On the evidence for axionlike particles from active galactic nuclei

Burrage, Davis, and Shaw [1] recently suggested exploiting the correlations between high and low energy luminosities of astrophysical objects to probe possible mixing between photons and axionlike particles (ALP) in magnetic field regions. They also presented evidence for the existence of ALP’s by analyzing the optical/UV and x-ray monochromatic luminosities of active galactic nuclei. We extend their work by using the monochromatic luminosities of 320 unobscured active galactic nuclei from the Sloan Digital Sky Survey/Xmm-Newton Quasar Survey [2] which allows the exploration of 18 different combinations of optical/UV and x-ray monochromatic luminosities. However, we do not find compelling evidence for the existence of ALPs. Moreover, it appears that the signal reported by Burrage et al. is more likely due to x-ray absorption rather than to photon-ALP oscillation.

Figure 1

Probability distribution function of $C$, the ratio between the photon and the total intensity, when $p_0=0.$, 0.5, 1.

Figure 2

Probability distribution of the scatter in the ALP-mixing model for various $P_{\mathrm{mix}}$ values. We assumed $p_0=0.1$ and ${\sigma }_{\mathrm{in}}=0.2$. For small $P_{\mathrm{mix}}$, the distribution is nearly Gaussian, apart from the low intensity tail. Here and below we focus on the low end tail, where the relative deviations are largest.

Figure 3

Probability distribution of the scatter taking into account shot noise for various values of collected x-ray photons. We assumed ${\sigma }_{\mathrm{in}}=0.2$ for the intrinsic Gaussian noise. Like the ALP-mixing model, shot noise can increase the low luminosity tail.

Figure 4

Cumulative distribution functions of the scatter for the BDS-77 catalog (left panel) and BDS-203 (right panel). Each empirical CDF is plotted along with the ALP-mixing model and Gaussian model theoretical CDFs. We assumed $p_0=0.1$ and $P_{\mathrm{mix}}=1$.

Figure 5

Fingerprint plots of variance versus skewness. Each point represents a bootstrap resampling of 77 data points (a) from the BDS-77 catalog of AGNs, (b) simulated with the ALP-mixing model and (c) simulated with the ALP-mixing model, but where there happens to be a large outlier.

Figure 6

Value of the $r$ statistic for the various wavelengths and catalogs ($P_{\mathrm{mix}}=1$). The evidence is significantly lower in those bands where the effects of absorption are expected to be less.

Figure 7

Cumulative distribution functions of the scatter for the Full (a) and High-$\Gamma$ (b) catalogs. The empirical CDF is plotted along with the ALP-mixing model and Gaussian model theoretical CDFs. For the ALP-mixing model we assumed $p_0=0.1$ and $P_{\mathrm{mix}}=1$. When $P_{\mathrm{mix}}<1$, the data curve stays the same while the CDF of the ALP-mixing model tends to that of the Gaussian model.

Figure 8

$r$ distribution for the BDS-203 catalog at $E=2\mathrm{keV}$ when the scatter is distributed according to the ALP-mixing model. The measured signal, $r$, is represented by the vertical black line, while its statistical significance is quoted in the legend as $p(r)$.

Figure 9

$r$ distribution for the Full catalog at $E=2\mathrm{keV}$ when the scatter is distributed according to the ALP-mixing model.

Figure 10

$r$ distribution for the High-$\Gamma$ catalog at $E=2\mathrm{keV}$ when the scatter is distributed according to the ALP-mixing model.

Figure 11

$r$ distribution for the High-$\Gamma$ catalog at $P_{\mathrm{mix}}=1$ for four different frequencies when the scatter is distributed according to the Gaussian model.

Figure 12

$r$ distribution for the High-$\Gamma$ catalog at $E=2\mathrm{keV}$ and $P_{\mathrm{mix}}=1$ for four different models. The shot-noise models stay in between the Gauss and ALP-mixing models.