Galactic Center Excess Overview

• Galactic Center Excess is an extended gamma-ray feature in the Milky Way’s center with a spectrum peaking at a few GeV and a morphology well described by generalized NFW profiles.
• Methodologies reveal that background-model systematics can alter its measured spectral normalization by up to 60%, highlighting challenges in precise characterization.
• Interpretations range from dark matter annihilation to unresolved millisecond pulsar populations, with recent studies favoring composite models to explain the emission.

The Galactic Center Excess (GCE) is an extended gamma-ray emission component in the central region of the Milky Way, observed in Fermi Large Area Telescope data and characterized by a spectrum that peaks at a few GeV. Its interpretation remains contested because its spectral shape, radial profile, centroid, and symmetry can be described either by centrally concentrated smooth templates motivated by annihilating dark matter or by unresolved stellar-population tracers associated with the Galactic bulge, especially millisecond pulsars (MSPs). The GCE has therefore become a focal problem in gamma-ray astrophysics, indirect dark-matter searches, Galactic diffuse-emission modeling, and population-synthesis studies (Mauro, 2021, Ploeg, 2021, Mauro, 21 May 2026).

1. Observational phenomenology

The GCE is robustly detected as a GeV-peaked excess toward the inner Galaxy. Analyses using 11 years of Fermi-LAT data found that its spectral energy distribution is bumpy, peaks at a few GeV, and is well fit by a log-parabola, while the excess is significantly detected over $0.6$–$30$ GeV and extends significantly out to about $12^\circ$ from the Galactic center (Mauro, 2021). A later analysis designed to reduce interstellar-emission and source-modeling systematics reported that the updated spectrum from $0.5$ to $1000$ GeV again confirms a peak at a few GeV and yields only upper limits above tens of GeV at roughly $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$ (Mauro, 21 May 2026).

Several measured properties recur across studies. The GCE centroid is very close to the Galactic center, with best-fit positions in the range $l=[-0.3^\circ,0.0^\circ]$, $b=[-0.1^\circ,0.0^\circ]$, and quadrant-based fits show similar spectra in different regions around the center (Mauro, 2021). At the same time, the inferred normalization is not fully stable: changing interstellar emission models (IEMs), event selections, and analysis choices can shift the measured GCE spectral normalization by roughly $60\%$, making diffuse-model systematics the dominant limitation for interpretation (Mauro, 2021).

The morphology is usually summarized with generalized Navarro-Frenk-White-like templates. In one precision study the reconstructed surface-brightness profile is well described by an approximately spherical generalized Navarro-Frenk-White morphology with inner slope $\gamma \simeq 1.15$, whereas an 11-year analysis found typical best-fit slopes $30$0–$30$1 with a global average $30$2 (Mauro, 21 May 2026, Mauro, 2021). No significant energy evolution of the morphology was measured in $30$3–$30$4 GeV at a level larger than about $30$5 of the average $30$6, which is itself around $30$7 (Mauro, 2021).

2. Morphology and the spatial-template controversy

The central controversy in the GCE literature is whether the excess is better described by a roughly spherical, centrally concentrated component or by stellar-bulge-tracing templates. One influential analysis using hydrodynamical gas modeling found that the gamma rays are statistically better described by the stellar over-density in the Galactic bulge and the nuclear stellar bulge than by a spherical excess, with the X-bulge and nuclear bulge improving the fit more than an NFW-squared dark-matter template and rendering the NFW component not significantly detected once the bulge templates are included (Macias et al., 2016). In that framework, the non-spherical morphology was taken as evidence against a purely dark-matter interpretation.

Other analyses reach a different conclusion. A 2021 study using 11 years of Fermi-LAT data concluded that the GCE is compatible with a spherical symmetric morphology, with an ellipsoidal fit yielding a major-to-minor axis ratio between $30$8 and $30$9, and found that the dark-matter-like template fits better than the boxy bulge template by roughly $12^\circ$0–$12^\circ$1, depending on the IEM (Mauro, 2021). A later precision measurement likewise reported that bulge-tracing templates consisting of the nuclear bulge plus boxy bulge fail to reproduce the full radial morphology, most notably around $12^\circ$2–$12^\circ$3 and at $12^\circ$4, while the DM-motivated component provides a good description over the full angular range and remains highly significant even when the bulge templates are fit simultaneously (Mauro, 21 May 2026).

Masking studies sharpen but do not eliminate the ambiguity. Using different point-source and disk masks, one analysis found that above $12^\circ$5–$12^\circ$6 GeV the GCE morphology systematically favors an approximately spherical shape, with $12^\circ$7 and $12^\circ$8 for regular masks, while the spherical NFW template generally outperforms the Boxy Bulge, Boxy Bulge plus Nuclear Bulge, and X-shaped Bulge (Zhong et al., 2024). The exception is the stellar bulge profile from Coleman et al. (2020), which provides a similar fit; in fact, a two-component model combining a spherical NFW component with the Coleman Bulge outperforms any tested single component, although the fractional contribution of each component remains background-model dependent (Zhong et al., 2024).

The morphology debate has also expanded beyond purely spherical halos. A 2026 study of triaxial and tilted dark-matter halos found that the GCE spectrum and inner cuspiness are robust against halo triaxiality and tilt, but that the overall morphology can discriminate between halo configurations and is more compatible with a triaxial and tilted dark-matter halo than with a triaxial and tilted stellar halo (Hu et al., 23 Feb 2026). This reframes the spatial question: the dispute is no longer simply “spherical dark matter versus boxy bulge,” but increasingly “which non-spherical structure best survives diffuse-model systematics.”

3. Millisecond pulsars and unified population models

The MSP interpretation is motivated by two empirical facts: the GCE spectrum resembles that of Fermi-detected MSPs, and unresolved point sources can produce diffuse-looking emission. Recent population-synthesis work has pushed this interpretation beyond phenomenological luminosity functions toward unified evolutionary modeling of resolved disk MSPs and unresolved boxy-bulge and nuclear-bulge populations (Ploeg et al., 2020, Ploeg, 2021).

In this framework, the Galactic MSP population is modeled as a unified, data-driven population whose resolved members in the Galactic disk and whose unresolved members in the boxy bulge and nuclear bulge can all be described by the same underlying evolutionary framework (Ploeg, 2021). The preferred gamma-ray luminosity law is

$12^\circ$9

where $0.5$0 is the spectral cutoff energy, $0.5$1 is the magnetic field strength, and $0.5$2 is the spin-down power (Ploeg, 2021). A closely related parameterization recovered from the Fermi-LAT MSP sample is

$0.5$3

which the authors interpret as support for curvature-radiation-like emission and as close to the fundamental-plane relation previously found by Kalapotharakos et al. (Ploeg et al., 2020).

A defining feature of these analyses is explicit MSP evolution between birth and observation. The period $0.5$4 is evolved forward in time using magnetic-dipole spin-down, with $0.5$5 once the magnetic field is fixed, and observed $0.5$6 values are corrected for the Shklovskii effect and Galactic acceleration (Ploeg, 2021, Ploeg et al., 2020). This produces an age dependence in the luminosity distribution. Because the Galactic bulge and Galactic disk have different star-formation histories, bulge MSPs are expected to be older and therefore dimmer than disk MSPs. However, the data do not require a large intrinsic bulge–disk offset: the 95% credible interval for the bulge-versus-disk luminosity difference includes zero (Ploeg et al., 2020).

The unified framework also quantifies the number of MSPs required to explain the GCE. For the best model in one study, roughly $0.5$7 to $0.5$8 MSPs are needed in the bulge at 68% confidence, while about $0.5$9 to $1000$0 MSPs with $1000$1 are required at 95% confidence (Ploeg et al., 2020). The same analysis identifies three resolved LAT MSPs with significant probabilities of belonging to the bulge population: PSR J1747-4036, PSR J1811-2405, and PSR J1855-1436 (Ploeg et al., 2020).

Natal kicks constrain the spatial morphology of any bulge MSP population. Using resolved gamma-ray MSP proper motions, one study inferred MSP velocities relative to circular motion of $1000$2 and found that natal kicks increase each bulge MSP spatial-distribution dimension by approximately $1000$3, make the bar less boxy, and shift the boxy-bulge axis ratios from roughly $1000$4 to $1000$5 (Ploeg, 2021). The important conclusion is that natal kicks smooth the bulge distribution only modestly; they do not isotropize it, and the distribution remains far from spherical (Ploeg, 2021).

4. Dark-matter interpretations and external constraints

Dark matter remains a major interpretation because the GCE is compatible with generalized NFW-like annihilation templates with $1000$6–$1000$7, and because its spectrum can be fit by weak-scale annihilation models (Mauro et al., 2021, Mauro, 2021). A multimessenger study found that the GCE morphology and spectrum are compatible with dark matter annihilating in the main halo of the Galaxy, using either a generalized NFW profile with $1000$8–$1000$9 or an Einasto profile with $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$0 (Mauro et al., 2021).

That same study also showed that external constraints severely restrict simple dark-matter explanations. A combined analysis of 48 Milky Way dwarf spheroidal galaxies found no significant gamma-ray signal, but the resulting upper limits remain broadly compatible with the GCE-favored dark-matter points once halo uncertainties are taken into account (Mauro et al., 2021). The stronger exclusions come from AMS-02 charged cosmic rays: hadronic and semi-hadronic annihilation channels are excluded by the antiproton data unless the vertical size of the diffusion halo is smaller than about $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$1 kpc, which the authors describe as being in tension with radioactive cosmic-ray data and radio observations, while any GCE candidate with a significant $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$2 component is ruled out in the optimistic positron-background treatment (Mauro et al., 2021). Under those assumptions, the only surviving candidate is a nearly pure $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$3 annihilator with mass around $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$4 GeV and roughly thermal cross section (Mauro et al., 2021).

Recent GC-only analyses have increasingly favored mixed stellar-bulge plus dark-matter fits rather than pure dark-matter templates. A 2025 study combined adaptive template fitting with skyFACT and non-Poissonian photon-count statistics via the 1-point probability distribution function, fitting the data with both a stellar bulge and a dark-matter component simultaneously (Manconi et al., 5 Nov 2025). In those fits the stellar bulge dominates, the dark-matter normalization is usually driven to zero, and no significant dark-matter detection is found. Nonetheless, the residual space for annihilation is quantified with 95% C.L. upper limits, strongest for a contracted NFW126 profile and weakest for Burkert, with constraints that are stringent for $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$5 GeV (Manconi et al., 5 Nov 2025). The Burkert limits are more than two orders of magnitude weaker than NFW126, while NFW100 and Einasto are similar within about a factor of two (Manconi et al., 5 Nov 2025).

The GCE has also been a benchmark for concrete particle models. In a Pass 8 covariance-matrix fit to the MSSM, neutralinos with masses between $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$6 and $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$7 GeV were found to describe the excess via annihilation into $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$8-boson pairs or top quarks (Achterberg et al., 2017). In the NMSSM, a scan that included Higgs data, relic-density, LUX, and dwarf-spheroidal constraints found that the GCE can be well explained by pure $$E^2\Phi \lesssim 10^{-8}\ \mathrm{GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$$9, pure $l=[-0.3^\circ,0.0^\circ]$0, and mixed $l=[-0.3^\circ,0.0^\circ]$1, with the $l=[-0.3^\circ,0.0^\circ]$2 channel providing the best interpretation and $l=[-0.3^\circ,0.0^\circ]$3-value reaching 0.55 (Cao et al., 2015).

5. Point-source statistics, secondary emission, and inference methodology

Methodological disputes have been nearly as important as physical interpretations. Early work emphasized that models injecting electrons and positrons can generate secondary inverse Compton and bremsstrahlung emission whose morphology differs from the prompt template because the leptons diffuse and lose energy before radiating (Lacroix et al., 2015). A full 3D treatment with separate spatial templates for prompt, IC, and bremsstrahlung components showed that a broadband spectral fit alone can be insufficient to determine whether secondaries are required. In the representative MSP model tested there, the data did not require secondaries, with a secondary-to-primary ratio $l=[-0.3^\circ,0.0^\circ]$4 (Lacroix et al., 2015).

Non-Poissonian template-fitting analyses of unresolved point sources were later challenged by signal-mismodeling studies. In a $l=[-0.3^\circ,0.0^\circ]$5-radius region around the Galactic Center, one paper showed analytically and in simulations that if a smooth signal is modeled too rigidly, the fit can inflate the variance and thereby mimic a point-source population in non-Poissonian template fitting (Leane et al., 2020). In simulations containing no true GCE point sources at all, spurious GCE point-source detections were recovered in all 25 realizations of one setup, with Bayes factors as large as $l=[-0.3^\circ,0.0^\circ]$6–$l=[-0.3^\circ,0.0^\circ]$7, and even larger in a signal-only case (Leane et al., 2020). The same work argued that allowing north–south asymmetry in the smooth GCE template causes the apparent preference for GCE point sources to collapse in the $l=[-0.3^\circ,0.0^\circ]$8 region, implying that a dominantly smooth origin remains consistent with existing non-Poissonian template fits (Leane et al., 2020).

Newer simulation-based inference methods have attempted to combine spectral and spatial information rather than treating them separately. One 2024 study developed a neural-posterior-estimation framework that jointly analyzes photon directional and energy information while simulating the Fermi-LAT point-spread function and energy dispersion photon by photon (Christy et al., 2024). In mock tests, the joint energy-dependent counts-in-pixels statistic produced tighter posteriors than energy-independent spatial statistics and could distinguish DM-only, MSP-only, and mixed cases even when the average DM and MSP spectra were deliberately made similar (Christy et al., 2024).

An even more direct challenge to the bright-point-source picture came from an energy-aware CNN analysis of Fermi-LAT data from 2–20 GeV. Incorporating 10 energy bins and using spectral as well as spatial information, the inferred GCE source-count distribution shifted dramatically to lower fluxes, becoming essentially consistent with Poisson emission for the best-fit background model (List et al., 23 Jul 2025). If the excess is nevertheless due to point sources, the median prediction becomes $l=[-0.3^\circ,0.0^\circ]$9 sources in the Galactic Center, with more than 35,000 sources at 90% confidence, rather than the hundreds of sources preferred by earlier spatial-only analyses (List et al., 23 Jul 2025). This result substantially weakens the claim that spatial non-Poissonity alone decisively favors unresolved pulsars.

Sensitivity studies underscore that these inference disputes are inseparable from diffuse-model systematics. Simulations of Fermi-LAT observations showed that with perfect background knowledge the injected GCE spectrum, centroid, symmetry, and NFW slope $b=[-0.1^\circ,0.0^\circ]$0 are properly recovered, but mismodeling of the IEM introduces systematics of about $b=[-0.1^\circ,0.0^\circ]$1–$b=[-0.1^\circ,0.0^\circ]$2 in the GCE energy spectrum between 1–10 GeV and about $b=[-0.1^\circ,0.0^\circ]$3 in the recovered NFW slope (Mauro, 2020). An unmodeled low-latitude Fermi-bubbles component can also bias the morphology and spectrum once its flux reaches only about $b=[-0.1^\circ,0.0^\circ]$4–$b=[-0.1^\circ,0.0^\circ]$5 of the relevant reference scale (Mauro, 2020).

6. Composite scenarios, high-energy structure, and present status

The contemporary literature increasingly treats the GCE as a precision systematics problem rather than a single-template anomaly. One multi-messenger reanalysis that rebuilt Galactic diffuse templates from AMS-02, Voyager 1, and Fermi-LAT data found that the broad properties of the GCE are qualitatively unchanged, but reported a more significant high-energy tail than previous work, very prominent in the northern hemisphere and less so in the south (Cholis et al., 2021). Because the stacked known-MSP spectrum used there has $b=[-0.1^\circ,0.0^\circ]$6 and $b=[-0.1^\circ,0.0^\circ]$7 GeV, the authors argued that known MSPs are incapable of producing the high-energy emission and are therefore disfavored as the sole explanation (Cholis et al., 2021). In the relatively cleaner southern sky, dark matter annihilation to $b=[-0.1^\circ,0.0^\circ]$8 with

$b=[-0.1^\circ,0.0^\circ]$9

was reported to provide a good fit (Cholis et al., 2021).

At the same time, several studies favor explicitly composite descriptions. The masking analysis that found near-spherical morphology also reported that a two-component model consisting of a spherical NFW component plus the Coleman Bulge outperforms any single component, although the precise fraction attributed to each remains uncertain across masks and background models (Zhong et al., 2024). The 2025 skyFACT-plus-1pPDF analysis similarly fit the GCE with a mixed stellar-bulge plus dark-matter model and concluded that the stellar bulge dominates while a subdominant dark-matter contribution is still allowed and constrained (Manconi et al., 5 Nov 2025).

Additional astrophysical components have also been proposed. A 2026 model linking the GCE to fossil activity of Sgr A$60\%$0 argued for an irreducible hadronic contribution from a precessing Blandford–Znajek jet, giving a spin-dependent floor of roughly $60\%$1–$60\%$2 of the observed GCE surface brightness across the inner ten degrees for the EHT-favored spin $60\%$3 (Rodriguez, 12 May 2026). This component is too shallow and too spectrally displaced to explain the full excess, but it was proposed as a relevant subcomponent for precision modeling (Rodriguez, 12 May 2026).

The current state of the field is therefore not a simple binary between dark matter and MSPs. Some analyses report that bulge-tracing templates outperform spherical halos, others that a nearly spherical generalized NFW-like component remains highly significant even when bulge templates are included, and still others that the best description is mixed (Macias et al., 2016, Mauro, 21 May 2026, Zhong et al., 2024). A plausible implication is that the empirical “GCE” may not be a single physical component with a single optimal template, but a superposition whose decomposition remains limited by Galactic diffuse-emission uncertainties, source confusion, and the choice of statistical framework. What is not in dispute is that the central few to tens of degrees of the Milky Way contain a GeV-peaked excess whose interpretation continues to drive developments in gamma-ray morphology, population synthesis, multimessenger dark-matter constraints, and simulation-based inference (Mauro, 2021, Ploeg, 2021, Manconi et al., 5 Nov 2025).