A shared explanation for web-like patterns in black-hole X-ray variability

Looking beyond the strength of a signal

Black-hole X-ray binaries show quasi-periodic oscillations, or QPOs: brightness variations that repeat approximately rather than with perfect regularity. These appear as relatively narrow peaks in a power spectrum, which describes how variability strength is distributed across frequencies. Broader fluctuations occupy a wider frequency range.

A power spectrum alone does not directly measure how different frequencies interact. The researchers instead examine the bispectrum, which combines three Fourier components: two frequencies and their sum. Its magnitude is called the bispectral modulus; its phase is the biphase. Bicoherence is a normalized measure of this coupling. Together, these quantities reveal relationships that variability strength alone can miss.

One interaction, two patterns

The proposed model multiplies the underlying variability by a factor that changes in step with the QPO. This modulator includes the fundamental oscillation and its second harmonic, at twice the fundamental frequency. Each modulation component maintains a fixed phase offset from its corresponding QPO component.

Multiplication mixes frequencies, creating shifted copies of the original variability. In a two-dimensional bispectral map, the calculation produces ridges where one frequency equals the QPO frequency, and a diagonal “hypotenuse” where two frequencies add to it. Together, these structures form a web-like pattern. The same reasoning applies to the second harmonic.

The model's central prediction is that each ridge and its corresponding hypotenuse arise together, with opposite biphases: approximately minus and plus the relevant phase offset. Negative-frequency components, mathematically required for real-valued signals, are essential to deriving the hypotenuse contribution.

Four observational datasets

The team analysed NICER data comprising three datasets for Swift J1727.8−1613 and one for MAXI J1820+070. The Swift datasets had QPO frequencies of approximately 0.88, 1.25 and 2.54 hertz. The MAXI dataset had a QPO near 0.75 hertz.

The researchers used photons between 0.5 and 12 kiloelectronvolts for Swift, and between 1.5 and 12 kiloelectronvolts for MAXI. They divided the light curves into 24-second segments, calculated Fourier transforms, and averaged the resulting bispectral measurements. Effective exposures ranged from about 5.3 to 12.3 kiloseconds. The analysis compared bicoherence, bispectral modulus and biphase rather than relying on a single map.

What the observations showed

The reported maps broadly support the predicted ridge–hypotenuse relationships. Some structures were difficult to identify in bicoherence but appeared more clearly in the modulus or biphase. In MAXI J1820+070, for example, bicoherence showed no clear conventional pattern, while biphase revealed a negative fundamental hypotenuse and a positive corresponding ridge.

For Swift's 1.25-hertz dataset, the fundamental ridge had a biphase around positive π/2 radians, while its hypotenuse lay approximately between negative π/2 and negative π. This supports an opposite-sign tendency, not an exact numerical match. The second-harmonic pair lay around positive and negative π/4. The authors explicitly treat these phase comparisons as qualitative; a detailed statistical analysis is deferred.

Normalization also matters. A large normalization denominator can suppress features in bicoherence, whereas the unnormalized modulus can emphasize strong Fourier amplitudes. Neither map's visual prominence alone establishes coupling strength.

Agreement without a unique explanation

The team compared measured modulus profiles with the modulation prediction, a Fourier-power comparison term, and a phenomenological approximation that suppresses shifts from higher to lower frequencies. Each received an independently fitted scale factor. These fits tested frequency dependence, not absolute modulation amplitudes.

All three comparisons broadly reproduced the observed trends, and some became nearly indistinguishable. Agreement in magnitude therefore does not uniquely identify the proposed mechanism. The linked structures and their phase relationship offer a more specific diagnostic.

Deviations remained near the fundamental and harmonic frequencies. Several contributions overlap there and must be added with their phases before calculating their combined magnitude. Such overlap could explain some discrepancies, but the study does not quantitatively fit all contributions together.

Important limits

The calculation assumes weak modulation, retains first-order terms and treats broadband components as independent Gaussian processes. Intrinsic nonlinear coupling in that background could add signals absent from the model. The simple oscillation representation also neglects finite QPO width. In the highest-frequency Swift dataset, the QPO drifted between roughly 2 and 3 hertz.

Ridge measurements became increasingly noisy above about 10 hertz. Subtracting the average modulus between 30 and 50 hertz provided only an approximate noise correction, not an unbiased intrinsic estimate.

Finally, multiplication cannot establish whether the QPO modulates broadband variability or vice versa. Proposed changes in coronal properties are physical interpretations, not demonstrated causes. This version is a preprint submitted to MNRAS, not evidence of completed peer review.