RED: Multi-Domain Concepts in Astronomy & Engineering

• RED is a multifaceted term designating distinct concepts in astronomy—such as dusty star-forming galaxies and red quasars—and in engineering, including energy-optimized PIM systems and adaptive scheduling frameworks.
• In astronomy, RED uses rigorous selection criteria (e.g., color cuts and model fitting) to classify objects based on dust obscuration, evolved stellar populations, or anomalous spectral features.
• In engineering, RED frameworks optimize system performance through retention-aware voltage scaling, dynamic task scheduling in robotics, and noise-robust RF classification for rogue emitter detection.

In recent arXiv literature, RED denotes a family of technically distinct concepts rather than a single object. In astronomy and astrophysics, “red” functions as an observational descriptor tied to specific color cuts, spectral-energy-distribution morphologies, or red-sequence selection, as in the extremely red host of GRB 080207, red and extremely red quasars, Herschel-selected dusty star-forming galaxies, luminous red galaxies, and JWST-selected Red Emission line Galaxies (Hunt et al., 2011, Hamann et al., 2016, Duivenvoorden et al., 2018, Rozo et al., 2015, Withers et al., 4 Jun 2026). In engineering and computer systems, RED is an acronymic title for frameworks in eDRAM-based processing-in-memory, robotic real-time scheduling, and rogue emitter detection (Kim et al., 13 Feb 2025, Li et al., 21 May 2026, Yang et al., 2022). This range of usage makes the term notable less for a single invariant definition than for a recurrent emphasis on classification under constrained observables: color in astrophysics, and operating-state or decision-space structure in engineered systems.

1. Red as an observational classification in astronomy

In the astronomical papers represented here, “red” is always tied to a measurable selection rule rather than a purely qualitative appearance. Hunt et al. describe the host of GRB 080207 as “extremely red” because it has $(R-K)_{\rm Vega}\sim6.3$ and $F_{24}/F_R\sim994$, placing it simultaneously in the categories of extremely red objects (EROs) and dust-obscured galaxies (DOGs) (Hunt et al., 2011). In Herschel studies of dusty star-forming galaxies, “red” means far-infrared/submillimeter colors satisfying $S_{500}>S_{350}>S_{250}$, a criterion that preferentially selects high-$z$ DSFGs but is also entangled with dust temperature, confusion noise, flux boosting, blending, and lensing (Duivenvoorden et al., 2018). In red-quasar work, the term can refer either to optically defined reddened Type 1 QSOs, such as the reddest 10% in redshift-binned $g-i$ color, or to more restrictive intrinsic-reddening and UV-to-mid-IR criteria such as $E(B-V)>0.25$ or $i-W3\ge4.6$ (Rosario et al., 2021, Glikman et al., 2022, Hamann et al., 2016).

Other astronomical usages are more explicitly model-based. redMaGiC selects luminous red galaxies from the red sequence by fitting an empirical color model and imposing a redshift-dependent $\chi^2$ threshold tuned to constant comoving density, so “red” here denotes membership in a calibrated red-sequence population optimized for photometric redshift performance rather than dust obscuration (Rozo et al., 2015). The JWST medium-band study of Red Emission line Galaxies defines REGs as strong-line galaxies whose continuum colors are at least $2\sigma$ redder than the fitted line-color/continuum-color relation, with thresholds of $\geq0.82$ mag or $F_{24}/F_R\sim994$0 mag depending on the reference sample (Withers et al., 4 Jun 2026).

Taken together, these usages show that “red” can index at least four different physical situations already explicit in the literature: dust-obscured star formation, dust-reddened AGN continua, evolved stellar populations on the red sequence, and high-$F_{24}/F_R\sim994$1 systems whose continuum color is anomalous relative to their nebular-line strength. This suggests that the term is observationally precise but physically non-unique.

2. Red quasars and extremely red quasars

In quasar studies, “red” has become a structured subfield with increasingly fine distinctions between mild optical reddening, heavily dust-reddened broad-line QSOs, and the more spectroscopically peculiar ERQ population. The e-MERLIN study of red QSOs defines rQSOs as the upper 10th percentile of the Galactic-extinction-corrected $F_{24}/F_R\sim994$2 distribution in redshift bins of 1000 QSOs, with control QSOs drawn from the central 50% around the median. In a matched sample of $F_{24}/F_R\sim994$3 rQSOs and $F_{24}/F_R\sim994$4 cQSOs at $F_{24}/F_R\sim994$5, $F_{24}/F_R\sim994$6 rQSOs show extended kpc-scale radio structure versus $F_{24}/F_R\sim994$7 cQSOs, and the excess radio emission is concentrated on host-galaxy scales $F_{24}/F_R\sim994$8 kpc rather than on classical $F_{24}/F_R\sim994$9 kpc jet/lobe scales (Rosario et al., 2021). The paper interprets this as evidence that red QSOs are not merely blue QSOs seen through slightly more dust, but a radio-distinct sub-population whose enhanced radio emission likely arises from AGN-driven jets or winds on galaxy scales.

The WISE-2MASS survey adopts a stricter physical definition: red QSOs are broad-line Type 1 QSOs with host-subtracted intrinsic reddening $S_{500}>S_{350}>S_{250}$0. After GANDALF host subtraction for $S_{500}>S_{350}>S_{250}$1, the authors identify $S_{500}>S_{350}>S_{250}$2 red QSOs within a mid-infrared-selected sample of $S_{500}>S_{350}>S_{250}$3 Type 1 QSOs. In the luminosity-restricted comparison, red QSOs are radio-detected more often than blue QSOs in both FIRST and VLASS, are brighter in median radio stacks, and have steeper stacked radio spectra, with $S_{500}>S_{350}>S_{250}$4 versus $S_{500}>S_{350}>S_{250}$5 under $S_{500}>S_{350}>S_{250}$6 (Glikman et al., 2022). The paper argues that dusty AGN-driven winds provide the most plausible joint explanation for both the reddening and the radio excess.

The multiwavelength SDSS study of red QSOs reaches a related conclusion from UV-to-FIR SED modeling and emission-line diagnostics. In a matched sample of $S_{500}>S_{350}>S_{250}$7 red QSOs and $S_{500}>S_{350}>S_{250}$8 control QSOs, the fitted accretion-disk reddening differs strongly, with $S_{500}>S_{350}>S_{250}$9 mag for red QSOs and $z$0 mag for controls, while torus properties and host-galaxy ISM indicators show no comparably strong separation. The key residual is a significant rest-frame $z$1–$z$2m excess that correlates with optical reddening and is strongest in red QSOs with broad [O III] wings or large C IV shifts, leading the authors to propose dusty winds at nuclear scales as the physical origin of the colors (Calistro-Rivera et al., 2021).

Kim et al. address the causal question directly with optical and near-infrared line ratios. For $z$3 red quasars at $z$4, the Paschen-to-Balmer broad-line ratios are about $z$5 times higher than in unobscured Type 1 quasars, implying $z$6 mag, and CLOUDY modeling shows that most of these ratios are difficult to reproduce with plausible BLR physical conditions without dust extinction. The same study finds Eddington ratios higher than those of unobscured Type 1 quasars by factors of $z$7, disfavoring a moderate-viewing-angle explanation (Kim et al., 2017).

Hamann et al. identify an even more restrictive population of extremely red quasars. Their “core” sample consists of $z$8 objects at $z$9 selected by $g-i$0 (AB) and $g-i$1 Å. These quasars have unusually large C IV equivalent widths, high $g-i$2 “wingless” profiles, large $g-i$3, $g-i$4, and $g-i$5 ratios, frequent BAL or BAL-like signatures, and in follow-up near-infrared spectroscopy very broad and blueshifted [O III] $g-i$6 (Hamann et al., 2016). Their flat or blue rest-UV continua despite very red $g-i$7 colors lead the authors to reject a simple foreground dust-screen model and to favor patchy obscuration by small dusty clouds embedded in powerful outflows.

3. Red galaxies, dusty star formation, and high-redshift red continua

The host of GRB 080207 is an important benchmark for “red” as a galaxy-scale descriptor. Hunt et al. identify the host at the Chandra afterglow position as a very faint source with $g-i$8 and $g-i$9, corresponding to $E(B-V)>0.25$0. Its SED shows a clear $E(B-V)>0.25$1m stellar bump, yielding $E(B-V)>0.25$2, and the best fit is an M82-like starburst with $E(B-V)>0.25$3, stellar mass $E(B-V)>0.25$4, and $E(B-V)>0.25$5–$E(B-V)>0.25$6 mag (Hunt et al., 2011). The paper stresses that the redness is not the signature of a passive galaxy but of a dusty, massive, evolved star-forming system, and suggests that such ERO/DOG hosts may define a distinct subpopulation of dark-GRB environments.

In Herschel-selected DSFG work, “red” describes the observed rise of the SPIRE SED toward longer wavelength. The SCUBA-2 follow-up of $E(B-V)>0.25$7 bright HeLMS red sources selected by $E(B-V)>0.25$8 and $E(B-V)>0.25$9 mJy detected $i-W3\ge4.6$0 of the sample at $i-W3\ge4.6$1 at $i-W3\ge4.6$2m. The mean photometric redshift is $i-W3\ge4.6$3 with dispersion $i-W3\ge4.6$4, and $i-W3\ge4.6$5 sources have a high probability of lying at $i-W3\ge4.6$6 (Duivenvoorden et al., 2018). The same paper introduces a confusion-noise sampling method for SPIRE SED fitting and shows that redness is not a pure redshift indicator: cooler dust SEDs, correlated confusion, flux boosting, blending, and lensing can all contribute to apparently extreme colors.

The JWST medium-band study of REGs redefines red high-$i-W3\ge4.6$7 galaxies in a different way. Starting from a large ELG sample at $i-W3\ge4.6$8, the authors identify $i-W3\ge4.6$9 outliers whose continua are much redder than expected for their line strengths, over $\chi^2$0. They divide these REGs into $\chi^2$1 REG LRDs, $\chi^2$2 extended REGs, and $\chi^2$3 compact REGs (Withers et al., 4 Jun 2026). The extended REGs are resolved in $\chi^2$4 and have average $\chi^2$5 mag, average $\chi^2$6, and average $\chi^2$7, consistent with dusty star-forming galaxies. The compact REGs are unresolved, very faint, often line-contaminated in the broad bands, and are interpreted as likely LRD-like systems missed by standard broad-band selections. This suggests that medium-band JWST imaging can recover a red high-$\chi^2$8 population that conventional LRD criteria incompletely sample.

4. Red-sequence galaxies and photometric large-scale structure

In cosmological survey methodology, “red” is formalized through the red sequence rather than through dust obscuration. redMaGiC, introduced by Rozo et al., is an algorithm for selecting luminous red galaxies with minimal photometric-redshift uncertainty by fitting an empirically calibrated red-sequence model and enforcing a luminosity threshold together with a redshift-dependent goodness-of-fit cut (Rozo et al., 2015). The red-sequence mean color is modeled as

$\chi^2$9

with total covariance

$2\sigma$0

and the photometric-redshift likelihood is

$2\sigma$1

Selection proceeds through the conditions

$2\sigma$2

where $2\sigma$3 is spline-parameterized and self-trained so that the final sample approximates a constant comoving density $2\sigma$4. The method is distinct from redMaPPer cluster finding: redMaGiC uses the red-sequence calibration infrastructure of redMaPPer, but its output is a catalog of field galaxies optimized for photometric large-scale-structure studies rather than clusters (Rozo et al., 2015).

Applied to DES Science Verification data, the fiducial redMaGiC sample spans $2\sigma$5, has comoving density $2\sigma$6, and achieves median photo-$2\sigma$7 bias $2\sigma$8, scatter $2\sigma$9, and $\geq0.82$0 outlier fraction $\geq0.82$1 (Rozo et al., 2015). The paper emphasizes that these photo-$\geq0.82$2 errors are very nearly Gaussian and well characterized. It also argues that redMaGiC avoids the extrapolation biases of machine-learning methods because it relies on a calibrated color–magnitude–redshift model of the red sequence rather than a purely data-driven mapping from a possibly non-representative spectroscopic training set.

5. RED as an energy-optimization framework for eDRAM-based PIM

Outside astronomy, RED is also the formal name of an architecture and scheduling framework: “Energy Optimization Framework for eDRAM-based PIM with Reconfigurable Voltage Swing and Retention-aware Scheduling” (Kim et al., 13 Feb 2025). The framework is motivated by Transformer-like AI workloads on digital processing-in-memory systems, where the dominant energy term shifts from arithmetic to memory access. In the representative PIM macro analyzed in the paper, memory access consumes $\geq0.82$3 of total power, and within a $\geq0.82$4T eDRAM macro the dominant access-power components are RWL/RBL voltage swing ($\geq0.82$5), pull-down driver ($\geq0.82$6), and sense amplifier ($\geq0.82$7) (Kim et al., 13 Feb 2025).

RED is built on two observations: memory access energy dominates in PIM, and read bitline voltage swing, sense-amplifier power, and retention time are in trade-off relations. Its scheduler evaluates all valid tiling schemes and candidate memory operations using the energy model

$\geq0.82$8

$\geq0.82$9

$F_{24}/F_R\sim994$00

The hardware substrate is a reconfigurable $F_{24}/F_R\sim994$01T eDRAM macro whose controller changes the pull-down voltage $F_{24}/F_R\sim994$02 at runtime, with evaluated operating points at $F_{24}/F_R\sim994$03, $F_{24}/F_R\sim994$04, $F_{24}/F_R\sim994$05, and $F_{24}/F_R\sim994$06 mV, while also applying refresh skipping and sense-amplifier power gating (Kim et al., 13 Feb 2025).

The evaluation uses BERT-Base, BERT-Large, T5-Base, and T5-Large on Samsung $F_{24}/F_R\sim994$07 nm at $F_{24}/F_R\sim994$08 MHz and $F_{24}/F_R\sim994$09 V, with INT8 activations and weights. The headline result is up to $F_{24}/F_R\sim994$10 higher energy efficiency than the prior SRAM-based PIM baseline Neural Cache, together with up to $F_{24}/F_R\sim994$11 reduction in eDRAM macro energy consumption, for $F_{24}/F_R\sim994$12 area overhead and $F_{24}/F_R\sim994$13 energy overhead for scheduling (Kim et al., 13 Feb 2025). The framework’s central claim is therefore not merely circuit tuning, but cross-layer co-optimization of memory operating point, retention behavior, and workload tiling.

6. RED in real-time robotic scheduling and in rogue-emitter detection

A second engineering usage is “Adaptive Real-Time DAG Scheduling for Robotic Inference under Environmental Dynamics”, a runtime framework for multi-task DNN inference on embedded robots (Li et al., 21 May 2026). RED models each application as a DAG

$F_{24}/F_R\sim994$14

with nodes

$F_{24}/F_R\sim994$15

and assigns proportional intermediate deadlines by level: $F_{24}/F_R\sim994$16 The framework also refines MIMONet workloads into shared encoders and task-specific decoders, supports online graph mutation, and uses on-demand synchronization and overload handling. Implemented on NVIDIA Jetson platforms and an Apple M-series MacBook, it reduces average latency by $F_{24}/F_R\sim994$17 versus EDF, lowers deadline miss rate by $F_{24}/F_R\sim994$18, and improves QoE by $F_{24}/F_R\sim994$19 across the main evaluation (Li et al., 21 May 2026). The paper’s emphasis is that robotic “environmental dynamics” can change not just execution times but the DAG itself, so fixed static scheduling assumptions are inadequate.

A third engineering usage is rogue emitter detection (RED) in physical-layer security (Yang et al., 2022). Here the received signal from emitter $F_{24}/F_R\sim994$20 is modeled as

$F_{24}/F_R\sim994$21

and the problem is to reject emitters outside the enrolled legitimate set. The proposed method combines a denoising autoencoder with center-based deep metric learning. Training uses the joint loss

$F_{24}/F_R\sim994$22

with $F_{24}/F_R\sim994$23, $F_{24}/F_R\sim994$24, and $F_{24}/F_R\sim994$25. After training, RED computes legitimate class centers and classifies a test signal by thresholding its distance to the nearest center (Yang et al., 2022). On ADS-B and IEEE 802.11 datasets at $F_{24}/F_R\sim994$26 dB SNR, the method achieves AUC $F_{24}/F_R\sim994$27 and $F_{24}/F_R\sim994$28, outperforming SR2CNN, Isolation Forest, and LOF. This suggests that, in this usage, RED denotes open-world rejection based on noise-robust and center-separable RF embeddings rather than ordinary closed-set classification.

Across these engineering papers, RED denotes systems that adapt runtime decisions to latent structure that a naive baseline would ignore: retention-aware memory behavior in PIM, graph mutation and shared-encoder structure in robotics, and embedding geometry under low SNR in RF security (Kim et al., 13 Feb 2025, Li et al., 21 May 2026, Yang et al., 2022).