Skip to content
AptifoldGet in touch

Aptifold Research · Gravitational waves

When gravitational waves scatter off a black hole

Gravitational waves stretch and squeeze space as they pass. When two waves of the same frequency scatter off a black hole, they can produce a much weaker wave at twice the original frequency. How strong is that new wave?

Aptifold’s AI research agent found two ways to mix the incoming waves that give exact formulas for the new wave in this model. One mixture reveals a small surprise. At low frequencies, one pattern grows a little less than the leading prediction as frequency rises, while another grows a little more.

A harder test for a black-hole calculation

At low frequencies, doubling the frequency makes the leading part of the outgoing amplitude four times as large. Divide out that predictable growth and a smaller effect appears. The two wave patterns in this study move in opposite directions.

The exact solutions give physicists a test for simulations. Feed in the specified waves and compare both the strength of the new wave and the timing of its peaks relative to the incoming waves. Then change the frequency. A simulation can reproduce the fourfold growth and still get the smaller effect backward.

This benchmark covers weak waves with equal frequencies scattering off a nonrotating black hole. How this effect changes with spin or unequal input frequencies is an open question.

Start with what was known.

The equations of general relativity let two waves interact. In the model here, a static, nonrotating black hole scatters two waves with the same frequency and angular pattern. The interaction drives a second harmonic. Predicting its strength means accounting for the source across the black hole’s exterior and applying the physical conditions at the horizon and far away.

Bucciotti and colleagues formulated this real-frequency scattering problem, including the outgoing strain and its quadratic scaling at low frequency. Chen and Xie developed a general way to reduce the source integrals. Their work gives us the boundary conditions, comparison quantities, and algebraic tools for a more specific question. Bucciotti et al. · Chen and Xie

The result

Two mixtures give a closed expression.

Relativity describes the angular pattern of the incoming wave using two components, called even and odd parity. Vary their relative strength and phase. For each angular label ℓ and each positive shared frequency in the channel studied here, exactly two mixtures give a rational local particular solution of the projected second-order equation. The formula uses the incoming wave fields and their derivatives at a single radius.

To predict the outgoing wave, add the solution that satisfies the conditions at the horizon and far away. The outgoing signal for the first mixture then uses one linear reflection amplitude. The second uses two contributions, with phases that let them partially cancel.

Parent waveSquare the even-parity reflection at ω.
Daughter waveUse the odd-parity reflection at 2ω.
Outgoing strainWeight and add the two complex amplitudes.

That gives two exact benchmarks for nonlinear scattering in this channel. An implementation can check the strength and phase it predicts against the closed expressions.

Computed scattering

Compare the two wave patterns

The label ℓ describes how an incoming wave varies around the black hole. Use the second exact solution for each pattern. It prescribes a slightly different mixture of the two parity components for ℓ = 4 and ℓ = 5. Each pattern keeps its own mixture fixed as you change the frequency.

Choose a highlighted interval. Both outgoing amplitudes grow. After dividing each amplitude by w², the ℓ = 4 result falls and the ℓ = 5 result rises. Here w = Mω is the frequency scaled by the black hole’s mass.

What changes over this interval?

Exactly fourfold growth would read 4× and 0%. The two patterns land on opposite sides of that prediction.

Wave pattern ℓ = 4
Amplitude vs. the start3.99919×
Change after dividing by w²−0.0203%
Wave pattern ℓ = 5
Amplitude vs. the start4.00077×
Change after dividing by w²+0.0194%
The two normalized responses move in opposite directionsSix computed points for the second polarization. As w increases, the parent ℓ equals 4 response falls and the ℓ equals 5 response rises. The selected interval is 1/2048 to 1/1024.Change from first point · 10⁴ log ratio−6−30+3+61/20481/10241/512w = Mωℓ 4, w 1/2048ℓ 4, w 1/1024ℓ 4, w 1/512ℓ = 4ℓ 5, w 1/2048ℓ 5, w 1/1024ℓ 5, w 1/512ℓ = 5
The graph magnifies the small change left after dividing by w². Each series is compared with its value at w = 1/2048. The vertical scale is 10⁴ times the natural log of that ratio.

ℓ = 4 · Falls

−0.415010Computed slopeAnalytic limit −0.414614

ℓ = 5 · Rises

+0.396542Computed slopeAnalytic limit +0.396163

Read the six saved values
Six full-domain source integrations
Parent ℓw|Q⁻₂| / w²Enclosure radiusResolution
41/20480.00020930530854942030931591.17e-26fine
41/10240.000209262898883585777095135.73e-27fine
41/5120.0002091780241518251.87e-19coarse
51/20480.00015323732871619579222562.59e-26fine
51/10240.000153267002010803626725748.46e-27fine
51/5120.0001533264226333122.31e-19coarse

A full exterior integration of the quadratic Einstein source checks the two signs independently of the closed reflection formula. At the lower interval, the slopes are within about 0.096% of their analytic limits. The normalized signal is slightly weaker at higher frequency for ℓ = 4 and slightly stronger for ℓ = 5.

The large-radius contribution was more than ten thousand times the final strain in the lowest-frequency checks. Two treatments of the integration cutoff gave overlapping complex enclosures. Resolving that cancellation is what makes the sign check useful.

Numerical sensitivity

A small net change can hide large shifts.

In the ℓ = 3 real cubic calculation, the outer grid chooses directions for the two incoming waves. The inner grid integrates over loop momenta for each pair of directions. Increase the inner resolution at a fixed pair and the contribution can move either way.

Refine the outer grid

Both grids use loop orders (5, 7). The third row substitutes finer loop values only at the 17 checked directions.

Later window · coefficient × 10³
Parent gridGeometriesReal part
6 × 6 × 4132+1.97321
8 × 8 × 4224+2.24381
8 × 8 × 4 with 17 replacements224+2.29126

Refine the inner grid at fixed directions

+0.475Net shift

+6.822Sum of magnitudes

All 17 checks · change in coefficient × 10⁴

  1. Point 141+1.449
  2. Point 149−0.941
  3. Point 137+0.901
  4. Point 109+0.750
  5. Point 117−0.648
  6. Point 121−0.610
  7. Point 104−0.608
  8. Point 173+0.299
  9. Point 122−0.282
  10. Point 77+0.230
  11. Point 45−0.063
  12. Point 205+0.015
  13. Point 17−0.013
  14. Point 223−0.008
  15. Point 195+0.004
  16. Point 0−0.000
  17. Point 192−0.000

With the same angular cuts and damping widths, the finer parent grid moves the later estimate from +1.973 to +2.244 in units of 10⁻³. For the 17 loop checks, the absolute shifts add to 6.82 × 10⁻⁴ while their net is +4.75 × 10⁻⁵. The positive interior estimate persists. The 13.7% gap between the outer grids still needs a matched comparison with higher inner resolution on both grids.

The cut region covers 86.4% of the angular measure. The plot concerns that region. The omitted region’s contribution and the full-domain real six-point calculation remain open.

How the investigation developed

Reproduce a result with public strain

For GW150914, the agent reproduced seven notebook checks on Hanford and Livingston data. Both fitted peak times moved when the input timestamp was repaired, even though their separation had already matched. The follow-up calculated a 7.051453 ms gap with interpolation and 7.080078 ms on the notebook’s discrete grid. These are two readings of a fixed-template filter output. Inspect the detector analysis

Check uncertainty with known answers

In synthetic ringdown, applying a finite-window operation to both the signal and noise description repaired the tested coverage failure. Reducing the data discarded some information. A later example showed that improving a waveform fit can also move the inferred mass. Try the ringdown checks

Find a question precise enough to prove

An early frequency-map screen found no nonzero winding in 17,464 trusted grid cells. The scan could miss unresolved pairs. In the scattering work, a solution for one mixture led to a search for every mixture with the same local property. The resulting factorization gave the pair and a way to exclude other ratios in the declared class.

A separate population pilot checked whether one event’s finite sample table supported reweighting. A catalog-level comparison and a faithful reproduction of a detector microglitch remain open.

Cross-check the nonlinear response

An independent five-point point-mass projection recovers both signed leading coefficients at ℓ = 5 and 6. In the low-frequency limit, the ratio of the second mixture’s response magnitude to the first’s goes from 0.66015 to 1.54467 across those labels. The frequency chart above follows the second mixture alone and compares its slope at ℓ = 4 and 5.

For technical review

Reconstruct the claim from the equations.

The main claim is an exact classification within a specified rational local class. The formula, the direct radial integrations, and the standalone loop estimate each have a declared domain.

Work through the pole bounds, run the obstruction matrix check, and recalculate the analytic slope from the reflection phases. The replay download contains the exact certificate and saved numerical inputs.

Read and verify the full derivation Download code and data · ZIP
Channel, normalization and boundary data

Take G = c = M = 1 for the formulas below. The parents are two identical (ℓ, ℓ) modes at the same real frequency ω > 0 on Schwarzschild, with integer ℓ ≥ 2. The daughter is (2ℓ, 2ℓ) at 2ω. Let p = Oᵢₙ/Eᵢₙ be the ratio of odd and even Regge–Wheeler–Zerilli master amplitudes. The time dependence is e−iωt and the tortoise coordinate is x = r + 2 log(r/2 − 1).

γⱼ = (j − 1)j(j + 1)(j + 2) / 12
g = γℓ,  Γ = γ₂ℓ / 2,  k = iω
p₁ = −i (g + k) / (g − k)
p₂ = −i (Γ + g) / (Γ − g)
   = −i (9ℓ² + ℓ − 4) / (7ℓ² − ℓ)

Q⁻ is the outgoing second-order spin-minus strain amplitude divided by the square of the incoming first-order spin-minus strain amplitude in orthonormal harmonics, in units M = 1. The future horizon is ingoing and the freely specified incoming physical daughter strain is zero. RE is the horizon-ingoing even parent’s linear reflection at ω and RO is the corresponding odd daughter reflection at 2ω.

Q⁻₁ = [Kℓ k² / (4g²(k − Γ))] Rₒ
Q⁻₂ = [Kℓ / (4Γ)] { [k(k − g²/Γ)/(k + g)²] Rₑ²
                    + [k/(k − Γ)] Rₒ }
Kℓ = (2ℓ+1)! (20ℓ³+35ℓ²+ℓ−8) / [144 (ℓ−1)! (ℓ−2)!]
     × √{ (2ℓ+2)! (2ℓ−2)! / [π (4ℓ+1)!] }
Completeness, physical extraction and a prior-art test

With J = iO and J = ρJref, reduce local expressions to three quadratic products of the parent and its first derivative. The projected source is SEE + ρSEO + ρ²SOO, with both mixed orderings in SEO. Construct two local particular solutions at ρ = 1 and ρ = ρ2. Subtracting their operator image leaves (ρ − 1)(ρ − ρ2)SOO.

Bounds at r = 0, 2, −3/λ2ℓ and infinity restrict the rational ansatz. A polynomial left-null witness in the resulting 41-by-30 linear system is nonzero on the odd–odd source throughout the stated domain. The pure odd input has the same obstruction. Independent fixed-label checks rebuild the source. The physical extraction checks the ten daughter Einstein equations and retains the full quadratic asymptotic coordinate change and the homogeneous boundary completion.

Bucciotti et al. show two zeros of the outgoing amplitude. At ℓ = 2 and w = 3/8, both exact local mixtures give nonzero outgoing strain, with magnitudes about 0.00480 and 0.00155. Chen and Xie’s weighted derivative recovers the two quadrupole roots when applied to our source. Read the literature check on both results

Threshold slope and direct source comparison

For the second mixture, the relative phase of the two reflected contributions determines the first correction to |Q⁻₂|/w². The real elastic Compton kernel gives bⱼ = (15L² + 13L + 24)/[(2j + 1)L(4L − 3)], with L = j(j + 1). The real cubic term is Bᵣ = πKℓ(bℓ − 2b₂ℓ)/Γ². The threshold derivative of log(|Q⁻₂|/w²) is Bᵣ/C₂, where C₂ is the signed quadratic coefficient.

For ℓ = 3, after removing exp[8iw log(4w)] in the fixed phase convention, C₂ = −9.27414085149 × 10⁻⁴, Bᵣ = (1615/84084)√(π/715) = +0.00127315440960…, and the imaginary cubic term is +0.0109208754138…. The real coefficient and magnitude slope are invariant under a constant tortoise-origin shift. The direct full-domain ℓ = 4 and 5 checks use the reconstructed quadratic source, ordered infinity phases and the analytic ingoing horizon term. The CSV gives the numerical enclosure radii and the source-result hashes. Elastic amplitude input

Finite-part loop and numerical limits

The direct ℓ = 3 calculation combines 26 bulk diagrams and three point-mass recoil placements. It integrates positive-width retarded radial poles and subtracts the large-radius linear term. The normalization from the real stripped density to strain is −56π/√(715π). For unit parent directions a and b, U = |a + b| and Q = |2ẑ − a − b|. Retain Q ≥ 0.25, 2 − U ≥ 0.10, U ≥ 0.10 and min(1 − az, 1 − bz) ≥ 0.04.

At ε = 0.0025, 0.00125, 0.000625 and 0.0003125, the screening parameter μ equals 0.7ε. Each three-width window uses I(ε) − 4I(ε/2) + 4I(ε/4), which cancels the leading ε and ε log ε terms. The refinement CSV gives normalized, weighted shifts at each width and in both windows. The six sensitive points were chosen using their damping changes. Treat the spread as a diagnostic of sensitivity. An error bound for the full integral remains open.

Expert review should start with the completeness witness and physical strain convention, then check the matched published boundary problem. Unequal frequencies, generic angular channels, rotating holes, and application of this scattering result to observed ringdown are open directions.

Sources and antecedents

  1. Bucciotti, Redondo-Yuste, Zhong, Kuntz and Cardoso. Same Schwarzschild scattering observable, parity-dependent strain zeros, boundary completion and leading threshold scaling.
  2. Chen and Xie. General integration by parts for nonlinear Regge–Wheeler–Zerilli response kernels.
  3. Brizuela, Martín-García and Tiglio. General second-order Schwarzschild perturbations and radiative reconstruction.
  4. Ben Achour and Roussille and Bourg et al. Exact algebraically special perturbations and a neighboring regime of quadratic quasinormal modes.
  5. Bjerrum-Bohr, Chen and Skowronek and Bjerrum-Bohr et al. Five-point tree and elastic Compton amplitude inputs.
  6. LIGO Scientific Collaboration and Virgo and GWOSC. Discovery and public strain. Siegel, Isi and Farr and Wang et al. Conditioning and gating/inpainting comparators for the inference study.