PRIME: NIR Microlensing Survey

• PRIME is a dedicated near-infrared microlensing survey that uses a 1.8-m telescope and a wide-field NIR camera to access the heavily extincted central Galactic bulge.
• It employs high-cadence imaging, specialized optics, and real-time difference image analysis to detect microlensing events and a variety of NIR variable stars in crowded fields.
• The survey complements optical programs and supports joint Earth-space parallax measurements with NASA’s Roman Space Telescope to refine exoplanet demographics and Galactic structure insights.

Searching arXiv for PRIME-related papers and instrumentation context. The PRime-focus Infrared Microlensing Experiment (PRIME) is a dedicated near-infrared (NIR) prime-focus microlensing survey centered on the Galactic bulge and operated at the South African Astronomical Observatory (SAAO) in Sutherland, South Africa. It combines a 1.8-m telescope, a wide-field NIR camera using four Roman-class HgCdTe detectors, and a high-cadence survey strategy designed to access the highly extincted central Galactic bulge, including the region around Sgr A*, that is largely inaccessible to optical microlensing programs. Its stated goals are to measure the frequency and mass function of planets, to compare inner-bulge results with outer-bulge optical surveys, and to enable space-based microlens parallax measurements in conjunction with NASA’s Nancy Grace Roman Space Telescope. The first bulge survey began in February 2024, and analysis through June 1, 2025 yielded 486 microlensing candidates and more than a thousand variable stars (Sumi et al., 20 Aug 2025).

1. Scientific scope and observational niche

PRIME was conceived as the first dedicated high-cadence wide-field bulge microlensing survey in the NIR. The scientific rationale is straightforward: dust extinction at low Galactic latitude suppresses the reach of optical surveys toward the Galactic center, whereas NIR observations can probe the denser and more extinguished inner bulge. PRIME therefore targets a parameter space complementary to optical programs such as OGLE, MOA, and KMTNet, with an explicit emphasis on the central bulge and the Galactic plane (Sumi et al., 20 Aug 2025).

Its main science program has several coupled components. One is exoplanet demographics: PRIME aims to measure the frequency and mass function of bound exoplanets and free-floating planets in the central bulge, then compare those results with outer-bulge measurements from optical microlensing surveys. A second is Galactic-structure work: the survey is intended to map the microlensing event rate and optical depth in the inner bulge, thereby constraining bulge density and kinematics and informing Roman Galactic Bulge Time Domain Survey field selection. A third is source-population science: the survey detects numerous NIR variables, including Mira variables, RR Lyrae, and Cepheids, in highly reddened regions where existing catalogs are incomplete (Sumi et al., 20 Aug 2025).

A recurrent misconception is that PRIME simply reproduces optical microlensing in a different band. The technical and scientific positioning is narrower and more specific. PRIME is designed to open the central, highly extincted bulge, where higher event rates coexist with stronger extinction, shorter timescale distributions, and severe crowding. The simulation work indicates that these competing effects matter: higher inner-bulge source density and event rate do not translate trivially into maximal planet yield, because extinction and shorter $t_E$ can reduce detection efficiency, especially at lower cadence (Kondo et al., 2023).

2. Telescope, optics, and focal-plane system

The telescope is a dedicated 1.8-m near-infrared prime-focus microlensing survey telescope at SAAO, Sutherland, at altitude 1780 m. Its primary mirror is a 1.8-m AstroSitall paraboloid with focal length 3792 mm ($f/2.11$), coated with Al+SiO$_2$ and specified to give 92–95% reflectivity from 1000–1800 nm. A four-element fused-silica prime-focus corrector yields a combined focal length of 4125 mm ($f/2.29$). The optics were designed for an 80% encircled-energy diameter of 14 $\mu$m, corresponding to 1.4 pixels or 0.7 arcsec across the field, matched to typical Sutherland seeing of $\sim 1.4$ arcsec (Sumi et al., 20 Aug 2025).

PRIME-Cam, built at NASA GSFC and UMd, uses four Teledyne H4RG-10 HgCdTe detectors with $4096 \times 4096$ pixels each, 10 $\mu$m pixels, and a 2.5 $\mu$m cutoff. The focal plane is 86.83 mm $\times$ 86.83 mm with 4.91 mm gaps between detectors, and the pixel scale is 0.5 arcsec/pixel. Readout uses ACADIA electronics in the dewar and MACIE electronics outside, with Sample Up The Ramp frames every 2.86 s and read noise of 5 ADU per read. The cold window is BK7 and blocks out-of-band thermal radiation with OD $f/2.11$0 from 1950–2200 nm and OD $f/2.11$1 from 2200–3000 nm in combination with the filters; the warm dewar window was changed from fused silica to sapphire in 2025 to mitigate condensation (Sumi et al., 20 Aug 2025).

The field of view is 1.14$f/2.11$21.14$f/2.11$3 = 1.29 deg$f/2.11$4 excluding gaps, or 1.21$f/2.11$51.21$f/2.11$6 = 1.45 deg$f/2.11$7 including gaps. Sgr A* falls in chip 2 of field GB94. Two filter wheels provide $f/2.11$8 (850–930 nm), $f/2.11$9 (970–1075 nm), $_2$0 (1170–1330 nm), $_2$1 (1490–1780 nm), three narrow bands (NB1063, NB1243, NB1630 in low-OH windows), and blocked/open positions. For 100 s exposures, the quoted 5$_2$2 limiting magnitudes in Vega are $_2$3, $_2$4, $_2$5, and $_2$6. Typical sky backgrounds per 2.86 s frame are $_2$7 ADU in $_2$8, $_2$9 in $f/2.29$0, $f/2.29$1 in $f/2.29$2, and $f/2.29$3 in $f/2.29$4; because the detector well is $f/2.29$5, equivalent to $f/2.29$6 ADU at $f/2.29$7/ADU, an $f/2.29$8-band exposure is limited to 4 frames and therefore 8.58 s per exposure (Sumi et al., 20 Aug 2025).

These instrument choices define PRIME’s core capability. NIR-optimized optics from 830–1800 nm, Roman-class HgCdTe detectors, and a wide field together enable time-domain observations in a central-bulge regime inaccessible to optical microlensing surveys and difficult for wide-field $f/2.29$9-band surveys because bright NIR variables can saturate there (Sumi et al., 20 Aug 2025).

3. Alignment architecture and image-quality control

The alignment strategy was developed as a three-step procedure: preliminary alignment by a laser tracker, fine alignment by intra- and extra-focal image analysis, and complementary and fine alignment by the Hartmann test. The telescope was installed at the SAAO Sutherland site in August 2022, PRIME-Cam was mounted at prime focus in October 2022, and the on-site alignment campaign demonstrated that the optical system satisfied the operational requirement (Yama et al., 2023).

Operational tolerances were derived from optical simulation. The key tolerances are PFU tip/tilt $\mu$0 arcsec and PFU decenter $\mu$1m; these correspond to an 80% encircled-energy radius $\mu$2 arcsec, allowing a 15% degradation relative to the nominal design target of $\mu$3 arcsec. Under these tolerances, the worst-case simulated Hartmann constant in $\mu$4 band, averaged over the inner field, is 0.314 arcsec. The dominant alignment-sensitive aberrations are coma and astigmatism, and because the field is wide, minimizing global astigmatism and preserving PSF uniformity across the focal plane is essential for crowded-field survey photometry (Yama et al., 2023).

The preliminary alignment used a FARO Vantage laser tracker to establish the primary optical axis and the Prime Focus Unit axis through fiducial measurements and iterative PFU decenter and tip/tilt adjustment. In the Japan laboratory test in 2020, this achieved PFU decenter 38.6 $\mu$5m and PFU tip/tilt 2.0 arcsec. On site in August 2022, the corresponding values were 46.4 $\mu$6m and 12.4 arcsec, with a post-FARO $\mu$7-band Hartmann constant of 0.57 arcsec (Yama et al., 2023).

Fine alignment was then pursued with two complementary methods. The $\mu$8-band intra- and extra-focal procedure used three CMOS sensors and a basis-image fitting framework related to curvature sensing and the Transport of Intensity Equation,

$\mu$9

with wavefront expanded as

$\sim 1.4$0

The method worked in the Japan test but did not converge on site because of software issues. The team therefore switched to a pseudo Hartmann test in $\sim 1.4$1 band using “Yama-Cam,” a compact InGaAs test camera. On site, the $\sim 1.4$2-band pseudo Hartmann improved the Hartmann constant from 0.92 arcsec to 0.44 arcsec. The $\sim 1.4$3-band pseudo Hartmann then improved the center Hartmann constant from 0.509 arcsec to 0.262 arcsec, with an inner-field average of 0.295 arcsec; after subsequent mechanical work and re-alignment, the inner-field average was 0.322 arcsec. Although 0.322 arcsec is marginally above the simulated 0.314 arcsec worst-case requirement, the authors noted that the nominal model excluded several systematics and concluded that the telescope satisfies the operational tolerance (Yama et al., 2023).

4. Survey footprint, cadence, and observatory operations

The Galactic bulge is observable from February to October, and approximately 50% of yearly telescope time is devoted to Galactic bulge time-domain surveying; the remaining 50% is shared among partner institutions for other programs, including Targets of Opportunity. Installation was completed in July 2022, first observations occurred in mid-2023, and the dedicated bulge survey started in February 2024 (Sumi et al., 20 Aug 2025).

The survey targets 35 primary Galactic bulge fields spanning approximately $\sim 1.4$4 and $\sim 1.4$5, plus 9 low-extinction outer-bulge fields overlapping MOA fields. The field design explicitly overlays Roman Galactic Bulge Time Domain Survey candidate regions for synergy and field selection. Sgr A* lies in GB94 (Sumi et al., 20 Aug 2025).

From February 2024 through June 1, 2025, primary operations were carried out in the $\sim 1.4$6 band. Each visit used six exposures with a 10 arcsec dithering radius; each exposure comprised four SUTR frames, of which frames 1–3 were used after bias removal from frame 0, giving an effective exposure per visit of $\sim 1.4$7 s. Including overhead, each visit took about 140 s. A full cycle through 35 fields took about 82 minutes, corresponding to roughly 4–9 visits per field per night. Prior to August 2024, the survey used 12 exposures per visit, with effective exposure 103 s, no dithering, and about 164-minute cadence. $\sim 1.4$8-band observations were obtained at about 3-night cadence to measure source colors $\sim 1.4$9 for angular-radius estimates, and the nine low-extinction outer-bulge fields were observed in $4096 \times 4096$0 with about 3-day cadence to measure $4096 \times 4096$1-band source magnitudes (Sumi et al., 20 Aug 2025).

Since June 2, 2025, the strategy has been revised toward higher-cadence monitoring concentrated along the Galactic plane to enhance exoplanet detection efficiency, described as analogous to MOA’s strategy. The cadence groups are 21.6 min for GB76, 77, 78, 94, 110, and 111; 43.1 min for GB75, 79, 109, and 112; 86.2 min for GB60, 92, 93, 95, and 113; and 1 day for GB74, 80, 91, 96, 97, 108, and 114 (Sumi et al., 20 Aug 2025).

Operationally, PRIME is also a high-volume time-domain facility. Raw 16-bit frames are 34.7 MB per chip and compress to 23–28 MB per chip; up-the-ramp-fitted 32-bit “slope” images are 67.1 MB per chip, and the annual data volume is about 120 TB. Data are compressed on site and transferred to servers in Japan. During off-bulge seasons, PRIME conducts a wide-area $4096 \times 4096$2-band all-sky grid survey; as of May 7, 2025, about 13,000 deg$4096 \times 4096$3 had been observed, corresponding to about 40.8% of the sky accessible to PRIME. The facility is also used for ToO observations of gravitational-wave events, gamma-ray bursts, and other transients (Sumi et al., 20 Aug 2025).

5. Reduction pipeline, alerting, and first detections

The reduction chain includes super-bias subtraction, reference-pixel correction, non-linearity correction, up-the-ramp fitting to produce “slope” images, and flat-fielding from nightly twilight flats. The science extraction is based on Difference Image Analysis using reference images built from stacks of 12 exposures, or 103 s total per field. The reported DIA implementation runs within about 5 minutes of acquisition. New events are identified in differenced images, light curves are generated, point-source point-lens fits are applied, and anomalies such as binary, planetary, or finite-source deviations are flagged (Sumi et al., 20 Aug 2025).

This processing chain is directly coupled to an alert system. PRIME issues real-time alerts for microlensing candidates to support exoplanet searches and bulge chemical-evolution studies, and preliminary candidate lists for the 2024 season and the 2025 season through June 1 were made publicly available (Sumi et al., 20 Aug 2025).

The first results cover data from February 2024 through June 1, 2025. In that interval, PRIME identified 486 microlensing candidates, of which 284 were found in the 2024 season and 202 up to June 1, 2025. In addition, the survey detected more than a thousand variable stars, including Miras. The spatial distribution of the candidates indicates relatively higher event rates within Roman’s candidate central fields. The survey also demonstrated successful imaging of dense stellar fields such as GB110 with 8.58 s exposure in $4096 \times 4096$4, which the authors present as evidence of PRIME’s crowding-handling capability with DIA (Sumi et al., 20 Aug 2025).

Among the detected microlensing candidates, 41 show deviations from the point-source point-lens model. An example binary event is PRIME-2024-BLG-066 in GB92, with best-fit parameters $4096 \times 4096$5, $4096 \times 4096$6, and $4096 \times 4096$7 d. Example PSPL events include PRIME-2024-BLG-124 in GB75, with $4096 \times 4096$8 d and $4096 \times 4096$9, and PRIME-2024-BLG-170 in GB78, with $\mu$0 d and $\mu$1 (Sumi et al., 20 Aug 2025).

The variable-star component is not ancillary in a trivial sense. The NIR time series detect pulsators in central, highly reddened regions that are too faint for 2MASS and too bright for VVV, while multi-band $\mu$2 data enable variability detection and extinction characterization. This suggests a dual role for PRIME: a microlensing survey and a bulge stellar-population survey in a wavelength and latitude regime with sparse prior time-domain coverage (Sumi et al., 20 Aug 2025).

6. Microlensing formalism, Roman parallax, and anticipated yield

For a single point lens, the light-curve model used in the survey description is the Paczyński magnification curve,

$\mu$3

with

$\mu$4

Here $\mu$5 is the impact parameter, $\mu$6 is the time of peak, and $\mu$7 is the Einstein timescale. The Einstein radius satisfies

$\mu$8

where $\mu$9 and $\mu$0, while $\mu$1. Without additional constraints, $\mu$2, $\mu$3, and $\mu$4 remain degenerate. Space-based microlens parallax is intended to break part of that degeneracy:

$\mu$5

For Earth-space observations, the parallax vector can be approximated as

$\mu$6

If finite-source effects yield $\mu$7 and the source angular radius $\mu$8 is estimated from PRIME’s NIR colors such as $\mu$9, then $\times$0 and the lens mass follows directly (Sumi et al., 20 Aug 2025).

This Roman connection is structurally central to PRIME rather than auxiliary. Roman plans six 70.5-day seasons at 12.1-minute cadence, totaling 438 days over its 5-year prime mission, plus four additional lower-cadence seasons, and is expected to detect about 27,000 microlensing events, about 1,400 exoplanets, including about 200 with masses below $\times$1, and about 1,000 free-floating planets. Concurrent PRIME–Roman observations are intended to capture the subtle Earth-space light-curve differences required for space-based parallax, making possible lens mass and distance determinations, including for free-floating planets down to Earth mass. PRIME also supplies longer-season monitoring and NIR color information when Roman cannot observe (Sumi et al., 20 Aug 2025).

Before survey operations, end-to-end simulations were used to estimate likely exoplanet yield and optimize cadence. Assuming the Cassan et al. (2012) mass function as modified by Penny et al. (2019), the predicted yield was 42–52 planets per year depending on strategy, including 1–2 planets with $\times$2, 22–25 planets with $\times$3, and 19–25 planets with $\times$4. Four strategies were studied, from a 6-field 16-minute survey to 18-field wider-footprint strategies including a hybrid cadence. The simulation emphasized the cadence-footprint trade-off: low-mass planets favor higher cadence, while giant-planet yields increase with larger footprint (Kondo et al., 2023).

PRIME therefore occupies a specific place in microlensing survey design. It complements optical outer-bulge surveys, rather than replacing them; it extends NIR time-domain coverage into the central bulge; and it is configured to provide the ground-based half of an Earth-space parallax baseline with Roman. A plausible implication is that its long-term significance will depend not only on raw event counts, but also on how effectively those counts are converted into lens-mass measurements, inner-bulge event-rate maps, and NIR-selected stellar samples in regions where previous survey coverage has been sparse (Sumi et al., 20 Aug 2025).