Second-Order Self-Force and Nonlinear Regularization

Second-order gravitational self-force theory is no longer only a formal program: complete calculations of binding energy, energy flux, and quasicircular nonspinning waveforms exist, and the effective source underlying those calculations has been documented and validated component by component. The frontier remains sharply restricted. Nonlinear singular products, second-order gauge, low multipoles, asymptotic frames, and multiscale orbital evolution are controlled in specialized settings, but a first-principles generic eccentric and inclined Kerr implementation is not yet available.

Evidence cutoff: 2026-09-19. Claims about available calculations, software, and unresolved problems below refer to evidence public by this date. The September Teukolsky result discussed below is a preprint.

Required background. PDEs and Green Functions for Gravitation supplies distributions and retarded boundary-value problems. Helpful background. The Einstein Equation supplies the nonlinear expansion; Weak Metric Perturbations and Gauge Freedom supplies first- and second-order gauge maps; Presymplectic Potentials and Currents supplies charge and boundary variations; Black-Hole Perturbation Theory supplies the mode equations and radiative boundaries.

Green-function bridge. At first order, a delta-supported source defines a distribution and the retarded Green function gives a well-defined field away from the worldline. At second order, the source contains products of first-order fields and derivatives. Products of arbitrary distributions are not automatically defined. One must either work in a gauge where a distributional extension is proved or subtract a nonlinear puncture before taking the worldline limit; writing a formal product of two singular distributions does not define a source.

The second-order field equation

Let $\varepsilon=m/M\ll1$, where $m$ and $M$ are physical masses and $M$ is held fixed in the small-body family. Expand the exterior metric as

$$g_{\mu\nu}^{\mathrm{full}} =g_{\mu\nu} +\varepsilon h_{\mu\nu}^{(1)} +\varepsilon^2 h_{\mu\nu}^{(2)} +O(\varepsilon^3).$$

Expanding the Einstein tensor gives, schematically,

$$\delta G_{\mu\nu}[h^{(1)}] =8\pi G T_{\mu\nu}^{(1)},$$

and

$$\delta G_{\mu\nu}[h^{(2)}] =8\pi G T_{\mu\nu}^{(2)} -\delta^2G_{\mu\nu}[h^{(1)},h^{(1)}].$$

Here $T=\varepsilon T^{(1)}+\varepsilon^2T^{(2)}+\cdots$, and $\delta^2G[h,h]$ denotes the coefficient of $\lambda^2$ in $G[g+\lambda h]$, including the Taylor factor $1/2$. This fixes its normalization independently of how another author labels second variations. The symbol $T_{\mu\nu}^{(2)}$ is schematic: it contains worldline and monopole corrections, with spin or higher multipoles when their scaling places them at this order.

Near a nonspinning monopolar body, the actual first-order perturbation is $H^{(1)}\equiv\varepsilon h^{(1)}=O(Gm/s)$, whereas its coefficient is $h^{(1)}=O(GM/s)$. Keeping this distinction avoids counting the small mass twice. The physical quadratic contribution therefore contains

$$\varepsilon^2\delta^2G[h^{(1)},h^{(1)}] =\delta^2G[H^{(1)},H^{(1)}] =O\!\left(\frac{G^2m^2}{s^4}\right).$$

For any small worldtube radius $s_0\gt 0$, its naive spatial integral behaves as $\int_0^{s_0}s^2s^{-4}ds=\int_0^{s_0}s^{-2}ds$, which diverges. The quadratic term is therefore not an ordinary locally integrable function and cannot simply be paired with a test function. This is the central nonlinear regularization problem.

Matched small-body expansions nevertheless yield a smooth effective metric

$$\tilde g_{\mu\nu} =g_{\mu\nu} +\varepsilon h_{\mu\nu}^{\mathrm R(1)} +\varepsilon^2 h_{\mu\nu}^{\mathrm R(2)}$$

in which a nonspinning, effectively monopolar body moves geodesically through the retained order. Pound derives the second-order equation of motion and the hypotheses on the body and gauge in Pound 2017, §§ II–VI, especially the effective-metric result in § V. The statement is local and perturbative; it does not make the retarded metric globally vacuum or eliminate finite-size forces whose moments enter at the same order.

Nonlinear punctures and source cancellation

Decompose each perturbation coefficient into puncture and residual pieces,

$$h_{\mu\nu}^{(n)} =h_{\mu\nu}^{\mathrm P(n)} +h_{\mu\nu}^{\mathrm{res}(n)}.$$

The residual agrees with the regular field and its required derivatives on the worldline only when the puncture has sufficient local order. It is not identical to the exact regular field throughout the neighborhood. In a Lorenz-type gauge, the physical second-order puncture begins more strongly than its first-order counterpart,

$$\varepsilon^2h_{\mu\nu}^{\mathrm P(2)} \sim\frac{G^2m^2}{s^2} +\frac{G\Delta m_{\mu\nu}}{s} +\text{direction-dependent and logarithmic terms}.$$

where $\Delta m_{\mu\nu}$ denotes a physical monopole correction of order $m^2/M$. The coefficients encode the nonlinear monopole, tidal couplings, worldline acceleration, and any spin or quadrupole retained by the scaling. At the level of Einstein-tensor coefficients the residual equation is

$$\delta G[h^{\mathrm{res}(2)}] =S_{\mathrm{eff}}^{(2)} \equiv 8\pi G T^{(2)} -\delta^2G[h^{(1)},h^{(1)}] -\delta G[h^{\mathrm P(2)}].$$

All terms on the right are schematic until a gauge and distributional prescription are fixed. In particular, the Lorenz wave operator used on the first-order pages is $-2\delta G$ after trace reversal and imposing the constraint; its source must be translated by the same factor. In an implementation, nonintegrable powers cancel analytically before numerical evaluation. First-order puncture products, slow-time derivatives, mode couplings, the second-order puncture, and low-multipole mass changes are distinct contributions and should be tested separately.

For quasicircular Schwarzschild motion, Upton and collaborators give the multiscale Lorenz-gauge source, including precisely those pieces, and validate them with analytical and numerical checks in Upton et al. 2026, §§ II–VI, especially the source construction in §§ III–V and validation in § VI. This is strong evidence for that source in that setting. It is not a generic-Kerr proof, and a finite-order puncture remains an approximation to the local singular field rather than an exact global Detweiler–Whiting field.

Mass and spin renormalization cannot be postponed to a final parameter fit. A $1/s$ second-order monopole can be interpreted differently if the background mass, the small body’s mass, and the Bondi mass are not distinguished. Odd-parity dipoles encode angular-momentum shifts, while even-parity dipoles are tied to the choice of center-of-mass worldline. Their completion determines physical charges and can feed the waveform phase. A convergent radiative $\ell\ge2$ calculation with incorrect monopole or dipole data is incomplete.

Two regularization strategies

Position A: Lorenz-gauge puncture and effective source. This approach keeps a familiar hyperbolic operator and subtracts sufficiently many terms from the singular field so that a finite residual equation can be solved. It has produced the existing quasicircular Schwarzschild second-order results and fits naturally with worldtubes and multiscale evolution. Its cost is a large, finitely differentiable source with many coupled modes, and it must control cancellation between individually divergent expressions.

Position B: highly regular gauges and distributional sources. In a specially constructed gauge, the second-order singularity is milder. Upton and Pound prove a distributional second-order source and a pointlike effective stress-energy in this class, opening a route to mode-sum rather than full puncture regularization Upton and Pound 2021, §§ III–VI, especially the distributional construction in § V. This can simplify local regularization, but transforming a globally retarded solution into a highly regular gauge while controlling boundaries and reconstruction is a substantial task.

These positions are compatible formulations, not contradictory physical theories. It is not yet established that one dominates for generic Kerr orbits. A convincing comparison would calculate the same second-order invariant using both, with independent source generation and matched asymptotic charges.

At the field-equation level, gauge-invariant second-order Schwarzschild master variables provide another cross-check. Spiers, Pound, and Wardell derive covariant spherical decompositions of the Einstein, Regge–Wheeler–Zerilli, and Teukolsky equations and supply raw and gauge-invariant formulas in Spiers, Pound, and Wardell 2024, §§ II–VII and Appendices C–E. Those formulas organize mode coupling, but they do not by themselves regularize a point-particle source.

Second-order gauge and worldline identification

With the convention used above, a two-stage gauge transformation generated by $\xi_1^\mu$ and $\xi_2^\mu$ has the schematic action

$$h^{(1)} \mapsto h^{(1)}+\mathcal L_{\xi_1}g,$$ $$h^{(2)} \mapsto h^{(2)} +\mathcal L_{\xi_2}g +\mathcal L_{\xi_1}h^{(1)} +\frac12\mathcal L_{\xi_1}^2g.$$

Signs and factors change with the active/passive and Taylor conventions. The quadratic terms cannot be omitted. The worldline also transforms at both orders, and a singular $\xi_1^\mu$ can move terms between the nominal first- and second-order singular structures. Thus “the second-order self-force” is not a gauge-invariant vector.

Second-order observables must be relational and must include the first-order shift of their arguments. For example,

$$Q(\Omega,\varepsilon) =Q_0(\Omega) +\varepsilon Q_1(\Omega) +\varepsilon^2Q_2(\Omega)+\cdots$$

requires holding the physical $\Omega$ fixed, not a background radius. The term $Q_2$ contains the explicit $h^{\mathrm R(2)}$ contribution, pieces quadratic in $h^{\mathrm R(1)}$, and the first-order displacement of the orbit. It also requires a fixed asymptotic time, background-mass convention, and Bondi or ADM frame.

What has actually been calculated

It is useful to separate four levels of maturity.

Formal equations. Second-order motion, local singular expansions, second-order gauge transformations, and effective-metric motion are established for a small compact body under declared smoothness and multipole assumptions. The formal framework is not the main uncertainty.

Regularized implementation. A complete quasicircular Schwarzschild scheme has produced the second-order binding energy. Pound and collaborators compare their result with a first-law construction and quantify the distinction between Bondi and local mechanical energies in Pound et al. 2020, pp. 1–5, eqs. (1)–(14) and Fig. 1. Second-order energy fluxes for the same broad nonspinning quasicircular sector agree with post-Newtonian and numerical-relativity comparisons Warburton et al. 2021, pp. 2–5, especially eqs. (1)–(12).

Concrete waveforms. A two-timescale construction has produced fast first-postadiabatic waveforms for nonspinning quasicircular binaries, with striking agreement against numerical relativity even beyond the nominal extreme-mass-ratio regime Wardell et al. 2023, pp. 2–5, especially eqs. (1)–(8). That agreement is evidence for useful resummation; it is not proof that the raw mass-ratio series is uniformly accurate for all comparable-mass configurations.

Extensions, not generic coverage. A 2026 model includes a slowly spinning primary and a generically oriented precessing secondary spin, but remains quasicircular, assumes small primary-spin misalignment, and is validated in the reported domain $q\gtrsim5$ and $|\chi_1|\lesssim0.1$ Mathews et al. 2026, §§ I and VI–VII. This is meaningful progress beyond nonspinning motion. It does not supply first-principles second-order fields for generic eccentric, inclined, rapidly spinning Kerr inspirals or transient resonances.

A second field-equation solver

The 8 September 2026 preprint by Leather and collaborators computes the second-order outgoing energy-flux coefficients for the $(\ell,m)=(2,1)$ and $(2,2)$ modes of quasicircular Schwarzschild binaries using the Teukolsky equation. At 22 directly tabulated comparison radii, the median relative differences from the Lorenz-gauge calculation are $8.65\times10^{-6}$ and $2.21\times10^{-4}$, respectively. Five additional radii use interpolated comparators and are excluded from those statistics Leather et al. 2026, § VII.2, Fig. 5, and Appendix D.

This tests a different second-order field equation, but the methods share first-order inputs and Lorenz-gauge effective-source infrastructure. Their agreement cannot detect every common input error. The implementation also omits specified memory and memory-distortion effects; its numerical error bars summarize tested sensitivities, not complete bounds. These qualifications are explicit in Leather et al. 2026, § IV.1.1 and Appendix D.6. The calculation is progress within the quasicircular Schwarzschild sector; it supplies neither all radiative modes nor a generic Kerr waveform.

Interpreting progress and remaining limitations

Optimistic assessment. The convergence of formal derivations, binding energy, flux, waveform comparisons, public waveform tools, and the 2026 component-by-component source validation makes quasicircular Schwarzschild second-order self-force a functioning computational theory. Its unexpectedly good agreement with numerical relativity suggests that suitable resummations can inform binaries well outside the strict EMRI limit.

Conservative assessment. LISA’s generic sources inhabit a larger space: Kerr spin, eccentricity, inclination, resonances, long-time phase evolution, secondary spin and finite size, and an unambiguous asymptotic waveform frame. Success in quasicircular Schwarzschild does not bound errors in those directions. Comparable-mass agreement can also conceal cancellations introduced by resummation, so it should not be used as a substitute for order-by-order convergence or generic small-mass-ratio validation.

Both assessments are credible because they answer different questions. The optimistic claim concerns demonstrated utility in a restricted sector. The conservative claim concerns completeness and controlled errors for generic EMRIs. The evidence cutoff supports the first and does not yet support the second.

Uncertainties and failure modes

Several uncertainties remain consequential. A nonlinear source expression can pass pointwise tests while losing accuracy through cancellation near the worldline. Mode coupling creates double sums whose truncation error is not diagnosed by increasing only the output $\ell$. Finite differentiability can limit spectral convergence. Monopole and dipole completion can shift masses, angular momentum, or the center-of-mass worldline. A gauge transformation regular locally may fail to preserve retarded boundaries globally.

At infinity, second-order radiation interacts with Bondi mass loss, memory, and the choice of Bondi–Metzner–Sachs frame. Waveforms compared in different frames can disagree in phase or memory even when local source calculations agree. During inspiral, a two-timescale expansion can fail near transient orbital resonances, the separatrix, or plunge. Spin and quadrupole terms can enter at the same nominal order as second-order monopolar self-force and must be included according to their scaling.

Recognizable failures include residual sources growing as the tube shrinks, constraint violations that converge more slowly than the field, invariants that depend on puncture order after extrapolation, flux imbalance, or an $O(\varepsilon^2)$ phase term that changes under asymptotically nontrivial gauge transformations. Agreement with a single post-Newtonian coefficient is not enough; independent gauge, source, and boundary checks are needed.

What could change the conclusions

This frontier assessment should be updated when any of the following occurs:

• a new, independently validated second-order invariant is computed in two gauges or by puncture and highly regular methods;
• a convergent, reproducible generic-Kerr implementation handles eccentric and inclined motion with documented mode, worldtube, boundary, and precision errors;
• a first-principles generic waveform includes transient resonances and an explicit Bondi–Metzner–Sachs frame and is validated against independent calculations;
• a published nonlinear-source correction, erratum, or low-multipole reinterpretation changes an existing binding-energy, flux, or waveform result;
• rapidly spinning primaries, generic misalignment, or finite-size multipoles enter a first-postadiabatic model with a calibrated error budget.

The September comparison advances the first trigger while retaining shared inputs and the stated memory restriction. “Second order is solved” should still be read only as a claim about demonstrated quasicircular sectors, not the generic EMRI problem.

Exercise

Assume the physical first-order perturbation satisfies $H^{(1)}\sim Gm/s$ near the worldline. Show why $(\nabla H^{(1)})^2$ is not locally integrable in three spatial dimensions. If a second-order puncture cancels that source through all powers more singular than $s^{-1}$, what improvement follows? Does it alone establish a generic second-order waveform model?

Solution

Since $\nabla H^{(1)}=O(Gm/s^2)$,

$$(\nabla H^{(1)})^2=O(G^2m^2/s^4).$$

The spatial volume measure contributes $s^2ds\,d\Omega$, so the absolute radial integral behaves as

$$\int_0^{s_0} s^2\frac{ds}{s^4} =\int_0^{s_0}\frac{ds}{s^2},$$

which diverges. The naive quadratic source is not locally integrable.

If the puncture cancels all terms through $s^{-2}$ and leaves $O(s^{-1})$, the effective-source integral behaves as

$$\int_0^{s_0} s^2\frac{ds}{s} =\int_0^{s_0} s\,ds,$$

which is finite. This establishes local integrability of that residual source, not differentiability beyond what the remainder supplies.

It does not establish a generic waveform model. One still needs a globally retarded solution, gauge and low-mode completion, boundary and asymptotic-frame control, convergent mode coupling, second-order orbital evolution, and coverage of generic Kerr frequencies and resonances. Local source regularity is one necessary layer of the calculation.

References

• Leather, Benjamin, Andrew Spiers, Adam Pound, Samuel D. Upton, Barry Wardell, Leanne Durkan, and Niels Warburton. “Second-Order Teukolsky Calculations for Nonspinning, Quasicircular Binaries.” arXiv:2609.08995v1 [gr-qc] (8 September 2026), preprint. Versioned open article.
• Mathews, Josh, Barry Wardell, Adam Pound, and Niels Warburton. “Postadiabatic Self-Force Waveforms: Slowly Spinning Primary and Precessing Secondary.” Physical Review D 113, 064034 (2026). doi:10.1103/ph3p-mscl.
• Pound, Adam, Barry Wardell, Niels Warburton, and Jeremy Miller. “Second-Order Self-Force Calculation of Gravitational Binding Energy in Compact Binaries.” Physical Review Letters 124, 021101 (2020). doi:10.1103/PhysRevLett.124.021101.
• Pound, Adam. “Nonlinear Gravitational Self-Force: Second-Order Equation of Motion.” Physical Review D 95, 104056 (2017). doi:10.1103/PhysRevD.95.104056.
• Spiers, Andrew, Adam Pound, and Barry Wardell. “Second-Order Perturbations of the Schwarzschild Spacetime: Practical, Covariant, and Gauge-Invariant Formalisms.” Physical Review D 110, 064030 (2024). doi:10.1103/PhysRevD.110.064030.
• Upton, Samuel D., and Adam Pound. “Second-Order Gravitational Self-Force in a Highly Regular Gauge.” Physical Review D 103, 124016 (2021). doi:10.1103/PhysRevD.103.124016.
• Upton, Samuel D., Barry Wardell, Adam Pound, Niels Warburton, and Leor Barack. “Effective Source for Second-Order Self-Force Calculations: Quasicircular Orbits in Schwarzschild Spacetime.” Physical Review D 113, 064013 (2026). doi:10.1103/f9j2-k64r.
• Warburton, Niels, Adam Pound, Barry Wardell, Jeremy Miller, and Leanne Durkan. “Gravitational-Wave Energy Flux for Compact Binaries through Second Order in the Mass Ratio.” Physical Review Letters 127, 151102 (2021). doi:10.1103/PhysRevLett.127.151102.
• Wardell, Barry, Adam Pound, Niels Warburton, Jeremy Miller, Leanne Durkan, and Alexandre Le Tiec. “Gravitational Waveforms for Compact Binaries from Second-Order Self-Force Theory.” Physical Review Letters 130, 241402 (2023). doi:10.1103/PhysRevLett.130.241402.

Dated Research assessment. For open questions and recent results, follow the link below. Each assessment states the date through which it considers results and the limits of its conclusions.

• Second-Order Self-Force and EMRI Accuracy