MNRAS 546, 1–23 (2026) https://doi.org/10.1093/mnras/staf2255 
Advance Access publication 2025 December 22 

JWST /MIRI coronagraphic search for planets in systems with gapped 

exoKuiper belts and proper-motion anomalies 

R. Bendahan-West , 1 ‹ S. Marino , 1 A. L. Carter , 2 V. Squicciarini , 1 A. D. James , 1 

A. A. Sefilian , 3 T. D. Pearce , 4 M. F. Friebe , 5 C. Lazzoni , 1 , 6 B. Lakeland , 7 S. Ray , 8 

M. C. Wyatt , 9 L. Matr ̀a , 10 J. Milli , 11 V. C. Faramaz , 3 Th. Henning, 12 S. Hinkley , 1 

G. M. Kennedy , 4 , 13 D. Mesa 

5 and A. Zurlo 

14 , 15 

1 Department of Physics and Astronomy, University of Exeter, Stocker Road, Exeter EX4 4QL, UK 

2 Space Telescope Science Institute (STScI), 3700 San Martin Drive, Baltimore, MD 21218, USA 

3 Department of Astronomy and Steward Observatory, University of Arizona, 933 N Cherry Avenue, Tucson AZ 85721, USA 

4 Department of Physics, University of Warwick, Gibbet Hill Road, Coventry CV4 7AL, UK 

5 Astrophysikalisches Institut und Universit ̈atssternwarte, Friedrich-Schiller-Universit ̈at Jena, Schillerg ̈aßchen 2-3, D-07745 Jena, Germany 
6 INAF, Astronomical Observatory of Padua, Vicolo dell’Osservatorio 5, I-35122, Padua, Italy 
7 School of Physics and Astronomy, University of Birmingham, Edgbaston, Birmingham B15 2TT,UK 

8 School of Mathematics and Physics, University of Queensland, St Lucia, QLD 4072, Australia 
9 Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK 

10 School of Physics, Trinity College Dublin, the University of Dublin, College Green, Dublin 2, Ireland 
11 Univ. Grenoble Alpes, CNRS, IPAG, F-38000 Grenoble, France 
12 Max-Planck-Insitut f ̈ur Astronomie, K ̈onigstuhl 17, D-69117 Heidelberg, Germany 
13 Malaghan Institute of Medical Research, Gate 7, Victoria University, Kelburn Parade, Wellington, New Zealand 
14 Instituto de Estudios Astrof ́ısicos, Facultad de Ingenier ́ıa y Ciencias, Universidad Diego Portales, Av. Ej ́ercito Libertador 441, Santiago, Chile 
15 Millennium Nucleus on Young Exoplanets and their Moons (YEMS), Chile 

Accepted 2025 December 19. Received 2025 December 19; in original form 2025 August 25 

A B S T R A C T 

Over the past decade, ALMA has unco v ered a range of substructures within e xoK uiper belts, pointing to a population of 
undetected planets. With James Webb Space Telescope ( JWST )’s sensitivity, we now have the opportunity to identify these 
planets thought to be responsible for the observed substructures in debris discs. We present Cycle 1 JWST /MIRI 11 . 4 μm 

coronagraphic observations of three e xoK uiper belts that exhibit gaps in their radial structures: HD 92945, HD 107146, and 

HD 206893, to determine whether planets are responsible for carving these structures, as seen in our Solar system with the gas 
giants. We reduce the JWST /MIRI data using SPACEKLIP , and introduce new routines to mitigate the Brighter-Fatter effect and 

persistence. We do not detect any planet candidates, and all detected objects in the field of view are consistent with background 

stars or galaxies. Ho we ver, by combining JWST mass limits, archi v al observ ational constraints, and astrometric accelerations, 
we rule out a significant portion of planet parameter space, placing tight constraints on the planets possibly responsible for 
these gaps. To interpret these results, we explore multiple gap-carving scenarios in discs, either massless or with non-zero mass, 
including clearing by in situ planet(s), as well as shaping by inner planets through mean-motion or secular apsidal resonances. 
Finally, we conclude that the planets causing the proper-motion anomaly in these systems must reside within the inner 20 au. 

Key words: planet and satellites: detection – planet-disc interactions – infrared: planetary systems. 

1

A
o
s  

J  

O
g

�

i
E  

G  

f
w  

N  

2
b

©
P
C
p

 I N T RO D U C T I O N  

lthough thousands of exoplanets have been discovered at small 
rbital radii through radial velocity (RV) and photometric transit 
urv e ys, only a limited number of planets more massive than
upiter have been detected beyond ∼10 au (e.g. R. Cloutier 2024 ).
ver the past two decades, the number of wide-orbit planets have 
radually increased, largely thanks to advances in high-contrast 
 E-mail: rb941@e x eter.ac.uk 

W  

m  

M  

The Author(s) 2025. 
ublished by Oxford University Press on behalf of Royal Astronomical Society. Th
ommons Attribution License ( https:// creativecommons.org/ licenses/ by/ 4.0/ ), whic
rovided the original work is properly cited. 
maging instruments such as the Spectro-Polarimetric High-contrast 
xoplanet REsearch (SPHERE; J. L. Beuzit et al. 2019 ) and the
emini Planet Imager (GPI; B. Macintosh et al. 2014 ). These

acilities have enabled detections of increasingly lower-mass planets, 
ith sensiti vities no w reaching do wn to ∼2 - 3 M Jup (e.g. E. L.
ielsen et al. 2019 ; A. Vigan et al. 2021 ; V. Squicciarini et al.
025 ). Today, these ground-based capabilities are being surpassed 
y the unprecedented contrast and sensitivity offered by James 
ebb Space Telescope ( JWST ), now reaching down to Saturn
asses and lower in some cases (A. L. Carter et al. 2021a ; A.
. Lagrange et al. 2025 ). Pushing these detection limits further
is is an Open Access article distributed under the terms of the Creative 
h permits unrestricted reuse, distribution, and reproduction in any medium, 

http://orcid.org/0009-0000-0303-2145
http://orcid.org/0000-0002-5352-2924
http://orcid.org/0000-0001-5365-4815
http://orcid.org/0000-0002-3122-6809
http://orcid.org/0009-0005-6943-6819
http://orcid.org/0000-0003-4623-1165
http://orcid.org/0000-0001-5653-5635
http://orcid.org/0009-0004-5747-7827
http://orcid.org/0000-0001-7819-9003
http://orcid.org/0000-0002-8122-2240
http://orcid.org/0000-0003-2259-3911
http://orcid.org/0000-0001-9064-5598
http://orcid.org/0000-0003-4705-3188
http://orcid.org/0000-0001-9325-2511
http://orcid.org/0000-0001-6403-841X
http://orcid.org/0000-0001-8074-2562
http://orcid.org/0000-0001-6831-7547
http://orcid.org/0000-0001-8467-1933
http://orcid.org/0000-0002-5903-8316
mailto:rb941@exeter.ac.uk
https://creativecommons.org/licenses/by/4.0/


2 R. Bendahan-West et al. 

M

i  

s
 

w  

p  

c  

M  

m  

a  

a  

a  

2  

b  

e  

o  

p  

p
 

i  

s  

g  

t  

o  

T  

e  

4  

o  

m  

C  

2  

2  

t  

t  

M  

o  

g  

t  

r  

a  

e  

r  

N  

K  

p  

2
 

o  

p  

t  

a  

l  

S  

t  

S  

r  

e  

e  

r  

(  

M  

a  

g  

C  

T  

o  

c  

e  

o  

a  

c  

2  

b  

p  

e  

t
 

c  

d  

i  

s  

(  

h  

e  

b
 

a  

i  

s  

g  

z  

c  

p  

a  

M  

d  

t  

d  

M  

i  

c  

w  

2  

a  

c  

a  

2  

c  

i  

c  

2  

R  

b  

C  

M  

d  

i  

a  

c
 

c  

H  

c  

t  

1  
s crucial for fully characterizing the outer regions of planetary
ystems. 

While high-contrast imaging has advanced our understanding of
ide-orbit planets, an alternative and complementary approach to
robing these outer regions involves studying exoKuiper belts, i.e.
old debris discs at tens of au (A. M. Hughes, G. Duchene & B.
atthews 2018 ; S. Marino 2022 ). These dusty belts, composed of
aterial ranging from observed μm-sized grains to inferred km-sized

nd larger planetesimals, are located at tens of au from their host stars
nd are considered a common feature of planetary systems, detected
round ∼20 per cent of nearby AFGK-type stars (B. Sibthorpe et al.
018 ). Through a collisional cascade, the large bodies in these cold
elts grind down into smaller dust grains, which produce the infrared
 xcess observ ed in these systems. This continuous replenishment
f dust counteracts removal processes such as collisions, radiation
ressure and Poynting–Robertson (PR) drag, allowing the discs to
ersist o v er ∼ Myr −Gyr time-scales (M. C. Wyatt 2008 ). 
Observations of debris discs at different wavelengths can provide

nformation regarding the distribution of different-sized grains. At
horter wavelengths (e.g. optical, near-infrared), smaller μm-sized
rains are probed as they scatter the light from the star, making
he disc visible. Such scattered light observations have been carried
ut with different instruments; for instance, on board Hubble Space
elescope ( HST ; e.g. D. A. Golimowski et al. 2011 ; G. Schneider
t al. 2014 ), with ground-based instruments like SPHERE (e.g. HR
796, TWA 7; J. Milli et al. 2017b ; B. Ren et al. 2021 , respectively)
r GPI (e.g. T. M. Esposito et al. 2020 ; K. A. Crotts et al. 2024 ), and
ore recently with JWST (e.g. F omalhaut, Ve ga, β Pic, F omalhault
, HD 181327, ε Eridani; A. G ́asp ́ar et al. 2023 ; K. Lawson et al.
024 ; I. Rebollido et al. 2024 ; K. Y. L. Su et al. 2024 ; S. G. Wolff et al.
025 ; C. Xie et al. 2025 , respectively). At millimetre wavelengths,
he thermal emission from larger mm-sized grains is detected using
he high sensitivity and resolution of ALMA (e.g. REASONS; L.

atr ̀a et al. 2025 and references therein). While the distribution
f smaller grains is affected by non-gravitational forces, the larger
rains remain largely unperturbed by these and are therefore thought
o trace the parent planetesimal population more accurately. As a
esult, the structure and morphology of debris discs can be imaged
nd studied to infer valuable insight into the formation and dynamical
volution of planetary systems. In the Solar system, for example, the
esonant populations within the Kuiper belt preserve evidence of
eptune’s migration history, while the edges of both the asteroid and
uiper belts encode information about the masses and orbits of giant
lanets (R. Malhotra 1995 ; S. Ida et al. 2000 ; A. Morbidelli et al.
005 ; A. Morbidelli & D. Nesvorn ́y 2020 ). 
The evidence for substructures and asymmetries in disc morphol-

gy is commonly interpreted as an indirect signature of unseen
lanetary companions. One of the most well-known examples is
he warp observed in the β Pic disc, which hinted at the presence of
 massive planet (D. Mouillet et al. 1997 ; J. C. Augereau et al. 2001 )
ater confirmed through direct imaging (A. M. Lagrange et al. 2009 ).
imilar warps have since been detected in other systems, suggesting

he presence of undetected perturbers (e.g. HD 110058, HD 111520;
. Stasevic et al. 2023 ; K. A. Crotts & B. C. Matthews 2024 ,
espectively). A minority of debris discs appear to be narrow and
ccentric (e.g. Fomalhaut, HR 4796, HD 202628; M. A. MacGregor
t al. 2017 ; G. M. Kennedy et al. 2018 ; V. Faramaz et al. 2019 ,
espectively), which may indicate perturbations by eccentric planets
M. C. Wyatt et al. 1999 ; T. D. Pearce & M. C. Wyatt 2014b ; G.

. Kennedy 2020 ; L. Rodet & D. Lai 2022 ). Other discs show
symmetric clumps that may result from resonant trapping or recent
iant collisions (e.g. β Pic; C. M. Telesco et al. 2005 ; Y. Han, M.
NRAS 546, 1–23 (2026) 
. Wyatt & W. R. F. Dent 2023 , ε Eri; M. Booth et al. 2023 ).
he sharpness of the inner edge can also provide insights into its
rigin, i.e. whether edges are sculpted by planets or consistent with
ollisional evolution alone (e.g. S. Marino 2021 ; A. Imaz Blanco
t al. 2023 ; R. R. Rafikov 2023 ; T. D. Pearce et al. 2024 ). As
bservational resolution continues to impro v e, substructures such
s gaps are increasingly being detected in debris discs, akin to those
ommonly observed in protoplanetary discs (S. M. Andrews et al.
018 ). Such observed gaps in debris discs are analogous to the gap
etween the asteroid and Kuiper belt in the Solar system, carved and
opulated by the g as giants. These g ap structures provide compelling
vidence for planetary companions that sculpt the morphology of
hese discs. 

HD 92945, HD 107146, and HD 206 893 are three well-
haracterized debris disc systems with observed gaps in their dust
istribution (S. Marino et al. 2018 , 2019 , 2020 , respectively). Direct
maging campaigns have searched for gap-carving planets in these
ystems, ruling out companions abo v e 2–5 M Jup at the gap locations
J. Milli et al. 2017b ; D. Mesa et al. 2021 ), though inner planets
ave been detected in HD 206 893 (J. Milli et al. 2017b ; S. Hinkley
t al. 2023 ). Therefore, identifying the putative planets predicted to
e embedded in these gaps remains an observational challenge. 
Nevertheless, the origin of gaps in debris discs remains uncertain

nd can be explained by multiple different dynamical mechanisms
nvolving the presence of unseen planets. The most straightforward
cenario involves a single planet embedded within the gap, which
ravitationally perturbs and scatters nearby debris within its chaotic
one. The amount of material being cleared at the gap location
an also vary. In some scenarios, Trojans can be captured on the
lanet orbit and can create a detectable ring of material (as seen
round TWA 7b, B. Ren et al. 2021 ; K. A. Crotts et al. 2025 ; A.
. Lagrange et al. 2025 ). The mass required for an in situ planet

epends on the assumptions about the disc mass. In the simplest case,
he disc is treated as massless, which does not capture planetesimal-
riven planetary migration (A. J. Mustill & M. C. Wyatt 2012 ; S.
arino et al. 2018 ). Ho we ver, if the disc mass is taken into account,

t can induce planet migration, allowing a less-massive planet to
arve a gap of comparable width as a single non-migrating planet
ould (D. R. Kirsh et al. 2009 ; S. J. Morrison & K. M. Kratter
018 ; M. F. Friebe, T. D. Pearce & T. L ̈ohne 2022 ). Alternatively,
 chain of multiple lower-mass planets distributed across the gap
ould collectively clear the region in the disc, whether the latter is
ssumed to be massless (A. Shannon et al. 2016 ; C. Lazzoni et al.
018 ) or massive (S. J. Morrison & K. M. Kratter 2018 ). In some
ases, a planet does not need to be located within the gap itself, and
nteractions from a planet located in the regions interior to the disc
an carve a gap further out. For example, gaps could arise from the
:1 mean-motion resonance (M. Tabeshian & P. A. Wiegert 2016 ; Z.
eg ́aly et al. 2018 ), or through secular apsidal resonances induced
y one precessing planet in a non-zero mass disc (T. D. Pearce & M.
. Wyatt 2015 ; X. Zheng et al. 2017 ; A. A. Sefilian, R. R. Rafikov &
. C. Wyatt 2021 , 2023 ) or two precessing planets in a massless

isc (B. Yelverton & G. M. K ennedy 2018 ). These v arious scenarios
mply different planet configurations and evolutionary histories, but
ll remain valid degenerate solutions until ruled out by observational
onstraints. 

This paper aims to provide a comprehensive view of the planet
onstraints that can be placed for the three systems HD 92945,
D 107146 and HD 206893, and to e v aluate, based on these, the

urrent understanding of possible gap-carving scenarios. In Sec-
ion 2 , we present new JWST /MIRI coronagraphic observations at
1 . 4 μm and describe the best practices for reducing JWST /MIRI



Gap-carving planets in exoKuiper belts 3 

c
t
p
p  

p
t  

t
t
o
t
a

2

T
t
a  

S
t
s
u

2

H  

s
d  

e  

2  

s
a  

M  

c
l
b
p
a
b
g

 

i
o  

a
o  

a
T  

t  

a

a
c
g
d  

t
a  

F  

p
 

T  

1  

A  

G  

Table 1. Stellar and debris disc properties for HD 92945, HD 107146, and 
HD 206893. The disc inner and outer edges are denoted by r in and r out , 
respectively, while the centre of the gap(s) is indicated by r gap . The gap 
widths ( w gap ) are calculated as the FWHM, based on the 1 σ values given in 
A. Imaz Blanco et al. ( 2023 ). 

Parameters HD 92945 HD 107146 HD 206893 

Distance (pc) 21 . 51 ± 0 . 01 (1) 27 . 47 ± 0 . 02 (1) 40 . 77 ± 0 . 06 (1) 

Spectral type K1V 

(2) G2V 

(3) F5V 

(4) 

Star mass (M �) 0 . 86 ± 0 . 01 (5) 1 . 03 + 0 . 02 
−0 . 04 

(6) 1 . 32 + 0 . 07 
−0 . 06 

(7) 

Age (Myr) 200 ± 100 (8 , 9 , 10) 150 + 100 
−50 

(8 , 9 , 10) 155 ± 15 (11) 

Inclination ( ◦) 65 . 4 ± 0 . 6 (12) 19 . 9 ± 0 . 6 (12) 40 ± 3 (12) 

PA ( ◦) 100 . 0 ± 0 . 6 (12) 153 . 3 ± 1 . 5 (12) 62 ± 4 (12) 

r in (au) 54 ± 2 (13) 44 ± 2 (13) 35 ± 8 (13) 

r out (au) 133 ± 6 (13) 144 . 3 ± 1 . 0 (13) 120 ± 20 (13) 

r gap , 1 (au) 72 . 0 ± 1 . 5 (13) 56 . 0 ± 0 . 7 (13) 69 ± 3 (13) 

w gap , 1 (au) 19 ± 9 (13) 7 . 7 ± 1 . 4 (13) 40 ± 9 (13) 

r gap , 2 (au) ... 78 . 3 ± 1 . 2 (13) ... 
w gap , 2 (au) ... 42 ± 6 (13) ... 

Sources: (1) Gaia Collaboration et al. ( 2023 ), (2) C. A. O. Torres et al. ( 2006 ), 
(3) E. A. Harlan & D. C. Taylor ( 1970 ), (4) R. O. Gray et al. ( 2006 ), (5) P. 
Plavchan et al. ( 2009 ), (6) T. D. Pearce et al. ( 2022 ), (7) P. Kervella et al. 
( 2004 ), (8) C. H. Chen et al. ( 2014 ), (9) W. S. Holland et al. ( 2017 ), (10) S. 
A. Stanford-Moore et al. ( 2020 ), (11) S. Hinkley et al. ( 2023 ), (12) S. Marino 
( 2021 ), (13) A. Imaz Blanco et al. ( 2023 ). 

h  

2  

s  

w  

g  

e  

B  

4  

a  

i  

r
t  

d  

2
p  

s

2

W
m  

c  

C  

d
p
a  

M  

a  

d
2

C  

m  

i
s

o

oronagraphic data using the SPACEKLIP package. Section 3 outlines 
he candidate vetting process and the construction of detection 
robability maps (DPMs), which are then used to rule out planetary 
arameters in each system. In Section 4 , we use these DPMs to
lace limits on potential planets, focusing on companions near 
he inner edge of the e xoK uiper belts and those responsible for
he observed proper-motion anomaly (PMa). Section 5 explores 
he various constraints on the planets responsible for carving the 
bserved gaps in these discs, with each gap-carving scenario tailored 
o the system’s specific characteristics. Finally, the main conclusions 
re summarized in Section 6 . 

 OBSERVATIONS  A N D  DATA  R E D U C T I O N  

his section presents JWST /MIRI coronagraphy observations of 
hree debris disc systems with known gaps: HD 92945, HD 107146, 
nd HD 206893, as part of the Cycle 1 programme GO 1668 (PI:
. Marino). Below, we outline the sample selection process for 

his programme, detail the used MIRI coronagraphy observation 
trategy, and finally describe the data reduction process performed 
sing SPACEKLIP , incorporating its latest upgrades. 

.1 Sample 

D 92945, HD 107146, and HD 206 893 originate from a larger
ample containing five debris discs with gaps identified in ALMA 

ata (S. Marino et al. 2018 ; C. Daley et al. 2019 ; M. A. MacGregor
t al. 2019 ; S. Marino et al. 2019 , 2020 ; A. Nederlander et al.
021 ). Gaps have been detected in four additional systems using
cattered light imaging alone (HD 141569, HD 131835, HD 120326, 
nd HD 141943; C. Perrot et al. 2016 ; M. Bonnefoy et al. 2017 ;
. Feldt et al. 2017 ; A. Boccaletti et al. 2019 , respectively). We

onsider discs resolved by ALMA because, contrary to scattered 
ight detections, ALMA probes the mm-sized grains unaffected 
y non-gravitational forces that can distort the true extent of the 
lanetesimal disc. Additionally, since the mm-sized dust grains 
re by-products of continuous planetesimal collisions, the distri- 
ution of planetesimals must enclose the presence of the observed 
aps. 

To image planets within the gaps of the discs, a face-on orientation
s optimal. In face-on systems, the JWST inner working angle 
bscures only the innermost region, allowing the rest of the disc and
ny potential planets to remain fully visible. Conversely, in edge- 
n systems, parts of the disc are obscured by the inner working
ngle, limiting the ability to detect planets embedded within the disc. 
herefore, given the geometry of the systems, we excluded two of

he five discs with gaps due to their edge-on orientation (HD 15 115
nd AU Mic). 

HD 92945, HD 107146, and HD 206893, share comparable 
ges, stellar masses, and distances, ho we ver, their specific disc 
haracteristics differ slightly (e.g. gap location, number of gaps, 
ap symmetry, etc). A comprehensive summary of the stellar and 
isc parameters for these systems can be found in Table 1 . All three
argets show strong evidence for an inner companion with astrometric 
ccelerations also known as PMa (P. Kervella et al. 2019 ; P. Kervella,
. Arenou & F. Th ́evenin 2022 ), measured from the comparison of
roper motions between the Hipparcos and Gaia catalogues. 
In particular, HD 92 945 is a 100 - 300 Myr old K1V star at 21.5 pc.

his system’s disc spans from 54 to 133 au and contains a single
9 au gap centred at 72 au (A. Imaz Blanco et al. 2023 ). Previous
LMA and HST observations hinted at asymmetry in the gap (D. A.
olimowski et al. 2011 ; S. Marino et al. 2019 , respectively), which
as been confirmed by recent JWST /NIRCam data (C. Lazzoni et al.
025 ). HD 107 146 is a 100 - 250 Myr old G2V star at 27.5 pc. Its disc
pans from 44 to 144 au and features a double-gapped radial profile
ith a narrow 7.7 au gap centred at 56 au followed by a wider 42 au
ap centred at 78 au (A. Imaz Blanco et al. 2023 ). Both gaps do not
xhibit strong signs of asymmetry (S. Marino et al. 2018 ; A. Imaz
lanco et al. 2023 ). HD 206893, a 140 - 170 Myr old F5V star at
1 pc, has a disc spanning from 35 to 120 au with a 40 au gap centred
t 69 au (A. Imaz Blanco et al. 2023 ). Although ALMA observations
ndicate possible gap asymmetry (S. Marino et al. 2020 ), the gap
emains consistent with being axisymmetric. This system also hosts 
wo substellar companions: HD 206 893 B, a 28 . 0 ± 2 . 2 M Jup brown
warf located at 9 . 6 ± 0 . 3 au (J. Milli et al. 2017a ; J. Kammerer et al.
021 ; S. Hinkley et al. 2023 ), and HD 206 893 c, a 12 . 7 ± 1 . 1 M Jup 

lanet at 3 . 53 ± 0 . 07 au found to be responsible for the PMa in the
ystem (S. Hinkley et al. 2023 ). 

.2 Obser v ations 

e present JWST /MIRI observations using the 4-quadrant phase- 
ask (4QPM; D. Rouan et al. 2000 ; C.-P. Lajoie et al. 2014 ; A. Boc-

aletti et al. 2015 ) and the F1140C coronagraphic filter as part of the
ycle 1 programme GO 1668 (PI: S. Marino). The programme was
esigned to obtain an optimal data reduction, allowing for different 
oint-spread function (PSF) subtraction methods to be performed: 
ngular differential imaging (ADI; M. M ̈uller & G. Weigelt 1987 ; C.
arois et al. 2006 ), reference differential imaging (RDI; G. Ruane et

l. 2019 ), and a combination of ADI and RDI (ADI + RDI). All the
ata reduction was conducted using SPACEKLIP (J. Kammerer et al. 
022 ; A. L. Carter et al. 2023 ). 
Based on pre-launch estimates (from PANCAKE simulations; A. L. 

arter et al. 2021b ), the MIRI F1140C filter provided the best planet
ass sensitivity at the location of the gaps for the three systems. The

nstrument settings were optimized using the PANCAKE tool and are 
ummarized in Table 2 . 

Following pre-launch performance predictions, we opted for an 
bserving strategy allowing for two roll angles on the science 
MNRAS 546, 1–23 (2026) 



4 R. Bendahan-West et al. 

M

Table 2. Observing parameters for the GO 1668 JWST programme, with science targets in bold and corresponding PSF references in italics. Spectral 
types and Kmag are obtained from Simbad. Exposure times represent total durations per star; individual exposures require division by the number of 
dithers and rolls. Note that background observations were performed for every science roll and following the dithered reference observations, all using 
the same settings as their counterpart observations. 

Star SpT Kmag Readout N groups N ints t exp (s) N dithers N rolls Roll angle ( ◦) t total (s) 

HD 92945 K1V 5.660 FASTR1 1251 6 1800.236 1 2 7 3600.472 
HD 95234 M1III 1.537 FASTR1 145 2 69.747 9 1 – 627.722 

HD 107146 G2V 5.540 FASTR1 1251 6 1800.236 1 2 7 3600.472 
HD 111067 K4III 1.917 FASTR1 275 2 132.064 9 1 – 1188.573 

HD 206893 F5V 5.593 FASTR1 1251 6 1800.236 1 2 7 3600.472 
HD 208445 M4III 2.061 FASTR1 240 2 115.286 9 1 – 1037.575 

t  

o  

t  

P  

r  

t  

t
 

C  

t  

t  

s  

p  

s  

s  

t  

C  

a  

A  

o  

s
 

c  

B  

r  

n  

t  

t  

b  

a  

w  

o  

r  

m  

c

2

T  

b  

(  

d  

1

2

i
s
3

p  

p  

n  

L  

v  

D  

j
 

t  

L  

o  

t  

S  

(  

t
 

a  

r  

o
 

d  

L  

r  

e  

d  

M  

T  

t  

m
 

a  

r  

t  

c
 

w  

c  

b  

c  

t  

b  

p  
argets in addition to PSF reference stars. Each roll for the science
bservations is shifted by 7 ◦ with 30-min exposure times. To improve
he spatial diversity for the PSF subtraction, we performed nine-
OINT-SMALL-GRID dithers (C.-P. Lajoie et al. 2016 ) for the
eference stars. The exposure time of the PSF stars is scaled such that
he signal to noise achieved per dither observation is comparable to
he signal to noise of the science observations. 

The different reference stars have been selected using the Search-
al 1 tool. We looked for the brightest star within 20 ◦ of the science

arget and that would not saturate in less than five groups in
he target acquisition. Note that spectral type matching between
cience and reference targets is not problematic for observations
ast 5 μm, as the emission from the science targets and reference
tars is in the Rayleigh–Jeans regime. 2 We also verified that these
tars lack any stellar companions within ∼100 au, as indicated by
he Gaia Renormalized Unit Weight Error (RUWE) < 1 . 4 (Gaia
ollaboration et al. 2023 , and more details in Section 3.3.5 ) and
strometric accelerations larger than 3 σ (P. Kervella et al. 2019 ).
dditionally, we ensured there was no infrared excess indicative
f a discernible disc that could interfere with the chosen reference
tars. 

For MIRI coronagraphy, background observations are required to
ounter an inherent stray light artefact known as the ‘glow stick’ (A.
occaletti et al. 2022 ), which is seen to dominate any science or

eference observations. The recommended solution is to observe a
earby area of the sky without bright sources, using the same integra-
ion parameters (i.e. same number of groups, integrations, exposure
ime) as the corresponding science or reference observations. This
ackground observation will identically reproduce the ‘glow stick’,
llowing for an optimal subtraction of this artefact. The programme
as divided into three uninterrupted sequences, each dedicated to
ne of the science stars. Each sequence included the two science
olls, reference dithers, and background observations. This approach
inimizes potential wavefront drifts between observations, which

ould otherwize compromise the PSF subtractions. 

.3 Data reduction with SPACEKLIP 

he data reduction process follows the example and guidance given
y the ERS-01386 coronagraphic programme on HIP 65 426 b
S. Hinkley et al. 2022 ; A. L. Carter et al. 2023 ). The entire
ata processing uses the python package SPACEKLIP 3 , a custom
NRAS 546, 1–23 (2026) 

 https:// www.jmmc.fr/ english/ tools/ proposal-preparation/ search-cal/ 
 https:// jwst-docs.stsci.edu/ methods-and-roadmaps/ jwst-high-contrast- 
maging/jwst- high- contrast- imaging- proposal- planning/hci- psf- reference- 
tars 
 https:// github.com/ spacetelescope/ spaceKLIP 

4

5

1
6

ipeline combining coronagraphic tools such as the official JWST

ipeline 4 (H. Bushouse et al. 2022 ) and PYKLIP subtraction tech-
iques (J. J. Wang et al. 2015 ). The reduction used the SPACEK-
IP version 2.2.1.dev15 + gf97eb19 , the JWST pipeline
ersion 1.18.1, the PYKLIP version 2.8.2, the CAlibration REference
ata System (CRDS) version 12.1.10, and the CRDS context file
wst 1364.pmap . 
We downloaded the uncalibrated data files (version 2023 2a) from

he MAST archive 5 and followed through the different SPACEK-
IP reduction stages, as described below. Note that here we focus
n the key SPACEKLIP parameters and highlight any deviations from
he default pipeline. For a more comprehensive description of the
PACEKLIP steps, refer to J. Kammerer et al. ( 2022 ), A. L. Carter et al.
 2023 ), and the readthedocs page. 6 The next steps are presented in
he order they were applied during data reduction. 

I. Stage 1: Most of the Stage 1 parameters have not been changed
nd followed the default values used in the MIRI coronagraphy tuto-
ial available online on the SPACEKLIP readthedocs page 6 (accessed
n 20/08/2025). The only changes are the following: 
(a) LIKELY ramp fitting: The ramp fitting used in this paper

iffers from the default ramp fitting algorithm OLS C (Ordinary
east Squares). While OLS C fits straight-line segments to the

amp assuming uniform Gaussian noise, the LIKELY algorithm we
mployed instead maximizes a likelihood function that models the
etector’s true noise properties, including both photon and read noise.
ore details about this method are found in T. D. Brandt ( 2024a , b ).

his approach provides more accurate slope estimates and impro v es
he identification of cosmic-ray events, leading to more reliable flux

easurements. 
We found that in all the uncalibrated files, the first ≈10 groups

nd the last 2 groups of each inte gration behav e differently from the
emaining groups (see Fig. A1 ). To account for this, we excluded
hese 12 groups during ramp fitting, which results in only a ≈1 per
ent loss of usage groups. 

(b) Brighter-Fatter effect (BFE) correction: The BFE appears
hen the detector response becomes non-linear due to strong flux

ontrasts between neighbouring pixels, resulting in distorted and
roadened PSFs near bright pixels (I. Argyriou et al. 2023 ). In our
ase, the reference stars have a higher detector count than the science
argets, producing noticeable BFE and causing the reference PSF to
e different and broader near those brighter pix els. F or the ERS
rogramme observing HIP 65 426 (A. L. Carter et al. 2023 ), the BFE
 https://jwst-pipeline.readthedocs.io 
 The data described here may be obtained from the MAST archive at doi: 
0.17909/gsf1-2q34. 
 https:// spaceklip.readthedocs.io/ en/ latest/ index.html 

https://www.jmmc.fr/english/tools/proposal-preparation/search-cal/
https://jwst-docs.stsci.edu/methods-and-roadmaps/jwst-high-contrast-imaging/jwst-high-contrast-imaging-proposal-planning/hci-psf-reference-stars
https://github.com/spacetelescope/spaceKLIP
https://jwst-pipeline.readthedocs.io
https://doi.org/10.17909/gsf1-2q34
https://spaceklip.readthedocs.io/en/latest/index.html


Gap-carving planets in exoKuiper belts 5 

w
a
d
p
B  

u

i
t  

t
b
a
‘
n
m  

o  

t
b
s
b

 

i  

p

p  

c
m  

w
t
t

Q  

i  

i  

u
r  

3  

m  

f
a  

t
c  

a  

t
d

s
c
v  

o
fi
m
c  

c
b
s
i

t
i  

(  

r
a

T  

s
 

n  

c

o
i  

b  

i
a  

f  

J  

i

i
w
e  

m
p  

i
f

t
i  

n  

(  

u  

H

P  

m  

n  

w  

1  

m  

w  

K  

p  

o  

(
 

p
i
s  

c
a
i
s  

c  

s

3

T  

e
w
c
a  

7 https:// jwst-docs.stsci.edu/ mid-infrared-instrument/ miri-operations/ miri- 
target- acquisition/miri- coronagraphic- imaging- target- acquisition 
as negligible as the detector count levels between the reference 
nd science observations were better matched. This discrepancy was 
ue to the observing parameter optimization in PANCAKE used for 
rogramme GO 1668, which preferred long ramps o v er matching 
FE. A new functionality to match the detector counts in PANCAKE is
nder development. 
To correct for this effect, we use the mask groups option 

mplemented in the Stage 1 pipeline of SPACEKLIP . This function 
rims the necessary amount of groups during the ramp fitting of
he brighter images, in our case the reference and corresponding 
ackground observations, to match the BFE between the reference 
nd science PSFs. Three masking methods are available: ‘custom’, 
basic’ and ‘advanced’. The ‘custom’ method uniformly trims a fixed 
umber of groups across all pixels and integrations. The ‘basic’ 
ethod optimizes the number of groups to trim o v er a central subset

f the image, where the PSF lobes and BFE are strongest. We adopt
he ‘advanced’ method, which performs the optimization on a pixel- 
y-pixel basis and yields the most accurate correction for this data 
et. More details about the ‘advanced’ method and the differences 
etween the methods are provided in Appendix B . 

II. Stage 2: The main goal of this step is to calibrate each
ntegration of the images from counts/s to MJy/sr. We kept the default
arameters. 
III. Image processing: The following steps describe the steps 

erformed before the PSF subtraction. We build on the online MIRI
oronagraphy tutorial 6 , with parameters empirically adjusted to best 
atch this data set. Here, we only focus on the parameters that
ere modified from their default values. In addition, we introduce 

wo new functions: background observation cleaning and persistence 
rimming. 

(a) Bad pixel cleaning: First, the pixels flagged in the Data 
uality array during the ramp fitting stage were filled in with the
nterp2d method with a kernel size of 7 pixels. We found that

ncreasing the kernel size from the default 3 pixels to 7 while
sing interp2d produced less residuals in the final PSF-subtracted 
eductions. We then used sigma-clipping ( sigclip ) with a sigma of
 to find additional bad pixels and cleaned them using the interp2d
ethod with a kernel size of 7 pixels. We then identified bad pixels

rom temporal variations across integrations using timeints and 
 sigma of 4, which we then cleaned with timemed . Finally,
he remaining ∼5 - 10 bad pixels were cleaned manually with the 
ustom method, using an interp2d kernel of 7 pixels. It is not
l w ays possible to apply an accurate correction to all bad pixels (e.g.
hose in PSF lobes), which can leave some visible residuals (see 
iscussion in Section 3.1 ). 
(b) Background observation cleaning: Although background ob- 

ervations are designed to sample an empty region of the sky, 
ontaminating sources can still fall within the background field of 
iew. This was observed in the science backgrounds of the HD 92945
bservations, which resulted in ne gativ e sources contaminating the 
nal reductions for this star (Fig. C1 ). Since background observations 
ust be repeated and centred on diagonally opposed quadrant, we 

an compare the two backgrounds to identify and remo v e such
ontaminants. This is done automatically by flagging ±3 σ outliers 
etween the two background images, yielding cleaner final PSF- 
ubtracted reductions (see Appendix C ). The method is implemented 
n SPACEKLIP as clean backgrounds . 

(c) Background subtraction: We apply the sub- 
ract background godoy function from SPACEKLIP , described 

n N. Godoy et al. ( 2024 ). Consequently, unlike A. L. Carter et al.
 2023 ), we do not discard the first integration of the science or
eference observations, which was previously done because of an 
pparent increase in noise explained by reset switch charge decay. 
his method therefore prevents from losing 16 per cent of the
cience observations and 50 per cent of the reference observations. 

(d) Ima g e alignment: Following A. L. Carter et al. ( 2023 ), we did
ot realign the MIRI images due to the challenging alignment of the
omplex MIRI 4QPM PSF structure. 

(e) Persistence trimming: Bright sources can leave a lasting imprint 
n the JWST detectors through a phenomenon known as persistence, 
n which the signal from earlier exposures is not completely cleared
efore the following ones (D. Dicken et al. 2024 ). Persistence can
ntroduce spurious structure into the data that may be misinterpreted 
s real sources. We detect this effect in all the MIRI observations
rom this programme (Fig. C2 ), as well as in other programs (e.g.
ames et al. in preparation; see also the source identified as a speckle
n N. Godoy et al. 2024 ). 

During MIRI coronagraphic observations, the star is first placed 
n two different target acquisition (TA) regions, 7 which coincide 
ith the locations where persistence is observed. To mitigate this 

ffect, we identify the stellar positions in both TA images and
ask the corresponding pixels in any integrations affected by 

ersistence. The full process is described in Appendix D and is
mplemented in SPACEKLIP as the persistence trimming 
unction. 

IV. PSF subtraction: The PSF subtraction techniques used follow 

hree different principal component analysis (PCA)-based methods 
mplemented in SPACEKLIP through PYKLIP : ADI, RDI and a combi-
ation of ADI + RDI. More details are provided in A. L. Carter et al.
 2023 , Section 2.6). Fig. 1 shows an o v erview of the resulting images
sing all PSF subtraction strategies for HD 92945, HD 107146, and
D 206893. 
There are two main additional parameters to include during the 

SF subtraction: the number of KLIP PCA modes (also known as KL
odes) used to determine ho w aggressi ve the subtraction is, and the

umber of annuli/subsections (A. L. Carter et al. 2023 ). In our case,
e are limited to 6 KL modes for ADI (6 integrations in a single roll),
8 KL modes for RDI (9 dithers of 2 integrations each), and 24 KL
odes for ADI + RDI (sum of ADI and RDI). The data reduction
as done on the full range of KL modes [1, max], and we scan all the
L modes to look for significant point sources. The ability to find
oint sources depends on the number of KL modes used (e.g. level
f residuals and o v ersubtraction) and ultimately affects the contrast
see Section 3.2 ). 

PSF subtraction can be performed on images as a whole or images
artitioned into different annuli and subsections. Splitting the image 
nto different annuli/subsections allows the possibility to target PSF 

ubtraction in specific regions of the image. While the image gets
learer (i.e. fewer residuals) when increasing the number of annuli 
nd subsections, the resulting contrast deteriorates through a decrease 
n the algorithmic throughput of the subtraction process (i.e. any 
ignal suffers from stronger o v ersubtraction). We deriv ed the deepest
ontrast when using the image as a whole i.e. a single annulus and a
ingle subsection. 

 ANALYSI S  

his section details the analysis of the PSF-subtracted images taken to
 xtract an y planet candidate detection. We first examine in Section 3.1 
hether the sources observed in the processed images are planet 

andidates or background contaminants based on multiwavelength 
nd multi-epoch archi v al observ ations of the three systems. After the
MNRAS 546, 1–23 (2026) 

https://jwst-docs.stsci.edu/mid-infrared-instrument/miri-operations/miri-target-acquisition/miri-coronagraphic-imaging-target-acquisition#gsc.tab=0


6 R. Bendahan-West et al. 

M

Figure 1. Grid displaying the unsubtracted images and the results of the different PSF subtraction techniques (ADI, RDI, and ADI + RDI) used for HD 92945, 
HD 107146, and HD 206893. All unsubtracted data share the same colour scale, while all the PSF subtracted data share a different colour scale, both with units 
of MJy/sr. The data reduction was done using SPACEKLIP as described in Section 2.3 , with the images shown using the maximum KL modes. The full MIRI field 
of view (24 arcsec × 24 arcsec ) is shown with all images rotated to north-east. The scale bar at the bottom right corner represents a projected distance of 50 au. 

s  

t  

o  

a

3

A  

c  

w  

b  

a  

f  

s
 

i  

T  

t  

a  

c  

B  

a

 

e  

w  

c  

s  

s  

g  

d  

a  

w  

S  

s  

g  

s  

o  

o
 

l  

C  

(  

d  

b  
ource vetting, we quantify in Section 3.2 the contrast sensitivity of
he observations and constrain the potential mass and semimajor axis
f planets in the systems using detection probability maps (DPMs)
nd PMa in Section 3.3 . 

.1 Source vetting 

s illustrated in Fig. 1 , the processed images contain several
andidate sources. Ho we ver, we found that no sources are consistent
ith a companion planet. All the observed candidates are found to be
ackground objects. From the PSF-fitting, the majority of the sources
re extended, indicating that they are likely background galaxies. The
ollowing section describes the vetting process characterizing each
ource. 

A prominent residual at ≈1 arcsec from the image centre is
dentified in the RDI panels of HD 92945 and HD 107146 (Fig. 1 ).
his residual is caused by a bad pixel within a PSF lobe, which makes

he lower-left lobe (in the de-rotated frame) consistently the brightest
cross all observations. This asymmetry in the PSF is difficult to
orrect during the BFE correction and PSF subtraction steps (Fig.
1 ). We do not consider this feature as a candidate source in our
nalysis. 
NRAS 546, 1–23 (2026) 
F or the v etting process, we e xamined multiwav elength and multi-
poch archi v al data for all three systems to verify if the same sources
ere detected previously. For data collected at different epochs, we

onsidered the star’s proper motion to determine whether an observed
ource is co-moving with the star or is a background object. Most
ources detected in the JWST /MIRI 11.4 μm observations are back-
round contaminants identified in archi v al ALMA (mm and sub-mm)
ata, archi v al high-contrast imaging data with HST (optical and IR),
nd the Gaia archive. Additionally, we forward modelled each source
ith a point-like PSF using the extract companions function in

PACEKLIP (see resulting fits in Fig. E1 ). Planets and background stars
hould be well fitted with a point-like model whereas background
alaxies should be better fitted with an extended PSF model. Table 3
ummarizes the properties of the sources observed within the field
f view of the observations, while Fig. 2 compares the JWST /MIRI
bservations with archival ALMA observations. 
Four sources are detected in the HD 92945 observations (the

eft panels of Fig. 2 ). By tracing back the stellar proper motion,
1, C2, and C3 align with counterparts in archi v al HST /ACS data

F606W and F814W; D. A. Golimowski et al. 2011 ) and in ALMA
ata (0.86 mm, S. Marino et al. 2019 ). Their expected positions as
ackground objects are marked by yellow circles in the lower panel



Gap-carving planets in exoKuiper belts 7 

Table 3. Properties of the sources detected within the field of view of the observations. Sources are numbered from north to south, as shown in Fig. 2 . Columns 
4–6 list the coordinates derived from the PSF-fitting, with the corresponding observation epoch given in column 2. The source morphology (extended or 
point-like) is determined from the PSF-fitting results shown in Fig. E1 . For extended sources, fluxes are measured via aperture photometry, whereas the flux of 
HD 92945 C4 is derived from point-source fitting. We note that the flux of HD 107146 C1 may be affected by its location between two quadrants of the 4QPM. 

Target Obs. date Source � RA, � Dec RA Dec Flux 11 μm Extended? Archi v al detection? 
(arcsec) (s) (arcsec) ( μJy) 

HD 92945 13 June 2023 C1 −2.6, 11.4 10:43:27.69 −29.03.41.21 103.5 Y ALMA + HST 
C2 −5.3, 1.2 10:43:27.48 −29.03.51.41 19.8 Y ALMA + HST 
C3 −4.9, −1.9 10:43:27.51 −29.03.54.51 2.9 Y ALMA + HST 
C4 −4.3, −4.5 10:43:27.56 −29.03.57.11 6.8 N Gaia 

HD 107146 15 June 2023 C1 −1.7, −3.2 12:19:06.10 16.32.47.17 35.5 Y HST 

HD 206893 19 June 2023 C1 1.0, 10.9 21:45:22.12 −12.46.49.17 8.7 Y N 

C2 4.2, 4.9 21:45:22.34 −12.46.55.17 8.5 Y N 

C3 −5.9, −10.1 21:45:21.65 −12.47.10.17 92.3 Y HST 

Figure 2. Comparison between the 11 . 4 μm JWST /MIRI data (top row) and archi v al ALMA observ ations (bottom ro w) for HD 92945, HD 107146, and 
HD 206893 obtained from S. Marino et al. ( 2018 , 2019 , 2020 , respectiv ely). The radial e xtent of the ALMA disc is o v erplotted on the JWST data to guide the eye. 
The positions of the sources in the JWST observations are traced back to account for the proper motion of these stars to determine whether they are co-moving 
or background objects. Solid yellow circles indicate sources observed in both JWST and ALMA (C1, C2, and C3 in HD 92945, and C3 in HD 206893), while 
dashed yellow circles denote sources without ALMA counterparts (C4 in HD 92945, C1 in HD 107146, and C1 and C2 in HD 206893). Only C4 in HD 92945 
is consistent with being a star, while the others are consistent with background galaxies.The scale bar in the bottom right corner represents a projected distance 
of 50 au. The white circle in the bottom left corner of the ALMA plots denotes the beam size used in the observations. 

o  

(  

s
t
o  

5
p  

c
G  

a

 

v  

A  

i  

g  

e  

(  

P  

s

f Fig. 2 , consistent with ALMA detections at > 3 σ significance
5.5 σ , 4.2 σ , 3.2 σ , for C1, C2, C3, respectively). PSF-fitting further
hows that all three sources are extended (Fig. E1 a), supporting 
heir interpretation as background galaxies. Within a 15 arcsec radius 
f HD 92945, the Gaia archive reports only one star, Gaia DR3
455707157212258048, which, after accounting for HD 92945’s 
roper motion, lies ∼0.6 arcsec from the position of C4. The offset
ould reflect the absence of a proper motion measurement for the 
aia star. PSF-fitting shows that C4 is consistent with a point source

nd is therefore most likely a background star. 
For HD 107146, only one source is detected within the field of
iew (middle panels of Fig. 2 ). Although it is not seen in archival
LMA observations (1.14 mm; S. Marino et al. 2018 ), the source

s clearly visible in archi v al HST /ACS imaging as a background
alaxy (F606W and F814W; D. R. Ardila et al. 2004 ; G. Schneider
t al. 2014 ). Its predicted location within the disc around 2020
G. Schneider et al. 2014 ), is consistent with current observations.
SF fitting (Fig. E1 b) confirms that the source is extended, further
upporting its classification as a background galaxy. We note that 
MNRAS 546, 1–23 (2026) 



8 R. Bendahan-West et al. 

M

i  

b
 

p  

p  

s  

w  

(  

m  

a  

H  

c  

t  

g  

p  

s

3

E  

b  

s  

S  

o  

fl  

c  

a  

o  

p  

a  

c
 

P  

i  

K  

1  

i  

a  

t  

a  

i
 

r  

c  

a  

B  

e  

d  

f  

t  

e  

t  

i  

i  

a  

w  

0  

i  

p  

E  

r  

Figure 3. Calibrated 5 σ contrast curves (left) for the MIRI 1140C obser- 
vations with the respective mass sensitivities in units of M Jup (right), as a 
function of projected separation. Black curves correspond to MIRI F1140C 

contrasts, blue curves correspond to PANCAKE contrast predictions, and for 
HD 92945 only, the red curve corresponds to the NIRCam F444W contrast 
(from GO 3989, C. Lazzoni et al. 2025 ). The different line styles represent 
contrast obtained using different PSF subtraction techniques (i.e. ADI, RDI, 
and ADI + RDI). All contrasts were calculated with SPACEKLIP using the 
maximum number of KL modes and with 1 annulus/1 subsection. 

o  

p
 

p  

e  

m  

(  

S  

e  

a  

(  

p  

i
 

b  

(  

8 https:// github.com/ vsquicciarini/ madys 
ts measured flux may be strongly attenuated by its position at the
oundary between two quadrants of the 4QPM. 
Three sources are visible in the HD 206893 observations (the right

anels of Fig. 2 ). None of the sources are consistent with the two
re viously kno wn companions in the system, HD 206893 B and c (at
eparations of 0.2 and 0.1 arcsec, respectively, S. Hinkley et al. 2023 ),
hich lie too close to the star within the MIRI inner working angle

0.36 arcsec, A. Boccaletti et al. 2022 ). By tracing back the proper
otion of HD 206893, we find that C3 aligns with counterparts in

rchi v al ALMA data (0.89 mm, S. Marino et al. 2020 ) and raw
ST /NICMOS data (F110W and F160W; J. Milli et al. 2017b ),

onfirming it as a background object. PSF-fitting (Fig. E1 c) shows
hat C3 is extended, consistent with being classified as a background
alaxy. The other two sources, C1 and C2, have no counterparts in
re vious observ ations. PSF-fitting indicates that both are extended,
uggesting they are also likely background galaxies. 

.2 Contrast cur v es 

ven though we did not detect any companions, non-detections can
e used to set upper limits on the presence of planets in these
ystems. For this, we use calibrated contrast curves, produced by the
PACEKLIP package, which also take into account the transmission
f the coronagraphic mask. These contrast curves correspond to the
ux level of a companion sufficient to produce a 5 σ detection. The
ontrast curve is calibrated by injecting f ak e companions at different
ngular separations and performing the PSF subtraction. As the flux
f the injected planets is known, the flux that is being lost while
erforming the PSF subtraction can be quantified and a correction
pplied to the contrast derived. Fig. 3 demonstrates the contrast
apabilities for the obtained MIRI observations. 

The contrast depth depends on the number of KL modes used for
SF subtraction. The deepest contrast for each subtraction method

s illustrated in Fig. 3 , corresponding to the maximum number of
L modes: 6 for ADI, 18 for RDI, and 24 for ADI + RDI. Beyond
 mode for ADI, 6 modes for RDI, and 6 modes for ADI + RDI,
ncreasing the number of KL modes produces little change in the
chieved contrast, although the deepest values are still obtained at
he maximum. We also note that RDI provides the best performance
t separations < 1 arcsec, while ADI + RDI yields deeper contrast
n the background-dominated regime at larger separations. 

Additionally, we compare the measured contrasts to literature
esults and predicted contrast performance. We reach a sensitivity
orresponding to contrast ratios of ≈5 × 10 −5 for separations � 1 . 5
rcsec, comparable to typical JWST /MIRI performance (e.g. A.
occaletti et al. 2022 ; E. C. Matthews et al. 2024 ; M. M ̂ alin
t al. 2025 ; A. Sanghi et al. 2025 ). Ho we ver, we measure a
iscrepancy between the observed contrast and predicted sensitivities
rom PANCAKE simulations (Fig. 3 ), which is more significant in
he background-dominated regime ( > 1 arcsec). This mismatch is
xpected for MIRI coronagraphic observations, as PANCAKE simula-
ions currently o v erpredict contrast capabilities. This o v erprediction
s due to several MIRI-specific artefacts that are not included
n PANCAKE simulations (including the ‘glow stick’ subtraction
nd BFE correction). Although the PANCAKE MIRI predictions
ere empirically scaled to match the performance from the ERS-
1386 programme on HIP 65 426 b (A. L. Carter et al. 2023 ), the
mpact of background subtraction and BFE appears to be more
ronounced in the presented GO 1668 observations than in the
RS programme. This therefore highlights the need for further

efinement of MIRI PANCAKE predictions to accurately reflect current
NRAS 546, 1–23 (2026) 
bservational limits and using empirical MIRI contrast curves may
rovide more reliable performance estimates. 
To convert contrast into mass sensitivity, as shown in the right

anels of Fig. 3 , we use a combination of ATMO-ceq (M. W. Phillips
t al. 2020 ) and BEX (E. F. Linder et al. 2019 ) planetary evolution
odels, joined together following the approach in A. L. Carter et al.

 2021a ). This conversion is done using the MADYS package 8 (V.
quicciarini & M. Bonavita 2022 ), using the bex-atmo2023-ceq
volution model. For Fig. 3 , the system ages are fixed to 200, 150,
nd 155 Myr for HD 92945, HD 107146, HD 206893, respectively
see Table 1 ), and the contrast and masses are shown as a function of
rojected separation (age uncertainty and de-projection will be taken
nto account in Section 3.3 ). 

For HD 92945, we compare the contrast and mass sensitivities
etween JWST /MIRI at 11 . 4 μm and JWST /NIRCam at 4 . 4 μm
C. Lazzoni et al. 2025 ). While NIRCam delivers significantly

https://github.com/vsquicciarini/madys


Gap-carving planets in exoKuiper belts 9 

d  

s
w
e
G  

e  

N
e
i  

e
M  

d
a

3

I  

w  

a  

p
f  

s  

p
a  

o
o  

g
t

1
l  

m
w
w
p  

a
a
m  

r  

a
s

 

a
i  

s
a
o
d

c  

c
t
p  

t
s

c
i

9

1

Figure 4. DPMs for HD 92945, HD 107146, and HD 206893 (top to bottom). 
The blue shading shows the probability of a 5 σ planet detection with MIRI at 
11.4 μm, with contours marking the 50, 95, and 99.7 per cent confidence 
lev els. F or HD 92945, the red dash–dotted line shows the 99.7 per cent 
detection limit from NIRCam at 4.4 μm. The darker grey-shaded regions 
denote constraints from archi v al direct imaging and RV data, and grey dotted 
regions from Gaia astrometry (RUWE). The orange hatching marks limits 
imposed by the disc morphology, where planets within 3R Hill of the disc edges 
would disrupt it (Section 3.3.2 ). The purple curve describes the planet mass 
and separation combinations required to explain the observed PMa signal. 
The green solid line highlights the orbital parameters for a planet shaping the 
eeper contrast than MIRI, the corresponding impro v ement in mass
ensitivity is less pronounced. There is also a significant caveat here, 
here NIRCam wavelengths are much more sensitive to potential 

ffects of disequilibrium chemistry and clouds (e.g. D. C. Bardalez 
agliuffi et al. 2025 ; K. A. Crotts et al. 2025 ), so the chemical

quilibrium model (ceq) used is likely the most optimistic case for
IRCam. We also strengthen that the calculated sensitivities are 

xpressed as a function of projected separation, and de-projection 
s required for robust constraints on planet mass and location. For
xample, the difference in mass sensitivity between NIRCam and 
IRI appears more prominent in Fig. 3 , whereas in Fig. 4 the

iscrepancy is reduced once we take into account inclination, stellar 
ge uncertainty and planet eccentricity (see Section 3.3 ). 

.3 Detection probability maps (DPM) 

n this section, we use nondetections and contrast curves to assess
hich types of planets can be ruled out as a function of their mass

nd semimajor axis. Since contrast curves are defined in terms of
rojected separation, we must de-project them by accounting for the 
ull range of orbital configurations that could place a planet at a given
eparation (e.g. M. Bonavita 2020 ). To do this, we build detection
robability maps (DPMs), calculated by combining the MADYS tool 
nd an impro v ed v ersion of EXODMC, 9 which is now a dependenc y
f MADYS . This tool generates DPMs and indicates the probability 
f detecting a planet of a given mass and semimajor axis if it exists,
iven the achieved contrast. Given non-detections, these DPMs allow 

o rule out planets with a high detection probability. 
To build such maps, we define a grid of planet masses (0.01–

00 M Jup ) and semimajor axes (0.1–12 arcsec) uniformly spaced in 
og-space. F or ev ery point in this grid, we first convert the planet

ass to a predicted magnitude using planetary evolution models, 
ith the stellar age and uncertainties sampled uniformly . Specifically , 
e use the same bex-atmo2023-ceq model in MADYS , as done 
reviously for Fig. 3 . While we define a broad parameter grid, the
chieved contrast limits only probe planet masses where the models 
re fully tabulated, requiring no extrapolation for the detectable 
ass range in any of the three systems. Ho we ver, such models

emain only partially validated for effects such as cloud co v erage
nd disequilibrium chemistry, which could affect the estimated mass 
ensitivities. 

For each point in the grid, we also convert semimajor axis into
 distribution of projected separations. We assume that any planet 
n the outer regions is coplanar with the disc and compute a set of
eparations by drawing random orbital configurations. Inclinations 
re sampled from a normal distribution constrained by ALMA 

bservations, while eccentricities are drawn from a half-Gaussian 
istribution 10 centred at 0 with a standard deviation of 0.1. 
These predicted magnitudes and projected separations are then 

ompared to the contrast limits of the observations. A grid point is
onsidered detectable if the planet’s expected magnitude is brighter 
han the 5 σ contrast limits at the corresponding separation. This 
rocess is repeated across the entire grid to compute the 5 σ de-
ection probability, incorporating random orbital configurations and 
ampling o v er the system’s age range to account for age uncertainties. 

When calculating the DPMs, we use the deepest available contrast 
urves to exclude the widest possible range of planet configurations 
n each system. For these MIRI observations, we have decided 
MNRAS 546, 1–23 (2026) 

 https:// github.com/ mbonav/ Exo DMC 

0 Ne gativ e values are discarded. 

disc inner edge (Section 4.1 ), and the black dot with a question mark marks 
the possible parameters for a planet located at the centre of the gaps in a 
massless disc (Section 5.2.1 ). For HD 206893, the light-grey region rules out 
planet parameters to ensure planet stability based on mutual R Hill . All DPM 

components are detailed in Section 3.3 . 

https://github.com/mbonav/Exo_DMC


10 R. Bendahan-West et al. 

M

t  

t  

a  

t  

m  

t  

i  

r  

t
 

D  

(  

i  

d  

r  

s  

t  

e  

t  

a  

s  

d  

r  

M  

L  

2  

l  

l
 

r  

t  

a  

d  

(

3

T  

c  

i  

a  

c  

r  

H  

a
 

C  

c  

a  

v  

a
 

s  

a  

(  

c  

D  

m  

1

1

(  

a  

s  

1
S  

n  

e  

p  

a  

e

3

I  

s  

v  

d  

h  

o  

w
(  

2  

e  

D

m

w
t  

d  

w
 

f  

w  

e  

a  

i

3

F  

o  

c  

t  

a  

t  

a
 

c  

s  

w  

(  

4
 

v
z  

s  
o combine the contrasts obtained from the three PSF subtraction
echniques illustrated in Fig. 3 . Since no planets were detected across
ll subtraction methods and KL modes, we select, at each separation,
he best (i.e. lowest) contrast achieved across all techniques and KL

odes. While this approach only yields a marginal impro v ement o v er
he contrast from ADI + RDI with maximum KL modes, it results
n a contrast that maximizes the sensitivity to potential planets. The
esulting DPMs are shown in Fig. 4 , where the shade of blue indicates
he probability of a 5 σ planet detection. 

For HD 92945, we have added the 99.7 per cent contour from the
PM calculated using JWST /NIRCam coronagraphic data at 4 . 4 μm

F444W) from C. Lazzoni et al. ( 2025 ) for comparison (red line
n Fig. 4 ). Compared to MIRI/F1140C, NIRCam/F444W achieves
eeper mass sensitivity for this system, with an impro v ement of
oughly a factor of 3 and particularly enhanced sensitivity at smaller
eparations. Ho we v er, this observ ed mass impro v ement assumes that
he planet has a cloudless atmosphere in chemical equilibrium. Planet
volution models at these shorter wavelengths (from the NIR and up
o ≈8 μm, E. C. Matthews et al. 2024 ; K. A. Crotts et al. 2025 )
re strongly affected by uncertainties in atmospheric properties,
uch as composition, cloud co v erage, but also vertical mixing, and
izequilibrium chemistry, which in turn significantly influence the
eliability of predicted fluxes (e.g. seen for Eps Ind Ab; E. C.

atthews et al. ( 2024 ), TWA 7b; K. A. Crotts et al. ( 2025 ); A. M.
agrange et al. ( 2025 ), and more details in R. Bowens-Rubin et al.
025 ). These effects are much less pronounced at longer wavelengths
ike MIRI’s 11 . 4 μm, making MIRI limits more robust, despite being
ess constraining. 

Beyond the JWST constraints on the DPMs, additional shaded
egions seen in Fig. 4 further exclude possible planet configurations in
hese systems. The following subsections describe how these regions
re calculated. The functions used to generate the curves based on
isc extent (Section 3.3.2 ), PMa (Section 3.3.4 ), and Gaia astrometry
Section 3.3.5 ) have been made available online. 11 

.3.1 Planet limits from archival direct imaging and RV data 

he grey-hatched regions in Fig. 4 rule out the presence of additional
ompanions based on previous upper limits from ground-based direct
maging (VLT/SPHERE) and RV monitoring (HARPS, SOPHIE,
nd HIRES). These archi v al data provide valuable complementary
onstraints to JWST /MIRI observations, particularly at smaller sepa-
ations (for RV) and in o v erlapping re gions (for SPHERE). Note that
D 206893 B was detected with SPHERE, ho we ver, the presence of

dditional companions was ruled out (J. Milli et al. 2017a ). 
SPHERE contrast curves were retrieved from the SPHERE High-

ontrast Data Center (HC-DC, P. Delorme et al. 2017 ). SPHERE
ontrast limits were converted into mass constraints using the same
pproach as for JWST (i.e. DPM using ATMO + BEX models
ia MADYS ). The 99.7 per cent contour from these DPMs is used
s the archi v al direct imaging upper limit. 

RV data were obtained from the HARPS, SOPHIE, and HIRES
pectrographs, with SOPHIE observations retrieved from the OHP
rchive 12 , and HIRES and HARPS data taken from L. Tal-Or et al.
 2019 ) and T. Trifonov et al. ( 2020 ), respectively, as those works
orrect the standard pipeline RVs for known small systematics.
etection limits were derived following the local power analysis
ethod outlined by N. Meunier, A. M. Lagrange & K. De Bondt
NRAS 546, 1–23 (2026) 

1 https:// github.com/ raphhbw/ ExePMa 
2 http:// atlas.obs-hp.fr/ sophie/ 

r  

z  

e  

w  
 2012 ), where planetary detectability is assessed on a semimajor
xis–mass grid. For each point on the grid, we calculate a Keplerian
ignal at the observation timestamps, with the relevant period, at
00 random phases and produce the rele v ant K eplerian-only Lomb–
cargle periodogram. A signal is considered ruled out if, within a
arro w windo w around the injected period, the periodogram peak
xceeds all peaks in the periodogram of the RV data, meaning a
lanet that should have been detected is absent. The detection limit
t a given semi-major axis is then defined as the lowest planet mass
xcluded across all 100 random phases. 

.3.2 Planet limits from the disc extent 

n addition to planet detection limits, we include dynamical con-
traints based on the observed disc extent and gap locations. The
ertical dashed lines in Fig. 4 mark the inner and outer edges of the
iscs, as well as the boundaries of the observed gaps. The orange-
atched regions represent dynamically unstable zones, where planets
n a circular orbit located within 3 Hill radii (R Hill ) of a disc edge
ould significantly disturb the surrounding material if m plt � m star 

B. Gladman 1993 ; S. Ida et al. 2000 ; I. de Pater & J. J. Lissauer
001 ; D. R. Kirsh et al. 2009 ; M. F. Friebe et al. 2022 ; T. D. Pearce
t al. 2024 ). The boundaries of the orange region in Fig. 4 follow T.
. Pearce & M. C. Wyatt ( 2014b ) and are defined by 

 plt = 3 m star 

∣∣∣∣ r edge 

a plt 
− 1 

∣∣∣∣
3 ( 1 

N R Hill 

)3 

, (1) 

here a plt and m plt are semimajor axis and mass of the planet, m star 

he stellar mass, and N R Hill (set to 3) specifies the distance from the
isc edges ( r edge ) in R Hill . The absolute value accounts for both cases
here the disc edge lies inside or outside of the planet’s orbit. 
Any planet within this 3 R Hill zone would likely clear material

rom the disc and shift the disc edges to locations inconsistent
ith current observations. Therefore, any planet sculpting the disc

dges cannot lie in these e xcluded re gions. Similar constraints can be
pplied to planets located within the gaps, which we discuss further
n Section 5.2.1 . 

.3.3 Planet limits from mutual interactions 

or systems with confirmed planets, as in HD 206893, the positions
f the known companions are marked in Fig. 4 by black dots. In such
ases, additional dynamical constraints can be applied by considering
he potential for mutual close encounters between the planets. To
 v oid such interactions that can lead to dynamical instability on short
ime-scales, planetary orbits must be sufficiently spaced, typically by
 minimum number of mutual R Hill . 

Following J. Chambers, G. Wetherill & A. Boss ( 1996 ), we
onsider two planets to be dynamically unstable if their orbital
eparation is less than a critical 2 

√ 

3 ≈ 3 . 5 times their mutual R Hill ,
here the mutual R Hill is defined as equation 1 in J. Chambers et al.

 1996 ). In our specific case, HD 206893 B and c are spaced out by
.3 mutual R Hill , which satisfies the stability criterion we set. 
To identify regions where no additional planets could exist without

iolating this stability criterion, we compute the 2 
√ 

3 mutual R Hill 

ones around each known companions. These e xcluded re gions are
hown as the light grey area in Fig. 4 and are combined into a single
egion, since the known planets do not lie within each other’s unstable
ones. Note that the stability criterion depends on factors we are not
xploring here, including mean-motion resonances between planets
hich can maintain stability, or planet eccentricity and additional

https://github.com/raphhbw/ExePMa
http://atlas.obs-hp.fr/sophie/


Gap-carving planets in exoKuiper belts 11 

p  

S  

r
r

3

A
a
i
G
a
c
(  

f  

w
t  

w  

t  

h  

c
 

c  

i  

a  

i
f  

V  

t
p
i  

(

3

F
E
i  

d  

a
c

 

t
r
c
c
b
i  

m  

b
o  

c  

l

4
D

U
t  

w
l

e  

e  

c
L  

b

D  

l  

v
s

4

A  

s  

e  

i  

t  

m  

1
s  

P  

w  

c  

F  

I  

s
p
p  

w
p

l  

p  

u  

o
p

 

w  

0  

p  

M  

w
 

2
f  

0  

d
 

3
F  

a  

i  

s
t  

r  

e  

b  

p
(  

t

lanets in the system (see details in J. Chambers et al. 1996 ; A. W.
mith & J. J. Lissauer 2009 ). Nevertheless, we use the excluded
egion to set upper limits on where dynamically stable planets could 
eside. 

.3.4 Planet limits from proper-motion anomaly 

ll three systems exhibit strong evidence for an inner companion, 
s indicated by astrometric accelerations, also known as PMa. PMa 
s defined as the difference between the proper motion measured by 
aia eDR3, interpreted as the instantaneous proper-motion vector 

t J2016.0, and a long-term proper-motion baseline (24.75 yr) 
omputed from the variation of sky coordinates between Hipparcos 
J1991.25) and Gaia eDR3 itself (P. Kervella et al. 2019 ). Starting
rom PMa components from the P. Kervella et al. ( 2022 ) catalogue,
e looked for orbits from unseen companions which could explain 

he observed PMa. As done in S. Marino et al. ( 2020 ), for each system
e assumed a single companion on a circular orbit seen coplanar to

he disc. The PMa information is not sufficient to fully solve the orbit;
o we v er, it pro vides valuable constraints, resulting in a de generate
urve in the mass–semimajor axis space. 

The purple line and shaded purple regions (68, 95, and 99.7 per cent
onfidence level) in Fig. 4 represent the mass of the planet needed
n order to explain the observed PMa. While this method assumes
 single planet on a circular orbit, the disc morphology does not
ndicate the presence of highly eccentric massive planets that would 
orce a disc eccentricity (M. C. Wyatt et al. 1999 ; M. C. Wyatt 2005 ;
. Faramaz et al. 2014 ; T. D. Pearce & M. C. Wyatt 2014b ). In

he case of HD 206893, where two massive substellar objects are 
resent, the derived purple curve remains consistent with the PMa 
n the system attributed to planet c, aligning with S. Hinkley et al.
 2023 ). 

.3.5 Planet limits from Gaia astrometry (RUWE) 

inally, we use Gaia astrometry and the Renormalized Unit Weight 
rror (RUWE) parameter to construct the grey dotted regions shown 

n Fig. 4 . The RUWE quantifies the excess noise in Gaia ’s astrometric
ata, with RUWE > 1 . 4 typically indicating a strong signal, often
ttributed to unresolved astrometric perturbations from an unseen 
ompanion (L. Lindegren et al. 2021 ). 

In all three systems studied here RUWE < 1 . 4, which allows us
o set upper limits on any possible astrometric signal and exclude 
egions of the planet mass–semimajor axis parameter space where a 
ompanion would have induced a detectable signal. For short-period 
ompanions, i.e. orbital periods shorter than Gaia DR3’s 1038-d 
aseline, we can exclude planet parameters following the approach 
n M. A. Limbach et al. ( 2024 , see their Section 3.3.1), under the
ain assumption of RUWE < 1 . 4. For periods longer than the Gaia

aseline, we apply the mass–separation relation using equation 32 
f F. Kiefer et al. ( 2025 ), to further rule out planet parameters. The
ombined e xcluded re gion is shown as the gre y dotted area in the top
eft corners of the DPMs in Fig. 4 . 

 PL A N ET  C O N S T R A I N T S  USING  T H E  

ETECTION  PR  O B  ABILITY  MAPS  

sing the mass sensitivity of JWST /MIRI observations, along with 
he additional constraints shown in the DPMs (Fig. 4 ), we assess
here potential planets could reside in these systems. Certain 

ocations are of particular interest, such as planets near the inner 
dge of the disc, which may be responsible for truncating the inner
dge (Section 4.1 ). The DPMs also help constrain the properties of
ompanions responsible for the strong PMa signals (Section 4.2 ). 
ater in Section 5 , we explore the possibility that the gaps are carved
y planets. 
Table 4 summarizes the updated planet limits derived from the 

PM contours at both the 99.7 per cent and 50 per cent confidence
evels for the location of interest. These values reflect the obser-
 ational sensiti vities, although dynamical arguments could impose 
tricter limits. 

.1 Planet truncating the disc inner edge 

ll three systems show inner edges that are consistent with being
teep (A. Imaz Blanco et al. 2023 ) and shaped by planets (T. D. Pearce
t al. 2024 ). The planet’s location is defined by its proximity to the
nner edge without destabilizing the disc, defined as 3 R Hill interior
o the disc’s inner edge (green curve in Fig. 4 ). The corresponding
inimum planet mass is set by requiring that clearing the disc o v er

0 diffusion time-scales occurs within the system’s age, ensuring 
ufficient time for the planet to carve the observed edge (T. D.
earce et al. 2022 , 2024 ). Adopting only 1 diffusion time-scale
ould yield a higher minimum mass, so using 10 provides a more

onserv ati ve estimate (T. Costa, T. D. Pearce & A. V. Krivov 2024 ).
 or consistenc y, we adopt the minimum planet masses deriv ed by A.
maz Blanco et al. ( 2023 , see their section 4.3), as we use the same
tellar and disc parameters. Two scenarios are explored: a single 
lanet truncating the inner edge, which requires a relatively massive 
lanet (tens of Earth masses), and a chain of equal-mass planets,
hich can achieve the same effect with significantly lower-mass 
lanets (a few Earth masses each). 
Although the minimum planet masses required for disc truncation 

ie below the detection limits of the JWST /MIRI observations, the
lanet masses ruled out by the DPMs (see Table 4 ) provide tighter
pper limits on the masses of these potential single sculptors. These
bservational constraints are obtained from the intersection of the 
robability contours with the green curve in Fig. 4 . 
Based on MIRI observations alone, a sculpting planet in HD 92945

ould need to be located at 40 - 47 au with a mass between
 . 26 and 6 . 5 M Jup . For HD 107146, the corresponding sculpting
lanet would be at 33 - 38 au with a mass ranging between 0 . 4 and 4 . 1
 Jup , and for HD 206893, the planet would need to be at 22 - 30 au
ith a mass between 0 . 35 and 10 M Jup . 
NIRCam 4.4 μm observations of HD 92945 (C. Lazzoni et al.

025 ) provide tighter constraints, narrowing the allowed parameters 
or a single inner edge sculptor to 44 - 47 au with a mass between
 . 26 and 1 . 3 M Jup . Ho we v er, NIRCam wav elengths are more model-
ependent than those from MIRI. 
While we have assumed that the disc is sculpted by a planet at

R Hill from its inner edge, several caveats should be considered. 
irst, this interpretation assumes that the inner edge of the disc is
ctively shaped by a planet that remains near this location. Ho we ver,
t is possible that a planet initially sculpted the edge earlier in the
ystem’s history and has since migrated inward, decoupling from 

he current position of the disc edge. In such cases, the planets
esponsible for the truncation may no longer reside near the inner
dge of these systems. Additionally, the disc edge could be shaped
y mechanisms that do not require a planet at all, and could be set by
lanetesimal formation alone or collisional evolution within the disc 
G. M. Kennedy & M. C. Wyatt 2010 ). These scenarios highlight
hat while our interpretation provides one possible explanation, it is 
MNRAS 546, 1–23 (2026) 



12 R. Bendahan-West et al. 

M

Table 4. Observational sensitivity to planets in HD 92945, HD 107146, and HD 206893, derived from the DPMs shown in Fig. 4 . Mass upper limits are given 
as a function of semimajor axis for three scenarios: a planet sculpting the inner edge of the disc, a planet located at the observed gap(s), and a planet responsible 
for the PMa signal. These values (and footnotes) represent the deepest observational limits currently attainable at these locations, though dynamical arguments 
may impose stricter constraints. Limits are quoted at the 99.7 per cent confidence level, with those in brackets indicating the 50 per cent le vel, deri ved from the 
contours in the DPMs. For HD 206893, dashes (–) indicate that MIRI observations lack the sensitivity to detect a planet sculpting the inner edge of the disc. The 
crosses ( ×) mark systems where no second gap has been observ ed. F or HD 92945, we also show the upper limits from JWST /NIRCam at 4.4 μm presented in 
C. Lazzoni et al. ( 2025 ). 

Star Inner edge Gap 1 Gap 2 PMa companion 

plt (au) m plt (M Jup ) a p (au) m p (M Jup ) a p (au) m p (M Jup ) a p (au) m p (M Jup ) 

HD 92945 (MIRI) 36 (43) a < 12 . 4 ( < 1 . 7) a 72 < 2 . 1 (1.3) × × 2 . 5 - 30 0 . 4 - 5 
HD 92945 (NIRCam) 44 (46) < 1 . 3 (0.6) 72 < 0 . 9 (0.4) × × 2 . 5 - 25 0 . 4 - 5 
HD 107146 33 (35) < 4 . 1 (2.1) 56 < 2 . 1 (1.6) 78 < 1 . 8 (1.4) 2 . 5 - 20 2 . 5 - 8 
HD 206893 – (–) b – (–) b 69 < 4 . 0 (2.8) × × 3 . 53 ± 0 . 07 (1) 12 . 7 ± 1 . 1 (1) 

Notes. a SPHERE observations more constraining at the inner edge, 6.5 M Jup at 40 au, given at the 3 σ confidence level. 
b SPHERE observations more constraining at the inner edge, 10 M Jup at 22.5 au, given at the 3 σ confidence level. 
Planet reference: (1) HD 206893 c; S. Hinkley et al. ( 2023 ). 

n  

s  

4

W  

c  

t  

a  

p  

m  

c  

a  

(  

J  

P
 

r  

e  

I  

a  

a  

s
8  

w  

e
 

t  

T  

a  

2  

a  

p  

p  

B  

t  

≈  

s  

a  

e  

a  

a  

s  

t

5

T  

h  

s  

u  

l
 

s  

p  

t  

e  

g  

r  

i  

o  

(  

e  

e  

k  

t  

p

5

T  

a  

U  

c  

r  

d  

2  

f
 

i  

b  

a  

r  

t  

d  

t  

u  

S  

i  
ot the only one. Further discussion of these caveats and alternative
cenarios can be found in T. D. Pearce et al. ( 2024 , see their section 6).

.2 Planet causing the PMa 

hile the PMa indicates the presence of a perturber, it does not
onstrain its orbital radius, as it depends on a degenerate combina-
ion of planet mass and separation. To break this de generac y, we
ssess how the constraints from the DPMs rule out different planet
arameters. At small separations (typically less than a few au), RV
onitoring and the Gaia RUWE parameter can exclude massive,

lose-in planets as the source of the PMa, provided no significant
strometric or spectroscopic signal is detected. At wide separations
tens of au), direct imaging limits from ground-based facilities or
WST rule out companions massive enough to induce the observed
Ma. 
The DPMs in Fig. 4 show that, for all three systems, the planets

esponsible for the PMa signal cannot reside near the disc’s inner
dge, and therefore likely does not play a role in actively shaping it.
nstead, the observational constraints indicate that the perturber in
ll cases is confined to the inner regions of the planetary system. The
llowed range of semimajor axis and masses for such companion are
ummarized in Table 4 , and are typically around 2.5–30 au and 0.4–
 M Jup . In the case of HD 206893, this interpretation is consistent
ith the detection of planet c at 3.5 au, which has been shown to

xplain the observed PMa signal (S. Hinkley et al. 2023 ). 
In recent years, a broader trend has emerged in which PMa signals

end to trace planets located in the inner regions of planetary systems.
his pattern is not only seen in the three systems presented here, but
lso in AF Lep (AF Lep b, R. J. De Rosa et al. 2023 ; K. Franson et al.
023 ; D. Mesa et al. 2023 ), β Pic ( β Pic c, G. M. Brandt et al. 2021a ),
nd HR 8799 (HR 8799 e, G. M. Brandt et al. 2021b ). Ho we ver, the
resence of a PMa perturber at small separations does not rule out the
resence of additional, further-out planets. For example, HD 206 893
, β Pic b, and HR 8799 b,c and d all orbit at wider separations than

heir corresponding PMa planets, with HR 8799 b located as far out as
70 au. Another compelling example for planet formation at wider

eparations is TWA 7 b, a planet candidate detected within the gap of
 debris disc around 50 au (K. A. Crotts et al. 2025 ; A. M. Lagrange
t al. 2025 ). These cases highlight that while PMa signals typically
rise from inner planets, other planets may form at larger separations
nd play a key role in shaping the observed disc morphology. A
imilar configuration may therefore apply to the systems studied in
his work. 
NRAS 546, 1–23 (2026) 
 G A P  C A RV I N G  S C E NA R I O S  

here are multiple gap carving scenarios to consider, and while
ypothesizing a planet embedded within the gap itself is the most
traightforward solution, it is not the only one. Different mechanisms
ltimately predict different planetary parameters, including their
ocation relative to the gap, as well as evolutionary histories. 

In the following subsections, we assess four distinct gap-carving
cenarios for each of the systems. The first scenario involves a single
lanet embedded within the disc, carving a gap around its orbit
hrough scattering (Section 5.2 ). The second scenario (Section 5.3 )
xamines whether a planet located near the disc’s inner edge could
enerate a gap via the 2:1 mean-motion resonance (MMR). The
emaining two scenarios focus on secular apsidal resonances: one
nvolving two inner planets and a massless disc (Section 5.4 ), and the
ther with a single inner planet and a massive (non-zero mass) disc
Section 5.5 ). Overall, these four gap-carving scenarios are adapted to
ach system individually, with slight modifications where necessary,
.g. to account for the presence of two gaps in HD 107146 or the two
nown companions in HD 206893. Table 5 provides a summary of
he validity of each gap-carving scenario with corresponding planet
redictions. 

.1 General caveats 

hroughout this section, we consider the gap location, the gap width
nd the level of gap asymmetry to e v aluate the dif ferent scenarios.
sing the gap depth could be a useful tool to exclude some gap-

arving mechanisms; ho we ver, current constraints allo w for a wide
ange of possible depths, from 50 per cent full to more than 90 per cent
epleted, and therefore are not constraining (A. Imaz Blanco et al.
023 ). The gap depth is therefore excluded as a criterion from the
ollowing discussion. 

When interpreting the planet predictions in this work, it is
mportant to keep in mind the assumptions on which they are
ased. The constraints on gap-carving planets are based on analytical
rguments. More detailed N -body simulations could refine these
esults, ho we ver, such tailored modelling lies beyond the scope of
his paper. Additionally, the predictions depend strongly on both the
isc parameters and the observational upper limits obtained from
he DPMs. Disc parameters can vary depending on the wavelength
sed to observe the disc as well as the models used to fit the data.
mall changes in disc parameters could lead to large differences

n predicted planet properties due to the scaling laws that we use.



Gap-carving planets in exoKuiper belts 13 

Table 5. Summary of the gap-carving scenarios for HD 92945, HD 107146, and HD 206893 discussed in Section 5 . Some theoretical predictions remain 
unaffected by the new JWST observational constraints, and references to the original studies with still valid predictions are pro vided. Howev er, note that these 
values might need updating with new disc parameters. For HD 92945’s secular resonance in a massless disc + 2 planets scenario, the JWST /NIRCam constraints 
are given in brackets. 

Gap-carving scenarios HD 92945 HD 107146 HD 206893 
Gap 1 Gap 2 

• Embedded planet: 
Single planet + massless disc a plt : 72 au a plt : 56 au a plt : 78 au a plt : 69 au 
(Section 5.2.1) m plt : 0 . 03 − 0 . 75 M Jup m plt : 0 . 02 - 0 . 06 M Jup m plt : 1 . 4 − 1 . 8 M Jup m plt : 1 . 6 − 4 . 0 M Jup 

Chain of planets + massless ∼ 2 planets ≈0 . 02 M Jup 
a If single wide gap: ∼2–3 planets ≈0 . 04 M Jup 

a 

disc (Section 5.2.2) ∼2–3 planets with ≈0 . 04 M Jup 
a 

Single planet + massive disc m plt � 0 . 26 M Jup 
(1) If single wide gap: m plt � 2 . 5 M Jup 

(1) 

(Section 5.2.3) M disc � 75 M ⊕(1) m plt � 1 . 5 M Jup 
(1) M disc � 190 M ⊕(1) 

M disc � 50 M ⊕(1) 

• 2:1 MMR (Section 5.3 ): Ruled out Ruled out Ruled out Ruled out 

• Sec. res. in massless disc a 1 : 11 - 20 au (11 - 15 au) If single wide gap: Ruled out 
+ 2 planets (Section 5.4): m 1 : 1 . 4 - 3 . 1 M Jup 

(1 . 4 - 1 . 9 M Jup ) 
Ruled out b (2) 

a 2 : 40 - 47 au (44 - 47 au) 
m 2 : 0 . 3 - 7 . 8 M Jup 

(0 . 3 - 1 . 5 M Jup ) 

• Sec. res. in massi v e disc a plt : 10 - 25 au If single wide gap: With HD 206893 B only c : 
+ 1 planet (Section 5.5): m plt : 0 . 5 - 3 M Jup a plt : 8 - 15 au M disc : ≈ 400 M ⊕

M disc : 10 - 100 M ⊕ m plt : 2 - 5 M Jup 

M disc : 30 - 100 M ⊕

Notes. a Assuming an inter-planet spacing of 16 R Hill . More compact configurations would lead to lower planet masses. 
b Two gaps could be opened in specific planet configurations (Friebe & Sefilian in preparation). 
c Friebe & Sefilian (in preparation) explore the addition of HD 206893 c, but find it difficult to reproduce the observed gap. 
References: (1) M. F. Friebe et al. ( 2022 ), (2) B. Yelverton & G. M. Kennedy ( 2018 ). 

S
t
c
f
g

5

T  

d  

s
t  

s
e
t
o  

e
p
(  

o

5

W  

p  

s

R  

2  

a
 

m

m

w  

w  

t  

T  

s
a

 

r  

0  

i  

J  

l  

T  

p  

2  

13 We use the Hill radius criterion rather than resonance o v erlap (chaotic 
zones; J. Wisdom 1980 ), as it is more robust regarding planet eccentricity (T. 
D. Pearce & M. C. Wyatt 2014a ). Approximating the gap width as the chaotic 
zone has been done in the past and produces results consistent with N -body 
simulations (e.g. fig. 5 in S. Marino et al. 2018 ). 
imilarly, impro v ements in observational sensitivity would impact 
he validity of predicted planet properties. Despite these caveats, the 
onstraints presented here represent the most up-to-date estimates 
or the planets that could be responsible for carving the observed 
aps, given current theoretical understanding. 

.2 Planet(s) embedded in the disc 

he most commonly used mechanism for carving a gap in a debris
isc involves a planet on a circular orbit embedded within the disc,
cattering nearby planetesimals. Non-resonant planetesimals close 
o the planet are expected to be cleared, resulting in an azimuthally
ymmetric gap. For a more asymmetric gap, some level of planet 
ccentricity would be required. In this first scenario, we investigate 
hree configurations: a massive single planet located at the centre 
f the gap in a massless disc (Section 5.2.1 ), a chain of multiple
qual-mass planets populating the gap (Section 5.2.2 ), and a single 
lanet clearing a gap through migration within a massive disc 
Section 5.2.3 ). Note that this list is not e xhaustiv e, and variants
r combinations of these scenarios could be possible. 

.2.1 Single planet in a massless disc 

e start by assuming a massless disc ( M disc � m plt ) and a single
lanet on a circular orbit located at the centre of the gap. In this
cenario, the planet is expected to scatter material found within 3 
 Hill 
13 on either side of its orbit (B. Gladman 1993 ; S. Ida et al.

000 ; D. R. Kirsh et al. 2009 ; M. F. Friebe et al. 2022 ), resulting in
 gap width of 6 R Hill . 

The mass of a planet ( m plt ) required to carve such a gap, assuming
 plt � m star , is given by 

 plt = 3 m star 

(
w gap 

6 a plt 

)3 

, (2) 

here a plt is the planet’s semimajor axis, which in this scenario
ould be equal to the central radius of the gap ( r gap ), and w gap is

he observed width of the gap. Using the disc parameters listed in
able 1 , we estimate the mass of the gap-carving planet for each
ystem. These predicted planet masses and locations are visualized 
s the black dots with a question mark in Fig. 4 . 

Using equation ( 2 ), for HD 92945, the estimated planet mass
equired to carve the observed 19 au gap at 72 au, ranges from 0.03 to
.75 M Jup . The range of planet mass is visible by the black error bars
n Fig. 4 and is calculated from the uncertainty on w gap , r gap and m star .
WST /MIRI data excludes planets more massive than 2 M Jup at this
ocation, but this sensitivity does not reach the predicted mass range.
he same can be said about the JWST /NIRCam data that excludes
lanets more massive than 0.9 M Jup at this location (C. Lazzoni et al.
025 ). Since both observational limits remain abo v e the predicted
MNRAS 546, 1–23 (2026) 



14 R. Bendahan-West et al. 

M

m  

f
 

a  

t  

c  

r  

s  

a
 

d  

W  

r  

w  

c  

l  

1  

l  

n  

w  

a  

m  

R  

p
 

c  

P  

≥  

m  

(  

f  

d  

R  

t  

p
 

r  

i  

s  

c  

A  

e  

r  

p  

2  

r  

c  

a  

b  

S

5

A  

p  

t  

s  

m  

t  

e
 

a  

t  

c  

M  

m  

f  

f  

c
 

m  

A  

c  

c  

y
s  

g
(  

∼
 

l  

s  

W  

s  

r  

r

5

A  

s  

p  

d  

t  

2  

a
 

d  

p  

t  

d  

t  

s  

p  

t  

a  

2
 

t  

i  

g  

w  

q  

s  

e
 

t  

p  

c  

14 The effect will vary if multiple planets are considered (M. C. Wyatt 2003 ; 
A. Bonsor et al. 2014 ) 
15 In the case of a more asymmetric gap, the planet would likely need to follow 

a more eccentric orbit. 
ass range, the single planet scenario remains a possible explanation
or the observed gap in HD 92945, if the gap is truly axisymmetric. 

Applying the same analysis to HD 206893, a planet at 69 au with
 mass between 1.6 and 6.9 M Jup would be required to explain
he observed 40 au gap. In this case, the JWST /MIRI observations
an rule out planets more massive than 4 M Jup , which in turn
efines the predicted mass range to 1 . 6 - 4 . 0 M Jup . Therefore, this
cenario remains a possible cause for the observed gap, assuming an
xisymmetric gap. 

For HD 107146, we consider two embedded planets in a massless
isc, each located at the centre of one of the two observed gaps.
e would therefore require a first planet at 56 au with a mass

anging from 0.02 to 0.06 M Jup and a second planet at 78 au
ith a mass between 1.4 and 3.3 M Jup . JWST /MIRI constraints

an rule out planets more massive than 1.8 M Jup at the outer gap
ocation, tightening the predicted mass for that second planet to
 . 4 - 1 . 8 M Jup . Conversely, the predicted mass range at the first gap
ocation is significantly lower than the observ ational sensiti vity. We
ote that a configuration with both planets centred within the gaps
ould be considered dynamically stable following the planet stability

rgument made in Section 3.3.3 , although the planet separation (3.7
utual R Hill ) lies close to the critical threshold of 2 

√ 

3 ≈ 3 . 5 mutual
 Hill . A more detailed e v aluation of the dynamical stability of these
redictions would require dedicated N -body simulations. 
When e v aluating this gap-carving scenario, it is important to

onsider the time-scale necessary to open a gap in these discs.
revious N -body simulations have shown that planets with masses
10 M ⊕ can carve gaps by depleting at least 50 per cent of the

aterial in less than 10 Myr for a ∼1M � at separations of ∼75 au
e.g. fig. 5 in S. Marino et al. 2018 ). These time-scale estimates are
ound to be shorter but comparable to those inferred from scattering
iffusion and clearing time-scale (S. Tremaine 1993 ; S. Morrison &
. Malhotra 2015 ). Even allowing for uncertainties in stellar ages,

hese results indicate that there is enough time for the predicted
lanets to have sculpted the gaps seen in the observed discs. 
While the gap is assumed to be cleared of non-resonant material,

esonant material could still remain trapped within the gap and make
t appear less depleted than predicted. This is seen in the TWA 7
ystem, where a planet is found embedded in the gap alongside
o-orbital material (B. Ren et al. 2021 ; K. A. Crotts et al. 2025 ;
. M. Lagrange et al. 2025 ). Although the gap depths can be

stimated from observations and disc modelling, the measurements
emain uncertain. Current uncertainties allow for both shallow ( < 50
er cent) and deep ( > 90 per cent) gaps (A. Imaz Blanco et al.
023 ). Consequently, gap depth alone does not yet provide sufficient
eliability to place strong constraints on planet masses. Tighter
onstraints would therefore require higher sensitivity observations
nd dedicated N -body simulations accounting for the clearing of
oth the resonant and non-resonant material (see e.g. figs 6 and 7 in
. Marino et al. 2018 ). 

.2.2 Chain of multiple planets in a massless disc 

 chain of lower-mass planets distributed across the gap could
roduce comparable gap widths if the disc mass is much smaller
han the planet mass. In this case, the system’s age and gap width
et a lower limit on the planet masses required to ef fecti vely clear
aterial via scattering, while dynamical stability places a limit on

he number of planets that can coexist within the gap (A. Shannon
t al. 2016 ). 

To e v aluate this scenario, we use equation 4 from A. Shannon et
l. ( 2016 ), which defines the minimum mass of the planets required
NRAS 546, 1–23 (2026) 
o carve the gap for a given inter-planet spacing. Whereas their
alculation assumed a typical spacing of 20R Hill (J. Fang & J.-L.
argot 2013 ), we adopt 16R Hill , which reduces the estimated planet
asses by a factor of two (A. Shannon et al. 2016 ) and allows

or more planets to populate the gap. Since no results are available
or more compact planet configurations, 16R Hill provides the most
onserv ati ve estimate. 

F or HD 92945, giv en the inter-planet spacing of 16R Hill , the mini-
um mass of the planets in the gap would be ∼7 M ⊕ ( ∼0.02 M Jup ).
 16R Hill spacing corresponds to ∼23 au, meaning only ∼2 planets

ould occupy the ∼20 au wide gap, one at each edge. For HD 107146,
ombining both depleted regions into a single wide gap of ∼50 au
ields a minimum planet mass of ∼14 M ⊕ ( ∼0.04 M Jup ). The 16R Hill 

pacing of ∼30 au would then allow ∼2–3 planets to populate the
ap. For HD 206893, the required minimum planet mass is ∼12 M ⊕
 ∼0.04 M Jup ), with an inter-planet spacing of ∼23 au permitting
2–3 planets to populate the gap width. 
In all three systems, the predicted mass limits are significantly

ower than the detection sensitivity of the observations, meaning that
uch planets cannot be ruled out, and this scenarios remains valid.
e note, ho we ver, that calculations are restricted to an inter-planet

pacing of 16R Hill , which for a more compact configuration, would
esult in lower planet mass predictions, but exploring such cases
equires tailored N -body simulations. 

.2.3 Single planet in a massive disc 

 debris disc with a mass comparable to the planet mass can
ignificantly influence the planet embedded in the disc. As a single
lanet interacts with surrounding material, it predominantly scatters
ebris outward and exchanges angular momentum, causing the planet
o lose energy and gradually migrate inward 14 (D. R. Kirsh et al.
009 ). The extent of this migration depends on both the disc mass
nd the planet mass. 

M. F. Friebe et al. ( 2022 ) demonstrated that the same gap width in a
ebris disc can be reproduced by either a massive, nearly stationary
lanet (akin to the massless disc scenario) or a lower-mass planet
hat carves the gap as it migrates via planetesimal scattering. This
e generac y, illustrated in fig. 5 of M. F. Friebe et al. ( 2022 ), highlights
hat different combinations of planet and disc mass can lead to the
ame gap width. In contrast, the depth would vary, as lower-mass
lanets would not eject material, leaving remaining scattered material
o populate a shallower gap. As a result, the gap depth may inform
bout the planet’s migration history and migration rate (M. C. Wyatt
003 ). 
With the current JWST observations , we cannot e xclude an y of

he planet and disc mass combinations predicted from the gap widths
n M. F. Friebe et al. ( 2022 ). Consequently, assuming axisymmetric
aps, 15 the results of M. F. Friebe et al. ( 2022 ) remain consistent
ith a migrating planet as the origin of the gap. While this would not
ualitatively change the conclusions, the associated mass estimates
hould be updated using revised disc parameters (A. Imaz Blanco
t al. 2023 ). 

In addition to scattering material, a migrating planet can also
rap surrounding debris as co-orbital material (e.g. Jupiter’s Trojan
opulation), leading to observable substructures in the discs. In the
ase of TWA 7, the planet candidate observed within the disc gap



Gap-carving planets in exoKuiper belts 15 

i
t  

R  

c
a
p  

F

c  

c  

K  

f  

A  

w  

d  

c
2  

d
p

5

A  

t  

T  

2  

i  

m  

M  

w

m

a  

p  

p  

M
 

2  

M  

h
t

r
i  

W  

g  

3  

c
t  

w
W  

w
r

 

2  

2  

w  

t
 

s
l  

o  

p  

t  

t  

s  

o
 

i
i  

w  

t  

e  

t  

A  

d  

w  

o
i  

p
s  

E

5

T
i  

m  

d
e
w
t
e  

r  

g
(  

2
d  

Y  

n
b  

g  

l  

a
 

(  

t
u  

B  

(

r

w  

(
 

r
Y  

e
o  

K  

I  

t

s accompanied by co-orbital material, strongly suggesting resonant 
rapping and giving the appearance of a double-gap disc structure (B.
en et al. 2021 ; A. M. Lagrange et al. 2025 ). A similar mechanism
ould explain the double gap morphology seen in HD 107146, where 
 single planet migrating inwards might have accumulated a Trojan 
opulation, producing the two apparent gaps (e.g. section 3.2.1 in M.
. Friebe et al. 2022 ). 

There are several extensions to this scenario. For example, there 
ould be two or more planets embedded in a massive disc. In such
ases, planets may e xperience div ergent migration (S. J. Morrison &
. M. Kratter 2018 ), clearing broad gaps, or convergent migration if,

or e xample, the y are locked in resonance (K. J. Walsh et al. 2011 ).
nother extension is the case of a single planet embedded in a disc
ith an evolving mass. A planet born embedded in a massive and
ispersing protoplanetary disc, may carve a gap much wider than its
haotic zone due to sweeping secular resonances (X. Zheng et al. 
017 ). Similar to the one planet case, testing these scenarios requires
edicated N -body simulations, which are beyond the scope of this 
aper. 

.3 2:1 mean-motion resonance 

 planet orbiting near the inner edge of the disc, rather than within
he gap itself, could open a gap at its external 2:1 MMR (e.g. M.
abeshian & P. A. Wiegert 2016 ; M. Tabeshian & P. A. Wiegert
017 ). In this scenario, the planet’s semimajor axis is set by requiring
ts 2:1 MMR to coincide with the observed gap location, while its

ass depends on the width of the gap. Rearranging equation 8 from
. Tabeshian & P. A. Wiegert ( 2016 ) and expressing it in terms of
idth and not fractional width, the planet mass is given by 

 plt = 

M Jup 

0 . 009 

(
w gap 

2 
√ 

2 ln (2) × r gap 
− 0 . 009 

)
, (3) 

ssuming the planet is on a circular orbit. We use this relation to
redict the planet mass and location, and compare these to the planet
arameters excluded by the DPMs (Fig. 4 ) to e v aluate whether this
MR-driven scenario can explain the observed gaps. 
In the case of HD 92945, if the observed gap were shaped by the

:1 MMR, the responsible planet would need a mass of 5 . 2 − 17 . 5
 Jup at 45 au. A planet with these predicted orbital parameters would

ave been detected by JWST /MIRI observations, therefore ruling out 
he 2:1 MMR mechanism as the origin of the gap in HD 92945. 

For HD 107146, we first test whether the innermost gap could 
esult from a 2:1 MMR, while assuming that a different mechanism 

s shaping the second gap (e.g. 3:1 MMR, M. Tabeshian & P. A.
iegert 2017 ). By setting the 2:1 MMR at the centre of the inner

ap, we find that a planet with a mass between 4.1 and 6.9 M Jup at
5 au would be capable of carving the width of the first gap. In this
onfiguration, the 3:1 MMR would be at ∼73 au, coinciding with 
he second gap, which can also be found to carve observable gaps
hen planet eccentricity becomes important (M. Tabeshian & P. A. 
iegert 2017 ). Ho we ver, a planet with these predicted parameters
ould have been detected with the MIRI observations, therefore 

uling out this scenario. 
Alternatively, if the second gap in HD 107146 were shaped by a

:1 MMR, the corresponding planet would have a mass ranging from
0.3 to 28.8M Jup at 49 au. Such a planet would have been detected
ith MIRI and can therefore rule out this scenario as the origin of

he second gap in HD 107146. 
Finally, applying the same analysis to HD 206893, the 2:1 MMR

cenario would require a planet with a mass between 20 and 33 M Jup 

ocated at 43 au to produce the observed gap. Ho we ver, a planet
f this mass would have been observed with MIRI. Additionally, a
lanet at 43 au would lie within an observed ring of material between
he inner edge (35 au) and the gap’s inner edge (49 au), and would
herefore truncate the disc in a way that contradicts the observed
tructure. The 2:1 MMR scenario is therefore ruled out as the origin
f the gap in HD 206893. 
Overall, the 2:1 MMR scenario fails to explain the observed gaps

n the three systems studied. Even when accounting for uncertainties 
n gap locations and widths, the required planet masses to carve such
ide gaps are generally too high and already excluded by observa-

ional limits. While we have focused on cases where the MMR was
xterior to the planet, we could also consider resonances interior to
he planet (M. Tabeshian & P. A. Wiegert 2016 ; M. Tabeshian & P.
. Wiegert 2017 ), therefore placing planets near the outer edge of the
iscs. Ho we ver, these configurations lead to the same conclusions,
here the required planets are too massive and are therefore ruled
ut by MIRI observations. Finally, as with the embedded planet 
n a massive disc scenario, a lower-mass migrating planet could in
rinciple open wider gaps through the sweeping of material into 
pecific MMRs (e.g. M. C. Wyatt 2003 ; M. F. Friebe et al. 2022 ).
valuating such scenario is outside the scope of this paper. 

.4 Secular resonances in a massless disc due to two planets 

he observed gaps could also be carved by secular apsidal resonances 
nduced by two inner planets on slightly eccentric orbits, assuming a

assless disc for simplicity. In this scenario, the gap is not shaped by
irect scattering, but through long-term gravitational interactions that 
xcite planetesimal eccentricities. These secular resonances occur 
hen a planetesimal’s pericentre precession rate matches one of 

he system’s eigenfrequencies, which go v erns the planet’s orbital 
volution (e.g. C. D. Murray & S. F. Dermott 1999 ). At these
esonance locations, eccentricities grow o v er time and material is
radually depleted locally, producing observable asymmetric gaps 
T. D. Pearce & M. C. Wyatt 2015 ; B. Yelverton & G. M. Kennedy
018 ). A two-planet system would generate two resonances, each 
ominated by one planet, but the resonances differ in width. In B.
elverton & G. M. Kennedy ( 2018 ), one resonance is typically too
arrow to generate a detectable gap, while the resonance dominated 
y the outer planet is generally broad enough to carve an observable
ap (see section 3 of B. Yelverton & G. M. Kennedy 2018 ). The
ocation and width of the gap depend on the planet’s semi-major axis
nd mass ratios, as well as the planet’s eccentricities. 

Following the approach used in B. Yelverton & G. M. Kennedy
 2018 ), we only focus on the most dominant secular resonance and
reat the second resonance as negligible. This simplification allows 
s to use the analytical expression (rearranged from equation 24 in
. Yelverton & G. M. Kennedy 2018 ) for the location of the gap
 r gap ) due to the secular resonance: 

 gap = 

(
m 2 

m 1 

)2 / 7 

a 
11 / 7 
2 a 

−4 / 7 
1 , (4) 

here a 1 and a 2 are semimajor axes of the inner (1) and outer planet
2), and m 1 and m 2 are their respective masses. 

In addition to predicting the gap location, the width of the
esonance depends on which resonance dominates. Table 3 in B. 
elverton & G. M. Kennedy ( 2018 ) provides the corresponding
xpressions for the resonance width. Here we approximate the width 
f the resonance to the width of the gap as in B. Yelverton & G. M.
ennedy ( 2018 ), but note that this is only a first-order approximation.

n the simpler cases, the width of the resonance only depends on
he outer planet’s semimajor axis and eccentricity, which can be 
MNRAS 546, 1–23 (2026) 



16 R. Bendahan-West et al. 

M

a

w

w  

t
a  

c

5

T  

r  

A  

d  

n  

t  

p  

b  

w  

Y  

c  

F
 

l  

l  

s  

a  

b

5

E  

m  

r  

t  

p  

o  

c  

(  

a  

F  

w  

2  

d
a  

f  

a  

t  

i
 

v  

f  

e  

p  

e  

e  

a  

T  

I  

e  

t  

Figure 5. Plots e v aluating the secular resonance in a massless disc due to 
two planets scenario for HD 92945 and HD 206893. The general structure of 
both plots follows the DPM shown in Fig. 4 . 

a  

c
 

o  

W  

(  

a  

a  

b  

o  
pproximated as 

 gap ≈ e 2 a 2 

0 . 07 
, (5) 

here w gap is the width of the gap (or resonance in this approxima-
ion) and e 2 is the eccentricity of the outer planet. Using both r gap 

nd w gap , we can assess whether a given two-planet configuration is
apable of producing the observed gap. 

.4.1 Caveats 

his scenario assumes that only a single gap is formed via secular
esonances, even though two resonant locations exist in principle.
s shown in B. Yelverton & G. M. Kennedy ( 2018 ), only the
ominant (widest) resonance is retained while the other is considered
egligible, resulting in the formation of just one gap. Ho we ver, when
he planet semimajor axes are comparable and the mass of the outer
lanet is smaller or comparable to the mass of the inner planet,
oth resonances have a comparable width, and only considering the
idest resonance is no longer applicable (see figs 4 and 5 in B.
elverton & G. M. Kennedy 2018 ). In this case, both resonances
ould be important enough to carve distinct gaps in the disc (see
riebe & Sefilian, in preparation). 
In this work, we use the simple approximation to find the gap

ocation rather than numerically solving for the two resonance
ocations and widths. Despite this approximation, the predicted
ecular resonance location remains consistent with the numerical
pproach (e.g. blue lines in fig. 9 of S. Marino et al. 2020 , comparing
oth methods for a similar planet configuration to HD 206893). 

.4.2 HD 92945 

quation ( 4 ) has four free parameters and only one constraint ( r gap ),
aking this a degenerate problem. Ho we ver, we can use additional

estrictions based on the PMa, disc extent, and direct imaging limits,
o restrict the de generac y in planet parameters. We place the inner
lanet on the PMa curve, as a planet is required in the inner regions
f the system to explain the observed PMa in the system. This choice
onsequently fixes m 1 as a function of a 1 . By rearranging equation
 4 ) for m 2 , we can determine the outer planet’s mass as a function of
 2 by fixing the secular resonance location at the gap location (72 au).
inally, the observed sharp inner edge (A. Imaz Blanco et al. 2023 ),
hich suggests the presence of a sculpting planet (T. D. Pearce et al.
024 ), allows us to place the outer planet as close as possible to the
isc’s inner edge (at 3 R Hill ), therefore lifting the de generac y on m 2 

nd a 2 . While this entire process yields a set of planet pairs as a
unction of a 1 that are capable of carving a gap via secular resonance
t the correct location, it represents only one possible simplification
o an otherwise highly degenerate problem. The e v aluation process
s illustrated in Fig. 5 (a). 

By shifting the inner planet along the PMa curve, we assess the
alidity of each planet pair against the constrained parameter space
rom DPMs, determining the range of planet parameters that could
xplain the observed gap. One of the main limiting factors for the
ossible planet configurations is the direct imaging sensitivity, which
xcludes a planet pair when at least one of them falls within these
 xcluded re gions (seen as a crossed out configuration in Fig. 5 b). In
ddition, we also consider the inner edge clearing time-scale (used in
. D. Pearce et al. 2022 , 2024 ) to limit the valid planet configurations.
f the mass of the outer planet is too low, the clearing timescale would
xceed the system’s age, allowing us to rule out the planet pair. Note
hat we have tested additional criteria, such as the mutual Hill radius
NRAS 546, 1–23 (2026) 
nd gap clearing timescales, though neither limits the possible planet
onfigurations for HD 92945. 

We apply the method outlined abo v e to quantify the population
f planet pairs capable of producing the observed gap in HD 92945.
e incorporate the 1 σ uncertainty on the PMa companion’s mass

i.e. m 1 ), to capture the resulting range in the outer planet’s location
nd mass (visible as the error bars in Fig. 5 a). Using the JWST /MIRI
nd SPHERE detection limits, the inner planet is expected to lie
etween ∼11 and 20 au with a mass of 1.4 to 3.1 M Jup , while the
uter planet would be located between 40 and 47 au with a mass



Gap-carving planets in exoKuiper belts 17 

o  

F  

1  

f  

a  

b
 

q
c
(
w
(  

t
a

5

F
m
t  

m
(  

s
a
a
o
t

5

E
s
(  

U  

t  

l  

a  

p
r  

(  

T  

a  

s
c
a

 

p
m  

i  

p  

l  

i
b
p
w

5

W
r
p  

b

o  

2  

e
m
g
r
t  

m
 

S  

t  

d  

o
c  

9
c  

t  

s  

1  

p
t
o  

s  

p

5

F  

p
d  

2  

r  

M  

l  

M  

(  

T  

H

5

F  

d  

i  

p
s  

w  

t  

∼  

t  

 

d  

g
b

16 For a somewhat similar result involving a disc that back-reacts on the planet 
but does not perturb itself, see T. D. Pearce & M. C. Wyatt ( 2015 ). In that 
case, the occurrence of a secular resonance requires a disc-to-planet mass 
ratio of M disc /m plt ∼ 1, in contrast to the scenario in A. A. Sefilian et al. 
( 2021 , 2023 ), where resonances occur when M disc /m plt � 1 (see also section 
6.5 in A. A. Sefilian et al. 2023 ). 
f 0.3–6.5 M Jup . Alternatively, using the JWST /NIRCam red line in
ig. 5 (a) would restrict the inner planet to 11 - 15 au with a mass of
 . 4 - 1 . 9 M Jup , and the outer planet to 44 - 47 au with a mass ranging
rom 0.3 to 1.5 M Jup . Ho we ver, for a more conserv ati ve assessment
nd for reasons discussed in Section 3.3 , we adopt the constraints
ased on the JWST /MIRI data. 
It should be noted that secular resonances in this scenario re-

uire the planets to have some eccentricity. Our analysis adopts 
omplementary dynamical arguments that assume circular orbits 
from PMa and inner-edge sculpting). While strictly circular orbits 
ould remo v e the quoted mass predictions, even slight eccentricities 

e.g. e = 0 . 05, B. Yelverton & G. M. Kennedy 2018 ) are sufficient
o generate gaps via secular resonances, meaning results remain 
pplicable under realistic orbital conditions. 

.4.3 HD 107146 

or HD 107146, the two-planet secular resonance scenario in a 
assless disc faces the key limitation that the observed gaps in 

his system do not exhibit the strong asymmetry expected from this
echanism. N -body simulations by B. Yelverton & G. M. Kennedy 

 2018 ), which explored a wide range of two planet configurations,
how that carving a gap without introducing significant azimuthal 
symmetry is challenging. Since the gaps generated by this scenario 
re inherently asymmetric, the results are inconsistent with the 
bservations, ruling out secular resonance in a massless disc due 
o two planets as the origin of the gaps in HD 107146. 

.4.4 HD 206893 

valuating this secular resonance scenario for HD 206893 is more 
traightforward than for the previous two systems, as two companions 
HD 206893 B and c) have already been detected in this system.
sing equation ( 4 ) with the orbital parameters of these planets (from

able 3 in S. Hinkley et al. 2023 ), we find the secular resonance
ocation to be at 21 au, which lies interior to the disc’s inner edge
t 35 au (see the vertical red line in Fig. 5 b). Given the observed
lanet configuration, the outer resonance dominates o v er the inner 
esonance, which following table 3 in B. Yelverton & G. M. Kennedy
 2018 ) allows us to estimate the resonance width using equation ( 5 ).
he depleted region produced by this resonance is then 21 ± 6 au,
s indicated by the red-shaded area in Fig. 5 (b). Since the calculated
ecular resonance location does not align with the observed gap, we 
onfidently rule out this secular resonance in a massless disc scenario 
s the origin of the gap in HD 206893. 

From HD 206893’s DPM (Fig. 4 ), we identify a narrow region in
lanet parameter space between the light-grey region excluded by 
utual R Hill and the disc’s inner edge, where a third planet could,

n principle, exist. Ho we ver, this region is seen to overlap with the
redicted secular resonance at 21 au (Fig. 5 b), which could affect the
ocation’s dynamical stability. This o v erlap suggests that the disc’s
nner edge may instead be carved by the secular resonance induced 
y the two known companions, making the presence of an additional 
lanet near the inner edge unlikely. Further assessing such scenario 
ould require dedicated dynamical analysis. 

.5 Secular resonances in a massi v e disc due to a single planet 

e explore a gap-carving mechanism driven by secular apsidal 
esonances in discs with non-zero mass located exterior to a single 
lanet. In this case the planetary precession is driven by the disc’s
ackreaction, with the disc’s self-gravity modulating the evolution 
f planetesimals in addition to the planet (A. A. Sefilian et al. 2021 ,
023 ). 16 A key difference introduced by including disc mass is the
volution of the gap morphology. While secular resonances in both 
assless and massive discs initially produce highly asymmetric 

aps, a planet interacting with a massive disc undergoes secular 
esonant friction (if M disc /m plt � 1), which gradually circularizes 
he planet’s orbit (S. Tremaine 1998 ), and consequently results in a

ore symmetric gap o v er time (A. A. Sefilian et al. 2023 ). 
To e v aluate the v alid planet parameters, equation 19 in A. A.

efilian et al. ( 2021 ) relates a planet’s semimajor axis and mass to
he location of the secular resonance (i.e. gap location), given a fixed
isc mass. Their figs. 7 and 12 illustrate how different combinations
f planetary parameters and disc masses can produce a resonance 
apable of explaining the gap locations in HD 107 146 and HD
2945, respectively. Using the constraints derived from DPMs, we 
an refine the previous viable planet parameter range. If we assume
he planet responsible for this secular resonance also drives the PMa
ignal, we can o v erlay the PMa curv e from the DPMs onto figs. 7 and
2 of A. A. Sefilian et al. ( 2021 ) to further narrow down the possible
lanet–disc configurations consistent with the observed gap. Note 
hat these configurations correspond to disc-to-planet mass ratios 
f M disc /m plt � 1; for higher ratios, the system would not feature
ecular resonances and the disc may not be sufficiently stirred to
roduce observable signatures (A. A. Sefilian 2024 ). 

.5.1 HD 92945 

or HD 92945, we identify the overlap between the valid planet–disc
arameter space that satisfies gap opening time-scales and maximum 

isc mass requirements (white region in fig. 12 of A. A. Sefilian et al.
021 ) and the PMa-derived planet constraints (given in Table 4 ). This
esults in a planet located at ∼10 - 25 au with a mass between ∼0 . 5 - 3
 Jup . Finally, by requiring that a secular resonance falls at the gap

ocation, this constrains the disc mass to lie between ∼10 and 100
 ⊕. Using the limits derived from the JWST /NIRCam observations

C. Lazzoni et al. 2025 ) does not pro vide an y tighter constraints.
his scenario therefore remains valid to explain the observed gap in
D 92945. 

.5.2 HD 107146 

ig. 7 in A. A. Sefilian et al. ( 2021 ) illustrates the range of planet and
isc parameters in which this scenario is capable of producing a gap
n HD 107146, assuming the disc has a single wide gap. Assuming the
lanet responsible for this secular resonance also produces the PMa 
ignal, we can constrain the viable planet parameter space in the same
ay as for HD 92945. Under this assumption, the planet is expected

o be located between ∼8 and 15 au with a mass ranging from
2 - 5 M Jup . To place the secular resonance at the observed gap loca-

ion, the corresponding disc mass must lie between ∼30 and 100 M ⊕.
Ho we v er, a cav eat in this analysis is that HD 107146 features two

istinct gaps in its disc, and not one single wide gap. This double-
ap profile complicates the interpretation as this mechanism has only 
een shown to produce single gaps. In principle, this complication 
MNRAS 546, 1–23 (2026) 



18 R. Bendahan-West et al. 

M

c  

T  

t  

a  

d  

r  

o  

5

E  

t  

s  

H  

o  

M  

a  

H  

t  

F  

a  

m  

f  

t  

d

5

5

A  

o  

c  

s  

s  

R  

A  

p  

s  

d  

o  

b

5

S  

s  

l  

s  

d  

7  

p  

p  

(  

m  

l  

s  

m  

e

6

I  

t  

H  

g  

s  

o  

w  

g
 

t  

r  

a  

m  

a  

m
 

s  

t  

A  

fl
 

p  

i  

w  

(  

i  

o  

o  

w  

a  

c
 

c  

m  

t  

s  

l  

o  

f  

t  

r
 

f  

c  

b  

w  

u  

p
 

e  

i
 

p  

p  

m  

t  

b  

p
 

E  

k  
an be addressed if the system hosts two planets interior to the disc.
hrough their secular interactions with a sufficiently massive disc,

he planets can sculpt two distinct axisymmetric gaps, each located at
 corresponding secular resonance. Friebe & Sefilian (in preparation)
emonstrate that a broad range of planetary configurations, that
emain below current detection limits, can sculpt two gaps at the
bserv ed locations, pro vided the disc mass does not exceed ∼100M ⊕.

.5.3 HD 206893 

valuating this scenario for HD 206893 is challenging due to
he presence of two already detected massive companions in the
ystem. A. A. Sefilian et al. ( 2021 ) considered the influence of only
D 206893 B, constraining the disc mass to ≈170 M ⊕ (with updated
rbital parameters for HD 206893 B, this estimate increases to ≈400
 ⊕). Ho we ver, the presence of a second massive planet significantly

ffects the system’s dynamics, making an analysis based solely on
D 206893 B incomplete. The impact of an additional planet in

his secular resonance scenario with a massive disc is explored in
riebe & Sefilian (in preparation). They find that, while considering
ny additional inner planet will lower the disc mass required to
aintain the resonance location at the observed location, accounting

or a non-zero disc mass shifts the secular resonances inward relative
o the results of B. Yelverton & G. M. Kennedy ( 2018 ), making it
ifficult to explain the observed gap. 

.6 Alternati v e scenarios 

.6.1 Gap carving planet(s) orbiting exterior to the disc 

n additional scenario worth considering is the influence of planets
n orbits exterior to the disc. Although gap formation by such
ompanions has not been studied in great detail, evidence from
ystems such as HD 106906 suggests that wide-orbit planets can
trongly perturb debris structures through secular perturbations (E.
. Nesvold, S. Naoz & M. P. Fitzgerald 2017 ; M. A. Farhat,
. A. Sefilian & J. R. Touma 2023 ). Exterior companions could
ro vide an alternativ e pathway to e xplain some of the observed disc
ubstructures, highlighting a mechanism worth exploring in future
ynamical studies. For the three systems studied in this work, MIRI
bservations rule out companions > 1 . 5 - 2 M Jup at 3 σ confidence
eyond the disc outer edges. 

.6.2 Planet-less scenario 

o far, we have explored four gap-carving scenarios driven by
ingle or multiple planets, but gaps could also form from planet-
ess mechanisms. As discussed in S. Marino et al. ( 2019 , 2020 ),
everal planetless processes may explain the observed gaps across the
ifferent systems, particularly since all three discs exhibit gaps near
0 au. Possible mechanisms include: hydrodynamical processes that
roduce dust gaps in protoplanetary discs, which are then inherited by
lanetesimals formed from that dust and the collisionally generated
secondary) dust; a different size distribution in the gap, reducing the
ass in dust, but not necessarily the local solid mass (e.g. by having

arger planetesimals that dominate the mass budget); a change in the
trength of solids or a higher dynamical excitation at the gap location
aking their collisional evolution faster (see discussion in S. Marino

t al. 2019 , 2020 , for alternative gap origins). 
NRAS 546, 1–23 (2026) 
 C O N C L U S I O N S  

n this paper, we present Cycle 1 JWST /MIRI coronagraphic observa-
ions at 11 . 4 μm (F1140C) targeting three e xoK uiper belt systems:
D 92945, HD 107146, and HD 206893, all of which show radial
aps in the discs, perhaps suggestive of planet–disc interactions. Each
ystem also shows a significant PMa, further supporting the presence
f unseen companions. The main goal of these JWST observations
as to detect the planets thought to be responsible for the observed
aps. 

The MIRI observations were processed with SPACEKLIP , using
he ADI, RDI, and ADI + RDI PSF subtraction techniques. The
eductions incorporate key improvements tailored to MIRI-specific
rtefacts, including corrections for the BFE and persistence trim-
ing. We achieved the highest mass sensitivity with the ADI + RDI

pproach, converting the magnitudes to mass using a planet evolution
odel combining both ATMO-ceq and BEX. 
No planet candidates were detected in the data. All detected

ources were confirmed to be either background galaxies (identified
hrough multiwavelength, multi-epoch data) or a background star.
dditionally, PSF-fitting was done to all the sources to catalogue the
uxes and positions of these background contaminants. 
To quantify the observational sensitivity, we constructed detection

robability maps (DPMs) for each system, translating contrast limits
nto de-projected separation and planet mass space. For HD 92945,
e also incorporated JWST /NIRCam coronagraphic data at 4 . 4 μm

F444W) from C. Lazzoni et al. ( 2025 ) for comparison. These DPMs
nclude not only the JWST limits but also constraints from archi v al
bservations (ground-based direct imaging and RV), disc morphol-
gy, Gaia astrometric noise (RUWE), and the PMa signal. While
e did not reach the predicted sensitivity due to unaccounted MIRI

rtefacts in the PANCAKE simulations, the DPMs provide valuable
onstraints on the population of potential planets in each system. 

We used these constraints to assess which planet configurations
ould explain the sculpting of the inner edge of the discs, the
easured PMa signals and the observed gaps in the discs. We show

hat the planets responsible for the PMa are not the same as the ones
haping the inner edge of the discs. Instead, the PMa planets are likely
ocated within the inner 20 au of the system, consistent with trends
bserved in other systems. Only the next generation of instruments,
rom the ELTs to future Gaia releases, will expand the search for
he planets responsible for the observed PMa with increased angular
esolution and refined astrometric constraints. 

Regarding planets responsible for the observed gaps, we e v aluate
our possible gap-carving scenarios for each system, using the
onstraints provided by the DPMs. Each mechanism was assessed
ased on its ability to reproduce the observed gap location, gap
idth, and observed gap symmetry. These JWST observations allow
s to rule out or place tighter constraints on the possible gap-carving
lanets. These results are summarized in Table 5 . 
• For HD 92945, three gap-carving scenarios remain possible:

mbedded planet(s) carving the gap, or planets interior to the disc
nner edge producing gaps through secular resonances. 

• For HD 107146, the gaps could be explained by embedded
lanets in each gap or by migrating planet(s) capturing Trojan
opulations. The presence of two gaps and their level of symmetry
ake it currently difficult to explain them with secular resonances,

hough more detailed work including N -body simulations would
e required to assess these scenarios further (Friebe & Sefilian, in
reparation). 
• For HD 206893, most gap-carving scenarios can be ruled out.

valuation of the secular resonance in a massless disc due to the two
nown companions shows that the secular resonance location does



Gap-carving planets in exoKuiper belts 19 

n  

t  

t

c
o
e
d
a
c
s
t

A

W  

h  

R
f
R
t
s
b
F
b
b
n
t  

t
A
b

N
A
I
R
f
T  

b
O
s
d

 

a

D

A  

d

R

A
A
A
A  

B
B
B
B
B
B

B
B  

B
B
B  

B  

B
B
B

C
C  

C
C
C  

C
C
C
C
C
D
D  

d  

D  

 

D
E
F
F
F
F
F
F
F
G
G
G
G
G
G  

H
H
H
H
H
H
I
I
K
K  

K
K
K  

K

ot coincide with the observed gap, though it may be responsible for
he sculpting of the disc’s inner edge. Instead, the observed gap in
his system is best explained by the presence of embedded planet(s). 

Across all three systems, while some gap-forming mechanisms 
an be ruled out, multiple viable pathways remain for producing the 
bserved disc substructures. Removing this degeneracy will require 
ither detecting the inner planets causing the PMa’s or using future 
irect imaging instruments with increased sensitivity to detect planets 
t the gap locations themselves. Disentangling the underlying gap- 
arving mechanisms will then require combining such future planet 
earches with dedicated N -body simulations to directly compare 
heoretical predictions with the observed disc morphologies. 

C K N OW L E D G E M E N T S  

e w ould lik e to thank the referee for their constructive feedback that
elped us impro v ed the quality of this paper. RBW is supported by a
oyal Society (grant number RF-ERE-221025). SM acknowledges 

unding by the Royal Society through a Royal Society University 
esearch Fellowship (URF-R1-221669 ) and the European Union 

hrough the FEED ERC project (grant number 101162711). AAS is 
upported by the Heising-Simons Foundation through a 51 Pegasi 
 Fellowship. TDP is supported by a UKRI Stephen Hawking 
ellowship and a Warwick Prize Fellowship, the latter made possible 
y a generous philanthropic donation. LM acknowledges funding 
y the European Union through the E-BEANS ERC project (grant 
umber 100117693). Views and opinions expressed are however 
hose of the author(s) only and do not necessarily reflect those of
he European Union or the European Research Council Ex ecutiv e 
gency. Neither the European Union nor the granting authority can 
e held responsible for them. 
This work is based on observations made with the 

ASA/ESA/CSA JWST . The data were obtained from the Mikulski 
rchive for Space Telescopes at the Space Telescope Science 

nstitute, which is operated by the Association of Universities for 
esearch in Astronomy, Inc., under NASA contract NAS 5-03127 

or JWST . These observations are associated with programme #1668. 
his work has made use of the SPHERE Data Centre, jointly operated
y OSUG/IPAG (Grenoble), PYTHEAS/LAM/CeSAM (Marseille), 
CA/Lagrange (Nice) and Observatoire de P aris/LESIA (P aris) and 

upported by a grant from Labex OSUG@2020 (Investissements 
’avenir – ANR10 LABX56). 
RBW thanks W. O. Balmer, E. Bogat, R. Ferrer -Cha vez, R. Kane,

nd K. Franson for helpful discussions. 

ATA  AVA ILA BILITY  

ll the JWST data used in this paper can be found in MAST at
oi:10.17909/gsf1-2q34 . 

EFER ENCES  

ndrews S. M. et al., 2018, ApJ , 869, L41 
rdila D. R. et al., 2004, ApJ , 617, L147 
rgyriou I. et al., 2023, A&A , 680, A96 
ugereau J. C. , Nelson R. P., Lagrange A. M., Papaloizou J. C. B., Mouillet

D., 2001, A&A , 370, 447 
ardalez Gagliuffi D. C. et al., 2025, ApJ , 988, L18 
euzit J. L. et al., 2019, A&A , 631, A155 
occaletti A. et al., 2015, PASP , 127, 633 
occaletti A. et al., 2019, A&A , 625, A21 
occaletti A. et al., 2022, A&A , 667, A165 
onavita M. , 2020, Astrophysics Source Code Library, record ascl:2010.008 
onnefoy M. et al., 2017, A&A , 597, L7 
onsor A. , Raymond S. N., Augereau J.-C., Ormel C. W., 2014, MNRAS ,

441, 2380 
ooth M. et al., 2023, MNRAS , 521, 6180 
owens-Rubin R. et al. 2025, ApJL, 986, L26 
randt G. M. , Brandt T. D., Dupuy T. J., Li Y., Michalik D., 2021a, AJ , 161,

179 
randt G. M. , Brandt T. D., Dupuy T. J., Michalik D., Marleau G.-D., 2021b,

ApJ , 915, L16 
randt T. D. , 2024a, PASP , 136, 045004 
randt T. D. , 2024b, PASP , 136, 045005 
ushouse H. et al., 2022, spacetelescope/jwst: JWST 1.6.2 , https://doi.org/ 

10.5281/zenodo.6984366 
arter A. L. et al., 2021a, MNRAS , 501, 1999 
arter A. L. et al., 2021b, in Shaklan S. B., Ruane G. J., eds, Proc. SPIE

Conf. Ser., Vol. 11823, Techniques and Instrumentation for Detection of 
Exoplanets X . SPIE, Bellingham, p. 118230H 

arter A. L. et al., 2023, ApJ , 951, L20 
hambers J. , Wetherill G., Boss A., 1996, Icarus , 119, 261 
hen C. H. , Mittal T., Kuchner M., Forrest W. J., Lisse C. M., Manoj P.,

Sargent B. A., Watson D. M., 2014, ApJS , 211, 25 
loutier R. , 2024, preprint ( arXiv:2409.13062 ) 
osta T. , Pearce T. D., Krivov A. V., 2024, MNRAS , 527, 7317 
rotts K. A. , Matthews B. C., 2024, ApJ , 975, 136 
rotts K. A. et al., 2024, ApJ , 961, 245 
rotts K. A. et al., 2025, ApJL, 987, L41 
aley C. et al., 2019, ApJ , 875, 87 
e Rosa R. J. , Nielsen E. L., Wahhaj Z., Ruffio J.-B., Kalas P. G., Peck A.

E., Hirsch L. A., Roberson W., 2023, A&A , 672, A94 
e Pater I. , Lissauer J. J., 2001, Planetary Sciences, Cambridge University

Press, Cambridge, UK 

elorme P. et al., 2017, in Reyl ́e C., Di Matteo P., Herpin F., Lagadec E.,
Lan c ¸on A., Meliani Z., Royer F., eds, SF2A-2017: Proceedings of the
Annual meeting of the French Society of Astronomy and Astrophysics, 
preprint (arXiv:1712.06948) 

icken D. et al., 2024, A&A , 689, A5 
sposito T. M. et al., 2020, AJ , 160, 24 
ang J. , Margot J.-L., 2013, ApJ , 767, 115 
aramaz V. et al., 2014, A&A , 563, A72 
aramaz V. et al., 2019, AJ , 158, 162 
arhat M. A. , Sefilian A. A., Touma J. R., 2023, MNRAS , 521, 2067 
eldt M. et al., 2017, A&A , 601, A7 
ranson K. et al., 2023, ApJ , 950, L19 
riebe M. F. , Pearce T. D., L ̈ohne T., 2022, MNRAS , 512, 4441 
aia Collaboration et al., 2023, A&A , 674, A1 
 ́asp ́ar A. et al., 2023, Nat. Astron. , 7, 790 
ladman B. , 1993, Icarus , 106, 247 
odoy N. et al., 2024, A&A , 689, A185 
olimowski D. A. et al., 2011, AJ , 142, 30 
ray R. O. , Corbally C. J., Garrison R. F., McFadden M. T., Bubar E. J.,

McGahee C. E., O’Donoghue A. A., Knox E. R., 2006, AJ , 132, 161 
an Y. , Wyatt M. C., Dent W. R. F., 2023, MNRAS , 519, 3257 
arlan E. A. , Taylor D. C., 1970, AJ , 75, 507 
inkley S. et al., 2022, PASP , 134, 095003 
inkley S. et al., 2023, A&A , 671, L5 
olland W. S. et al., 2017, MNRAS , 470, 3606 
ughes A. M. , Duchene G., Matthews B., 2018, ARA&A , 56, 541 

da S. , Bryden G., Lin D. N. C., Tanaka H., 2000, ApJ , 534, 428 
maz Blanco A. et al., 2023, MNRAS , 522, 6150 
ammerer J. et al., 2021, A&A , 652, A57 
ammerer J. et al., 2022, in Coyle L. E., Matsuura S., Perrin M. D. eds, Proc.

SPIE Conf. Ser. Vol. 12180, Space Telescopes and Instrumentation 2022: 
Optical, Infrared, and Millimeter Wave . SPIE, Bellingham, p. 121803N. 

ennedy G. M. , 2020, Royal Society Open Science , 7, 200063 
ennedy G. M. , Wyatt M. C., 2010, MNRAS , 405, 1253 
ennedy G. M. , Marino S., Matr ̀a L., Pani ́c O., Wilner D., Wyatt M. C.,

Yelverton B., 2018, MNRAS , 475, 4924 
ervella P. , Th ́evenin F., Di Folco E., S ́egransan D., 2004, A&A , 426, 297 
MNRAS 546, 1–23 (2026) 

https://doi.org/10.17909/gsf1-2q34
http://dx.doi.org/10.3847/2041-8213/aaf741
http://dx.doi.org/10.1086/427434
http://dx.doi.org/10.1051/0004-6361/202346490
http://dx.doi.org/10.1051/0004-6361:20010199
http://dx.doi.org/10.3847/2041-8213/ade30f
http://dx.doi.org/10.1051/0004-6361/201935251
http://dx.doi.org/10.1086/682256
http://dx.doi.org/10.1051/0004-6361/201935135
http://dx.doi.org/10.1051/0004-6361/202244578
http://dx.doi.org/10.1051/0004-6361/201628929
http://dx.doi.org/10.1093/mnras/stu721
http://dx.doi.org/10.1093/mnras/stad938
http://dx.doi.org/10.48550/arXiv.2505.15995
http://dx.doi.org/10.3847/1538-3881/abdc2e
http://dx.doi.org/10.3847/2041-8213/ac0540
http://dx.doi.org/10.1088/1538-3873/ad38d9
http://dx.doi.org/10.1088/1538-3873/ad38da
http://dx.doi.org/10.5281/zenodo.6984366
https://doi.org/10.5281/zenodo.6984366
http://dx.doi.org/10.1093/mnras/staa3579
http://dx.doi.org/10.1117/12.2594501
http://dx.doi.org/10.3847/2041-8213/acd93e
http://dx.doi.org/10.1006/icar.1996.0019
http://dx.doi.org/10.1088/0067-0049/211/2/25
http://arxiv.org/abs/2409.13062
http://dx.doi.org/10.1093/mnras/stad3582
http://dx.doi.org/10.3847/1538-4357/ad7b28
http://dx.doi.org/10.3847/1538-4357/ad0e69
http://dx.doi.org/10.3847/1538-4357/ab1074
http://dx.doi.org/10.1051/0004-6361/202345877
http://dx.doi.org/10.1051/0004-6361/202449451
http://dx.doi.org/10.3847/1538-3881/ab9199
http://dx.doi.org/10.1088/0004-637X/767/2/115
http://dx.doi.org/10.1051/0004-6361/201322469
http://dx.doi.org/10.3847/1538-3881/ab3ec1
http://dx.doi.org/10.1093/mnras/stad316
http://dx.doi.org/10.1051/0004-6361/201629261
http://dx.doi.org/10.3847/2041-8213/acd6f6
http://dx.doi.org/10.1093/mnras/stac664
http://dx.doi.org/10.1051/0004-6361/202243940
http://dx.doi.org/10.1038/s41550-023-01962-6
http://dx.doi.org/10.1006/icar.1993.1169
http://dx.doi.org/10.1051/0004-6361/202449951
http://dx.doi.org/10.1088/0004-6256/142/1/30
http://dx.doi.org/10.1086/504637
http://dx.doi.org/10.1093/mnras/stac3769
http://dx.doi.org/10.1086/110986
http://dx.doi.org/10.1088/1538-3873/ac77bd
http://dx.doi.org/10.1051/0004-6361/202244727
http://dx.doi.org/10.1093/mnras/stx1378
http://dx.doi.org/10.1146/annurev-astro-081817-052035
http://dx.doi.org/10.1086/308720
http://dx.doi.org/10.1093/mnras/stad1221
http://dx.doi.org/10.1051/0004-6361/202140749
http://dx.doi.org/10.1117/12.2628865
http://dx.doi.org/10.1098/rsos.200063
http://dx.doi.org/10.1111/j.1365-2966.2010.16528.x
http://dx.doi.org/10.1093/mnras/sty135
http://dx.doi.org/10.1051/0004-6361:20035930


20 R. Bendahan-West et al. 

M

K
K
K
K
L
L
L  

 

 

L  

 

 

 

L
L
L
L
L
L  

M
M
M
M
M
M
M
M
M  

M
M  

M
M
M
M
M
M
M
M  

M  

M
M
M  

M
M  

 

M
N
N
N
P
P
P
P
P
P
P
P  

R
R

R
R
R
R  

R
S
S
S
S
S
S
S  

S
S
S
S  

S
S
T
T
T  

T
T  

T  

 

T
T  

V
W  

W  

W
W  

W
W
W
W  

X
Y
Z  

A

D  

t  

a  

t  

l  

g  

l  

b  

o  

w  

t  

T  
ervella P. , Arenou F., Mignard F., Th ́evenin F., 2019, A&A , 623, A72 
ervella P. , Arenou F., Th ́evenin F., 2022, A&A , 657, A7 
iefer F. , Lagrange A.-M., Rubini P., Philipot F., 2025, A&A, 702, A76 
irsh D. R. , Duncan M., Brasser R., Levison H. F., 2009, Icarus , 199, 197 
agrange A. M. et al., 2009, A&A , 493, L21 
agrange A. M. et al., 2025, Nature, 642, 905 
ajoie C.-P. , Soummer R., Hines D. C., Rieke G. H., 2014, in Oschmann J. M.

Jr, Clampin M., Fazio G. G., MacEwen H. A. eds, Proc. SPIE Conf. Ser.
Vol. 9143, Space Telescopes and Instrumentation 2014: Optical, Infrared,
and Millimeter Wave . SPIE, Bellingham, p. 91433R 

ajoie C.-P. , Soummer R., Pueyo L., Hines D. C., Nelan E. P., Perrin M.,
Clampin M., Isaacs J. C., 2016, in MacEwen H. A., Fazio G. G., Lystrup
M., Batalha N., Siegler N., Tong E. C. eds, Proc. SPIE Conf. Ser. Vol.
9904, Space Telescopes and Instrumentation 2016: Optical, Infrared, and
Millimeter Wave . SPIE, Bellingham, p. 99045K 

awson K. et al., 2024, ApJ , 967, L8 
azzoni C. et al., 2018, A&A , 611, A43 
azzoni C. et al., 2025, A&A, 704, A176 
imbach M. A. et al., 2024, ApJ , 973, L11 
indegren L. et al., 2021, A&A , 649, A2 
inder E. F. , Mordasini C., Molli ̀ere P., Marleau G.-D., Malik M., Quanz S.

P., Meyer M. R., 2019, A&A , 623, A85 
acGregor M. A. et al., 2017, ApJ , 842, 8 
acGregor M. A. et al., 2019, ApJ , 877, L32 
acintosh B. et al., 2014, Proc. Natl. Acad. Sci. , 111, 12661 
alhotra R. , 1995, AJ , 110, 420 
 ̂ alin M. et al., 2025, A&A , 693, A315 
arino S. , 2021, MNRAS , 503, 5100 
arino S. , 2022, preprint ( arXiv:2202.03053 ) 
arino S. et al., 2018, MNRAS , 479, 5423 
arino S. , Yelverton B., Booth M., Faramaz V., Kennedy G. M., Matr ̀a L.,

Wyatt M. C., 2019, MNRAS , 484, 1257 
arino S. et al., 2020, MNRAS , 498, 1319 
arois C. , Lafreni ̀ere D., Doyon R., Macintosh B., Nadeau D., 2006, ApJ ,

641, 556 
atr ̀a L. et al., 2025, A&A , 693, A151 
atthews E. C. et al., 2024, Nature , 633, 789 
esa D. et al., 2021, MNRAS , 503, 1276 
esa D. et al., 2023, A&A , 672, A93 
eunier N. , Lagrange A. M., De Bondt K., 2012, A&A , 545, A87 
illi J. et al., 2017a, A&A , 597, L2 
illi J. et al., 2017b, A&A , 599, A108 
orbidelli A. , Nesvorn ́y D., 2020, in The Trans-Neptunian Solar System,

Prialnik D., Barucci M. A., Young L. eds, Elsevier, p. 25 
orbidelli A. , Levison H. F., Tsiganis K., Gomes R., 2005, Nature , 435, 462
orrison S. , Malhotra R., 2015, ApJ , 799, 41 
orrison S. J. , Kratter K. M., 2018, MNRAS , 481, 5180 
ouillet D. , Larwood J. D., Papaloizou J. C. B., Lagrange A. M., 1997,

MNRAS , 292, 896 
 ̈uller M. , Weigelt G., 1987, A&A, 175, 312 
urray C. D. , Dermott S. F., 1999, Solar SystemDynamics, Cambridge

University Press, Cambridge, UK, https:// ui.adsabs.harvard.edu/ abs/ 19
99ssd..book.....M 

ustill A. J. , Wyatt M. C., 2012, MNRAS , 419, 3074 
ederlander A. et al., 2021, ApJ , 917, 5 
esvold E. R. , Naoz S., Fitzgerald M. P., 2017, ApJ , 837, L6 
ielsen E. L. et al., 2019, AJ , 158, 13 
earce T. D. , Wyatt M. C., 2014a, MNRAS , 443, 2541 
earce T. D. , Wyatt M. C., 2014b, MNRAS , 443, 2541 
earce T. D. , Wyatt M. C., 2015, MNRAS , 453, 3329 
earce T. D. et al., 2022, A&A , 659, A135 
earce T. D. et al., 2024, MNRAS , 527, 3876 
errot C. et al., 2016, A&A , 590, L7 
hillips M. W. et al., 2020, A&A , 637, A38 
lavchan P. , Werner M. W., Chen C. H., Stapelfeldt K. R., Su K. Y. L., Stauffer

J. R., Song I., 2009, ApJ , 698, 1068 
afikov R. R. , 2023, MNRAS , 519, 5607 
ebollido I. et al., 2024, AJ , 167, 69 
NRAS 546, 1–23 (2026) 
eg ́aly Z. , Dencs Z., Mo ́or A., Kov ́acs T., 2018, MNRAS , 473, 3547 
en B. et al., 2021, ApJ , 914, 95 
odet L. , Lai D., 2022, MNRAS , 516, 5544 
ouan D. , Riaud P., Boccaletti A., Cl ́enet Y., Labeyrie A., 2000, PASP , 112,

1479 
uane G. et al., 2019, AJ , 157, 118 
anghi A. et al., 2025, ApJ , 989, L23 
chneider G. et al., 2014, AJ , 148, 59 
efilian A. A. , 2024, ApJ , 966, 140 
efilian A. A. , Rafikov R. R., Wyatt M. C., 2021, ApJ , 910, 13 
efilian A. A. , Rafikov R. R., Wyatt M. C., 2023, ApJ , 954, 100 
hannon A. , Bonsor A., Kral Q., Matthews E., 2016, MNRAS , 462, L116 
ibthorpe B. , Kennedy G. M., Wyatt M. C., Lestrade J. F., Greaves J. S.,

Matthews B. C., Duch ̂ ene G., 2018, MNRAS , 475, 3046 
mith A. W. , Lissauer J. J., 2009, Icarus , 201, 381 
quicciarini V. , Bonavita M., 2022, A&A , 666, A15 
quicciarini V. et al., 2025, A&A , 693, A54 
tanford-Moore S. A. , Nielsen E. L., De Rosa R. J., Macintosh B., Czekala

I., 2020, ApJ , 898, 27 
tasevic S. et al., 2023, A&A , 678, A8 
u K. Y. L. et al., 2024, ApJ , 977, 277 
abeshian M. , Wiegert P. A., 2016, ApJ , 818, 159 
abeshian M. , Wiegert P. A., 2017, ApJ , 847, 24 
al-Or L. , Trifonov T., Zucker S., Mazeh T., Zechmeister M., 2019, MNRAS ,

484, L8 
elesco C. M. et al., 2005, Nature , 433, 133 
orres C. A. O. , Quast G. R., da Silva L., de La Reza R., Melo C. H. F.,

Sterzik M., 2006, A&A , 460, 695 
remaine S. , 1993, in Phillips J. A., Thorsett S. E., Kulkarni S. R. eds,

ASP Conf. Ser. Vol. 36, Planets Around Pulsars. Astron. Soc. Pac., San
Francisco, p. 335 

remaine S. , 1998, AJ , 116, 2015 
rifonov T. , Tal-Or L., Zechmeister M., Kaminski A., Zucker S., Mazeh T.,

2020, A&A , 636, A74 
igan A. et al., 2021, A&A , 651, A72 
alsh K. J. , Morbidelli A., Raymond S. N., O’Brien D. P., Mandell A. M.,

2011, Nature , 475, 206 
ang J. J. , Ruffio J.-B., De Rosa R. J., Aguilar J., Wolff S. G., Pueyo L.,

2015, Astrophysics Source Code Library, record ascl:1506.001 
isdom J. , 1980, AJ , 85, 1122 
olff S. G. , G ́asp ́ar A., Rieke G., Leisenring J. M., Sefilian A. A., Ygouf M.,

Llop-Sayson J., 2025, AJ , 170, 244 
yatt M. C. , 2003, ApJ , 598, 1321 
yatt M. C. , 2005, A&A , 440, 937 
yatt M. C. , 2008, ARA&A , 46, 339 
yatt M. C. , Dermott S. F., Telesco C. M., Fisher R. S., Grogan K., Holmes

E. K., Pi ̃ na R. K., 1999, ApJ , 527, 918 
ie C. et al., 2025, Nature , 641, 608 
elverton B. , Kennedy G. M., 2018, MNRAS , 479, 2673 
heng X. , Lin D. N. C., Kouwenho v en M. B. N., Mao S., Zhang X., 2017,

ApJ , 849, 98 

PPENDI X  A :  LI KELY  R A M P  FITTING  

uring the ramp-fitting step described in Section 2.3 .a, we excluded
he first 10 and last 2 groups of each integration. The initial groups
re affected by detector settling effects following the reset, while
he end of the ramp shows a systematic drop in counts. Both effects
ead to deviations from the linear ramp expected for well-behaved
roups. Fig. A1 illustrates this for two representative pixels: one
ocated within a bright PSF lobe (red lines) and another in the
ackground region (black lines), both taken from the first integration
f the HD 92945 uncalibrated science observation. For comparison,
e also plot the median group values across all integrations for

he same pixels (dashed lines), which show the same behaviour.
his effect is present in all MIRI coronagraphy observations of this

http://dx.doi.org/10.1051/0004-6361/201834371
http://dx.doi.org/10.1051/0004-6361/202142146
http://dx.doi.org/10.48550/arXiv.2409.16992
http://dx.doi.org/10.1016/j.icarus.2008.05.028
http://dx.doi.org/10.1051/0004-6361:200811325
http://dx.doi.org/10.48550/arXiv.2502.15081
http://dx.doi.org/10.1117/12.2056284
http://dx.doi.org/10.1117/12.2233032
http://dx.doi.org/10.3847/2041-8213/ad4496
http://dx.doi.org/10.1051/0004-6361/201731426
http://dx.doi.org/10.48550/arXiv.2511.07561
http://dx.doi.org/10.3847/2041-8213/ad74ed
http://dx.doi.org/10.1051/0004-6361/202039709
http://dx.doi.org/10.1051/0004-6361/201833873
http://dx.doi.org/10.3847/1538-4357/aa71ae
http://dx.doi.org/10.3847/2041-8213/ab21c2
http://dx.doi.org/10.1073/pnas.1304215111
http://dx.doi.org/10.1086/117532
http://dx.doi.org/10.1051/0004-6361/202452695
http://dx.doi.org/10.1093/mnras/stab771
http://arxiv.org/abs/2202.03053
http://dx.doi.org/10.1093/mnras/sty1790
http://dx.doi.org/10.1093/mnras/stz049
http://dx.doi.org/10.1093/mnras/staa2386
http://dx.doi.org/10.1086/500401
http://dx.doi.org/10.1051/0004-6361/202451397
http://dx.doi.org/10.1038/s41586-024-07837-8
http://dx.doi.org/10.1093/mnras/stab438
http://dx.doi.org/10.1051/0004-6361/202345865
http://dx.doi.org/10.1051/0004-6361/201219163
http://dx.doi.org/10.1051/0004-6361/201629908
http://dx.doi.org/10.1051/0004-6361/201527838
http://dx.doi.org/10.1016/B978-0-12-816490-7.00002-3
http://dx.doi.org/10.1038/nature03540
http://dx.doi.org/10.1088/0004-637X/799/1/41
http://dx.doi.org/10.1093/mnras/sty2657
http://dx.doi.org/10.1093/mnras/292.4.896
https://ui.adsabs.harvard.edu/abs/1999ssd..book.....M
http://dx.doi.org/10.1111/j.1365-2966.2011.19948.x
http://dx.doi.org/10.3847/1538-4357/abdd32
http://dx.doi.org/10.3847/2041-8213/aa61a7
http://dx.doi.org/10.3847/1538-3881/ab16e9
http://dx.doi.org/10.1093/mnras/stu1302
http://dx.doi.org/10.1093/mnras/stu1302
http://dx.doi.org/10.1093/mnras/stv1847
http://dx.doi.org/10.1051/0004-6361/202142720
http://dx.doi.org/10.1093/mnras/stad3462
http://dx.doi.org/10.1051/0004-6361/201628396
http://dx.doi.org/10.1051/0004-6361/201937381
http://dx.doi.org/10.1088/0004-637X/698/2/1068
http://dx.doi.org/10.1093/mnras/stac3411
http://dx.doi.org/10.3847/1538-3881/ad1759
http://dx.doi.org/10.1093/mnras/stx2604
http://dx.doi.org/10.3847/1538-4357/ac03b9
http://dx.doi.org/10.1093/mnras/stac2621
http://dx.doi.org/10.1086/317707
http://dx.doi.org/10.3847/1538-3881/aafee2
http://dx.doi.org/10.3847/2041-8213/adf53e
http://dx.doi.org/10.1088/0004-6256/148/4/59
http://dx.doi.org/10.3847/1538-4357/ad32d1
http://dx.doi.org/10.3847/1538-4357/abda46
http://dx.doi.org/10.3847/1538-4357/ace68e
http://dx.doi.org/10.1093/mnrasl/slw143
http://dx.doi.org/10.1093/mnras/stx3188
http://dx.doi.org/10.1016/j.icarus.2008.12.027
http://dx.doi.org/10.1051/0004-6361/202244193
http://dx.doi.org/10.1051/0004-6361/202452310
http://dx.doi.org/10.3847/1538-4357/ab9a35
http://dx.doi.org/10.1051/0004-6361/202346720
http://dx.doi.org/10.3847/1538-4357/ad8cde
http://dx.doi.org/10.3847/0004-637X/818/2/159
http://dx.doi.org/10.3847/1538-4357/aa831f
http://dx.doi.org/10.1093/mnrasl/sly227
http://dx.doi.org/10.1038/nature03255
http://dx.doi.org/10.1051/0004-6361:20065602
http://dx.doi.org/10.1086/300567
http://dx.doi.org/10.1051/0004-6361/201936686
http://dx.doi.org/10.1051/0004-6361/202038107
http://dx.doi.org/10.1038/nature10201
http://dx.doi.org/10.1086/112778
http://dx.doi.org/10.3847/1538-3881/adfcd6
http://dx.doi.org/10.1086/379064
http://dx.doi.org/10.1051/0004-6361:20053391
http://dx.doi.org/10.1146/annurev.astro.45.051806.110525
http://dx.doi.org/10.1086/308093
http://dx.doi.org/10.1038/s41586-025-08920-4
http://dx.doi.org/10.1093/mnras/sty1678
http://dx.doi.org/10.3847/1538-4357/aa8ef3


Gap-carving planets in exoKuiper belts 21 

Figure A1. Counts per group for two representative pixels during the first 
integration of the HD 92945 uncalibrated science observation: one in a bright 
PSF lobe (red) and one in a background region (black). Solid lines show the 
first integration, while dashed lines indicate the median o v er all inte grations. 
The right-hand panels zoom in on the start and end of the ramp, with blue 
arrows and boxes marking the regions shown. The y -axis ticks in these zoomed 
panels correspond to 50 counts. 

p  

t  

e

A
C

T
i
‘
b
W
o  

l  

b  

m

o
‘

c  

a
d  

a  

8  

p
 

o  

r  

r
e
i
t  

a

o  

s  

c  

e  

a
g  

t  

t  

g  

o  

g

i
m
t  

f
a  

s
h
n
t  

f  

m

A
C

F
m

rogramme. The right-hand panels zoom in on the start and end of
he group sequence, highlighting the slope changes that moti v ate the
xclusion of these groups from the fit. 

PPEN D IX  B:  BRIGHTER-FATTER  EFFECT  

O R R E C T I O N  

he BFE correction can remo v e significant residuals that otherwise 
mpact the quality of the reductions. Three methods are available: 
custom’, ‘basic’, and ‘advanced’. Fig. B1 sho ws the dif ferences 
etween no BFE correction, the ‘basic’ and ‘advanced’ methods. 
ithout correction, significant residuals remain in the inner regions 

f the image. The ‘basic’ method reduces residuals near the PSF
obes, but the number of groups trimmed causes the ‘glow stick’ to
e o v ersubtracted. The best reduction is obtained with the ‘advanced’
ethod. 
The ‘basic’ and ‘advanced’ methods differ in the region of 

ptimization but follow the same principle. More specifically, the 
advanced’ group-masking method operates as follows. First, we 
T  

t

igure B1. HD 92945 PSF-subtracted images using ADI + RDI with different B
ethod. Different residual patterns are visible across methods, with the ‘advanced’
alculate the median of the last groups across all science integrations,
nd repeat this across all reference integrations. We then indepen- 
ently median subtract each of the median group images to remo v e
ny uniform background, and identify and mask all pixels below an
5 per cent percentile threshold to isolate the routine to the brightest
ixels likely to be affected by BFE. 
Each median group image is then split into a ‘central’ box region

f 20 ×20 pixels, centred on the 4QPM centre, and an ‘external’
egion of pixels outside of this box. Using the central and external
egions, we independently calculate the absolute differences between 
ach median reference group image and the median science group 
mage, and determine the corresponding reference groups at which 
he difference for each region is minimized. This defines a maximum
nd minimum number of groups to be masked. 

Next, for each reference independently, we calculate the median 
f the final reference group across all integrations, perform a median
ubtraction to match the earlier steps, and determine the peak pixel
ount in the central region, and the minimum pixel count in the
 xternal re gion. A linear interpolation between the pix el count values,
nd the previously identified maximum and minimum number of 
roups to be masked, is then used to identify the number of groups
hat must be masked during the ramp fitting for each pixel based on
he counts in the median subtracted final group image. The number of
roups to mask for pixels with counts outside the defined minimum
r maximum, are clamped to the minimum of maximum number of
roups to mask, respectively. 
We emphasize that the success of this BFE correction technique 

s only possible because the reference observations experience a 
ore significant BFE due to their higher detector counts compared 

o the science observations. Groups can be readily remo v ed as a
unction of pixel position from the references without significantly 
ffecting the o v erall signal to noise of the subtracted image. If the
ituation were reversed, and the science observations experienced 
igher detector counts, then to mimic this technique groups would 
eed to be trimmed from the science observations, directly impacting 
he achie v able SNR. A more general approach to a BFE correction,
or example using a deconvolution kernel (I. Argyriou et al. 2023 ),
ay be more appropriate in these situations. 

PPENDI X  C :  B  AC K G R  O U N D  OBSERVATIO N  

L E A N I N G  

o automatically identify and remo v e the contaminants observ ed in
he HD 92945 science background observations, we computed the 
MNRAS 546, 1–23 (2026) 

FE correction methods: no group masking, ‘basic’ method, and ‘advanced’ 
 approach producing the cleanest images. 



22 R. Bendahan-West et al. 

M

d  

a  

m  

i  

s

S

w  

a  

>  

S  

g  

b  

s  

C  

F  

s  

t  

p  

t  

b  

r  

p

F  

b  

b  

p  

f  

p  

S  

m  

b  

p

r  

s

A

T  

i  

t  

i  

b  

5  

T  

6  

i  

t  

p

A

F  

i  

w  

i  

g

ifference between the two available science background images
nd divided it by the noise maps added in quadrature. Each noise
ap was defined as the standard deviation of each pixel across all

ntegrations for the corresponding science background. The resulting
ignal-to-noise ratio (SNR) map is given by 

NR = 

bkg 1 − bkg 2 √ 

σ 2 
bkg 1 

+ σ 2 
bkg 2 

, (C1) 

here bkg 1 and bkg 2 are the two science background observations,
nd σbkg 1 and σbkg 2 their respective noise maps. Pixels with SNR
 3 σ represent sources observed in bkg 1 to be masked, while
NR < −3 σ identifies contaminants in bkg 2 . When multiple back-
round files of the same type are available (science or reference
ackgrounds), SPACEKLIP takes their median for the background
ubtraction step, yielding a single, cleaned background image. Fig.
1 illustrates the effect of the background cleaning process which

igure C1. Effect of the background cleaning process. The left column
hows the median of the HD 92945 science background observations, while
he right column presents the resulting RDI PSF-subtracted images. All the
anels are in the detectors reference frame and not rotated north-east. The
op row corresponds to the case without background cleaning, where white
oxes highlight background contaminants (left) and their associated negative
esidual (right). The bottom ro w sho ws the results after applying the cleaning
rocedure. 
NRAS 546, 1–23 (2026) 
igure C2. Effect of persistence trimming. The first two panels show
ackground-subtracted science images zoomed in on the location affected
y persistence, for the first (left) and last (centre) integrations. The right
anels show the corresponding RDI PSF-subtracted images. Examples are
or HD 92945 and are rotated north-east. The top row shows data without
ersistence trimming, while the bottom row includes the trimming process.
olid black circles mark regions affected by persistence, dashed black circles
ark regions where no persistence is observed, and the white areas in the

ottom left panel show the masked pixels during the persistence trimming
rocess. 

emo v es strong ne gativ e residual sources, resulting in cleaner PSF-
ubtracted images. 

PPENDI X  D :  PERSISTENCE  T R I M M I N G  

o locate the persistence, we use the position of the star in the two TA
mages. We observe that persistence produces significant residuals
hroughout the data reduction process (Fig. C2 ), most prominently
n the first integration and to a lesser extent in the second, before
ecoming negligible. We therefore mask all pixels within a radius of
 pixels from the persistence location in the first two integrations.
his step is applied only to the science observations, which have
 integrations in total, as the reference observations only have 2
ntegrations and the effect cannot be clearly identified. The impact of
his trimming can be seen in the PSF-subtracted images in the right
anels of Fig. C2 . 

PPENDI X  E:  PSF-FITTING  

ig. E1 illustrates the PSF-fitting process for all the sources observed
n the field-of-view of the three targets. The sources are each fitted
ith a point source, and residuals to the fit indicate whether a source

s more consistent with a star or planet (point-like) or a background
alaxy (extended). 



Gap-carving planets in exoKuiper belts 23 

MNRAS 546, 1–23 (2026) 

Figure E1. Point-source fitting to the candidates observed in the MIRI coronagraphic observations. For each candidate, the left panel shows the source as 
observed in the data, the middle panel shows the PSF model generated from SPACEKLIP , and the residuals from the PSF-fitting are seen in the right panel. 

This paper has been typeset from a T E 

X/L 

A T E 

X file prepared by the author. 

© The Author(s) 2025. 
Published by Oxford University Press on behalf of Royal Astronomical Society. This is an Open Access article distributed under the terms of the Creative Commons Attribution License 
( https://cr eativecommons.or g/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 

https://creativecommons.org/licenses/by/4.0/

	1 INTRODUCTION
	2 OBSERVATIONS AND DATA REDUCTION
	3 ANALYSIS
	4 PLANET CONSTRAINTS USING THE DETECTION PROBABILITY MAPS
	5 GAP CARVING SCENARIOS
	6 CONCLUSIONS
	ACKNOWLEDGEMENTS
	DATA AVAILABILITY
	REFERENCES
	A LIKELY RAMP FITTING
	B BRIGHTER-FATTER EFFECT CORRECTION
	C BACKGROUND OBSERVATION CLEANING
	D PERSISTENCE TRIMMING
	E PSF-FITTING