
[rad]

log(q) = 4.11+0.386
0.23
5 0 5 0 .0 .5 .0 .5 .0 5 0 5 0 .5 .0 .5 .0
0.4 0.3 0.1 0.0 1 0 0 0 10.04 0.06 0.07 0.09 1 3 4 6
log(q)

t * [days] = 0.056+0.00379
0.00388
t * [days]

E, N = 0.316+0.0214
0.0242
E, N

E, E = 0.236+0.0467
0.046
E, E

90
95
100
15
0.100
0.03
0.04
0.05
0.06
7
2.75
2.80
3.05
0.70
0.82
0.80
0.98
1.06
4
1.8
2.0
2.2
1..54
3.0
4.5
6.0

0.0 5
60
0.0 5
90
1.0
0.5
0.0
0.5
1.0
5
0
0.05
0
10

2.5

0.4
0.3
0.1
2

0.0

0.0

+4.2454e3 1e 3
tE [days] t0 [HJD ] u0 s [rad] log(q) t * [days] E, N E, E

Figure 10. Comparison of model parameter distributions from the light curve photometry only (light grey)
and light curve photometry plus HST /Keck imaging constraints (black). The two cases shown are for the
s < 1, u0 < 0 model. The constraints from the high-resolution imaging are tightened most for the North and
East components of the microlensing parallax (πE,N , πE,E ) as well as the source radius crossing time (t∗ ).
The median values given in the title headings (above each histogram) are for the constrained light curve +
imaging model. All other model parameters give consistent distributions between the two cases.