Quasi-Periodic Oscillations

A noisy clock at the edge of a black hole

Point an X-ray timing instrument at an accreting stellar-mass black hole and the brightness does not hold steady. It flickers — chaotically on most timescales, but with one or two preferred rhythms buried in the noise. Fold those rhythms out with a Fourier transform and they show up as broad humps in the power spectrum: the source is brightening and dimming dozens, sometimes hundreds, of times a second, at a rate that is sharp enough to call a frequency but loose enough that the phase wanders. That is a quasi-periodic oscillation, or QPO.

The word "quasi" is doing real work. A radio pulsar is a true clock; its spin keeps phase for billions of cycles, so its signal is a delta-function spike. A QPO instead jitters. It ticks at a well-defined average rate and then drifts, losing phase memory after a handful to a few hundred cycles. That blur is not a measurement defect — it is the physical signature of turbulent, fluctuating gas orbiting at the very inner edge of an accretion disk, where matter circles a compact object at a fair fraction of the speed of light. The frequencies we hear in the X-rays are, quite literally, the orbital and oscillation frequencies of gas a few gravitational radii above a black hole's horizon or a neutron star's surface. There is no closer probe of strong-field motion that does not require a gravitational-wave detector.

How a QPO appears in the power spectrum

QPOs live in the frequency domain, not the time domain. You take a long X-ray light curve, slice it into segments, compute the power spectral density (PSD) of each, and average. Broadband noise from the accretion flow fills the PSD; a QPO is a Lorentzian peak riding on top of that continuum. Each QPO is described by three numbers:

ν₀    centroid frequency (the tick rate)
FWHM  full width at half maximum (Δν)
Q  =  ν₀ / Δν        quality factor (how sharp)

A high Q means a coherent, long-lived oscillation; a low Q means a broad, blurry one. The amplitude is quoted as the fractional root-mean-square variability — the rms power in the peak as a percentage of the source flux — typically a few percent up to ~15% for strong low-frequency QPOs. The convention that turned QPOs into a quantitative science is to model the entire PSD as a sum of Lorentzians, fitting the broadband noise and the QPOs together rather than treating the QPO as an isolated bump. That decomposition, championed by Tomaso Belloni, Mariano Méndez and others, is what lets two telescopes at two epochs be compared on the same footing.

The frequencies the inner disk can produce

Why should there be preferred frequencies at all? Because near a compact object, an orbiting parcel of gas has not one but three natural frequencies, and general relativity splits them apart. In Newtonian gravity a slightly perturbed circular orbit closes on itself, so the orbital frequency, the radial oscillation frequency, and the vertical oscillation frequency are all equal. In the strong field of a spinning (Kerr) black hole they differ. For a prograde circular orbit at radius r (in gravitational units r_g = GM/c²) around a hole of spin parameter a, the frequencies are:

Orbital (Keplerian)   ν_φ  =  (c³ / 2πGM) · 1 / (r^{3/2} + a)
Radial epicyclic       ν_r  =  ν_φ · √(1 − 6/r + 8a/r^{3/2} − 3a²/r²)
Vertical epicyclic     ν_θ  =  ν_φ · √(1 − 4a/r^{3/2} + 3a²/r²)

Three facts follow that explain almost everything observed. First, all three frequencies scale as 1/M, so heavier compact objects tick more slowly — the same physical orbit produces kilohertz signals around a 1.4 M☉ neutron star but only tens of hertz around a 10⁶ M☉ supermassive black hole. Second, the radial epicyclic frequency ν_r falls to zero exactly at the innermost stable circular orbit (the radius where 1 − 6/r + … vanishes), because the ISCO is precisely the orbit where radial oscillations stop being stable. Third, the difference between the orbital and vertical frequencies, ν_φ − ν_θ, is the Lense-Thirring precession rate — the rate at which frame dragging makes a tilted orbit wobble like a slow gyroscope. These three quantities — Keplerian, epicyclic, and precession frequencies — are the raw material of every QPO model.

The key numbers

To anchor the scales, here is the inner edge of a non-spinning 10 M☉ black hole. The gravitational radius is r_g = GM/c² ≈ 1.48 km × (M/M☉) ≈ 15 km. The ISCO sits at 6 r_g ≈ 89 km. The orbital frequency at the ISCO of a Schwarzschild hole is the well-known closed form:

ν_ISCO = c³ / (2π · 6^{3/2} · GM)
       ≈ 2198 Hz × (M☉ / M)
       ≈ 220 Hz   for M = 10 M☉

So a few-hundred-hertz high-frequency QPO is exactly what you expect from gas orbiting near the ISCO of a ten-solar-mass black hole — and that is what is observed. The numbers across the relevant systems:

Quantity	Neutron star (1.4 M☉)	Stellar BH (10 M☉)	Supermassive (10⁶ M☉)
Gravitational radius r_g	2.1 km	15 km	1.5 × 10⁶ km
ISCO radius (a=0)	12.4 km	89 km	9 × 10⁶ km
ν at ISCO (a=0)	~1570 Hz	~220 Hz	~2.2 mHz
Observed HFQPO range	200 – 1300 Hz	40 – 450 Hz	~10⁻⁴ Hz (rare)
Observed LFQPO range	1 – 60 Hz	0.1 – 30 Hz	—

The match between the predicted ISCO frequency and the observed high-frequency QPOs across four orders of magnitude in mass is the central piece of evidence that QPOs really do come from the innermost flow. Even AGN show analogous behaviour: the active galaxy RE J1034+396 has a ~1-hour X-ray QPO, exactly the timescale a few-million-solar-mass black hole would produce if its physics scaled down from the stellar-mass systems.

The leading models

No single model is universally accepted, and the honest state of the field is that the low-frequency and high-frequency QPOs may have different origins. The contenders cluster into two ideas.

Model	Applies to	Mechanism	Key prediction
Relativistic precession (Stella & Vietri 1998)	HF + type-C LF together	ν_φ, periastron precession ν_φ−ν_r, nodal precession ν_φ−ν_θ from one radius	Three QPOs from one radius; inverts for M and a
Lense-Thirring precession (Ingram, Done & Fragile 2009)	Type-C LFQPO	Frame dragging precesses a hot, misaligned inner flow	QPO phase tracks reflection / iron-line shape
Epicyclic resonance (Abramowicz & Kluźniak 2001)	HFQPO 3:2 pairs	Nonlinear resonance between ν_r and ν_θ at a fixed radius	Twin peaks lock at 3:2 regardless of luminosity
Sonic-point / beat frequency (Miller, Lamb & Psaltis 1998)	Neutron-star kHz pairs	Upper peak = inner-disk Keplerian; lower = beat with spin	Peak separation ≈ spin frequency or half of it
Accretion-ejection / disk oscillation modes	Various LFQPO	Magnetically driven standing waves or instabilities in the disk	Frequency tracks inner-disk truncation radius

The most influential modern result is observational, not theoretical: NICER and earlier RXTE data show that as the type-C low-frequency QPO sweeps up in frequency, the iron Kα emission line from the inner disk rocks back and forth in a way consistent with a precessing inner flow illuminating different parts of the disk. That iron-line "see-saw," reported by Adam Ingram and collaborators for H1743-322 in 2016, is the strongest direct evidence that at least the type-C LFQPO is a geometric precession effect rather than an intrinsic brightness oscillation.

Worked example: weighing GRO J1655-40

The black hole binary GRO J1655-40 is the textbook case because it shows the rare triple: two high-frequency QPOs at 450 Hz and 300 Hz (the 3:2 pair) plus a type-C low-frequency QPO near 17 Hz, all detected by RXTE. The relativistic precession model identifies these with the orbital frequency, the periastron-precession frequency, and the nodal-precession frequency at a single radius, giving three equations for three unknowns: the radius r, the mass M, and the spin a*.

ν_φ        =  450 Hz      (orbital)
ν_φ − ν_r  =  300 Hz      (periastron precession → HF lower peak)
ν_φ − ν_θ  =   17 Hz      (nodal / Lense-Thirring → type-C LFQPO)

Solving the Kerr frequency relations simultaneously (Motta et al. 2014) yields

M   ≈  5.3 M☉
a*  ≈  0.29
r   ≈  5.7 r_g

The decisive check: the optical radial-velocity curve of the companion star independently gives a dynamical mass of 5.4 ± 0.3 M☉. The timing-only solution and the optical-only solution agree to within the errors on the same object — a genuine cross-validation of using QPO frequencies to weigh a black hole and measure its spin. It is rare to be able to do strong-field timing physics and Newtonian binary dynamics on one source and have them concur.

How QPOs are observed

QPO science is timing science, which means it lives or dies on collecting area and time resolution rather than on imaging. The instrument needs to register individual X-ray photons with timestamps good to microseconds and gather enough photons per cycle to see the modulation above Poisson noise. The lineage of missions:

• EXOSAT (1983–1986). ESA's first long-orbit X-ray observatory; its uninterrupted ~80-hour viewing windows let van der Klis's team find the first clear QPOs in 1985.
• RXTE (1995–2012). NASA's Rossi X-ray Timing Explorer, with the 6,500 cm² Proportional Counter Array and 1-microsecond timing, was built for this. It discovered kHz QPOs in more than 20 neutron stars and the high-frequency 3:2 pairs in black holes. Essentially the entire QPO catalog rests on RXTE.
• NICER (2017–). The Neutron star Interior Composition Explorer on the ISS brings soft-band (0.2–12 keV) sensitivity and 100-nanosecond timing, refining LFQPO and kHz studies and connecting them to spectral state.
• NuSTAR (2012–). Hard X-ray focusing (3–79 keV) ties QPO phase to the reflection spectrum and the iron line, enabling the see-saw test of precession models.
• IXPE (2021–) and the proposed eXTP / STROBE-X. Polarization adds a new axis: a precessing flow should sweep its polarization angle in step with the QPO, a test now beginning with IXPE.

Variants and related phenomena

• Type-A, -B, -C low-frequency QPOs. In black hole transients the LFQPO is sorted by shape and by the noise it rides on. Type-C (strong, variable, on flat-top noise) tracks the spectral hardening and is tied to Lense-Thirring precession; type-B (weak, on weak red noise) appears in the soft-intermediate state and may signal the jet; type-A is broad and rare.
• kHz QPO pairs in neutron stars. Twin peaks at hundreds of hertz to ~1300 Hz that drift in frequency while keeping a roughly constant separation near the spin frequency or half of it — the beat-frequency signature that requires a stellar surface.
• Heartbeat oscillations. The 1–10 Hz "ρ" and "κ" patterns of GRS 1915+105 and IGR J17091-3624 are limit-cycle radiation-pressure instabilities of the disk, distinct from but often discussed alongside QPOs.
• Tidal-disruption-event QPOs. A handful of TDEs (e.g. ASASSN-14li) show X-ray QPOs at ~100–400 s, hinting that the same inner-disk clock turns on briefly when a star is shredded onto a supermassive hole.
• AGN QPOs. RE J1034+396's ~1-hour signal and a growing list of candidate "quasi-periodic eruptions" extend QPO phenomenology to supermassive black holes, with timescales scaled up by the mass ratio.

Common misconceptions and subtleties

• A QPO is not a pulsation. A pulsar's beat is a rigid rotating beam with permanent phase coherence. A QPO is a fluctuating fluid oscillation that drifts; its finite width is intrinsic, not instrumental. Confusing the two leads to treating the QPO as a body clock when it is a flow clock.
• Centroid frequency ≠ "the ISCO frequency" directly. Most QPO frequencies correspond to a radius somewhat outside the ISCO (the inner flow truncates above it, and precession frequencies are differences of large numbers). Reading off the ISCO frequency from a QPO requires a model, not a one-line conversion.
• The 3:2 is a ratio, not a fixed pair of frequencies. Different sources show 3:2 at very different absolute frequencies (450/300 vs 67/41 Hz) because the absolute scale is set by mass while the ratio is set by relativistic dynamics. The commensurability survives; the numbers do not.
• QPOs are a state phenomenon. Type-C LFQPOs strengthen in the hard and hard-intermediate states and fade as a source softens; HFQPOs appear in the steep-power-law / very-high state. A QPO present last week can be gone this week not because the model failed but because the accretion geometry changed.
• "Quasi-periodic" does not mean "weak evidence." A 6σ Lorentzian in a well-sampled PSD is a robust detection; the quasi-periodicity refers to the signal's coherence, not the confidence of the measurement.