arXiv subjects

Junhan Kim

Publications and source records attributed to Junhan Kim.

At least 19 recordsLinked to original sources

arXiv2026

CHEX-MATE: AMALGAM weak-lensing analysis of 41 Planck Sunyaev-Zel'dovich-selected galaxy clusters

We present a weak-lensing shear analysis of 41 Planck SZ-selected galaxy clusters at $0.11\le z\le 0.55$ from the CHEX-MATE sample, using wide-field Subaru/Suprime-Cam and CFHT/MegaPrime imaging from the AMALGAM project. We detect the azimuthally averaged weak-lensing signal around the X-ray peak of each cluster, achieving a median S/N of 6.5 per cluster. The $45^\circ$-rotated component has a median S/N of -0.1 and ranges from -1.8 to +1.8, consistent with zero. We model the excess surface mass density profile of each cluster with an NFW profile to infer weak-lensing mass and concentration constraints. The total systematic uncertainty in the weak-lensing mass calibration is assessed to be $8\%$. Using a hierarchical Bayesian framework, we then derive weak-lensing-calibrated scaling relations for the halo concentration, $c_{200}$, as a function of $M_{200}$ and redshift, and for the Planck SZ mass proxy, $M_{SZ}$, as a function of $M_{500}$ and redshift, while accounting for sample selection effects, weak-lensing modelling biases, and residual calibration uncertainty. At $M_{200}=10^{15}M_\odot$ and $z=0.25$, we find $c_{200}=3.53\pm0.71$ with an intrinsic scatter of $0.22\pm0.04$ dex. The inferred normalisation and scatter are consistent with recent $\Lambda$CDM predictions for massive haloes, with no significant mass or redshift dependence over the probed range. For the Planck mass proxy, our baseline regression yields $M_{SZ}/M_{500}=0.83\pm0.09$ at $M_{500}=7\times10^{14}M_\odot$ and $z=0.25$, with an intrinsic scatter of $0.10\pm0.02$ dex. A restricted model with fixed unit mass slope and no redshift evolution gives $1-b=0.72\pm0.11$. We also provide weak-lensing-calibrated posterior estimates of $M_{500}$ for the sample based on the baseline $M_{SZ}$--$M_{500}$--$z$ relation. These results provide an initial weak-lensing mass calibration for CHEX-MATE multi-probe cluster studies.

astro-ph.CO↗

arXiv2026

On the limits and opportunities of AI reviewers: Reviewing the reviews of Nature-family papers with 45 expert scientists

With the advancement of AI capabilities, AI reviewers are beginning to be deployed in scientific peer review, yet their capability and credibility remain in question: many scientists simply view them as probabilistic systems without the expertise to evaluate research, while other researchers are more optimistic about their readiness without concrete evidence. Understanding what AI reviewers do well, where they fall short, and what challenges remain is essential. However, existing evaluations of AI reviewers have focused on whether their verdicts match human verdicts (e.g., score alignment, acceptance prediction), which is insufficient to characterize their capabilities and limits. In this paper, we close this gap through a large-scale expert annotation study, in which 45 domain scientists in Physical, Biological, and Health Sciences spent 469 hours rating 2,960 individual criticisms (each targeting one specific aspect of a paper) from human-written and AI-generated reviews of 82 Nature-family papers on correctness, significance, and sufficiency of evidence. On a composite of all three dimensions, a reviewing agent powered by GPT-5.2 scores above each paper's top-rated human reviewer (60.0% vs. 48.2%, p = 0.009), while all three AI reviewers (including Gemini 3.0 Pro and Claude Opus 4.5) exceed the lowest-rated human across every dimension. AI reviewers' accurate criticisms are also more often rated significant and well-evidenced, and surface a distinct 26% of issues no human raises. However, AI reviewers overlap far more than humans do (21% vs. 3% for cross-reviewer pairs), and exhibit 16 recurring weaknesses humans do not share, such as limited subfield knowledge, lack of long context management over multiple files, and overly critical stance on minor issues. Overall, our results position current AI reviewers as complements to, not substitutes for, human reviewers.

cs.CL↗

arXiv2026

LookaheadKV: Fast and Accurate KV Cache Eviction by Glimpsing into the Future without Generation

Transformer-based large language models (LLMs) rely on key-value (KV) caching to avoid redundant computation during autoregressive inference. While this mechanism greatly improves efficiency, the cache size grows linearly with the input sequence length, quickly becoming a bottleneck for long-context tasks. Existing solutions mitigate this problem by evicting prompt KV that are deemed unimportant, guided by estimated importance scores. Notably, a recent line of work proposes to improve eviction quality by "glimpsing into the future", in which a draft generator produces a surrogate future response approximating the target model's true response, and this surrogate is subsequently used to estimate the importance of cached KV more accurately. However, these approaches rely on computationally expensive draft generation, which introduces substantial prefilling overhead and limits their practicality in real-world deployment. To address this challenge, we propose LookaheadKV, a lightweight eviction framework that leverages the strength of surrogate future response without requiring explicit draft generation. LookaheadKV augments transformer layers with parameter-efficient modules trained to predict true importance scores with high accuracy. Our design ensures negligible runtime overhead comparable to existing inexpensive heuristics, while achieving accuracy superior to more costly approximation methods. Extensive experiments on long-context understanding benchmarks, across a wide range of models, demonstrate that our method not only outperforms recent competitive baselines in various long-context understanding tasks, but also reduces the eviction cost by up to 14.5x, leading to significantly faster time-to-first-token. Our code is available at https://github.com/SamsungLabs/LookaheadKV.

cs.LG↗

arXiv2026

On the Importance of a Multi-Scale Calibration for Quantization

Post-training quantization (PTQ) is a cornerstone for efficiently deploying large language models (LLMs), where a small calibration set critically affects quantization performance. However, conventional practices rely on random sequences of fixed length, overlooking the variable-length nature of LLM inputs. Input length directly influences the activation distribution and, consequently, the weight importance captured by the Hessian, which in turn affects quantization outcomes. As a result, Hessian estimates derived from fixed-length calibration may fail to represent the true importance of weights across diverse input scenarios. We propose MaCa (Matryoshka Calibration), a simple yet effective method for length-aware Hessian construction. MaCa (i) incorporates multi-scale sequence length information into Hessian estimation and (ii) regularizes each sequence as an independent sample, yielding a more stable and fruitful Hessian for accurate quantization. Experiments on state-of-the-art LLMs (e.g., Qwen3, Gemma3, LLaMA3) demonstrate that MaCa consistently improves accuracy under low bit quantization, offering a lightweight enhancement compatible with existing PTQ frameworks. To the best of our knowledge, this is the first work to systematically highlight the role of multi-scale calibration in LLM quantization.

cs.LG↗

arXiv2026

TurboBoA: Faster and Exact Attention-aware Quantization without Backpropagation

The rapid growth of large language models (LLMs) has heightened the importance of post-training quantization (PTQ) for reducing memory and computation costs. Among PTQ methods, GPTQ has gained significant attention for its efficiency, enabling billion-scale LLMs to be quantized within a few GPU hours. However, GPTQ's assumption of layer-wise independence leads to severe accuracy drops in low-bit regimes. Recently, BoA improved upon GPTQ by incorporating inter-layer dependencies within attention modules, but its reliance on sequential quantization across all out-channels makes it substantially less efficient. In this paper, we propose TurboBoA, a new backpropagation-free PTQ algorithm that preserves the accuracy benefits of BoA while significantly accelerating the process. The proposed TurboBoA introduces three key innovations: (i) joint quantization of multiple out-channels with a closed-form error compensation rule, which reduces sequential bottlenecks and yields more than a three-fold speedup; (ii) a correction mechanism for errors propagated from preceding quantized layers; and (iii) adaptive grid computation with coordinate descent refinement to maintain alignment during iterative updates. Extensive experiments demonstrate that TurboBoA delivers substantial acceleration over BoA while consistently improving accuracy. When combined with outlier suppression techniques, it achieves state-of-the-art results in both weight-only and weight-activation quantization. The code will be available at https://github.com/SamsungLabs/TurboBoA.

cs.LG↗

arXiv2026

Two-Stage Grid Optimization for Group-wise Quantization of LLMs

Group-wise quantization is an effective strategy for mitigating accuracy degradation in low-bit quantization of large language models (LLMs). Among existing methods, GPTQ has been widely adopted due to its efficiency; however, it neglects input statistics and inter-group correlations when determining group scales, leading to a mismatch with its goal of minimizing layer-wise reconstruction loss. In this work, we propose a two-stage optimization framework for group scales that explicitly minimizes the layer-wise reconstruction loss. In the first stage, performed prior to GPTQ, we initialize each group scale to minimize the group-wise reconstruction loss, thereby incorporating input statistics. In the second stage, we freeze the integer weights obtained via GPTQ and refine the group scales to minimize the layer-wise reconstruction loss. To this end, we employ the coordinate descent algorithm and derive a closed-form update rule, which enables efficient refinement without costly numerical optimization. Notably, our derivation incorporates the quantization errors from preceding layers to prevent error accumulation. Experimental results demonstrate that our method consistently enhances group-wise quantization, achieving higher accuracy with negligible overhead.

cs.LG↗

arXiv2025

Low-Frequency Noise Performance of Microstrip-Coupled Lumped-Element Aluminum KIDs using Hydrogenated Amorphous Silicon Parallel-Plate Capacitors for NEW-MUSIC

We present measurements of the low-frequency noise of microstrip-coupled, lumped-element aluminum kinetic inductance detectors that use hydrogenated amorphous silicon parallel-plate capacitors (Al/a-Si:H MS-PPC-LEKIDs), which are under development for the Next-generation Extended Wavelength Multiband Submillimeter Inductance Camera (NEW-MUSIC). We show that, under dark conditions, these devices are generation recombination (GR) noise dominated down to 0.1 Hz and, under optical load, they are likely dominated by GR and photon noise down to tenths of a Hz and possibly lower, both in spite of the use of a-Si:H PPCs. Our measurements set limits on the low-frequency two-level-system (TLS) noise of the a-Si:H material that are consistent with higher frequency measurements in the 0.1-10 kHz regime. These results establish that our MS-PPC-LEKID design for NEW-MUSIC will be photon-noise-limited under a range of observing conditions and, more generally, that a-Si:H PPC-KIDs are a viable new detector technology for even low modulation-rate applications such as astronomy.

astro-ph.IM↗

arXiv2025

Improved Modeling of Quasi-Static Thermal and Optical Response of Lumped-Element Aluminum Manganese KIDs

We report on the optical characterization of the AlMn kinetic inductance detectors (KIDs) in development for use in the Next-generation Extended Wavelength-MUltiband Sub/millimeter Inductance Camera (NEW-MUSIC) on the Leighton Chajnantor Telescope (LCT). NEW-MUSIC will cover 80-420 GHz, split into six spectral bands, with polarimetry. This broad spectral coverage will enable study of a range of scientific topics such as the accretion and feedback in galaxies and galaxy cluster evolution via the Sunyaev-Zeldovich effect, the transient synchrotron emission from the explosive deaths of massive stars and other time-domain phenomena, and dusty sources from low to high redshift (with polarization). Al KIDs have already been demonstrated for bands 2-5. AlMn KIDs will be used for the 90~GHz band, as Al's pair-breaking energy is too high. However, AlMn has only barely been explored as a KID material. To this end, we first improved the modeling techniques used for Al KIDs within BCS theory by eliminating the use of analytical approximations for the expressions of the complex conductivity and found these changes reduced fit parameter degeneracy in the analysis of AlMn. Then, we tested the addition of a gap smearing parameter, a standard extension to BCS theory in use for high kinetic inductance materials, and found it did not improve the fits.

astro-ph.IM↗

arXiv2025

CHEX-MATE: Cluster Multi-Probes in Three Dimensions (CLUMP-3D) II. Combined Gas and Dark Matter Analysis from X-ray, SZE, and WL

Under the standard model of hierarchical structure formation, the overall geometry of galaxy clusters is better described by a triaxial ellipse than a sphere. As a result, applying spherically-symmetric models can result in significant biases. These biases can be mitigated by fitting a triaxial model, requiring deep multiprobe data and a set of physically motivated models to describe them. Here we present a multiprobe triaxial analysis methodology based on the data available for galaxy clusters in the Cluster Heritage project with XMM-Newton - Mass Assembly and Thermodynamics at Endpoint of structure formation (CHEX-MATE), which includes X-ray data from XMM-Newton, SZ data from Planck and ACT, and WL data from Subaru. This work builds on our previous development of a gas-only X-ray and SZ triaxial fitting formalism in Paper I. We apply our approach to the CHEX-MATE cluster PSZ2 G313.33+61.13 (Abell 1689) and find that it is elongated along the line of sight relative to the plane of sky by a factor of $\mathcal{R}_{LP} = 1.27 \pm 0.02$. As a result, the WL mass obtained from our triaxial fit, $\text{M}_{200c}=(13.69_{-1.41}^{+1.56})\times10^{14} \text{M}_{\odot}$, is significantly lower than the value of $(17.77_{-1.75}^{+2.00})\times10^{14} \text{M}_{\odot}$ obtained from a spherically-symmetric fit that otherwise employs the same methodology. Our triaxial fit finds a concentration of $c_{200c}=8.55_{-1.61}^{+2.20}$, consistent with the spherically-symmetric value of $9.99_{-1.78}^{+2.26}$, which suggests that the unexpectedly high concentration in Abell 1689 is not due to triaxiality and orientation. We also measure the non-thermal pressure fraction at radii between 0.18-1.37 Mpc, finding a minimum of approximately 20 per cent at intermediate radii increasing to near 30 per cent at both the smallest and largest radii, and with a typical measurement precision of $\pm$5 per cent.

astro-ph.CO↗

arXiv2025

Enhanced imaging of M87*: Simulations with the EHT and extended-KVN

The Event Horizon Telescope (EHT) has successfully revealed the shadow of the supermassive black hole, M87*, with an unprecedented angular resolution of approximately 20 uas at 230 GHz. However, because of limited short baseline lengths, the EHT has been constrained in its ability to recover larger scale jet structures. The extended Korean VLBI Network (eKVN) is committed to joining the EHT from 2024 that can improve short baseline coverage. This study evaluates the impact of the participation of eKVN in the EHT on the recovery of the M87* jet. Synthetic data, derived from a simulated M87* model, were observed using both the EHT and the combined EHT+eKVN arrays, followed by image reconstructions from both configurations. The results indicate that the inclusion of eKVN significantly improves the recovery of jet structures by reducing residual noise. Furthermore, jackknife tests, in which one or two EHT telescopes were omitted - simulating potential data loss due to poor weather - demonstrate that eKVN effectively compensates for these missing telescopes, particularly in short baseline coverage. Multi-frequency synthesis imaging at 86-230 GHz shows that the EHT+eKVN array enhances the recovered spectral index distribution compared to the EHT alone and improves image reconstruction at each frequency over single-frequency imaging. As the EHT continues to expand its array configuration and observing capabilities to probe black hole physics more in depth, the integration of eKVN into the EHT will significantly enhance the stability of observational results and improve image fidelity. This advancement will be particularly valuable for future regular monitoring observations, where consistent data quality is essential.

astro-ph.IM↗

arXiv2024

NEW-MUSIC: The Next-generation Extended-Wavelength Multiband Sub/millimeter Inductance Camera

The Next-generation Extended Wavelength-MUltiband Sub/millimeter Inductance Camera (NEW-MUSIC) on the Leighton Chajnantor Telescope (LCT) will be a first-of-its-kind, six-band, transmillimeter-wave ("trans-mm") polarimeter covering 2.4 octaves of spectral bandwidth to open a new window on the trans-mm time-domain frontier, in particular new frontiers in energy, density, time, and magnetic field. NEW-MUSIC's broad spectral coverage will also enable the use of the Sunyaev-Zeldovich effects to study accretion, feedback, and dust content in the hot gaseous haloes of galaxies and galaxy clusters. Six-band spectral energy distributions, with polarization information, will yield new insights into stellar and planetary nurseries. NEW-MUSIC will employ hierarchical, phased arrays of polarization-sensitive superconducting slot-dipole antennas, coupled to photolithographic bandpass filters, to nearly optimally populate LCT's 14' field-of-view with six spectral bands over 80-420 GHz (1:5.25 spectral dynamic range; 2.4 octaves). Light will be routed to Al or AlMn microstripline-coupled, parallel-plate capacitor, lumped-element kinetic inductance detectors (MS-PPC-LEKIDs), an entirely new KID architecture that substantially enhances design flexibility while providing background-limited performance. Innovative, wide-bandwidth, etched silicon structures will be used to antireflection-treat the back-illuminated focal plane. NEW-MUSIC will cost-effectively reuse much of the MUSIC instrument, initially deploying a quarter-scale focal plane capable of the bulk of NEW-MUSIC science followed later by a full-FoV focal plane needed for NEW-MUSIC wide-area survey science.

astro-ph.IM↗

arXiv2024

BoA: Attention-aware Post-training Quantization without Backpropagation

Post-training quantization (PTQ) is a promising solution for deploying large language models (LLMs) on resource-constrained devices. Early methods developed for small-scale networks, such as ResNet, rely on gradient-based optimization, which becomes impractical for hyper-scale LLMs with billions of parameters. While recently proposed backpropagation-free or transformation-based methods alleviate this issue, they ignore inter-layer interactions or use the naive nearest-rounding-based quantized weight assignment to save the heavy computational cost of weight optimization. In this paper, we introduce a novel backpropagation-free PTQ algorithm that optimizes quantized weights by considering inter-layer dependencies. The key innovation is the development of attention-aware Hessian matrices that capture inter-layer interactions within the attention module. Extensive experiments demonstrate that our approach not only outperforms existing weight quantization methods but also shows good synergy with conventional methods to suppress activation outliers, leading to state-of-the-art weight-activation quantization performance. The code will be available at https://github.com/SamsungLabs/BoA.

cs.LG↗

arXiv2024

Towards Next-Level Post-Training Quantization of Hyper-Scale Transformers

With the increasing complexity of generative AI models, post-training quantization (PTQ) has emerged as a promising solution for deploying hyper-scale models on edge devices such as mobile and TVs. Existing PTQ schemes, however, consume considerable time and resources, which could be a bottleneck in real situations where frequent model updates and multiple hyperparameter tunings are required. As a cost-effective alternative, learning-free PTQ schemes have been proposed. However, the performance is somewhat limited because they cannot consider the inter-layer dependency within the attention module, which is a significant feature of Transformers. In this paper, we thus propose a novel PTQ algorithm that balances accuracy and efficiency. The key idea of the proposed algorithm called aespa is to perform quantization layer-wise for efficiency while targeting attention-wise reconstruction to consider the cross-layer dependency. Through extensive experiments on various language models and complexity analysis, we demonstrate that aespa is accurate and efficient in quantizing Transformer models.

cs.LG↗

arXiv2024

Hierarchical phased-array antennas coupled to Al KIDs: a scalable architecture for multi-band mm/submm focal planes

We present the optical characterization of two-scale hierarchical phased-array antenna kinetic inductance detectors (KIDs) for millimeter/submillimeter wavelengths. Our KIDs have a lumped-element architecture with parallel plate capacitors and aluminum inductors. The incoming light is received with a hierarchical phased array of slot-dipole antennas, split into 4 frequency bands (between 125 GHz and 365 GHz) with on-chip lumped-element band-pass filters, and routed to different KIDs using microstriplines. Individual pixels detect light for the 3 higher frequency bands (190-365 GHz) and the signals from four individual pixels are coherently summed to create a larger pixel detecting light for the lowest-frequency band (125-175 GHz). The spectral response of the band-pass filters was measured using Fourier transform spectroscopy (FTS), the far-field beam pattern of the phased-array antennas was obtained using an infrared source mounted on a 2-axis translating stage, and the optical efficiency of the KIDs was characterized by observing loads at 294 K and 77 K. We report on the results of these three measurements.

astro-ph.IM↗

arXiv2023

CHEX-MATE: CLUster Multi-Probes in Three Dimensions (CLUMP-3D), I. Gas Analysis Method using X-ray and Sunyaev-Zel'dovich Effect Data

Galaxy clusters are the products of structure formation through myriad physical processes that affect their growth and evolution throughout cosmic history. As a result, the matter distribution within galaxy clusters, or their shape, is influenced by cosmology and astrophysical processes, in particular the accretion of new material due to gravity. We introduce an analysis method to investigate the 3D triaxial shapes of galaxy clusters from the Cluster HEritage project with XMM-Newton -- Mass Assembly and Thermodynamics at the Endpoint of structure formation (CHEX-MATE). In this work, the first paper of a CHEX-MATE triaxial analysis series, we focus on utilizing X-ray data from XMM and Sunyaev-Zel'dovich (SZ) effect maps from Planck and ACT to obtain a three dimensional triaxial description of the intracluster medium (ICM) gas. We present the forward modeling formalism of our technique, which projects a triaxial ellipsoidal model for the gas density and pressure to compare directly with the observed two dimensional distributions in X-rays and the SZ effect. A Markov chain Monte Carlo is used to estimate the posterior distributions of the model parameters. Using mock X-ray and SZ observations of a smooth model, we demonstrate that the method can reliably recover the true parameter values. In addition, we apply the analysis to reconstruct the gas shape from the observed data of one CHEX-MATE galaxy cluster, Abell 1689, to illustrate the technique. The inferred parameters are in agreement with previous analyses for that cluster, and our results indicate that the geometrical properties, including the axial ratios of the ICM distribution, are constrained to within a few percent. With much better precision than previous studies, we thus further establish that Abell 1689 is significantly elongated along the line of sight, resulting in its exceptional gravitational lensing properties.

astro-ph.CO↗

arXiv2023

COMAP Early Science: VIII. A Joint Stacking Analysis with eBOSS Quasars

We present a new upper limit on the cosmic molecular gas density at $z=2.4-3.4$ obtained using the first year of observations from the CO Mapping Array Project (COMAP). COMAP data cubes are stacked on the 3D positions of 243 quasars selected from the Extended Baryon Oscillation Spectroscopic Survey (eBOSS) catalog, yielding a 95% upper limit for flux from CO(1-0) line emission of 0.129 Jy km/s. Depending on the balance of the emission between the quasar host and its environment, this value can be interpreted as an average CO line luminosity $L'_\mathrm{CO}$ of eBOSS quasars of $\leq 1.26\times10^{11}$ K km pc$^2$ s$^{-1}$, or an average molecular gas density $\rho_\mathrm{H_2}$ in regions of the universe containing a quasar of $\leq 1.52\times10^8$ M$_\odot$ cMpc$^{-3}$. The $L'_\mathrm{CO}$ upper limit falls among CO line luminosities obtained from individually-targeted quasars in the COMAP redshift range, and the $\rho_\mathrm{H_2}$ value is comparable to upper limits obtained from other Line Intensity Mapping (LIM) surveys and their joint analyses. Further, we forecast the values obtainable with the COMAP/eBOSS stack after the full 5-year COMAP Pathfinder survey. We predict that a detection is probable with this method, depending on the CO properties of the quasar sample. Based on the achieved sensitivity, we believe that this technique of stacking LIM data on the positions of traditional galaxy or quasar catalogs is extremely promising, both as a technique for investigating large galaxy catalogs efficiently at high redshift and as a technique for bolstering the sensitivity of LIM experiments, even with a fraction of their total expected survey data.

astro-ph.GA↗

arXiv2023

Vision Transformer-based Feature Extraction for Generalized Zero-Shot Learning

Generalized zero-shot learning (GZSL) is a technique to train a deep learning model to identify unseen classes using the image attribute. In this paper, we put forth a new GZSL approach exploiting Vision Transformer (ViT) to maximize the attribute-related information contained in the image feature. In ViT, the entire image region is processed without the degradation of the image resolution and the local image information is preserved in patch features. To fully enjoy these benefits of ViT, we exploit patch features as well as the CLS feature in extracting the attribute-related image feature. In particular, we propose a novel attention-based module, called attribute attention module (AAM), to aggregate the attribute-related information in patch features. In AAM, the correlation between each patch feature and the synthetic image attribute is used as the importance weight for each patch. From extensive experiments on benchmark datasets, we demonstrate that the proposed technique outperforms the state-of-the-art GZSL approaches by a large margin.

cs.CV↗

arXiv2022

COMAP Early Science: IV. Power Spectrum Methodology and Results

We present the power spectrum methodology used for the first-season COMAP analysis, and assess the quality of the current data set. The main results are derived through the Feed-feed Pseudo-Cross-Spectrum (FPXS) method, which is a robust estimator with respect to both noise modeling errors and experimental systematics. We use effective transfer functions to take into account the effects of instrumental beam smoothing and various filter operations applied during the low-level data processing. The power spectra estimated in this way have allowed us to identify a systematic error associated with one of our two scanning strategies, believed to be due to residual ground or atmospheric contamination. We omit these data from our analysis and no longer use this scanning technique for observations. We present the power spectra from our first season of observing and demonstrate that the uncertainties are integrating as expected for uncorrelated noise, with any residual systematics suppressed to a level below the noise. Using the FPXS method, and combining data on scales $k=0.051-0.62 \,\mathrm{Mpc}^{-1}$ we estimate $P_\mathrm{CO}(k) = -2.7 \pm 1.7 \times 10^4μ\textrm{K}^2\mathrm{Mpc}^3$, the first direct 3D constraint on the clustering component of the CO(1-0) power spectrum in the literature.

astro-ph.CO↗