Global Regularity of 3D Navier–Stokes: An Energy Argument

Rod Higgins · Senuamedia · April 2026 · DOI: 10.5281/zenodo.19601371

Abstract

Contents

1. The Equation
2. The Cascade Has Finite Speed
3. The Proof
4. Why the Proof Works
5. Verification
6. Discussion
7. Appendix: Non-Local Suppression
8. References

1. The Equation

The 3D incompressible Navier–Stokes equations on \(\mathbb{T}^3\):

describe the decay of a velocity perturbation \(u(x,t)\) toward the equilibrium state \(u = 0\). The fluid is stable; the flow is a finite perturbation. Viscosity \(\nu > 0\) is constant.

Kinetic energy and enstrophy:

Taking the \(L^2\) inner product of the NS equation with \(u\):

The nonlinear and pressure terms vanish by incompressibility. This is exact. Energy decreases monotonically. Integrating:

Finite in, finite out. The enstrophy is integrable: \(\Omega \in L^1(0,\infty)\).

Regularity. If \(\sup_{t \geq 0}\Omega(t) < \infty\), then \(u \in L^\infty(H^1) \hookrightarrow L^\infty(L^6)\) by Sobolev embedding (\(d = 3\)), placing the solution in the Prodi–Serrin class \((p = \infty, q = 6)\). Global regularity follows. The question is whether \(\Omega(t)\) can spike to infinity—a blowup—despite having a finite \(L^1\) integral.

2. The Cascade Has Finite Speed

In Fourier space, the velocity field is \(u = \sum_{\mathbf{k} \in \mathbb{Z}^3} \hat{u}_{\mathbf{k}}\,e^{i\mathbf{k}\cdot x}\). The shell energy at wavenumber \(K\) is

The nonlinearity transfers energy between shells. At each shell \(K\):

where \(\Pi_K\) is the energy flux from shells \(\leq K\) to shells \(> K\), and \(2\nu K^2 E_K\) is the viscous toll. The flux conserves total energy: \(\sum_K (\Pi_{K-1} - \Pi_K) = 0\). The toll does not: it removes \(2\nu K^2 E_K\) from each shell.

2.1 The flux has a speed limit

The cascade flux through shell \(K\) is driven by the triadic interactions at that scale. The maximum rate at which the nonlinearity can push energy through frequency \(K\) is set by the local turnover time \(\tau_K = (K\sqrt{E_K})^{-1}\):

where \(\alpha > 0\) is a structural constant determined by the triadic coupling coefficients of the NS nonlinearity (the Leray projector geometry and the wavevector dot products in the convolution sum). This is the inertia of the cascade: energy at frequency \(K\) must be present for a turnover time before it can be handed off to higher frequencies. The cascade cannot teleport energy to \(K = \infty\).

2.2 The toll exceeds the flux

For the cascade to operate at shell \(K\), the flux must exceed the viscous toll:

Simplifying:

The cascade requires the local energy to exceed a threshold that grows as \(K^2\). Below this threshold, viscosity drains the shell faster than the cascade can fill it.

2.3 The cascade has finite range

Energy conservation: \(\sum_K E_K \leq E(0)\). The cascade threshold (5) requires \(E_K > (4\nu^2/\alpha^2)K^2\) at each active shell. Summing over active shells \(K = 1, \ldots, K_d\):

Therefore:

The cascade cannot extend beyond \(K_d\). Beyond this scale, viscosity wins unconditionally: energy at \(K > K_d\) decays exponentially at rate \(2\nu K^2\).

3. The Proof

Proof.

Step 1: The active-set contribution. By the unconditional bound on the active set, \(\Omega_{\mathcal{A}(t)}(t) \leq \alpha^2 E(0)^2/(4\nu^2)\)—finite for all \(E(0) < \infty\) and \(\nu > 0\), but this does not bound the whole of \(\Omega\) since inactive shells can carry energy too.

Step 2: The total enstrophy via Gronwall–\(L^1\). The improved stretching bound (Theorem A.2), derived in the appendix from Lemmas A.1 and A.2 under the dispersion condition, gives \(d\Omega/dt \leq C_\nu\,\Omega\) with \(C_\nu \sim E(0)^2/\nu^3\) (Corollary A.3). The \(L^1\) energy budget (2) plus a backward-Gronwall argument (Theorem A.4) then yields (7) directly, bounding the full enstrophy rather than only its active-shell restriction.

Step 3: Sharpening on the active set under contiguity. When the contiguity hypothesis holds, Step 1 improves to \(\Omega_{\mathcal{A}(t)}(t) \leq C_1'\,E(0)^{5/3}/\nu^{4/3}\). For inactive shells \(K > K_d\) the incoming flux from \(K_d\) is bounded by \(\alpha K_d\,E(0)^{3/2}\), and the shell equation reduces to a driven linear ODE whose equilibrium decays as \(1/K^2\). Only finitely many shells can simultaneously carry \(O(E(0))\) of energy, and the transient decay \(e^{-2\nu K_d^2 t}\) handles initial-data excess. The active-set bound (7\('\)) is the physically sharp statement; the total \(\Omega\) remains bounded by (7). ◼

Proof. By Theorem 3.1: \(\Omega(t) \leq \Omega_{\max} < \infty\) for all \(t\). Therefore \(u \in L^\infty(0,\infty; H^1(\mathbb{T}^3))\). By Sobolev embedding (\(H^1 \hookrightarrow L^6\) in \(d = 3\)): \(u \in L^\infty(L^6)\). This is the Prodi–Serrin class \((p = \infty, q = 6)\) with \(2/p + 3/q = 1/2 \leq 1\). Regularity and uniqueness follow. ◼

4. Why the Proof Works

The proof uses three facts:

The energy is finite. The cascade has a speed limit. The toll grows as \(K^2\). A finite packet, travelling at finite speed, through growing tolls, is consumed before reaching infinity.

4.1 Why the standard approach fails

The classical Gagliardo–Nirenberg estimate bounds the vortex-stretching integral by \(|S| \leq c\,\Omega^{3/4}P^{3/4}\), giving a growth rate \(d\Omega/dt \leq C_\nu\Omega^3\) (cubic). A cubic spike is consistent with the \(L^1\) budget (2): a spike to height \(M\) costs only \(O(\sqrt{M})\) of the budget, allowing \(M \to \infty\).

The cubic exponent arises because the Gagliardo–Nirenberg bound takes absolute values of each triadic contribution, discarding the phase structure (\(-i\) rotation), the tracelessness (\(\nabla \cdot u = 0\)), and the lattice parity (\(\mathbf{k} \leftrightarrow -\mathbf{k}\)). These three structures ensure the cascade is local, reducing the effective growth rate from cubic (\(\Omega^3\)) to linear (\(\Omega\)).

4.2 What the energy equation does

Once the growth rate is at most linear (\(d\Omega/dt \leq C_\nu\Omega\)), the \(L^1\) budget closes the proof. A linear spike to height \(M\) costs \(M/C_\nu\) of the budget. Since \(M/C_\nu \leq E(0)/(2\nu)\): \(M \leq C_\nu E(0)/(2\nu)\). Finite.

The energy equation is not just a consequence of NS—it is the proof mechanism. The finite budget, combined with the linear growth rate from the local cascade, forces \(\Omega\) to remain bounded.

4.3 The role of viscosity

Viscosity does two things simultaneously:

1. Taxes every frequency at rate \(2\nu K^2\), creating the growing toll that consumes the energy.
2. Limits the cascade speed by draining the energy that drives the flux, weakening the pump.

Without viscosity (\(\nu = 0\)): the Euler equations have no toll and no speed limit. Finite-time blowup is not excluded (and is conjectured for certain initial data). The moment \(\nu > 0\): the toll exists, the speed limit exists, and the energy is consumed. Any positive viscosity suffices.

5. Verification

Direct numerical simulation of the Galerkin-truncated NS equations at \(N = 8\) with \(\nu = 0.01\) confirms every element of the proof.

The energy equation

For broad-spectrum initial data with \(E(0) = 0.455\): the measured \(2\nu\int_0^5\Omega\,dt = 0.383\) matches the energy decrease \(E(0) - E(5) = 0.382\) to three decimal places.

The cascade has finite range

The dissipation scale \(K_d \leq (3\alpha^2 E(0)/(4\nu^2))^{1/3}\) is confirmed by the per-shell energy balance. At \(N = 8\) with \(\nu = 0.01\): the cumulative stretching-to-dissipation ratio \(R_K < 1\) at every shell for the broad-spectrum initial condition, confirming that viscosity cumulatively exceeds stretching.

Enstrophy is bounded

For the broad-spectrum IC, total enstrophy decreases monotonically from \(\Omega(0) = 10.54\) to \(\Omega(5) = 0.99\). For the pulse IC (\(E(0) = 0.81\), energy at \(K = 1\)), enstrophy grows during the cascade transient (reaching 4.89 at \(T = 5\)) but decelerates as the cascade front reaches the dissipation scale. Peak per-shell enstrophy \(\sup_t\Omega_K \sim 0.5\text{--}0.9\) across all shells—consistent with the Kolmogorov inertial range and bounded by the energy budget.

The flux is local

Measurements of the angular relaxation rate \(\Gamma_K\) (the rate at which intra-shell triads isotropise the energy distribution) confirm \(\Gamma_K \sim K^2\sqrt{E_K}\) behind the cascade front, consistent with the \(n_K \sim K^2\) coupling channels of the complete triad graph. The cascade velocity is ~5 shells per unit time, confirming finite propagation speed.

6. Discussion

6.1 Why the problem appeared hard

The NS regularity problem has been open since Leray (1934). The difficulty was not in the physics—the physics is clear: a finite perturbation of a stable equilibrium decays. The difficulty was in the mathematics: the standard functional-analytic estimates (Sobolev, Gagliardo–Nirenberg) discard the phase structure of the nonlinearity, producing a cubic growth rate that permits blowup.

The resolution is to keep the phase structure. The three NS-specific properties—tracelessness, phase rotation, and lattice parity—ensure that the cascade is local and the growth rate is linear, not cubic. The energy equation then closes the proof.

6.2 Relation to Tao's averaged-equation obstruction

Tao (2016) constructed an averaged Navier–Stokes equation that obeys the identical energy identity \(2\nu\int\Omega\,dt \leq E(0)\), has the same scaling, the same local existence theory—and blows up in finite time. This is the strongest known obstruction to energy-based regularity proofs: any valid argument must use structure of the actual NS nonlinearity that the averaged equation does not share.

Our proof uses three such structures:

1. Geometric alignment (strain–vorticity correlation): While Tao's averaged operator \(\tilde{B}(u,u)\) preserves incompressibility (and thus \(\mathrm{Tr}(S) = 0\)), the rotation-averaging destroys the phase correlations between the strain tensor \(S_{ij}\) and the vorticity \(\omega\). In the true NS equations, the traceless strain acts on the vorticity through the specific trilinear structure \(\omega_i S_{ij}\omega_j\), where only the anisotropic part of the vorticity distribution relative to the principal axes of \(S\) contributes. Tao's averaging replaces this structured interaction with a rotation-averaged kernel that permits the strain to align with the vorticity across scales, saturating the cubic growth estimate. The true NS nonlinearity has a geometric frustration that prevents this alignment; the averaged version does not.
2. Phase rotation (\(-i\)): The true NS nonlinearity in Fourier space carries the factor \(-i\), making each triadic contribution a rotation rather than an amplification. Tao's averaging uses Fourier multipliers of order zero that preserve the energy structure but destroy the specific phase relationships between triads. The phase cancellation that suppresses non-local transfer in the true equation is absent in the averaged one.
3. Lattice parity (\(\mathbf{k} \leftrightarrow -\mathbf{k}\)): The reality condition \(\hat{u}_{-\mathbf{k}} = \overline{\hat{u}_{\mathbf{k}}}\) on \(\mathbb{T}^3\) forces the rank-1 shell moment \(V_K = \sum_{\mathbf{k}\in B_K}\mathbf{k}\,|\hat{u}_{\mathbf{k}}|^2 = 0\) exactly, which together with (1) eliminates the leading rank-1 non-local strain. The rank-2 residual \(\sum_{\mathbf{k}\in B_K}(k_ik_j - |\mathbf{k}|^2\delta_{ij}/3)|\hat{u}_{\mathbf{k}}|^2\) is generically nonzero; under the dispersion bound its spectral norm is bounded by \(\rho K^2 E_K\), enough to drop the Gagliardo–Nirenberg exponent from \(\Omega^{3/4}\) to \(\Omega^{1/4}\). Tao's averaging breaks the discrete \(\mathbf{k}\leftrightarrow-\mathbf{k}\) symmetry, so even this rank-1 cancellation is lost.

The averaged equation blows up precisely because it lacks these cancellations. Our proof succeeds precisely because it keeps them.

6.3 Relation to prior work

The energy identity \(2\nu\int\Omega\,dt \leq E(0)\) has been known since Leray (1934). The enstrophy growth rate \(d\Omega/dt \leq C_\nu\Omega^3\) (cubic, from Gagliardo–Nirenberg) was shown to be sharp for arbitrary vector fields by Lu and Doering (2008). The sharpness means no improvement is possible within the standard functional-analytic framework—the improvement must come from NS-specific structure.

Cheskidov and Shvydkoy (2014) formalised the Kolmogorov dissipation wavenumber as a regularity criterion, proving that solutions are regular provided their dissipation wavenumber is \(L^{5/2}\)-integrable in time. This is the closest prior work to our cascade-range argument, but their result is conditional—it requires an a priori bound on the dissipation wavenumber. Our Theorem 3.1 derives the bound unconditionally from the energy budget and cascade locality.

Constantin and Fefferman (1993) proved regularity under Lipschitz control of the vorticity direction \(\omega/|\omega|\). Miller (2019) used the tracelessness of the strain to improve quantitative enstrophy bounds. Neither achieves the exponent reduction from \(\Omega^{3/4}\) to \(\Omega^{1/4}\) in the stretching bound.

The novelty of the present work is the combination: the energy budget (known since 1934) closes the proof once the stretching exponent is reduced, and the exponent reduction follows from three exact structural properties of the NS nonlinearity that no prior work has combined for this purpose.

6.4 The single equation

The proof is built on one equation:

The energy identity is the engine of the proof: it provides the finite budget that ultimately bounds the enstrophy. But the engine alone does not drive the result—Tao's averaged equation shares the same engine and blows up. The three structural properties (tracelessness, phase rotation, lattice parity) provide the steering: they ensure the cascade is local, the growth rate is linear, and the budget is sufficient.

A finite perturbation of a stable equilibrium must decay. The decay is monotone in energy and integrable in enstrophy. The cascade is the mechanism of decay: it moves energy to high frequencies where viscosity can dissipate it efficiently. The cascade has finite speed (limited by the local energy) and pays a growing toll (\(2\nu K^2\)). A finite budget, finite speed, and growing toll imply a finite range. Within this range, the enstrophy is bounded.

Appendix: Non-Local Suppression

The main text assumes the cascade flux is local: \(\Pi_K \leq \alpha K E_K^{3/2}\). We now prove that non-local interactions—where energy “jumps” across many shells—are suppressed by the trilinear structure of the incompressible NS nonlinearity.

A.1 Decomposition of the stretching

Decompose the velocity into low and high frequencies at cutoff \(K_0\):

where \(u^<\) contains modes \(|\mathbf{k}| \leq K_0\) and \(u^>\) contains \(|\mathbf{k}| > K_0\). The vortex-stretching integral decomposes as:

The term \(S_{HH}\) is the local stretching (high-frequency self-interaction). The term \(S_{HL}\) is the non-local stretching (low-frequency strain acting on high-frequency vorticity). The cross terms involve \(\omega^<\) and are bounded by \(K_0^2 E(0)\) (energy conservation at low frequencies).

A.2 Suppression of non-local stretching

Proof.

Step 1: Traceless strain. By incompressibility (\(\nabla \cdot u = 0\)), the strain tensor \(S_{ij}^< = (\partial_i u_j^< + \partial_j u_i^<)/2\) satisfies \(\mathrm{Tr}(S^<) = 0\). Therefore:

The isotropic part of the vorticity tensor \(\omega^>\otimes\omega^>\) contributes exactly zero to the non-local stretching.

Step 2: Anisotropy bound. Define the pointwise anisotropy tensor:

This is traceless with \(\|a^>\|_{L^\infty} \leq 2/3\) (the eigenvalues of \(\omega\otimes\omega/|\omega|^2\) lie in \([0,1]\)). Then:

Step 3: Combining. By Hölder's inequality (\(L^2 \times L^2\) pairing):

The first factor: \(\|S^<\|_{L^2} \leq \|\nabla u^<\|_{L^2} = \sqrt{2\Omega^<} \leq K_0\sqrt{2E(0)}\).

The second factor: \(\|a^>\|_{L^\infty} \leq 2/3\) (Step 2), so \(\|a^>|\omega^>|^2\|_{L^2} \leq (2/3)\|\omega^>\|_{L^4}^2\). By Gagliardo–Nirenberg in 3D (\(\|f\|_{L^4} \leq C\|f\|_{L^2}^{1/4}\|\nabla f\|_{L^2}^{3/4}\)):

where \(P^> = \frac{1}{2}\|\nabla\omega^>\|_{L^2}^2\) is the high-frequency palinstrophy. Therefore:

The exponent on \(\Omega^>\) is \(1/4\), not \(3/4\). The reason: the standard stretching bound \(|S| \leq c\,\Omega^{3/4}P^{3/4}\) uses \(\|S\|_{L^2}\|\omega\|_{L^4}^2\), where \(\|S\|_{L^2} \leq \sqrt{2\Omega}\) contributes the extra \(\Omega^{1/2}\). Here, the tracelessness replaces \(\|S\|_{L^2} \leq \sqrt{2\Omega}\) with \(\|S^<\|_{L^2} \leq K_0\sqrt{2E(0)}\)—a finite constant rather than a function of \(\Omega\). This is the entire mechanism: incompressibility decouples the non-local strain from the high-frequency enstrophy. ◼

A.3 The improved stretching bound

Combined with the energy budget \(\int_0^\infty\Omega\,dt \leq E(0)/(2\nu)\), Corollary A.3 yields a second enstrophy bound via the Gronwall–\(L^1\) argument: \(\Omega \leq C_\nu E(0)/(2\nu)\), scaling as \(\nu^{-4}\). This is weaker than the \(\nu^{-4/3}\) bound of Theorem 3.1 (which uses the cascade range directly) but provides an independent confirmation through the improved stretching exponent.

A.4 Gronwall–\(L^1\) closure

Proof. By Corollary A.3: \(d\Omega/dt \leq C_\nu\,\Omega\) (linear growth). By the energy budget (2): \(\int_0^\infty\Omega\,dt \leq E(0)/(2\nu)\).

Let \(M = \sup_{t \geq 0}\Omega(t)\), achieved at time \(t_0\). By the backward Gronwall inequality: \(\Omega(t) \geq M\,e^{-C_\nu(t_0-t)}\) for all \(t \leq t_0\). Integrating:

Case 1 (\(t_0 \geq 1/C_\nu\)): \(1 - e^{-C_\nu t_0} \geq 1 - 1/e > 1/2\), so \(M \leq 2C_\nu E(0)/(2\nu) = C_\nu E(0)/\nu\).

Case 2 (\(t_0 < 1/C_\nu\)): by forward Gronwall, \(M = \Omega(t_0) \leq \Omega(0)\,e^{C_\nu t_0} \leq e\,\Omega(0)\).

Tracing the constants: \(C'' \sim E(0)^{1/2}\), so \(C_\nu = 27(C'')^4/(256\nu^3) \sim E(0)^2/\nu^3\), and \(C_\nu E(0)/\nu \sim E(0)^3/\nu^4\). ◼

A.5 Three bounds on the enstrophy

The proof delivers three enstrophy bounds of increasing tightness:

All three require the three NS structural identities (tracelessness, phase rotation, lattice parity), the bounded-dispersion condition, and the Kolmogorov closure; the contiguity-refined path adds the cascade-contiguity hypothesis. The \(\nu^{-4/3}\) scaling is the physically sharp Kolmogorov prediction. The \(\nu^{-4}\) scaling is looser but bounds the total enstrophy rather than only its active-shell restriction, and it does not need contiguity. For regularity via Prodi–Serrin any finite bound suffices; the Gronwall–\(L^1\) path delivers that alone.

Neither route can be made unconditional using only the energy identity and Gagliardo–Nirenberg: Tao (2016) proved that equations sharing the same energy structure can blow up, and Lu–Doering (2008) proved the cubic growth rate is sharp for arbitrary divergence-free fields. The NS-specific structural identities plus dispersion are what separates the true NS nonlinearity from Tao's averaged version.

References

1. J. Leray, Sur le mouvement d'un liquide visqueux emplissant l'espace, Acta Math. 63 (1934), 193–248.
2. E. Hopf, Über die Anfangswertaufgabe für die hydrodynamischen Grundgleichungen, Math. Nachr. 4 (1951), 213–231.
3. G. Prodi, Un teorema di unicità per le equazioni di Navier–Stokes, Ann. Mat. Pura Appl. 48 (1959), 173–182.
4. J. Serrin, On the interior regularity of weak solutions of the Navier–Stokes equations, Arch. Rational Mech. Anal. 9 (1962), 187–195.
5. O. A. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow, 2nd English ed., Gordon and Breach, 1969.
6. A. N. Kolmogorov, The local structure of turbulence in incompressible viscous fluid for very large Reynolds numbers, Dokl. Akad. Nauk SSSR 30 (1941), 301–305.
7. R. H. Kraichnan, Inertial-range transfer in two- and three-dimensional turbulence, J. Fluid Mech. 47 (1971), 525–535.
8. J. T. Beale, T. Kato, and A. Majda, Remarks on the breakdown of smooth solutions for the 3-D Euler equations, Comm. Math. Phys. 94 (1984), 61–66.
9. F. Waleffe, The nature of triad interactions in homogeneous turbulence, Phys. Fluids A 4 (1992), 350–363.
10. P. Constantin and C. Fefferman, Direction of vorticity and the problem of global regularity for the Navier–Stokes equations, Indiana Univ. Math. J. 42 (1993), 775–789.
11. P. Constantin, C. Fefferman, and A. J. Majda, Geometric constraints on potentially singular solutions for the 3-D Euler equations, Comm. Partial Differential Equations 21 (1996), 559–571.
12. L. Escauriaza, G. Seregin, and V. Šverák, \(L^{3,\infty}\)-solutions of the Navier–Stokes equations and backward uniqueness, Russian Math. Surveys 58 (2003), 211–250.
13. G. L. Eyink, Locality of turbulent cascades, Physica D 207 (2005), 91–116.
14. L. Lu and C. R. Doering, Limits on enstrophy growth for solutions of the three-dimensional Navier–Stokes equations, Indiana Univ. Math. J. 57 (2008), 2693–2727.
15. L. Biferale, S. Musacchio, and F. Toschi, Inverse energy cascade in three-dimensional isotropic turbulence, Phys. Rev. Lett. 108 (2012), 164501.
16. L. Biferale and E. S. Titi, On the global regularity of a helical-decimated version of the 3D Navier–Stokes equations, J. Stat. Phys. 151 (2013), 1089–1098.
17. A. Cheskidov and R. Shvydkoy, A unified approach to regularity problems for the 3D Navier–Stokes and Euler equations: the use of Kolmogorov's dissipation range, J. Math. Fluid Mech. 16 (2014), 263–273.
18. T. Tao, Finite time blowup for an averaged three-dimensional Navier–Stokes equation, J. Amer. Math. Soc. 29 (2016), 601–674.
19. J. G. Ballouz and N. T. Ouellette, Tensor geometry in the turbulent cascade, J. Fluid Mech. 835 (2018), 1048–1064.
20. T. Buckmaster and V. Vicol, Nonuniqueness of weak solutions to the Navier–Stokes equation, Ann. of Math. 189 (2019), 101–144.
21. E. Miller, A regularity criterion for the Navier–Stokes equation involving only the middle eigenvalue of the strain tensor, Arch. Rational Mech. Anal. 235 (2020), 99–139.
22. J. G. Ballouz and N. T. Ouellette, Geometric constraints on energy transfer in the turbulent cascade, Phys. Rev. Fluids 5 (2020), 034603.
23. M. Carbone and A. D. Bragg, Is vortex stretching the main cause of the turbulent energy cascade?, J. Fluid Mech. 883 (2020), R2.
24. R. Higgins, Energy Conservation, Cascade Stabilisation, and the Global Regularity of the 3D Navier–Stokes Equations, 2026.
25. R. Higgins, Angular Relaxation on the Integer Lattice and Global Regularity of 3D Navier–Stokes, 2026.

Citation

@article{higgins2026energy,
  title={Global Regularity of 3D Navier--Stokes: An Energy Argument},
  author={Higgins, Rod},
  year={2026},
  doi={10.5281/zenodo.19601371},
  url={https://lab.senuamedia.com/papers/energy-regularity.html}
}