Local Analogs of Little Red Dots: Clues to the X-ray Weakness

abstract · abstract

A major question about JWST's little red dots (LRDs) at $z\sim4-8$ is the origin of their apparent X-ray weakness. Even deep Chandra stacks yield only tentative detections or broad-band limits and cannot distinguish among super-Eddington accretion, intrinsic X-ray weakness, and Compton-thick (CT) obscuration. Local analogs offer a powerful complementary route. At $z\sim0.3$, similarly luminous sources require $\sim10^{3-4}$ less exposure for comparable signal-to-noise, suffer less source confusion, and directly probe the diagnostic rest-frame soft X-rays. Here we present a pilot Einstein Probe/FXT study of six "local little red dots" (LLRDs), V-shaped broad-line green pea galaxies selected as low-redshift counterparts of LRDs. These sources show X-ray diversity: "X-normal" LLRDs resemble type 1 AGNs in X-ray luminosity and optical variability, whereas the "X-weak" stack lies $\sim0.5-0.7$ dex below standard unobscured-quasar scaling relations while remaining compatible with the representative LRD stack and 95% of published individual limits. The X-weak LLRDs also lack detected optical variability and have redder optical continua than X-normal LLRDs, making their V-shaped SEDs closer to those of high-redshift LRDs. The detection of the X-weak stack at $\sim3.3σ$ with the soft-X-ray-sensitive EP-FXT provides rest-frame $\sim2-3$ keV constraints difficult to reconcile with a typical unobscured super-Eddington model or a local-AGN-like CT geometry examined here, although modified variants remain possible. The simplest alternative is a mildly X-ray-weak corona ($Δα_{\mathrm{ox}}\sim-0.1$ to $-0.2$) with Compton-thin absorption ($N_{\mathrm{H}}\sim10^{22-23}$ cm$^{-2}$), consistent with the independent Balmer decrement/break diagnostics. Together with growing samples of local LRD analogs, our results offer a practical path for testing the physical nature of LRD X-ray weakness.

S4: IV Summary and future prospects / block 1 · html

Targeted EP-FXT observations of six LLRDs demonstrate how local analogs can bypass the high-redshift LRD X-ray bottleneck. Using ∼ 1 / 1000 \sim\nicefrac{{1}}{{1000}} of the exposure for current LRD stacks, we obtain constraints tighter than those available for 95 % 95\text{\,}\% of individual LRDs with measured L 2500 L_{2500} while directly probing rest-frame soft X-rays inaccessible to Chandra at z ∼ 6 z\sim 6 .

S4: IV Summary and future prospects / block 1 · html

The observations reveal two X-ray classes. Two X-normal LLRDs exhibit normal to moderately enhanced X-ray emission, optical variability, and bluer optical continua. The X-weak subset, which most closely resembles high-redshift LRDs in optical/UV SED shape, Balmer diagnostics, and lack of detected variability, lies ∼ 0.65 dex \sim$0.65\text{\,}\mathrm{d}\mathrm{e}\mathrm{x}$ below the standard unobscured-quasar L 2 ​ k ​ e ​ V int L_{\mathrm{2keV}}^{\mathrm{int}} – L 2500 L_{2500} relation. Stacking three X-weak LLRDs yields a 3.3 ​ σ 3.3\sigma detection in the rest-frame 2 – 8 keV 28\text{\,}\mathrm{k}\mathrm{e}\mathrm{V} band, driven mainly by 2 – 3 keV 23\text{\,}\mathrm{k}\mathrm{e}\mathrm{V} photons. The rest-frame 0.5 – 2 keV 0.52\text{\,}\mathrm{k}\mathrm{e}\mathrm{V} band is not significantly detected ( 1.9 ​ σ 1.9\sigma ). The moderate deficit measured from the X-weak LLRD stack remains compatible with current individual and stacked LRD constraints, showing that extreme X-ray weakness need not be universal.

A1: Appendix A LLRD EP-FXT observation summary / block 1 · html

Table 2 summarizes the pointed EP-FXT observations obtained through our program (PI: J.-X. Wang) for the six LLRDs analyzed in this work. The exposure times were planned to yield at least a 3 ​ σ 3\sigma detection for an intrinsic 2 – 10 keV 210\text{\,}\mathrm{k}\mathrm{e}\mathrm{V} luminosity 0.5 dex 0.5\text{\,}\mathrm{d}\mathrm{e}\mathrm{x} below that expected from the standard broad- H ​ α \mathrm{H}\alpha AGN relation. J0850 was assigned a shorter exposure because its broad- H ​ α \mathrm{H}\alpha luminosity is approximately 1 dex 1\text{\,}\mathrm{d}\mathrm{e}\mathrm{x} lower than those of the other targets; for this source, the observational goal was to improve upon the archival eROSITA upper limit rather than to secure an individual detection.

A2: Appendix B eFXTDAS : enhanced data analysis tools for EP-FXT / block 1 · html

Figure 5: Example eFXTDAS quality-assessment summary for J115438.96+065025.0, generated with fxtcombine-quickview . From upper left to lower right, the panels show the stacked count-rate map and detected-source catalog, modeled background map, target-centered extraction geometry (overlaid with contaminating sources in the source/background extraction regions), stacked exposure map, 90%-encircled-energy PSF-radius map, and limiting-flux sensitivity map.

S4: IV Summary and future prospects / block 1 · html

The rest-frame soft-X-ray constraints are difficult to reconcile with either the typical unobscured super-Eddington model or the local-AGN-like CT geometry examined here, although absorbed super-Eddington spectra with stronger intrinsic X-ray emission, or CT configurations with substantially enhanced scattering or low-column-density leakage cannot be excluded. The simplest explanation consistent with both the X-ray and Balmer diagnostics is a mildly X-ray-weak corona, with Δ ​ α ox ∼ − 0.1 \Delta\alpha_{\mathrm{ox}}\sim-0.1 to − 0.2 -0.2 , viewed through Compton-thin material with N H ∼ 10 22 − 23 ​ cm − 2 N_{\mathrm{H}}\sim 10^{22-23}\,\mathrm{cm}^{-2} . Deeper X-ray observations and growing samples of local LRD analogs will test whether this interpretation is representative and clarify the diversity and physical origin of LRD X-ray emission beyond what is currently feasible at high redshift.

A1: Appendix A LLRD EP-FXT observation summary / block 1 · html

Table 2: Einstein Probe/FXT observation summary for the LLRDs

A1: Appendix A LLRD EP-FXT observation summary / block 1 · html

Note. — Dates are given in UTC. The exposure is the module-averaged value, ( t FXTA + t FXTB ) / 2 (t_{\mathrm{FXTA}}+t_{\mathrm{FXTB}})/2 , summed over the listed ObsIDs; it therefore represents the effective on-source exposure rather than the module-summed collecting time. “FF” and “01” denote the full-frame observing mode and thin-filter configuration, respectively.

article / block 1 · html

A major question about JWST’s little red dots (LRDs) at z ∼ 4 − 8 z\sim 4-8 is the origin of their apparent X-ray weakness. Even deep Chandra stacks yield only tentative detections or broad-band limits and cannot distinguish among super-Eddington accretion, intrinsic X-ray weakness, and Compton-thick (CT) obscuration. Local analogs offer a powerful complementary route. At z ∼ 0.3 z\sim 0.3 , similarly luminous sources require ∼ 10 3 − 4 \sim 10^{3-4} less exposure for comparable signal-to-noise, suffer less source confusion, and directly probe the diagnostic rest-frame soft X-rays. Here we present a pilot Einstein Probe/FXT study of six “local little red dots” (LLRDs), V-shaped broad-line green pea galaxies selected as low-redshift counterparts of LRDs. These sources show X-ray diversity: “X-normal” LLRDs resemble type 1 AGNs in X-ray luminosity and optical variability, whereas “X-weak” stack lies ∼ \sim 0.5 – 0.7 dex 0.50.7\text{\,}\mathrm{d}\mathrm{e}\mathrm{x} below standard unobscured-quasar scaling relations while remaining compatible with the representative LRD stack and 95 % 95\text{\,}\% of published individual limits. The X-weak LLRDs also lack detected optical variability and have redder optical continua than X-normal LLRDs, making their V-shaped SEDs closer to those of high-redshift LRDs. The detection of the X-weak stack at ∼ 3.3 ​ σ \sim 3.3\sigma with the soft-X-ray-sensitive EP-FXT provides rest-frame ∼ 2 − 3 ​ keV \sim 2-3\,\mathrm{keV} constraints difficult to reconcile with a typical unobscured super-Eddington model or a local-AGN-like CT geometry examined here, although modified variants remain possible. The simplest alternative is a mildly X-ray-weak corona ( Δ ​ α ox ∼ − 0.1 \Delta\alpha_{\mathrm{ox}}\sim-0.1 to − 0.2 -0.2 ) with Compton-thin absorption ( N H ∼ 10 22 − 23 ​ cm − 2 N_{\mathrm{H}}\sim 10^{22-23}\,\mathrm{cm}^{-2} ), consistent with the independent Balmer decrement/break diagnostics. Together with growing samples of local LRD analogs, our results offer a practical path for testing the physical nature of LRD X-ray weakness.

S2.SS2: II.2 Benchmarking LLRD X-ray emission against local AGNs and LRDs / block 1 · html

To benchmark LLRDs against AGNs and high-redshift LRDs, we fit their spectra with Xspec [ 5 ] using an absorbed AGN power-law plus fixed host-galaxy emission, attenuated by Galactic absorption: TBabs*(zTBabs agn *zpowerlw agn +host) . For X-weak LLRDs, we adopt N H = 10 22 ​ cm − 2 N_{\mathrm{H}}=10^{22}\,\mathrm{cm}^{-2} and Γ = 1.8 \Gamma=1.8 , consistent with Yue et al. [97] , and fit only the normalization. For the X-normal LLRDs, their higher SNR allows N H N_{\mathrm{H}} and, where warranted, the photon index to vary; J0928 additionally requires partial covering. Table 1 reports the fitted parameters; Appendix E presents details and the X-normal spectra (Figure 6 ). The host component includes X-ray binaries and hot gas normalized from the stellar mass and narrow- H ​ α \mathrm{H}\alpha -based SFR [ 57 , 72 , 51 , 65 ] . Its luminosity is only ∼ 10 40 ​ erg ​ s − 1 \sim 10^{40}\,\mathrm{erg}\,\mathrm{s}^{-1} , and even a dex uncertainty changes the reported nuclear luminosities or upper limits by < 5 % <5\% .

S2.SS3: II.3 Emergent SED and the Balmer break / block 1 · html

The observed monochromatic luminosity for a given inclination i i is obtained by integrating the emergent intensity over the visible surfaces of the near and far (relative to the observer) sides of the disk (see Appendix B ). The resulting SED for an SMBH with M BH = 10 7 ​ M ⊙ M_{\mathrm{BH}}=10^{7}M_{\odot} and m ˙ = 1000 \dot{m}=1000 is shown in Figure 4 . The spectrum is well-approximated by a blackbody with T eff = 6000 T_{\mathrm{eff}}=6000 K, and the Balmer break is clearly visible in high-inclination cases with i ≳ 60 ∘ i\gtrsim 60^{\circ} . This is due to severe geometric obscuration of the inner disk at large angles. The high-temperature UV emission is blocked by the “puffed-up” disk, making the Balmer break from the cooler outer radii more visible.

S2.SS4: II.4 The broad emission lines / block 1 · html

If the BLR clouds follow a disk-like Keplerian velocity field, a high inclination angle will naturally correspond to a large line-of-sight velocity, resulting in broader lines. Our model therefore predicts a positive correlation between Balmer break strength and the full width at half maximum (FWHM) of the BELs, as tentatively observed by [ 82 ] . Consequently, single-epoch black hole mass estimators calibrated on low-inclination type-1 AGNs may introduce an inclination-dependent bias when applied to LRDs. Assuming a mean inclination of 30 ∘ 30^{\circ} for type-1 AGNs and 75 ∘ 75^{\circ} for LRDs, the inclination effect induces an overestimation factor of ( sin ⁡ 30 ∘ / sin ⁡ 75 ∘ ) 2 ≈ 3.7 (\sin 30^{\circ}/\sin 75^{\circ})^{2}\approx 3.7 . For LRDs with Balmer breaks, the rest-frame 5100 ​ A ̊ 5100\ \mathring{\mathrm{A}} luminosity may be a useful empirical proxy for black hole mass.

S3: III Results / block 1 · html

By co-adding the X-ray counts across all LRDs, we obtained 1444 (3730) total counts in the soft ( 0.3 − 2 0.3-2 keV) and hard ( 2 − 7 2-7 keV) band and, after normalizing the background to the same extraction area, we got 1409 (3697) background counts in the soft and hard bands, respectively. This yields 35 ± 41 35\pm 41 net counts in the soft band and 33 ± 66 33\pm 66 in the hard band, corresponding to non-detections. In the absence of detections, we derive upper limits on the flux of the stacked LRD population. To this end, we adopt the exposure maps and convert the count rates into fluxes, obtaining 3 ​ σ 3\sigma upper limits of < 4.7 × 10 − 18 <4.7\times 10^{-18} erg/s/cm 2 in the soft and < 1.3 × 10 − 17 <1.3\times 10^{-17} erg/s/cm 2 in the hard band. We note that the choice of photon index adopted to convert the count rates into fluxes affects these latter by only ≈ 20 % \approx 20\% , leaving our main results unchanged.

S1: I Introduction / block 1 · html

This paper is organized as follows. In Section II we describe our procedure to analyze (stack) the X-ray data; in Section III we present our results; in Section IV we discuss the implications of our X-ray non-detections and discuss the compatibility of our results with current models of X-ray weakness, and we finally draw our conclusions in Section V . In this work we adopt a flat Λ \Lambda CDM cosmology with H 0 = 70 H_{0}=70 km/s/Mpc and Ω M = 0.3 \Omega_{M}=0.3 .

S5: V Summary and Conclusions / block 1 · html

A morphological analysis on a subset of LoRDs with HSC coverage shows that the majority of sources ( > 80 % >80\% ) are compact regardless of their spectral shape (see Section IV.1 ). We find that LoRDs with a V-shaped continuum are more likely to have a resolved morphology compared to sources without the V-shape. This suggests that some of the V-shaped spectra ( ∼ 40 % \sim 40\% ) are the result of an underlying stellar population and thus are unlikely to serve as proper LRD analogs. The remaining ∼ 60 % \sim 60\% of V-shaped LoRDs are proper LRD analog candidates, but space-based imaging is necessary to quantify their compactness.

S5: V Summary and Conclusions / block 1 · html

In this paper we have shown that the JWST filters used to select LRDs at z ∼ 5 z\sim 5 align well with the SDSS filters at both z = 0.38 z=0.38 and z = 0.78 z=0.78 . We leverage this mapping to photometrically select Local Red Dots (LoRDs) from SDSS using selection criteria similar to the red1/red2 criteria from Greene et al. [45] ; our selection is given in Equations 2 and 3 . In total we find ∼ 1300 \sim 1300 QSOs which share similar photometric colors to the LRDs (Section II.1 ). While not every LoRD we identify serves as a proper analog to LRDs, this is the first step towards studying LRD-like systems in the local Universe at the population level where archival data is plentiful.

S4: 4 Results / block 1 · html

The differences between fluxes and luminosities computed with an unabsorbed power law spectrum and with a heavily absorbed one with N H = 2 × 10 24 N_{H}=2\times 10^{24} or N H = 10 25 N_{H}=10^{25} cm − 2 \mathrm{cm^{-2}} are of the order of ∼ 20 − 30 % \sim 20-30\% in the SB and minimal in the MB and UHB, while are negligible between the two CT cases; therefore, we report only the value for the N H = 2 × 10 24 N_{H}=2\times 10^{24} CT case in 2 .

S4: 4 Results / block 1 · html

We note that for the 1-4 keV band, a fraction of the rest-frame band moves below the Chandra low energy limit of 0.3 keV for sources at z ≳ 2 z\gtrsim 2 (almost all our sources). When computing fluxes and luminosities, we must consider that only part of the intrinsic spectrum will be included in the observed band. We correct for this effect by computing the fraction of ’missing counts’ for each source for a given spectral shape. For example, if 70% of the intrinsic photons are expected to be missing below 0.3 keV, to recover the correct count rate, the exposure time is decreased by 70%. This correction is larger for the unobscured power-law spectrum - a factor ∼ 2 \sim 2 on the final stacked flux and luminosity values - than for the Compton thick obscuration case, since the fraction of soft photons is much larger in the first case.

S4: 4 Summary and discussions / block 1 · html

Our collision-based BH formation scenario provides a natural explanation for the detection of over massive BHs and reconciles their presence with the reported weakness of X-ray signals without requiring high levels of obscuration. Indeed, for our most extreme case, where all the BHs are continuously accreting and emitting 30 % 30\% of their Eddington luminosity in X-rays, at solar metallicities a column density of only N ave ∼ 3 × 10 18 N_{\rm ave}\sim 3\times 10^{18} cm -2 is required to be compatible with the observational constraints, while at lower metallicities ( Z = 10 − 2 Z=10^{-2} Z ⊙ ) a column density around N ave ∼ 2 × 10 21 N_{\rm ave}\sim 2\times 10^{21} cm -2 is enough to reduce the X-ray emission. However, as explored in Appendix C , the subsequent growth should then be well-regulated to be compatible with the X-ray constraints, at least for the typical population. Indeed, a moderate sub-Eddington accretion rate ( L / L Edd ≈ 0.1 L/L_{\rm Edd}\approx 0.1 ) —consistent with typical Type 1 AGNs—provides a plausible explanation to consider the existence of accretion. Nonetheless, we also point out that JWST has even found some more peculiar sources with BH to bulge mass ratios of order 40 % 40\% ( Juodžbalis et al., 2024a ) . In such cases, the combination of high seed masses plus very efficient accretion could be a promising way to explain the extreme outcomes in the more extreme sources. More extreme scenarios in which β \beta takes on higher values ( β ≳ 1.2 \beta\gtrsim 1.2 ), resulting in significantly earlier seed formation times and consequently altering the requirements for the subsequent accretion history and duty cycles, are discussed in detail in Appendix C .

A3: Appendix C Effect of accretion in final black hole masses / block 1 · html

We emphasize that the conclusion above—that moderate sub-Eddington accretion is sufficient—relies on the adoption of a collision-favorable evolutionary track ( α = 0.7 , β = 1.4 \alpha=0.7,\beta=1.4 ). In this regime, the BH seed forms with a significant fraction of the final mass. In contrast, for ’standard’ galaxy evolution scenarios (e.g., β ≈ 0.6 \beta\approx 0.6 , typical of local spirals), the collisional seeds are significantly smaller ( M seed ≪ M final M_{\rm seed}\ll M_{\rm final} ). Consequently, these scenarios require substantial mass growth via gas accretion to match the observed LRD BH masses. This implies a need for higher time-averaged accretion rates (higher k k ) or duty cycles approaching unity. As shown in our main analysis (Fig. 11 ), these accretion-dominated scenarios would result in X-ray fluxes that exceed the stacking upper limits unless extreme obscuration is invoked. Thus, the X-ray non-detections indirectly favor the ’heavy seed’ evolutionary paths where the duty cycle can remain low.

S4: IV AGN Origin of Local LRDs / block 1 · html

It is also useful to compare our results with previous variability studies of local LRDs. [ 11 ] investigated the optical variability of three local LRDs reported in [ 36 ] using light curves spanning ∼ 5 \sim 5 yrs and reported weak intrinsic variability . In contrast to our approach, they infer the intrinsic variability by modeling light curves that have relatively large photometric uncertainties. As their sources have g g -band magnitudes comparable to ours, the S/N of the light curves is similarly limited. In such a low S/N regime, intrinsic variability estimates can be sensitive to the treatment of photometric uncertainties and assumed variability models. Therefore, differences in methodology may contribute to the discrepancy between the inferred variability amplitudes.

S5.SS1: V.1 Mock Observation / block 1 · html

Although we adopt a specific epoch set corresponding to a particular LRD, this source has the longest baseline in the sample of [ 31 ] . We verify that using epoch sets drawn from other LRDs with shorter baselines yields nearly identical results (see Appendix A for details). This indicates that the dominant uncertainty arises from a short baseline of currently limited observation, rather than from the details of any individual object.

S2: 2 Sample presentation and Analysis / block 1 · html

Table 1 : Summary of the AGN subsamples used in this work, including selection criteria, sample size, and the method adopted to derive key physical parameters. 2 2 2 References. (1) Mejía-Restrepo et al. (2022) ;(2) Ricci et al. (2017) ; (3) Gupta et al. (2024) ; (4) Wu and Shen (2022) ; (5) Webb et al. (2020) ; (6) Jin et al. (2012b) ; (7) Du et al. (2014) ; (8) Du et al. (2015) ; (9) Castelló-Mor et al. (2016) ; (10) Tortosa et al. (2023) ; (11) Yue et al. (2024) ; (12) Maiolino et al. (2025) ; (13) Lin et al. (2026) .

S2: 2 Sample presentation and Analysis / block 1 · html

We note that in low-redshift comparison samples (BASS, SDSS–4XMM, NLS1s, and SEAMBHs), all sources have both broad H α \alpha and X-ray measurements as a natural consequence of the sample construction and survey depths, rather than as a result of additional selection cuts applied in this work. In particular, the SDSS–4XMM sources are optically classified as type I AGN based on the presence of a broad H α \alpha line component in their SDSS spectra, with no similarly broad [OIII] counterpart, while their X-ray detections arise from archival XMM- Newton observations of sufficient depth at low z z . Similarly, BASS AGN are bright, nearby systems with well-characterized optical broad-line emission and X-ray detections. In contrast, the LRDs and JWST-selected AGN are primarily selected via rest-frame UV and optical spectroscopy, and their X-ray measurements are almost exclusively limited to upper limits due to their observed X-ray weakness and current sensitivity constraints. Indeed, if their spectra are intrinsically steep (i.e., very soft), then at high z z much of the emission is redshifted out of the observed band, leading to a substantial loss of detected photons with current X-ray instruments. This difference reflects observational limitations rather than intrinsic selection biases and should be borne in mind when comparing the distributions.

S2: 2 Failed attempts to detect the radio emission of JWST-discovered AGN / block 1 · html

Figure 1: Visual representation of the series of failed attempts to detect the radio continuum emission in AGN discovered by JWST via stacking experiments. Blue points represent the results from stacking experiments of photometrically selected LRDs, while red markers indicate samples of spectroscopically confirmed Type I AGN. The average redshift of the samples are 5 < z < 6 5<z<6 .

S2: 2 Failed attempts to detect the radio emission of JWST-discovered AGN / block 1 · html

Table 1: Summary table of the stacking attempts performed so far to detect JWST-discovered AGN and LRDs in radio images. The first column refers to the literature work in which the stacking was performed, the second is ’S’ if the sample was spectroscopic and ’P’ if photometric, the third refers to the number of objects in the stack, the fourth to the frequency of the radio image, the fifth the the 3 ​ σ 3\sigma upper limit obtained from the stack.