Subaru Seminars

MANIFEST (Many-Instrument Fiber System) will be one of the groundbreaking instrument on the Giant Magellan Telescope (GMT), empowering astronomers to study the universe in incredible detail. This advanced fiber-optic positioning system allows the GMT to connect with multiple instruments and capture light from multiple objects simultaneously across the vast field of view of the GMT. For Australian astronomy, MANIFEST opens new opportunities to explore distant galaxies and stars with unmatched precision and resolution. This talk will delve into MANIFEST innovative design and capabilities, its integration with the GMT and implications for advancing cosmology, stellar evolution, and our quest to understand the galaxy evolution.

The U.S. National Science Foundation’s Daniel K. Inouye Solar Telescope (DKIST) is a four-meter solar telescope on the island of Maui, Hawai’i. It is the largest optical solar telescope in the world. The telescope was designed to measure the vector magnetic field not only in the solar photosphere and chromosphere with the highest spatial resolution, but also in the very faint solar corona. Four focal instruments are currently operational and another instrument is under implementation. Each has unique imaging and spectro-polarimetric capabilities covering a broad wavelength range. In this seminar, I provide a brief overview of key technologies for the solar telescope. In addition, I will show the highest-spatial-resolution solar images, a spatial distribution of measured coronal magnetic field, and measured electric fields, which has an unexpectedly large spatial scale.

The emergence of deep learning has significantly accelerated the advancement of AI. In astronomy, AI is now actively used in areas such as object classification and surrogate modeling, and is increasingly recognized as a fourth paradigm of science, complementing experiment, theory, and numerical simulation. From a technological perspective, image recognition and natural language processing align strongly with societal needs. The rapid development and deployment of innovations such as ChatGPT exemplify this accelerating trend. In this talk, I will cover the foundations of state-of-the-art models, with a focus on image recognition, a domain closely connected to astronomy, and I will present our recent R&D results. The base methodologies are not limited to specific application domains, making them broadly applicable and beneficial to the astronomy community. I will also discuss the relationship between astronomy and machine learning. While there are differences in academic culture between science and engineering, the underlying disciplines of physics and information science are deeply interconnected and mutually reinforcing. By highlighting recent astronomical research that incorporates advanced AI technologies, this talk aims to offer new perspectives and practical directions for future exploration in astronomy.

Wow, I can’t imagine when our telescope will be able to capture such magnificent images of the Orion nebula!” Many compliments to the SUBARU Telescope were featured in American newspapers published on January 29, 1999, just one day after the NAOJ’s press release about the first light. The released celestial images were taken by a near infrared camera, positioned at the prime focus, representing a single-piece primary mirror measuring 8.3-meters (27.2 feet) - the world largest on completion. Around the same time, the European Southern Observatory and the International Gemini Telescope Project also constructed 8-meter class telescopes, the former in Chile and the latter in Hawaii. In essence, these three competing projects can be regarded as the Olympic games of giant telescopes, as they battled for the best image and light collection power in the astronomy history, thereby enabling us to explore the most distant corners of the universe and unveil unknown celestial phenomena. In this short seminar, Dr. Izumi MIKAMI, former project engineer and manager at Mitsubishi Electric, presents how the SUBARU project team overcame the numerous challenges they encountered during the development of crucial elements, design, and progress control, based on his real experience from 1987 to 2020.

Galaxy clusters are key to probing environmental effects on galaxy evolution, yet their rarity makes clusters at z~2 scarce. In this talk, we report the discovery of seven cluster candidates with massive quiescent galaxies at z~2 across 3.5 deg² in the XMM-LSS field. If confirmed, this would roughly double the known sample at this frontier redshift. We construct photometric redshifts from deep, multi-wavelength imaging spanning from u* to K (i ~26, Ks ~24), compiled from HSC-SSP and collaborative/public surveys. We then map the number density of massive (log(M*/Msun)>10.5), quiescent (log sSFR [yr−¹]<-10) galaxies at z~2, identifying seven prominent overdensities. All candidates exhibit clear red sequences in z-H vs. H color–magnitude diagrams. One also shows extended X-ray emission, indicating its virialized nature. We will also present the physical properties of these candidates and outline prospects for spectroscopic follow-up with Subaru.

Recently, several studies analyzing the cosmic microwave background (CMB) have reported isotropic cosmic birefringence—the rotation of the polarization planes of photons (e.g., Minami & Komatsu 2020). Since cosmic birefringence violates parity symmetry, the detection of such a signal, if confirmed, would strongly suggest new physics beyond the standard framework of physics. In particular, axion-like particles (ALPs) have attracted attention as a plausible origin (e.g., Fujita et al. 2021). However, current analyses of cosmic birefringence using the CMB alone still face several concerns. To overcome the problems, it is important to establish methods for independent tests of the signal. In this seminar, I will introduce the methods using astrophysical sources, and also discuss how optical telescopes can play a key role in such projects (e.g., Carroll 1990, Naokawa 2025).

The evolution of galaxies is closely linked to the gas reservoirs surrounding them, known as the circumgalactic medium (CGM). Gas and metals are exchanged through inflows and outflows within the CGM, which serves as an interface between the interstellar medium (ISM) and the intergalactic medium (IGM). Optical integral-field units (IFUs) enable the individual detection of the CGM in emission, allowing us to map the gas and metals around host galaxies. In this talk, I will review recent advances in the observation and understanding of hydrogen gas and metals around galaxies at cosmic noon, particularly as extended Lyα emission (Lyα halos) and metal-line halos. I will also discuss the connection between CGM observations in emission and absorption, and compare these findings with simulations. Furthermore, I will present the structures and kinematics of the ionized CGM and galactic-scale outflows around dwarf starburst galaxies at z=0.

Hot Jupiters are a peculiar class of gas giant exoplanets orbiting extremely close to their host stars (P < 10 days), raising longstanding questions about their formation and inward migration. In-situ formation is widely considered implausible due to local disk conditions, suggesting instead that these planets formed at larger orbital distances and migrated inward. Two primary mechanisms have been proposed: (1) disk migration through planet-disk interactions, and (2) high-eccentricity migration (HEM), in which gravitational perturbations from a stellar or planetary companion excite large orbital eccentricities, followed by tidal circularization near periastron.

In this talk, I will present our two recent efforts that leverage tidal interactions to shed light on hot Jupiter migration pathways. First, we empirically constrain the efficiency of star–planet tidal dissipation and demonstrate that it is insufficient to circularize some hot Jupiters currently found on relatively wide orbits (4-10 days), challenging the universality of HEM. Second, we monitor orbital decay in the shortest-period hot Jupiters (1-2 days), aiming to directly measure ongoing tidal evolution. Continued timing measurements will allow us to reverse-engineer the end point of migration—whether it aligns with the inner edge of the protoplanetary disk or the pericenter of a once highly eccentric orbit.

Together, these tidal constraints offer new insight into the dynamical histories of hot Jupiters and the physics of planetary migration.

Redshift is a fundamental observable that allows us to access the three-dimensional information on objects. Efficiently deriving accurate redshift measurements is an important task for the large photometric and spectroscopic survey era. We develop a new Python package, RVSNUpy, to estimate spectroscopic redshift based on the cross-correlation technique. Based on the extensive test using synthetic and observed spectra, we demonstrate that RVSNUpy yields robust redshift measurements. RVSNUpy can be an efficient tool for deriving redshifts from large-scale spectroscopy, including Subaru/PFS. In addition to the development of RVSNUpy, we also built the photometric redshift estimation tool based on machine learning (ML), particularly based on the random forest model. The combination of these photometric and spectroscopic redshift measurements will prepare us for studying extragalactic science and cosmology in the large survey era.

In the hierarchical structure formation scenario predicted by the ΛCDM model, large spiral galaxies such as the Milky Way are thought to have been formed by the accretion and merger of smaller stellar systems. As these accreted systems are deposited into the stellar halo, the halo serves as a crucial observational region for unraveling a galaxy’s accretion history. The stellar halo of the Andromeda galaxy (M31) presents an ideal target for studying galaxy formation, owing to its proximity and the ability to observe its entire halo structure. However, because M31 is located at low Galactic latitude, its halo stars (red giant branch stars; RGB stars) are heavily contaminated by foreground stars (dwarf stars), making it challenging to resolve the detailed halo structure accurately. To overcome this problem, we investigate the structures of M31’s stellar halo using data obtained with Subaru/Hyper Suprime-Cam/NB515. The NB515 filter enables the separation of M31 RGB stars from foreground Galactic dwarf stars with an accuracy of approximately 90％. By analyzing the radial profile of the extracted RGB stars, we examine the properties of M31’s outer halo out to a projected radius of 300 kpc. Our key findings are as follows, with further details to be presented in this talk:
1. A single power-law profile well describes the stellar halo of M31 out to 100 kpc, and no clear transition between an inner and outer halo, as observed in the Milky Way, is found.
2. The halo structure up to 150 kpc is well approximated by a power-law with an index of ∼ −2, which is shallower than the typical power-law index of ∼ −3 estimated for many nearby spiral galaxies.
3. A break in the radial profile is identified at ∼150 kpc, a location comparable to that observed in the Milky Way.

Lyman Alpha Emitting galaxies (LAEs) are classically understood as young, low-mass, star-forming galaxies with low line-of-sight dust content. Due to a lack of significant absorption from dust, we observe strong Lya emission at a rest-frame wavelength of 121.6nm. This feature allows LAEs to be highly identifiable and to serve as profound beacons of the high-redshift universe. Because of this, LAEs are often used for large-scale statistical analyses in cosmology, galaxy formation, and galaxy evolution, however, there is still much we do not understand about their physical nature. The ODIN program uses three DECam narrowband filters to discover LAEs at z ~ 2, 3, and 4. In Firestone et al. (2024), we introduced improved techniques for narrowband LAE selection and interloper rejection. We presented stacked SEDs and Lya equivalent width (EW) distributions. High EW LAEs hinted that LAEs may have nontrivial radiative transfer and/or complex ISM conditions. In Firestone et al. (2025), we used ODIN LAEs and UVCANDELS photometry to test the conventional assumption that LAEs are experiencing their first major burst of star formation at the time of observation. We accomplished this using the Gaussian process-based non-parametric star formation history (SFH) reconstruction method, Dense Basis. We found that a strong majority (67%) of our LAE SFHs align with the conventional archetype of a first major SF burst. However, 28% exhibit earlier bursts of SF despite the ongoing burst having the highest SFR, and the final 5% experienced their highest SFR in the past. Since LAE stellar mass assembly is more complex than we thought and some LAEs have significant older stellar populations, this again begged the question of what radiative transfer mechanisms make Lya in LAEs easily detectable. Subsequently, we begin to characterize Lya radiative transfer by examining the relationship between dust extinction and EW. These results reveal that radiative transfer may be driven by gas velocities and support the hypothesis of a clumpy ISM. Overall, our results suggest that several evolutionary paths and radiative transfer mechanisms can produce galaxies with strong observed Lya emission.

The Advanced Instrumentation and Technology Centre (AITC), located at the Australian National University’s Research School of Astronomy and Astrophysics in Canberra, brings together engineers and scientists from Australia and abroad who are working at the forefront of astronomical and space instrumentation. In this talk, Dr. Martinez will provide an overview of AITC's main activities within the laser guide star adaptive optics (LGS-AO) program. The AITC LGS-AO group develops systems for astronomy, laser communications, and space situational awareness applications. Dr. Martinez will discuss current projects, including ANU's contribution to the ULTIMATE Ground Layer Adaptive Optics system for the Subaru Telescope, and the development of the Laser Tomography Adaptive Optics system for the 25-meter Giant Magellan Telescope, to be built in Las Campanas Observatory, Chile. Dr Martinez will also describe the R&D activities the LGS-AO group is pursuing, with focus on the tip-tilt indetermination challenge in Laser Guide Stars and novel wavefront sensing techniques. With a rapidly growing portfolio of projects, the speaker would like to use this opportunity to discuss potential areas of collaboration between ANU and Hawaii-based researchers. PhD and postdoc opportunities at ANU will also be presented.

Nebular emission provides a unique window into the physical properties of high-redshift galaxies. Leveraging JWST observations, we investigate three key aspects: hydrogen line anomalies, UV continuum slopes, and helium emission in galaxies at z > 6, revealing insights into their ionized regions and chemical enrichment.
Hydrogen Line Anomalies: We examine GS9422 and RXCJ2248 at z ∼ 6, which exhibit anomalous Balmer decrements inconsistent with Case B recombination and standard dust extinction. Photoionization models suggest two possible origins: density-bounded nebulae opaque only up to the Lyman-8 line or ionization-bounded nebulae enveloped by optically thick excited H I clouds.
UV Continuum Slopes: While ionization-bounded nebulae have been observed, density-bounded counterparts remain elusive. To identify potential density-bounded nebulae, we analyze UV slopes β of 974 galaxies at z = 4–14 using JWST archival spectra taken from major JWST GTO, ERS, and GO programs including JADES, CEERS, and UNCOVER. Among them, we find a galaxy at z=9.25, dubbed EBG-1, which shows an exceptionally blue UV slope of β = −2.99 ± 0.15. This extreme slope, together with a weak [OIII]5007 emission, indicates a high ionizing photon escape fraction (fesc > 0.5), suggesting density-bounded conditions.
Helium Emission and Abundance: Combining insights, we study He I emission in GS9422, RXCJ2248, and GLASS150008 at z~6, mostly with the spectroscopic coverage of rest-frame optical He I emission lines and hydrogen Balmer lines. We find that these high-z galaxies present strong He I emission compared to those of local dwarf galaxies. Photoionization model reveal that these high-z galaxies may have elevated He/H > 0.10, far exceeding local values. A positive correlation between He/H and N/O implies chemical enrichment dominated by CNO-cycle processing rather than standard core-collapse supernovae.
These findings provide crucial insights into the nebular environments and chemical evolution of early galaxies.

The primordial He abundance (Yp) is best determined by observations of metal-poor galaxies, while there are only a few known extremely metal-poor (< 0.1Z⊙) galaxies (EMPGs) having reliable He/H measurements. We conducted deep Subaru NIR spectroscopy for 10 EMPGs and determined their He/H values. Adding pre-existing galaxies with reliable He/H estimates to our sample, we obtain Yp = 0.23790+/-0.0034, which is slightly (∼ 1σ) smaller than the previous values. With our Yp constrain, we constrain the degeneracy parameter of electron-neutrino ξe, the effective number of neutrino species Neff. Our constraints may suggest a existence lepton asymmetry and allow for a high value of Neff up to Neff ~ 3.4 within the 1σ level, which could mitigate the Hubble tension.
We are also conducting cosmological analysis to constrain sigma_8 at high redshifts. Using the internal S21A data release product taken in the Subaru/Hyper Suprime-Cam (HSC) Subaru Strategic Program survey, we construct a z~4 galaxy catalog which consists of ~ 2 million Lyman break galaxies (LBGs). With this catalog, we calculate the angular correlation functions of the LBGs. We also measure the CMB lensing convergence profile around the LBGs with lensing products from the public data release of the Planck mission, detecting the signal at ∼6 σ. By fitting models to the angular correlation functions and lensing signal, we preliminary put constraints on the sigma_8, reducing the degeneracy between the sigma_8 and the galaxy bias.

Phase distribution induced by biological samples carries various important physical parameters. In this presentation, quantitative phase imaging using the transport of intensity equation under commercially available confocal microscopy is presented.

Determining the precise distance to Andromeda galaxy (M31), our nearest galactic neighbor, is classic yet significant and challenging work for astronomy. M31 has a potential to be an anchor galaxy in the first rung of the local distance ladder (geometry -> cepheids, TRGB, ...), but a geometrical distance with enough precision has not yet been achieved. I introduce our project, which aims to achieve a 1% measurement of M31 distance using the detached eclipsing method with Subaru HSC.

Deep near-infrared spectroscopy from JWST allows us to characterize the kinematics of high-z galaxies. We study outflows in 130 galaxies with -16<M_UV<-22 at z=3-9 identified in JWST spectroscopic data taken by the CEERS, ERO, FRESCO, GLASS, and JADES programs. We identify 30 out of the 130 galaxies with broad components of FWHM~150-800 km/s in the emission lines of Hα and [OIII]λ5007 that trace ionized outflows. We find that the outflow velocities as a function of star-formation rate are comparable to or higher than those of galaxies at z~1. Interestingly, these outflow velocities are typically not high enough to escape from the galactic potentials, suggestive of fountain-type outflows.
Without being disrupted by powerful outflows, rotating disk may form and sustain in the early universe. We investigate the dynamics of GN-z11, a luminous galaxy at z = 10.60, by analyzing the public deep integral field spectroscopy (IFS) data taken with JWST NIRSpec IFU. While the observations of the IFS data originally targeted a He II clump near GN-z11, we find that C III]λλ1907,1909 emission from ionized gas at GN-z11 is bright and spatially extended significantly beyond the point-spread function (PSF). The spatially extended C III] emission of GN-z11 shows a velocity gradient, red- and blue-shifted components in the northern and southern directions, respectively. We perform forward modeling with GalPak3D and find that the best-fit model is a nearly edge-on disk with a kinematic ratio of vrot/σv = 3.3, indicative of a rotation-dominated disk at z = 10.6. While the velocity gradient is consistent with the rotating disk solution, we recognize that galactic outflows can also explain the velocity gradient. Higher S/N and resolution data are necessary to conclude the physical origin of the velocity gradient in GN-z11.

Lya damping wing absorption measurement is a useful tool to constrain the key cosmic reionization parameters including neutral hydrogen fraction xHI, but the measurement had been limited only to the small number of extremely UV-continuum bright sources such as QSO/GRBs. For the first time, we constrain xHI and ionized bubble radii Rb by measuring Lya damping wing absorptions using bright galaxy spectra detected in JWST observations. We combine JWST/NIRSpec spectra taken by CEERS, GO-1433, DDT-2750, and JADES programs, and obtain 27 bright (M_UV<-18.5) UV-continuum galaxies at 7 < z < 12. We construct 4 composite spectra binned by redshift and find the clear transition from sharp to soft spectral break towards high redshift at the restframe 1216 Å suggesting the increase of Lyα damping wing absorption. We estimate Lya damping wing absorption in the galaxy spectra with a state-of-art stellar population synthesis model including Lya emission and host HI absorptions. Assuming the standard inside-out reionization picture having an ionized bubble with Rb around a galaxy in the intergalactic medium of xHI, we obtain xHI (Rb) values monotonically increasing (decreasing) trend from xHI = 0.54 to 0.95 (log Rb = 1.89 to -0.68 comoving Mpc) at redshift 7.12 to 9.91. We also estimate xHI at z<7 using the largest Lya emitter (LAE) catalog from Subaru Hyper Suprime-Cam photometric survey constructed by Kikuta et al. (2023). The z>5.7 Subaru LAE catalog consists of 8182 galaxies, and we estimate xHI from Lya luminosity function and LAE clustering properties using a prediction based on semi-numerical simulation 21cmFAST. The redshift evolution of xHI between z=6 to 12 inferred from JWST and Subaru result indicate a very rapid reionization scenario progressing mainly at z=7-8.

I will overview what the first year of observations with the James Webb Space Telescope tells us about the early growth history of supermassive black holes at the centers of galaxies. Key highlights include discovery of Seyfert-class AGN at z>4, both unobscured and obscured, and the successful detection of starlight from host galaxies down to what has been impossible with HST observations. I will specifically focus on JWST follow-up observations of z=6 HSC-SSP quasars, where we aim to study the relative growth of host galaxies and their black holes, and explore the connection between AGN activity and massive galaxy formation.

Motivated by the quantum description of gauge theories, I will discuss the cosmological effects of relaxing the Hamiltonian and momentum constraints in general relativity and Gauss' law in electromagnetism. The unconstrained theories have new source terms that mimic a pressureless dust and a charge density that only follows geodesics. The source terms may be the simplest explanation for dark matter and generically predict a charged component. Discovery of such terms would rule out inflation and be a direct probe of the initial conditions of the universe.

A special seminar talk by our Director. Subaru's role in the endeavours will be described.

The nocturnal terrestrial atmosphere glows brightly (i-band continuum ~ 20-22 mag/arcsec^2), mainly through a highly variable forest of lines from molecular emission and a few strong atomic lines, known jointly as *airglow*. Understanding the temporal and spatial variations of this emission therefore represents one of the major areas of concern for current and upcoming deep spectrographic surveys, as we push several magnitudes below the intensity levels set by the airglow foreground. The variations take place in both the overall intensity of each emitting species as well as in other properties of the emission, such as the temperature of the molecular emission. These variations have been observed with PFS and SuNSS, and clear trends have been identified and will be presented. However, the airglow is not just a source of foreground, it is also useful, since it provides a ubiquitous reference set of thousands of lines which is used for wavelength calibration. I will show how to improve airglow-based wavelength calibration by explicitly considering the temporal variability of each line, and present a new code which can produce robust linelists for wavelength calibration at arbitrary spectral resolution.

Low-mass galaxies at high redshifts are the building blocks of more massive galaxies at later times and are thus key populations for understanding galaxy formation and evolution. We have made deep narrow-band observations for two protoclusters and the general field in COSMOS at z~2. We find low-mass galaxies in a protocluster at z~2 show enhanced star-forming activity, deviating from the main sequence, indicating environmental effects. Moreover, within protocluster galaxies, star-forming activity is more centrally concentrated. This is probably due to the enhanced interactions between galaxies in protoclusters at cosmic noon, which can lose the angular momentum of the gas, drive it towards the galaxy center, and lead to a central starburst.

Identifying when/where massive galaxies first appeared in the Universe and knowing how the star-forming activities of those galaxies were quenched is critical for understanding galaxy formation and early evolution. It can also put strong constraints on the efficiency of galaxy formation in the early universe and, eventually, on the current standard theory of hierarchical structure formation. Recent observations have confirmed the existence of the massive galaxies (~10^11Msun) at z~4.6. In this study, we aim to go further back in time and search for protoclusters at z~5 to discover mature, massive galaxies of comparable mass that have already quenched star-forming activities. To search for such galaxies, we are conducting the Ruby-Rush project, which targets the primordial cluster candidate regions where many Lyman break galaxies are hosted, as found by the Gold-Rush project (Toshikawa et al. 2018). Quiescent galaxies have a strong Balmer break in their spectra. Using the near-infrared instrument, SWIMS on the Subaru, we have discovered massive quiescent galaxy candidates at z∼5 by neatly bracketing the break feature with two medium-band filters (K2 and K3). As a result, we have successfully identified several robust candidates. In this talk, I will report our discoveries of strong candidates of massive quiescent galaxies at z~5 and discuss how we can utilize medium band filters on Subaru to study the accelerated massive galaxy formation in the high-density regions only a billion years after the Big Bang.

Massive quiescent galaxies at z ~ 4 are likely progenitors of massive elliptical galaxies in the local Universe and have been the subject of many recent papers to constrain the physical drivers to form such massive galaxies and then stop their star formation activities in a short timescale. To date, only a few massive quiescent galaxies at z ~ 4 have been spectroscopically confirmed, and the detailed physics of their quenching mechanisms is still unclear. In particular, the role of surrounding environments for quenching at such a high redshift has remained totally unexplored. In this presentation, we report on the spectroscopic confirmation of a massive quiescent galaxy at z = 4.53 in the COSMOS field (Kakimoto et al. 2024, ApJ, 963, 49). The follow-up spectroscopy with Keck/MOSFIRE in the z-band reveals the Balmer break, indicative of evolved stellar populations. Its star formation history suggests that this galaxy experienced rapid quenching from z ~ 5. The galaxy is among the youngest quiescent galaxies confirmed so far at z > 3 with z_form ~ 5.2 (200 Myr ago) and is likely a galaxy in the process of being quenched. A unique aspect of the galaxy is that it is in an extremely dense region; there are four massive star-forming galaxies at 4.4 < z < 4.7 located within 150 physical kpc from the galaxy (assuming all galaxies are located at z=4.53). Interestingly, three of them have strongly overlapping virial radii with that of the central quiescent galaxy (~ 70 kpc), suggesting that the over-density region is likely the highest redshift candidate of a dense group with a spectroscopically confirmed quiescent galaxy at the center.
Our next observation is crucial to spectroscopically confirm this group with Subaru/FOCAS. This provides us with a unique opportunity to gain insights into the role of the group environment for quenching at z ~ 5, which corresponds to the formation epoch of massive elliptical galaxies in the local Universe.

As my term at Subaru draws to an end in a few months, in this seminar I will take the opportunity to talk about my various activities at the observatory over the years and summarise research results focused on characterising the stellar populations in early-type galaxies and their evolution over cosmic time. I will also talk about my instrument related work as part of the HSC queue and ULTIMATE-Subaru teams, in addition to several other roles associated with observatory functions.
Finally, I would like to highlight the amazing people I have been fortunate to work with and learn from at Subaru and some memorable events.

We are now coming to the final phase of the commissioning for the Subaru Prime Focus Spectrograph. For its successful science operation, high-quality sky subtraction is of crucial importance in order to accurately extract weak galaxy signals. To achieve that, we are developing modeling algorithms to determine the 2D point spread function (PSF) on the spectrograph detectors. The light coming into the detectors is affected by various components of the instrument before being observed as the final PSF. We are currently focusing mostly on the optical aberrations in the spectrographs, and developing optical models using Zernike polynomials. We exploit dedicatedly defocused images, where the light is widely spread on the detector pixels, which allows us to obtain more information about the optical system. Comparing them with the commissioning data and fitting the model parameters, we have achieved a 1% accuracy in 1d spectra for some part of the detector. In this talk, I will present these efforts and the current status of this PSF modeling.