Fermi Bubbles: Galactic Center Outflows

• Fermi Bubbles are giant, bilobate, nonthermal structures extending roughly 10 kpc from the Galactic Center with uniform gamma-ray brightness and sharp edges.
• Multiwavelength observations reveal complementary signatures in gamma-ray, microwave, radio, and X-ray bands that constrain cosmic ray dynamics, magnetic fields, and multiphase gas properties.
• Competing origin models—from long-duration hadronic winds to collimated outbursts and star-formation-driven winds—offer distinct explanations for the bubbles’ energetics, spectral features, and evolution.

The Fermi Bubbles (FBs) are two giant, bilobate, nonthermal structures centered on the Galactic Center (GC), extending to approximately $\pm 10$ kpc above and below the Galactic plane, or to angular heights of about $50^\circ$–$55^\circ$. They are defined observationally by hard gamma-ray emission with sharp edges and nearly uniform surface brightness, and they are closely associated with a microwave “haze,” polarized radio structures, and diffuse X-ray shells. Since their identification in Fermi-LAT data, the FBs have become a focal problem in Galactic feedback, cosmic-ray (CR) transport, and GC activity, with interpretations ranging from long-duration hadronic calorimetry to collimated AGN-like outbursts, star-formation-driven winds, and turbulence-driven in-situ leptonic acceleration (Crocker et al., 2010, Yang et al., 2018).

1. Discovery, geometry, and global phenomenology

The defining observational picture is stable across the literature. The FBs are bilateral lobes symmetric about the GC, with sharp outer edges, little internal substructure, and a remarkably uniform gamma-ray surface brightness. Their total solid angle is roughly $1\ \mathrm{sr}$, and their angular footprint is concentrated toward the GC, with the most intense gamma-ray emission at low Galactic latitudes. In physical terms, the lobes extend to heights of order $10$ kpc; in angular terms, they subtend about $50^\circ$ in latitude and several tens of degrees in width (Crocker et al., 2010, Yang et al., 2018).

Their integrated nonthermal energetics are likewise well constrained. The gamma-ray luminosity is $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$ in the GeV band, while the microwave haze luminosity over $20$–$60$ GHz is $L_{\rm haze}\approx (1$–$50^\circ$0. The diffuse X-ray luminosity associated with the same general region is at least an order of magnitude larger than the gamma-ray luminosity. This luminosity hierarchy is one of the basic empirical constraints on formation models, because any successful picture must reproduce the gamma-ray, microwave, and X-ray outputs simultaneously rather than in isolation (Crocker et al., 2010).

The low-latitude base is not simply a scaled version of the high-latitude lobes. It is brighter and spectrally harder than the higher-latitude FB emission, and it is displaced to negative Galactic longitudes relative to the GC. If that base emission is physically connected to the high-latitude FBs, the westward shift disfavors strictly symmetric SMBH-driven models; alternatively, the base may include a separate component aligned along the same lines of sight (Herold et al., 2019).

2. Spectral structure and multiwavelength counterparts

The gamma-ray spectrum is hard, with $50^\circ$1 as a useful first description, but the detailed phenomenology is latitude-dependent. Integrated analyses find a high-energy exponential cutoff near $50^\circ$2–$50^\circ$3 GeV, whereas the base at $50^\circ$4 remains hard and extends to $50^\circ$5 TeV in Fermi-LAT data without a significant cutoff; the corresponding $50^\circ$6 confidence lower limit on the cutoff energy is about $50^\circ$7 GeV. At higher latitudes, the intensity decreases and the spectrum softens or cuts off around $50^\circ$8 GeV (Dogiel et al., 2024, Xotta et al., 17 Dec 2025, Herold et al., 2019).

Spatially resolved analyses add further structure. A 60-month reanalysis found a robust relative deficit of low-energy gamma-ray flux toward the top of the South Bubble, together with an energy-dependent morphology in which the bubbles are more extended at higher energies. That study also found no spectral softening with longitude in either lobe. Interpreting the high-latitude spectra in either hadronic or leptonic terms requires a low-energy break in the parent CR population (Yang et al., 2014).

The microwave and radio counterparts are not incidental. The WMAP/Planck haze is morphologically coincident with the FBs, especially at low and intermediate latitudes, and polarized synchrotron ridges and filaments outline the lobes and their edges. Joint IC–synchrotron fits in a leptonic framework infer a characteristic magnetic field $50^\circ$9 with a plausible range of $55^\circ$0–$55^\circ$1. In a hadronic framework, secondary $55^\circ$2 can account for a microwave luminosity of order $55^\circ$3 for $55^\circ$4, but the spectral hardness of the haze at high latitudes remains a nontrivial constraint on purely secondary models (Dogiel et al., 2024, Crocker et al., 2010).

The X-ray counterpart is comparably important. ROSAT revealed broad arcs and shells, and eROSITA later traced larger X-ray bubbles extending to $55^\circ$5 kpc in height. Some analyses interpret the FB region as filled by hot, low-density plasma with $55^\circ$6–$55^\circ$7 and $55^\circ$8, while others emphasize shell-like structures with weak or strong shock diagnostics depending on the specific tracer and location. The X-ray morphology therefore constrains both the ambient halo and the dynamical state of the FB edges (Crocker et al., 2010, Dogiel et al., 2024).

3. Multiphase gas, magnetic structure, and the nuclear wind

Radio and UV spectroscopy show that the FBs are not only nonthermal gamma-ray lobes but also a multiphase nuclear-wind system. High-velocity absorption components inside the FBs span typical LSR ranges of roughly $55^\circ$9, while compact H I clouds in the inner $1\ \mathrm{sr}$0 reach velocities of a few hundred $1\ \mathrm{sr}$1. Spectroscopic analyses infer kinematic ages of $1\ \mathrm{sr}$2–$1\ \mathrm{sr}$3 Myr for the nuclear wind, warm-ionized mass outflow rates of $1\ \mathrm{sr}$4–$1\ \mathrm{sr}$5, neutral mass outflow rates of $1\ \mathrm{sr}$6, and a conservative combined lower limit $1\ \mathrm{sr}$7–$1\ \mathrm{sr}$8 for the cool and warm phases alone. A commonly used shell estimator is

$1\ \mathrm{sr}$9

with the hot X-ray phase likely carrying a substantial fraction of the total energy budget (Fox et al., 2019).

The cool clouds embedded in FB directions do not form a chemically homogeneous population. The first metallicity survey of FB high-velocity clouds found values ranging from less than $10$0 solar to about $10$1 solar. Most sampled clouds are sub-solar, but a minority are near-solar or super-solar. This directly challenges the earlier assumption that all cool FB clouds are launched from the GC with solar or super-solar metallicity. Instead, the cloud population appears to have dual origins in both the Milky Way disk and halo, implying that the expanding FB system both entrains disk material and sweeps up circumgalactic gas (Ashley et al., 2022).

Magnetic structure is likewise part of the FB phenomenology rather than a secondary detail. Polarization measurements reveal a partly ordered magnetic-field geometry extending beyond the gamma-ray lobes. One MHD explosive model reproduces this by stretching and twisting a magnetic loop anchored in a rotating molecular torus around the GC, generating a large-scale spiral field consistent with the observed polarized structures. In that scenario, either confinement of gamma-ray-emitting electrons to deeper ejecta regions or consecutive explosive episodes is required to reconcile the larger radio/X-ray extent with the smaller GeV lobes (Barkov et al., 2013).

4. Physical origin scenarios

The origin debate is not a binary choice between “AGN” and “starburst.” The literature contains several physically distinct classes of models, each reproducing some subset of the phenomenology and confronting specific tensions.

Scenario	Core ingredients	Representative tensions or signatures
Saturation-based hadronic wind	Long-term GC star formation, nuclear wind, pp losses, secondaries	Requires multi-Gyr confinement; predicts neutrinos
Mixed hadronic–leptonic model	pp gamma rays plus primary-electron IC/synchrotron	Secondary radio alone is too soft without additional primaries
Routine TDE-driven turbulent shell	Repeated TDEs, RT turbulence, stochastic electron acceleration	Weak outer shock; hadronic-only microwave fit is disfavored
Collimated outburst / jet	Narrow, near-vertical injection, forward shocks	Isotropic injection becomes too spherical
Star-formation Galactic wind	Sustained central SFR, weak forward shock, in-situ electron acceleration	Requires extended episode and CGM conditions consistent with X-rays

In the long-duration hadronic picture, CR protons and heavier ions injected by GC star formation are advected into the bubble volumes and confined for multi-Gyr timescales. The relevant transport equation is

$10$2

with hadronic loss time

$10$3

For $10$4–$10$5, one obtains $10$6–$10$7. In the saturation limit, $10$8 and $10$9, so the observed luminosity follows naturally if $50^\circ$0. In that framework, the FBs are multi-billion-year-old reservoirs of GC CRs, with $50^\circ$1 and $50^\circ$2 as representative values (Crocker et al., 2010).

A major criticism of purely hadronic models is the radio/microwave slope. If secondary electrons cool radiatively through synchrotron and IC, their steady-state spectrum is too soft to reproduce the observed radio index; a mixed model with a primary electron component is then required. One such analysis found that, for a primary spectrum $50^\circ$3, the permitted magnetic-field range is $50^\circ$4–$50^\circ$5, and the pp contribution can reach about $50^\circ$6 of the total gamma-ray flux. Outside that $50^\circ$7 range, the model fails to reproduce the radio and gamma-ray data simultaneously. The same work noted that a purely hadronic solution could be recovered only if adiabatic losses dominate, but then the required magnetic field strength and CR source power become much higher than those followed from observations (Cheng et al., 2014).

A different class of models treats the FBs as the product of repeated TDEs at Sgr A*. In one recent implementation, TDEs at a rate $50^\circ$8–$50^\circ$9 and energies $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$0–$$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$1 supply an average power $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$2. The outer shock is weak, with $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$3 and Mach number $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$4, so the observed nonthermal emission is attributed to RT-driven turbulence, MHD fluctuations via the Lighthill mechanism, and stochastic electron re-acceleration in a thin shell. That model fits the gamma-ray spectrum with $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$5 and a leptonic electron cutoff $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$6, while finding that a purely hadronic origin is disfavored because the secondary synchrotron is too soft and too faint by a factor $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$7–$$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$8 to account for the microwave haze (Dogiel et al., 2024).

Collimated-outburst models push in a different direction. If the FB edges are treated as strong forward shocks, isotropic or non-directed energy release becomes too round and too wide once the shock expands into the halo. In contrast, a narrow jet launched nearly perpendicular to the Galactic disc can reproduce the thin, elongated edges. One such study derived favored ballistic parameters $$L_\gamma \approx 4\times 10^{37}\ \mathrm{erg\ s^{-1}}$$9, $20$0, $20$1, and age $20$2, with slowdown solutions requiring a ballistic stage reaching $20$3. A related MHD explosive scenario found that a few-Myr-old event, or consecutive episodes, can also explain the large-scale magnetic structure, with total injected energy $20$4 and a lifetime estimate $20$5 (Mondal et al., 2021, Barkov et al., 2013).

Star-formation-driven wind models remain viable as well. In one hydrodynamical realization, an episode of star formation with $20$6 for $20$7 Myr drives a Galactic wind of mechanical luminosity $20$8 into a hot CGM with $20$9–$60$0. The forward shock is weak, with Mach number $60$1, and the gamma rays and microwave haze are attributed to in-situ accelerated electrons in a leptonic IC/synchrotron framework; in that model the hadronic contribution is $60$2 of the observed gamma-ray signal (Sarkar et al., 2015).

5. High-energy and multimessenger diagnostics

Hadronic models generically predict neutrinos, and the predicted fluxes are highly sensitive to the proton index and cutoff. One IceCube analysis showed that, if the FB gamma rays are hadronic and the parent proton spectrum is hard, with $60$3 and $60$4 PeV, then up to about $60$5–$60$6 of IceCube’s $60$7 three-year HESE events could originate from the FBs. In the same framework, a 10-year exposure at $60$8 TeV and $60$9 PeV yields about $L_{\rm haze}\approx (1$0 FB signal events versus about $L_{\rm haze}\approx (1$1 atmospheric events, corresponding to a $L_{\rm haze}\approx (1$2 excess by simple counting; for $L_{\rm haze}\approx (1$3 or lower cutoffs, the significance drops sharply (Lunardini et al., 2014).

Earlier hadronic estimates were even more optimistic for idealized conditions. For a proton spectrum $L_{\rm haze}\approx (1$4 with a cutoff at $L_{\rm haze}\approx (1$5 PeV, one prediction gave $L_{\rm haze}\approx (1$6 signal events per year above $L_{\rm haze}\approx (1$7 TeV in a Northern Hemisphere $L_{\rm haze}\approx (1$8-class detector against $L_{\rm haze}\approx (1$9 background events, or about $50^\circ$00 in one year. Any steeper spectrum or lower cutoff strongly suppresses that rate. A low neutrino flux, conversely, strengthens the case for predominantly leptonic emission (Crocker et al., 2010).

The MeV band is a particularly sharp hadronic–leptonic discriminator because the two scenarios diverge strongly there. A hadronic origin predicts the broad $50^\circ$01-decay feature in the $50^\circ$02–$50^\circ$03 MeV range, whereas a leptonic IC origin predicts a smooth continuum through the MeV gap. One forecast concluded that COSI can detect a hadronic component from the FBs at approximately $50^\circ$04–$50^\circ$05 MeV for magnetic fields $50^\circ$06, or set informative upper limits below about $50^\circ$07 MeV. The same study found that a survey Compton–pair telescope such as AMEGO-X would detect a pure leptonic IC component across almost the entire MeV gap and could identify spectral breaks in hybrid scenarios (Negro et al., 2021).

At TeV energies, present constraints are already informative. H.E.S.S. and HAWC have not detected the FBs, and the low-latitude H.E.S.S. limits suggest a possible cutoff near $50^\circ$08 TeV. CTAO simulations indicate that the low-latitude base is the highest-value target: with the $50^\circ$09-hour GC survey, CTAO-South should recover the base spectrum from $50^\circ$10 GeV up to $50^\circ$11 TeV for an injected power law with exponential cutoff at $50^\circ$12 TeV, and it should distinguish a PLEC from a pure power law at $50^\circ$13 for true cutoff energies in the approximate range $50^\circ$14 TeV to $50^\circ$15 TeV, depending on spectral index. Even a $50^\circ$16-hour exposure is expected to detect the base from $50^\circ$17 GeV to $50^\circ$18 TeV (Xotta et al., 17 Dec 2025).

6. Current synthesis, disputes, and open problems

Several points now appear robust. The FBs are GC-centered, sharp-edged, multiwavelength structures coupled to a multiphase nuclear wind. They are associated with hot X-ray-emitting plasma, warm and neutral high-velocity clouds, synchrotron-emitting electrons, and a hard GeV gamma-ray component. Spectroscopic work has shown that the cool gas is chemically diverse rather than uniformly metal-rich, so the FB system must interact dynamically with both disk material and pre-existing halo gas (Fox et al., 2019, Ashley et al., 2022).

The central disputes concern age, engine, and dominant radiation channel. Age estimates range from $50^\circ$19 yr in some explosive MHD scenarios, to a few Myr in collimated jet models, to $50^\circ$20–$50^\circ$21 Myr from spectroscopy of the nuclear wind, to $50^\circ$22–$50^\circ$23 Myr in some star-formation wind or X-ray-shell interpretations, and up to $50^\circ$24 yr in saturation-based hadronic calorimetry (Barkov et al., 2013, Mondal et al., 2021, Fox et al., 2019, Sarkar et al., 2015, Crocker et al., 2010). These are not minor parameter shifts; they encode fundamentally different transport regimes and different meanings of the gamma-ray luminosity.

There is also no single uncontested view of the edges. Some studies infer weak shocks, with $50^\circ$25 and Mach number $50^\circ$26, whereas others model the edges explicitly as strong forward shocks with Mach number $50^\circ$27–$50^\circ$28. A recent nested-bubble interpretation extends the strong-shock picture to both the FBs and the larger eROSITA bubbles, arguing that both structures can be produced by very similar collimated GC outbursts, each with $50^\circ$29 erg, half-opening angle $50^\circ$30, and launch speed near $50^\circ$31 at $50^\circ$32 pc, separated by $50^\circ$33 Myr. In that framework, the east–west asymmetry favors an eastern ambient-density gradient rather than a western crosswind (Dogiel et al., 2024, Mondal et al., 2021, Ghosh et al., 30 Jan 2026).

A further misconception corrected by recent work is that the microwave haze automatically validates a purely leptonic or purely hadronic model. In reality, the microwave spectrum and normalization are among the strongest discriminants. Purely hadronic models can match the gamma rays only if they also solve the radio problem, and purely leptonic models must still reproduce the luminosity ratios, edge sharpness, latitude dependence, and cooling constraints without excessive fine-tuning. This is why MeV spectroscopy, TeV measurements of the base, spatially resolved X-ray spectroscopy, and neutrino searches remain decisive rather than supplementary (Cheng et al., 2014, Negro et al., 2021, Xotta et al., 17 Dec 2025).

The FBs therefore remain a composite problem in CR transport, shock and turbulence physics, GC feeding and feedback, and multiphase halo structure. What is established is the phenomenology: giant bilateral lobes, hard gamma rays, microwave synchrotron, X-ray shells, sharp boundaries, and chemically diverse high-velocity clouds. What remains unsettled is whether the dominant engine is a long-lived nuclear wind, repeated TDEs, a collimated AGN-like outburst, or some hybrid sequence, and whether the gamma-ray power is mainly hadronic, mainly leptonic, or mixed in an energy- and latitude-dependent way.