Perseus: Astronomy & POMDP Applications

• Perseus is a multifaceted designation encompassing astronomical entities—such as galaxy clusters, molecular clouds, and dwarf galaxies—as well as a randomized value-iteration algorithm for POMDPs.
• In astronomy, the Perseus cluster serves as a benchmark for studying cool-core dynamics, merger-induced structures, and non-thermal high-energy phenomena through multiwavelength observations.
• In computer science, the Perseus algorithm employs randomized point-based value iteration in POMDPs, reducing computational load while sustaining solution quality in large-scale planning.

Perseus is a designation used in contemporary research for several distinct objects and methods: a nearby massive galaxy cluster centered on NGC 1275, a nearby molecular-cloud and star-forming complex, the dwarf galaxy Perseus I/Andromeda XXXIII near M31, and a randomized point-based value-iteration algorithm for partially observable Markov decision processes (POMDPs). In astronomy, the name is attached to systems that are unusually well resolved and therefore frequently used to study intracluster-medium thermodynamics, merger dynamics, radio and high-energy non-thermal processes, star formation, and compact stellar systems; in AI, it denotes an approximate solver that exploits piecewise-linear convex value geometry in belief space for large-scale planning under uncertainty (HyeongHan et al., 2024, Arce et al., 2010, Spaan et al., 2011).

1. Major scientific referents

The current literature uses the name across several scales, from Galactic star formation to cluster cosmology and decision-theoretic planning. The principal usages represented in recent work are summarized below.

Referent	Domain	Defining properties
Perseus cluster (Abell 426)	Galaxy cluster	brightest galaxy cluster in the X-ray sky; canonical cool core with AGN feedback; merger-perturbed outskirts
Perseus–Pisces supercluster	Large-scale structure	22-member X-ray-luminous chain; length 115.7 Mpc; total mass $\approx 2.5\times10^{16}\,M_\odot$
Perseus molecular cloud	Star formation	distance $\sim 250$ pc; total molecular mass $\simeq 7\times10^3\,M_\odot$; includes NGC 1333 and IC 348
Perseus I / Andromeda XXXIII	Dwarf galaxy	heliocentric distance $785\pm65$ kpc; likely satellite of M31
Perseus	POMDP solver	randomized point-based value iteration with improve-only backups

The astrophysical uses span at least four nested environments: cloud, star-forming region, galaxy cluster, and supercluster. In CLASSIX, the Perseus–Pisces supercluster was reconstructed with a friends-of-friends procedure as the largest superstructure in the Universe at redshifts $z\le 0.03$, with 22 members, volume $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$, and a total mass estimate $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$ (Boehringer et al., 2021).

2. The Perseus cluster as a cool-core system and merger remnant

The Perseus cluster is a nearby massive galaxy cluster whose central region hosts the AGN in NGC 1275. A panoramic X-ray reconstruction combining SRG/eROSITA, XMM-Newton, and Chandra adopts $z_{\rm cl}=0.01767$, $D_A=74.1$ Mpc, $M_{200c}=6.65\times10^{14}\,M_\odot$, and $\sim 250$0 Mpc, with corresponding circular velocity $\sim 250$1. In the core, Chandra and XMM show a temperature decline from $\sim 250$2 keV at $\sim 250$3 kpc to $\sim 250$4 keV in the innermost 20 kpc, low entropy $\sim 250$5–$\sim 250$6, nominal cooling time $\sim 250$7 Gyr, and central X-ray cavities of $\sim 250$8–20 kpc surrounded by weak shocks and ripples out to $\sim 250$9 kpc, each carrying $\simeq 7\times10^3\,M_\odot$0–$\simeq 7\times10^3\,M_\odot$1 erg of enthalpy (Churazov et al., 26 Jul 2025).

A recurrent misconception is that Perseus is simply a fully relaxed cool-core cluster. Multiple lines of evidence contradict that simplification. Chandra and XMM imaging reveal an ancient cold front $\simeq 7\times10^3\,M_\odot$2 kpc east of the core, east-west asymmetry to $\simeq 7\times10^3\,M_\odot$3 Mpc scales, and a filamentary galaxy distribution spanning $\simeq 7\times10^3\,M_\odot$4 Mpc. Weak-lensing analysis identified a subcluster halo at the $\simeq 7\times10^3\,M_\odot$5 level centered on NGC 1264, located $\simeq 7\times10^3\,M_\odot$6 kpc west of NGC 1275, with $\simeq 7\times10^3\,M_\odot$7, together with a $\simeq 7\times10^3\,M_\odot$8 mass bridge between the main and subcluster halos. Idealized N-body + ideal MHD simulations show that a $\simeq 7\times10^3\,M_\odot$9 off-axis major merger can generate both the observed $785\pm65$0 kpc cold front and the bridge through multiple core crossings, placing the system near its third apocentre and $785\pm65$1–6 Gyr post-pericentre (HyeongHan et al., 2024).

Panoramic X-ray work extends the disturbed morphology to the outskirts. Outside $785\pm65$2 kpc, division by a best-fit $785\pm65$3-model with $785\pm65$4 and $785\pm65$5 kpc shows east-west asymmetries out to $785\pm65$6 and beyond. To the west, a $785\pm65$7–2 Mpc-scale X-ray excess is centered on IC 310, 37′ or 800 kpc from NGC 1275. That analysis interprets IC 310 as the main galaxy of a subcluster in a merger over the past $785\pm65$8 Gyr, while NGC 1265 is modeled as a high-velocity galaxy with $785\pm65$9 and an extreme, nearly line-of-sight configuration. Both IC 310 and NGC 1265 show sharply bent radio tails described as “Radio-Uroboros” structures (Churazov et al., 26 Jul 2025). This suggests that the western perturbation is being reconstructed with different tracers—weak lensing, X-ray morphology, and radio-tailed galaxies—rather than with a single observational diagnostic.

3. Non-thermal, radio, and high-energy phenomenology of the cluster

Hard X-ray analysis with NuSTAR has become possible in the central region through the open-source package nucrossarf, which generates separate point-source and diffuse-source ARFs for the AGN and ICM, so that

$z\le 0.03$0

Using three archival observations, diffuse 3–25 keV emission from the ICM shows a $z\le 0.03$1 excess beyond 20 keV that is not describable by purely thermal models. The AGN is well fitted by an absorbed power law with photon index $z\le 0.03$2, and a dedicated PSF-scaling calibration gives a 3.4% systematic uncertainty on the AGN flux. Explaining the entire excess as scattered AGN light would require boosting the AGN normalization by $z\le 0.03$3, far beyond that systematic allowance. Thermal 1T, 2T, and DEM models all under-predict the $z\le 0.03$4 keV counts, and an inverse-Compton interpretation tied to the radio mini-halo would require $z\le 0.03$5 and is described as unconvincing. The derived 90% upper limit on inverse-Compton flux is $z\le 0.03$6 in 3–25 keV, implying a lower bound on the volume-averaged magnetic field of $z\le 0.03$7 (Creech et al., 2024).

LOFAR high-band observations at 120–168 MHz show that Perseus is also a multi-scale radio laboratory. At 0.3″ resolution, the northern extension of the 3C 84 lobe resolves into several parallel strands 1.5–3 kpc wide, with sub-filaments or “fibers” only 350–500 pc wide. Steep-spectrum filaments with $z\le 0.03$8 to $z\le 0.03$9 fill two known X-ray ghost cavities. At 7″ resolution, the central mini-halo is imaged as a highly asymmetric structure with sharp edges, filaments, and a 170 kpc eastern filament only $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$0 kpc wide; its spectral index is typically $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$1 to $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$2. At 26″–80″ resolution, compact-source subtraction reveals diffuse emission extending to a projected radius of $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$3 kpc, or 1.1 Mpc across, classified as a giant radio halo distinct from the inner mini-halo. The measured flux densities are $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$4 Jy for the 380 kpc mini-halo and $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$5 Jy for the 1.10 Mpc giant halo, with spectral indices $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$6 and $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$7, respectively. A 0.9 Mpc trail from IC 310 bends by $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$8 and connects directly into the giant halo (Weeren et al., 2024). One implication drawn in that work is that cluster-wide radio emission may be more common in cool-core clusters than had often been assumed.

At gamma-ray energies, the MAGIC stereoscopic campaign accumulated 85 hr of data, detecting VHE emission from NGC 1275 and IC 310 while placing limits on diffuse cluster emission. For IC 310, the spectrum from 2 GeV to 7 TeV is consistent with a single power law with $\simeq 2.9\times10^5\,\mathrm{Mpc}^3$9 and integral flux above 300 GeV of $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$0. For NGC 1275, MAGIC measured an excess of $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$1 events above 100 GeV at $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$2, with flux above 100 GeV of about 2.5% of the Crab Nebula and no signal above $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$3 GeV. The non-detection of diffuse cluster emission constrained the volume-averaged CR-to-thermal pressure ratio to $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$4 within the core and $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$5 over the virial region (Lombardi et al., 2011).

Prospective CTA analyses push the cluster further into the multi-messenger regime. For a 300 h CTA-North observation, template fitting predicts approximately $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$6 significance for a baseline CR model and $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$7 for a pure-hadronic best fit. In the absence of a detection, CTA should constrain the CRp-to-thermal energy ratio within $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$8 to $M_{\rm tot}\approx 2.5\times10^{16}\,M_\odot$9 for $z_{\rm cl}=0.01767$0 and $z_{\rm cl}=0.01767$1; in dark-matter decay it should reach $z_{\rm cl}=0.01767$2 s for masses above 1 TeV (Consortium et al., 2023). A 2026 3D MHD + Monte Carlo study further argues that CRs with $z_{\rm cl}=0.01767$3 eV can remain confined over cosmological timescales in Perseus-like clusters, yielding diffuse gamma-ray and neutrino fluxes just below current MAGIC, LHAASO, and IceCube limits but within CTA and next-generation neutrino-telescope discovery space (Hussain et al., 12 May 2026).

4. Compact stellar systems, intracluster populations, and environmentally transformed galaxies

HST imaging of the Perseus cluster core has identified 84 ultra-compact dwarf candidates with half-light radii $z_{\rm cl}=0.01767$4 out to projected radii of $z_{\rm cl}=0.01767$5 kpc. Most are brighter than $z_{\rm cl}=0.01767$6 and occupy the same size-luminosity locus as confirmed UCDs in other nearby environments. The majority lie on an extrapolation of the red sequence followed by Perseus dwarf ellipticals, but three UCD candidates near NGC 1275 have very blue colors, $z_{\rm cl}=0.01767$7, implying cessation of star formation within the past 100 Myr. The study argues that multiple formation channels are plausible in Perseus: in situ formation in the star-forming filaments around NGC 1275, tidal stripping of dwarf galaxies in a shell-rich central environment, and continuity with the high-mass end of the globular-cluster population (Penny et al., 2012).

The cluster also hosts a substantial stellar intracluster medium traced by globular clusters. HST ACS/WFC and WFPC2 photometry found a detectable population of intragalactic globular clusters extending to at least 500 kpc from the cluster center. Their luminosity function is well fitted by a Gaussian with turnover $z_{\rm cl}=0.01767$8 and dispersion $z_{\rm cl}=0.01767$9, and their color distribution is normal for GC systems. Extrapolation from the sampled fields yields a rough estimate of $D_A=74.1$0 or more intracluster GCs for the cluster as a whole. For the major ellipticals, the inferred specific frequencies are $D_A=74.1$1 for NGC 1272 and $D_A=74.1$2 for NGC 1275 (Harris et al., 2017).

Environmental processing is also visible in star-forming cluster members. LOFAR 144 MHz imaging led to the first identification of four jellyfish galaxies in Perseus within the central $D_A=74.1$3 region. Their one-sided radio tails, asymmetric optical morphologies, truncated nebular emission on the leading side, and compact $D_A=74.1$4 sources are consistent with ram-pressure stripping. Disk spectral indices are relatively flat, while tails are systematically steeper; for example, MCG +07-07-070 has $D_A=74.1$5 and $D_A=74.1$6. Estimated ram pressures range from $D_A=74.1$7 to $D_A=74.1$8, and all four galaxies lie a factor of $D_A=74.1$9–3.2 above the standard $M_{200c}=6.65\times10^{14}\,M_\odot$0–SFR relation (Roberts et al., 2021). A common oversimplification is to treat the Perseus environment only through its diffuse ICM and AGN; the UCD, IGC, and jellyfish populations show that galaxy transformation and compact-system assembly are equally prominent aspects of the cluster.

5. The Perseus molecular cloud and the star-forming region

The Perseus molecular cloud complex is a nearby low- to intermediate-mass star-forming region at a distance of roughly 250 pc, spanning about $M_{200c}=6.65\times10^{14}\,M_\odot$1 on the sky and containing total molecular mass $M_{200c}=6.65\times10^{14}\,M_\odot$2. It includes the rich protoclusters NGC 1333 and IC 348 as well as smaller groups in L1448, L1455, B1, and B5 (Arce et al., 2010). Spitzer IRAC + MIPS observations identified 369 YSOs. Of these, 67% are associated with NGC 1333 and IC 348, while the western half of the cloud contains three-quarters of the embedded YSOs. The extinction-corrected census comprises 70 Class 0+I, 32 Flat-spectrum, 231 Class II, and 36 Class III objects, and 91% of Class 0+I + Flat sources lie at $M_{200c}=6.65\times10^{14}\,M_\odot$3 mag, supporting an empirical threshold $M_{200c}=6.65\times10^{14}\,M_\odot$4 mag for protostellar-core formation. The corresponding global star-formation rate estimate is $M_{200c}=6.65\times10^{14}\,M_\odot$5 (Young et al., 2015).

Cloud energetics in Perseus have been quantified with COMPLETE survey CO mapping. A 3D RA-DEC-velocity search found 218 high-velocity spikes, of which 60 satisfied the criteria for COMPLETE Perseus Outflow Candidates, more than doubling the known outflow inventory. After standard opacity and inclination corrections, the total outflow budgets are $M_{200c}=6.65\times10^{14}\,M_\odot$6, $M_{200c}=6.65\times10^{14}\,M_\odot$7, and $M_{200c}=6.65\times10^{14}\,M_\odot$8 erg. These outflows can sustain turbulence locally in most active 1–4 pc regions, but across the entire cloud their kinetic energy is only about 13% of the turbulent reservoir, so they cannot by themselves maintain the cloud-wide motions (Arce et al., 2010). A complementary shell survey identified twelve shells with radii from about 0.1 to 3 pc and argued that quasi-spherical winds with mass-loss rates $M_{200c}=6.65\times10^{14}\,M_\odot$9–$\sim 250$00 are required. Their combined wind luminosity, $\sim 250$01–$\sim 250$02, is comparable to the estimated turbulence-dissipation rate $\sim 250$03 (Arce et al., 2011). The two studies together indicate that collimated outflows and wider-angle winds play distinct roles.

Dust and star-formation scaling relations are unusually well constrained. Herschel-Planck-2MASS mapping produced optical-depth and temperature maps at 36″ resolution in the Herschel-covered regions and 5′ elsewhere, with equivalent $\sim 250$04-band extinction ranging from 0.01 to 20 mag. The low-extinction calibration is

$\sim 250$05

and the cumulative area function follows $\sim 250$06 while the differential form scales approximately as $\sim 250$07. Using an updated Class I/0 catalog, the local Schmidt law was measured as

$\sim 250$08

with the additional conclusion that the area-extinction relation is essential for setting the total star-formation rate (Zari et al., 2015).

Radio and astrometric surveys resolve the region’s younger stellar content and its internal substructure. A multi-epoch VLA survey at 4.5 and 7.5 GHz detected 206 radio sources, 42 of them known YSOs; about 60% of those YSOs show indicators of non-thermal emission, and the fraction rises with evolutionary class from about 25% for Class 0 to about 80% for Class III. For 22 YSOs with X-ray counterparts, the radio/X-ray relation is consistent with a Güdel–Benz scaling with $\sim 250$09 (Pech et al., 2015). Gaia DR2 then identified, besides IC 348 and NGC 1333, five new co-moving groups—Autochthe, Alcaeus, Heleus, Electryon, and Mestor—with ages between 1 and 5 Myr and disk fractions between 10% and 50%; the older groups are located at higher Galactic latitude (Pavlidou et al., 2021). On a larger scale, Gaia DR3 cluster orbits in the Perseus complex indicate that the complex is not dispersing and that some apparent expansion can be a projection effect caused by stars orbiting the Galaxy at different velocities (Croce et al., 22 Apr 2025).

6. Peripheral astronomical and algorithmic uses of the name

Perseus I, also designated Andromeda XXXIII, is a distant dwarf galaxy discovered in Pan-STARRS1 stacked imaging as a $\sim 250$10 overdensity of candidate red-giant-branch stars. Its measured parameters are $\sim 250$11, heliocentric distance $\sim 250$12 kpc, three-dimensional distance from M31 of $\sim 250$13 kpc, absolute magnitude $\sim 250$14, half-light radius $\sim 250$15 pc, ellipticity $\sim 250$16, and central surface brightness $\sim 250$17. Its location far east of M31 was noted as potentially constraining the bulk motion of the M31 satellite system if spectroscopic association were confirmed (Martin et al., 2013).

In decision theory, Perseus is a randomized point-based value-iteration algorithm for POMDPs. For a POMDP $\sim 250$18, the method works on a finite sampled belief set $\sim 250$19 and uses the piecewise-linear convex representation

$\sim 250$20

Its defining feature is that it does not back up every belief point at every stage. Instead, it randomly selects beliefs from the unimproved set, performs a backup, and retains either the new vector or the old maximizer so that the value of each belief in $\sim 250$21 is non-decreasing. Because a single $\sim 250$22-vector often improves many nearby beliefs simultaneously, Perseus typically uses far fewer vectors than methods such as PBVI. The paper reports that on the Tag benchmark Perseus used 280 vectors versus over 1300 in PBVI, while matching or exceeding the solution quality of PBVI, HSVI, and bounded policy iteration on Tiger-grid, Hallway, Hallway2, and Tag. The same improve-only idea extends to continuous action spaces by maximizing over a sampled action set $\sim 250$23 rather than over all actions exactly (Spaan et al., 2011).

Across these uses, “Perseus” does not denote a single physical or formal object. It denotes, rather, a set of systems and methods that have become technically important because they are close enough, structured enough, or geometrically tractable enough to expose otherwise inaccessible phenomena: cluster-core heating and merger memory in Abell 426, cloud-scale feedback and hierarchical star formation in the molecular complex, faint-satellite structure in the M31 system, and scalable approximate planning in high-dimensional belief spaces.