betaFeedback

WorksPhysics and Astronomy

Generating Kolmogorov spectra and structure functions with a Rosenhead profile trefoil vortex knot

Abstract

Motivated by recent experimental observations by Matsuzawa et al. (2026) of colliding several vortex rings within a water tank, this paper extends a compact trefoil vortex simulation to later times and generates a persistent ==energy spectrum that obeys Kolmogorov scaling, together with other statistics characteristic of fully developed turbulence, including structure function scaling. This temporal extension uses a larger (4pi)3 periodic domain, with the improved compactness being achieved by the use of an algebraic Rosenhead vorticity profile, from which viscosity(nu)-independent finite energy dissipation is generated and pre-reconnection convergence of nu1/4-dependent higher-order vorticity moments. These properties suggest, perhaps for the first time, how decaying Navier-Stokes simulations can generate fully-developed turbulent flow from an unforced initial state that is far from boundaries, thereby suggesting how further simulations at higher values of Re could be harnessed for experimental comparisons. Moreover, the exponents for the energy decay and the integral scale growth, E(t)~(t+t0)-10/7 and and ell(t)~(t+t0)2/7, follow those based upon preservation of the Loitsyanskii integral. On a related question, an integral scale growth rate ell that is close to the experiment rate of ellX~(t+t0)(0.16or1/7)is only obtained if the Fourier modes at the scale of the periodic domain are not considered.

Background

One of the greatest challenges in improving our understanding of three-dimensional incompressible turbulence is producing, without external forcing, a turbulent flow that is far from boundaries, has a cascade of energy to small scales, finite dissipation rates and a E(k,t)∼k−5/3E(k,t) \sim k^{-5/3} wavenumber spectrum. Unlike turbulence associated with boundary layers [22], or with thermal convection [2], a clear experimental equivalent has not been found, with only measurements from wind tunnels and wakes available for comparative purposes. Moreover, numerical comparisons have been limited to isotropic calculations in periodic domains across periodic boundaries that require the use of artificial forcing to sustain any turbulence.

In a recent significant experiment, turbulence far from boundaries has been generated by the simultaneous firing of multiple vortex rings into a water tank [15], from which E(k,t)∼k−5/3E(k,t) \sim k^{-5/3} spectra have been observed over two wavenumber decades that persist for several characteristic time-scales, or eddy turnover times.

To replicate these results numerically, this paper extends in time and domain size a modest Reynolds number trefoil vortex knot calculation [10] using the more compact Rosenhead algebraic vorticity profile. Initialisation is done in the larger (4π)3(4\pi)^3 domain by putting the original initial condition into a (2π)3(2\pi)^3 subdomain, applying a hyperviscous filter, and then remeshing onto 102431024^3 resolution. Figure 1 presents the new numerical spectra in two ways. For comparison with the new experiment [15], the first illustrates the formation of one decade of a E(k,t)∼k−5/3E(k,t) \sim k^{-5/3} spectrum that persists for roughly four eddy turnover times. The second demonstrates Kolmogorov similarity.

Two rescalings of the wavenumber spectra and wavenumbers

Figure 1. Two rescalings of the wavenumber spectra and wavenumbers. Note first that both E(k)E(k) and ϵ(t)\epsilon(t) are rescaled by a factor of 8 to E8(k)=8E(k)E_8(k)=8E(k) and ϵ8=8ϵ\epsilon_8=8\epsilon with η8=(ν3/ϵ8)1/4\eta_8=(\nu^3/\epsilon_8)^{1/4} because the initial trefoil fits within the inner (2π)3(2\pi)^3 domain, and past numerical experience going from Yeung et al. [23] back to Kerr [7] comes from forced HIT calculations in (2π)3(2\pi)^3 domains Second, both frames use Kolmogorov scaled η8k\eta_8 k wavenumbers and (ϵ8ν5)1/4(\epsilon_8\nu^5)^{1/4} scaled E(k)E(k). The differences are: (a) The E8(k)E_8(k) spectra are not multiplied by (η8)5/3(\eta_8)^{5/3} so as to show the similarities to the new experimental spectra of Matsuzawa et al. [15]. Note that without rescaling, all of the spectra decay with time. (b) The same data but multiplied by (η8k)5/3(\eta_8 k)^{5/3} and including the k∼0k\sim0 shell. A plausible Kolmogorov constant of 1.25 runs through the spectra. The rescaled (η8k)5/3E8(k)≤0.005(\eta_8 k)^{5/3}E_8(k)\leq0.005 is relatively large, with the inset showing how E(ηk∼0.01)E(\eta k\sim0.01) (and all greater kk) are decaying with time.

The numerical method is based upon earlier vortex knot simulations [8 b] that replicated the topological changes of the series innovative vortex knot experiments [19, 20] that has now led to the more recent experiment by Matsuzawa et al. [15]. Trefoil knot simulations using algebraic vorticity profiles have several mathematical advantages. First, they are compact in the sense that their velocity (1) and vorticity (2) decay exponentially from their centres, thus minimizing the influence of the periodic boundaries. A trefoil also has strong, non-trivial helicity due to its topology.

However, a disadvantage of using trefoil knots comes from how the vorticity of the strongly curved vortices is mapped onto a Cartesian mesh. Using recent stability analysis [4] and simulations, Kerr [10] has shown why algebraic profiles are a better choice than Gaussian profiles, mainly because they allow the shedding of finer scale structures such as h<0h<0 helical vortex sheets, whose primary reconnection-scale properties have now been identified for several algebraic profile simulations [8, 10, 11]. Figure 3a,Figure 3d show the sheets’ remnants and their primary associated properties include temporally local scaling with a set of ν\nu-independent convergences of the higher-order vorticity moments [11], as illustrated in Figure 2c, with the sheets’ post-reconnection interactions generating finite energy dissipation rates and eventually temporally-integrated finite dissipation as in 2a.

Global diagnostics showing continuity with the $(2\pi)^3$ Rosenhead calculations

Figure 2. Global diagnostics designed to show continuity with the (2π)3(2\pi)^3 Rosenhead calculations in Kerr [10]. (a) The ν=8.35×10−5\nu= 8.35 \times10^{-5} evolution of the maximum vorticity ω∞(t)=∥ω(t)∥∞\omega_\infty(t) = \lVert\boldsymbol{\omega}(t)\rVert_\infty, then the dissipation rates ϵ(t)=νZ(t)\epsilon(t) = \nu Z(t) for ν=8.35×10−5, 1.667×10−4, 3.334×10−4\nu= 8.35 \times10^{-5},\ 1.667 \times10^{-4},\ 3.334 \times10^{-4} (ZZ is the enstrophy (4)). The spike in ω∞\omega_\infty at t=4t = 4 is followed by the maxima of the dissipation rates for all ν\nu at t∼0.5t \sim0.5, as found for the (2π)3(2\pi)^3 calculation [10], and represents roughly when the first reconnection ends. (b) For ν=8.35×10−5\nu= 8.35 \times10^{-5}, ϵ(t)\epsilon(t), a ϵ~=0.5E3/2/ℓ\tilde{\epsilon} = 0.5E^{3/2}/\ell estimate, and a scaled helicity H(t)/2E(t)Z(t)\mathcal{H}(t)/\sqrt{2E(t)Z(t)}. After t≥15t \ge15, the decaying ϵ(t)\epsilon(t) and the scaled helicity both settle down. That the post-reconnection scaled helicity is relatively constant, suggests the preservation of some aspect of the global topological structure. (c) For all ν\nu, early time ν\nu-independent convergences of ν1/4∥ω(t)∥∞\nu^{1/4}\lVert\boldsymbol{\omega}(t)\rVert_\infty and νZ(t)=(ν1/4ΩV1(t))2\sqrt{\nu Z(t)} = (\nu^{1/4}\Omega_{V1}(t))^2 at t∞=4t_\infty= 4 and t=ttx=4.6t = t_{t_x} = 4.6 respectively, with txt_x showing when the first reconnection ends. Not shown is the full ν1/4ΩVm(t)\nu^{1/4}\Omega_{V_m}(t) hierarchy with convergences at their tmt_m, as in Kerr [11].

Three-dimensional visualizations of the splitting of a trefoil into two rings, with vertical enstrophy profiles and temporal separation

Figure 3. Splitting of trefoil into two rings is shown using three-dimensional hh-mapped isosurfaces (dominated by green to blue h≳0h \gtrsim0), local ωm\omega_m locations at X’s that seed the upper (deep red) and lower (blue) centrelines, plus vertical profiles.. (a) t=4.4t = 4.4 shows ∣ω∣=6.3|\omega| = 6.3 just before reconnection finishes with only one vortex line.. (b) Vertical profiles of the enstrophy Z(z,t)Z(z,t) showing that the vertical separation of the new rings begins at t≈5t \approx5, as the two rings propagate to −z-z, with zz-mid the average zz between the two rings. The z(ωm(t))z(\omega_m(t)) line charts how the maximum of vorticity jumps between the upper and lower rings. (c) This charts the separation between the two rings, which provides one mechanism and measure for how the large scales increase. Δz=0\Delta z = 0 only at t=5t = 5. Frames d-f follow the ω\omega-rings as both are propagating to −z-z. (d) t=5.2t = 5.2, just after the first reconnection with an additional rose ω=10.8\omega= 10.8 isosurface to represent a h∼−10h \sim-10 remnant sheet, through which h=±20h = \pm20 braids poke. Plus an inset showing two closed grey uu-streamlines. (e) t=6.4t = 6.4 is similar to t=5.2t = 5.2 with additional braids that connect the rings. (f) t=30.4t = 30.4, with with the inset showing the rings final, off-set triangular shapes. A yellow curve in both projections shows how the two rings are connected.

This paper focuses more on the evolution after the first reconnection and is organized as follows. The equations and numerics are outlined in §1.1 followed by the definition of the trefoil vortex knot. After that, in §2 the spectral forms are plotted in Figure 1. Then §2.1 discusses the evolution of several global vorticity, energy and helicity measures that connect these results to recent work [10, 11], and which also provide insight into the times chosen for the three-dimensional structures in Figure 3. The first of these times appears just before the trefoil separates into two rings, before evolving into two triangular rings at a later time. §2.2 contains a discussion of how the large scale diagnostics grow.

Governing equations

The governing equations are the three-dimensional, incompressible Navier-Stokes equations for the velocity (div⁡u=0\operatorname{div} \boldsymbol{u}=0) and the vorticity ω=∇×u\boldsymbol{\omega}=\nabla\times\boldsymbol{u}.

∂u∂t+(u⋅∇)u=−∇p+νΔu,(1)\frac{\partial\boldsymbol{u}}{\partial t}+(\boldsymbol{u}\cdot\nabla)\boldsymbol{u}=-\nabla p+\nu\Delta\boldsymbol{u}, \tag*{(1)}
∂ω∂t+(u⋅∇)ω=(ω⋅∇)u+νΔω.(2)\frac{\partial\boldsymbol{\omega}}{\partial t}+(\mathbf{u}\cdot\nabla)\boldsymbol{\omega}=(\boldsymbol{\omega}\cdot\nabla)\mathbf{u}+\nu\Delta\boldsymbol{\omega}. \tag*{(2)}

These equations have three quadratic diagnostics, the continuum densities of the energy e=12∣u∣2e=\frac{1}{2}\lvert\mathbf{u}\rvert^2, enstrophy ζ=∣ω∣2\zeta=\lvert\boldsymbol{\omega}\rvert^2 and helicity h=u⋅ωh=\mathbf{u}\cdot\boldsymbol{\omega}, which can take either sign. The equations for ee and ζ\zeta, plus their integrated values, are

∂e∂t+(u⋅∇)e=−∇⋅(up)+νΔe−ν(∇u)2,E=12∫u2 dV,(3)\frac{\partial e}{\partial t}+(\mathbf{u}\cdot\nabla)e=-\nabla\cdot(\mathbf{u}p)+\nu\Delta e-\nu(\nabla\mathbf{u})^2,\qquad E=\frac{1}{2}\int u^2\,dV, \tag*{(3)}
∂ζ∂t+(u⋅∇)∣ω∣2=2ωSω+νΔ∣ω∣2−2ν(∇ω)2,Z=∫ω2 dV,(4)\frac{\partial\zeta}{\partial t}+(\mathbf{u}\cdot\nabla)\lvert\boldsymbol{\omega}\rvert^2=2\boldsymbol{\omega}\mathbf{S}\boldsymbol{\omega}+\nu\Delta\lvert\boldsymbol{\omega}\rvert^2-2\nu(\nabla\boldsymbol{\omega})^2,\qquad Z=\int\omega^2\,dV, \tag*{(4)}

where the local dissipation rates are ϵ=νζ=ν(∇u)2\epsilon=\nu\zeta=\nu(\nabla\mathbf{u})^2 and ϵζ=2ν(∇ω)2\epsilon_\zeta=2\nu(\nabla\boldsymbol{\omega})^2 and the enstrophy production rate is ζp=ω⋅Sω\zeta_p=\boldsymbol{\omega}\cdot\mathbf{S}\boldsymbol{\omega}.

Due to these robust relationships between the energy density ee and enstrophy density ζ\zeta, Reynolds number-independent energy dissipation, a dissipation anomaly, ΔEϵ=∫0Tϵϵ dt>0\Delta E_{\epsilon} = \int_{0}^{T_{\epsilon}} \epsilon\,\mathrm{d}t > 0, can form in a finite-time TϵT_{\epsilon} at large Reynolds numbers for another set of algebraic profile trefoils [11]. Negative h<0h < 0 vortex sheets play a crucial role in generating finite ΔEϵ\Delta E_{\epsilon} [10, 11], with the equation for hh, and the integrated helicity H=∫u⋅ω dV\mathcal{H} = \int\mathbf{u}\cdot\boldsymbol{\omega}\,\mathrm{d}V, obeying

∂h∂t+(u⋅∇)h=−ω⋅∇Π+νΔh−2νtr⁡(∇ω⋅∇uT)(5)\frac{\partial h}{\partial t} + (\mathbf{u}\cdot\nabla)h = -\boldsymbol{\omega}\cdot\nabla\Pi+ \nu\Delta h - 2\nu\operatorname{tr}(\nabla\boldsymbol{\omega}\cdot\nabla\mathbf{u}^{\mathsf{T}}) \tag*{(5)}

Another set of diagnostics associated with the vortex sheets are higher-order vorticity moments ΩVm(t)=(∫V∣ω∣2m dV)1/2m\Omega_{V_m}(t) = \left(\int_{V} |\boldsymbol{\omega}|^{2m}\,\mathrm{d}V\right)^{1/2m}. Those moments are being volume-integrated to allow comparisons between calculations done in different periodic domains [11], a task made easier here because all the new calculations use a V=(4π)3V=(4\pi)^3 domain. Discussed next is one way to address similar variable domain issues when comparing spectra from simulations performed in different domains.

Once a means of comparing variable domains was determined, a set of ν\nu-independent, tmt_m temporal convergences was found by taking ν1/4ΩVm(t)\nu^{1/4}\Omega_{V_m}(t), with this hierarchy of the tmt_m: t∞≤⋯tm⋯≤t1=txt_{\infty}\leq\cdots t_m\cdots\leq t_1=t_x. Figure 2c shows that scaling for ν1/4∥ω∥∞\nu^{1/4}\|\boldsymbol{\omega}\|_{\infty} and νZ=(ν1/4ΩV1)2\sqrt{\nu}Z=(\nu^{1/4}\Omega_{V_1})^2 for the Rosenhead profile calculations, with all the remaining scaled moments sandwiched between those. The origin of this new ν1/4\nu^{1/4} scaling is being investigated using a single orthogonal vortex pair with Rosenhead vorticity profiles.

The simulated viscosities are: ν=8.35×10−5\nu=8.35\times10^{-5} and ν=1.667×10−4\nu=1.667\times10^{-4} on n3=10243n^3=1024^3 periodic meshes and ν=3.334×10−4\nu=3.334\times10^{-4} on a 5123512^3 mesh. Following Bustamante & Kerr [3], the calculations are 2/3rds dealiased with a hyperviscous filter.

The vortex knot is initialised as in Kerr [10] with the two primary elements being the trajectory of the trefoil centreline and the mapping of the vorticity profile ωraw(ρ)\omega_{\mathrm{raw}}(\rho) onto the mesh using the distance ρ\rho and tangent ∂ξ0(ϕ)\partial\boldsymbol{\xi}_0(\phi) of the nearest point on the trajectory. The initial closed, unperturbed, double loop trajectory of all of the trefoils is ξ0(ϕ)=[x(ϕ),y(ϕ),z(ϕ)]=[r(ϕ)cos⁡(α),r(ϕ)sin⁡(α),acos⁡(α)]\boldsymbol{\xi}_0(\phi)=[x(\phi),y(\phi),z(\phi)]=[r(\phi)\cos(\alpha),r(\phi)\sin(\alpha),a\cos(\alpha)] with ϕ=1:4π\phi=1:4\pi and a=0.5a=0.5, w=1.5w=1.5, rf=1r_f=1, r(ϕ)=rf+asin⁡(wϕ)r(\phi)=r_f+a\sin(w\phi) and α=ϕ+acos⁡(wϕ)/(wrf)\alpha=\phi+a\cos(w\phi)/(wr_f). The algebraic profiles ωraw(ρ)\omega_{\mathrm{raw}}(\rho) are parameterized by a radius r∘r_{\circ}, its maximum/centerline vorticity ω∘\omega_{\circ} and a power prp_r, giving ωraw(ρ)=ω∘(r∘)pr/(ρ2+r∘2)pr\omega_{\mathrm{raw}}(\rho)=\omega_{\circ}(r_{\circ})^{p_r}/(\rho^2+r_{\circ}^2)^{p_r}, followed by a hyper-viscous smoothing. For the algebraic Rosenhead profile here, pr=2p_r=2 and r∘=0.05r_{\circ}=0.05.

Results: Three-dimensional spectra and temporal scaling

Energy spectra E(k,t)E(k,t) are formed by taking discrete Fourier transforms and for the L3=(4π)3L^3=(4\pi)^3 domain with n=1024n=1024, the minimum wavenumber shell is kmin⁡=min⁡k≠0(k)=2π/L=1/2k_{\min}=\min_{\mathbf{k}\ne0}(k)=2\pi/L=1/2 and when moderated by the η\eta microscale and plotted, min⁡(ηk)∼0.01\min(\eta k)\sim0.01. With the wavenumber shells E(k,t)E(k,t) filled using:

E(k,t)=12∑k∣uk(t)∣2overk≤∣k∣<k+kmin⁡.(6)E(k,t)=\frac{1}{2}\sum_k|\mathbf{u}_k(t)|^2 \quad\text{over}\quad k\leq|\mathbf{k}|<k+k_{\min}. \tag*{(6)}

Historically, attempts to generate persistent E(k,t)∼k−5/3E(k,t)\sim k^{-5/3} spectra from unforced decaying turbulence have used only a small set of particularly unstable Fourier modes, such as the Taylor-Green vortex Brachet et al. [1], or random initial conditions with a spectral peak at small or moderate wavenumbers [6] (Rodhiya & Sreenivasan 2026). For both, there is an inherent limitation when the developing turbulence reaches the periodic boundaries, reacts back upon itself, and inhibits the formation of −5/3-5/3 spectral cascade. This raises two questions.

a) Could an unstable vortical configuration do better? Kerr [11] showed that a dissipation anomaly (1.3) is possible if the periodic domain increases as the Reynolds number increased. However, the extent of the -5/3 regime was limited by the broad algebraic profile that was used.

b) And could the more compact algebraic Rosenhead profile from Kerr [10] do even better? Possibly, but only in a larger (4π)3(4\pi)^3 domain. The calculations in this paper contribute to that interpretation.

What is the effect of domain size? Because the initial condition was originally created for a (2π)3(2\pi)^3 domain, the initial condition is very inhomogeneous and the volume-averaged spectra are a factor of 23=82^3 = 8 smaller than those of the original (2π)3(2\pi)^3 calculation or forced calculations done in (2π)3(2\pi)^3 domains. To compensate, all the energies and enstrophies used for scaling the spectra are multiplied by 8=238 = 2^3. The result in Figure 1 is spectra that are similar to those from the earlier (2π)3(2\pi)^3 calculation (Kerr [10]) and spectra from homogeneous turbulent flows, including forced homogeneous, isotropic (HIT) calculations in (2π)3(2\pi)^3 domains going from Yeung et al. [23] back to Kerr [7].

With that background, Figure 1 presents two rescalings of the E(k,t)E(k,t) spectra, with both rescalings sharing Kolmogorov scaled wavenumbers kK=η8(t)kk_K = \eta_8(t)k, with η8=(ν3/(8ϵ(t)))1/4\eta_8 = (\nu^3/(8\epsilon(t)))^{1/4}, and with the 8E(k,t)8E(k,t) scaled by (8ϵ(t)ν5)1/4(8\epsilon(t)\nu^5)^{1/4}. With these differences:

(a) For qualitative comparisons with the similarly-scaled experimental spectra in Matsuzawa et al. [15], the E(kK8,t)/(8ϵ(t)ν5)1/4E(k_{K8},t)/(8\epsilon(t)\nu^5)^{1/4} spectra, without E(min⁡k)E(\min_k), are plotted.

(b) To provide more quantitative comparisons with HIT simulations and the (2π)3(2\pi)^3 spectra for the same initial condition, the min⁡k\min_k shells are included and the spectra are multiplied by kK5/3k_K^{5/3} to provide an estimate for the Kolmogorov constant CKC_K if 8E(kK8,t)8E(k_{K8},t) obeys

8E(kK8,t)∼CKϵ82/3kK8−5/3withCK∼1.25.(7)8E(k_{K8},t) \sim C_K\epsilon_8^{2/3}k_{K8}^{-5/3}\quad\text{with}\quad C_K \sim1.25 . \tag*{(7)}

Evolution of three-dimensional diagnostics and structure

The focus of this paper has been upon the period between when the first reconnection ends at tx≈4−4.6t_x \approx4 - 4.6, and tf∼40≳7txt_f \sim40 \gtrsim7t_x, with Kolmogorov similarity still seen in Figure 1 has lasted for the, with Figure 2a showing one of the peaks of ∥ω(t)∥∞\lVert\omega(t)\rVert_\infty near txt_x and strong ϵ(t)\epsilon(t) appearing for all t>txt > t_x.

Figure 2b shows two ratios that underlie aspects of the t>16t > 16 dynamics. The first is H(t)/2E(t)Z(t)\mathcal{H}(t)/\sqrt{2E(t)Z(t)}, which is surprisingly constant for t>7t > 7, despite the significant change in H(t)\mathcal{H}(t) (1.5) at t∼tx=4.6t \sim t_x = 4.6 due to reconnection. This steady H(t)/2E(t)Z(t)\mathcal{H}(t)/\sqrt{2E(t)Z(t)} might be due to an underlying, persistent structure between the new rings in Figure 3 as they propagate downwards.

The second is a classical turbulence estimate of the dissipation rate ϵ~=0.5E3/2/ℓ\tilde{\epsilon} = 0.5E^{3/2}/\ell that uses the integral scale ℓ\ell (2.3). This is compared to the directly determined dissipation rate ϵ(t)\epsilon(t) (1.3) in Figure 2b. That relatively good comparison would support the predictions for energy decay and integral scale growth based upon the Loitsyanskii integral (2.5) and given in Figure 4a,Figure 4b.

Time-dependencies of Rosenhead profile energy and velocity-based length scales

Figure 4. Time-dependencies of Rosenhead profile energy and velocity-based length scales. Most use ν=8.35×10−5\nu= 8.35 \times10^{-5}. (a) Two measures of the energy decay. Plausible fits for all are E∼(t+70)−10/7E \sim(t+70)^{-10/7}, which would be consistent with the scaling of based upon the Loitsyanskii integral (10). ET(t)E_T(t) is the full energy, with ETν=3.3×10−4∼(t+70)−10/7E_T^{\nu=3.3\times10^{-4}} \sim(t+70)^{-10/7} obeyed to t=92t=92. E−0=E−E(k=1)E_{-0}=E-E(k=1) does not include the ηk∼0.01\eta k\sim0.01 Fourier modes, for which another plausible, but less exact, fit would be (t+60)−6/5(t+60)^{-6/5}, the Saffman L\mathcal{L} prediction and what is observed for grid decay [13, 21]. (b) Two measures of the large, integral length scale ℓ\ell (8), one for ET(t)E_T(t) with ℓ∼t2/7±0.03\ell\sim t^{2/7\pm0.03}, the Loitsyanskii-based scaling, and the other for E−0E_{-0}, ℓ(−0)∼(t+70)0.11\ell_{(-0)}\sim(t+70)^{0.11} (dashed-red-orange), which is the same order as the ℓ∼t0.16\ell\sim t^{0.16} scaling of the new experiment [15]. Neither is near the ℓ∼t2/5\ell\sim t^{2/5} Saffman prediction [13]. (c) Root-mean-square radii based upon the energy and enstrophy, which are both controlled by the horizontal sizes/radii of the two rings. Plus the vertical separation into two swirling rings and the Taylor microscale λ≲0.5\lambda\lesssim0.5, with max⁡(R8λ(t≥5))≈500\max(R_{8\lambda}(t\ge5))\approx500. (d) Inset the ratio of ℓ\ell and rω,rmsr_{\omega,\mathrm{rms}}.

Figure 2c shows, for the three viscosities, the t→tmt \to t_m convergences of two vorticity diagnostics. For m=1m = 1, or equivalently for the reconnection-enstrophy νZ(t)=((ν1/4Ω1)2\sqrt{\nu Z(t)} = ((\nu^{1/4}\Omega_{1})^2 (after (1.4)) and m=∞m = \infty, ν1/4∥ω∥∞\nu^{1/4}\lVert\omega\rVert_\infty. These are at tx=t1=4.6t_x = t_1 = 4.6 and t∞=4t_\infty= 4 respectively. Additional Ωνm\Omega_{\nu m} have been calculated, with all of the tmt_m convergence times sandwiched between t∞=4t_\infty= 4 and tx=4.6t_x = 4.6, as in Kerr [10].

The evolution of the three-dimensional vorticity in Figure 3 uses helicity density, hh-mapped, vorticity isosurfaces. Using hh has previously helped in identifying the dynamical origins of the smallest scale vortex structures during reconnection, including showing how the hf=ωh_f = \omega-transport term (1.5)) leads to the h≤0h \le0 vortex sheets (Kerr [10]). As before (Kerr [10, 11]), those sheets allow the global enstrophy Z(t)Z(t) and the energy dissipation rates ϵ(t)\epsilon(t) to continue increasing after reconnection. An event that transforms the trefoil into two connected vortex rings, as illustrated using two types of visualisation in Figure 3, and observed in experiments [12]. This begins at t=4.4≲tx=4.6t = 4.4 \lesssim t_x = 4.6 in Figure 3a when only one centerline vortex can be identified, along with reddish and yellow h≲0h \lesssim0 sheets that are still being shed of off the three self-reconnecting local knots at the bottom.

The evolution of the new vortex rings from t=5.2t = 5.2 onwards is given in Figure 3d-Figure 3f. Also included in these frames are the post-reconnection vortex structures the develop out of the t<5.2t < 5.2 vortex sheets and new structures being shed by the rings. With one goal of these visualisations being to provide suggestions for how the k−5/3k^{-5/3} spectra and strong energy dissipation rates form.

At t=5.2t = 5.2, the primary ω=30\omega= 30 isosurfaces are shown in tones of h≳0h \gtrsim0 green that can be divided into three zones. First, the upper vortex ring with a red centreline. Second, the lower vortex ring with a blue centreline. And third, three braided bridges that are embedded within a second ∣ω∣=10.8|\omega| = 10.8, h<0h < 0 rose isosurface that is a remnant of the earlier h≲0h \lesssim0 vortex sheets. Where these bridges meet the rings are where the two rings had formed out of the trefoil. And near the bottom of one bridge in particular, new alternating red/blue (h±0h \pm0) braids are poking through the rose-coloured sheets before extending upwards.

Two additional grey velocity uu-streamlines are also shown, partially in the main figure and more fully in the inset that shows that these streamlines are closed Γ\Gamma-rings. Once identified, Γ=∮u⋅ds\Gamma= \oint\mathbf{u} \cdot\mathrm{d}\mathbf{s} integrals about the two rings generate circulations of Γ≈2\Gamma\approx2 (for the trefoil’s two loops). This property is preserved from t=0t = 0 to t∼10t \sim10 and could be a source for the Loitsyanskii integral I\mathcal{I} (10). However, for t>10t > 10 the closed loops want to (but cannot) extend beyond the periodic boundaries, making this method for calculating Γ\Gamma untenable.

Frame 3e at t=6.4t = 6.4 shows how the tiny (h=±20h = \pm20) t=5.2 braids evolve into larger transient structures between the rings, structures that over time are being continually renewed, stretched, then coiled around those rings. Frame 3f shows that this is a recurring pattern that could provide a mechanism for a continual refreshment of the k−5/3k^{-5/3} spectra that persist over these times. Note that over time, each ring becomes triangular with a relative rotation of 60° to the other, as shown by the inset. The yellow ss-line shows the path of one of those braids between the rings, beginning at the red + on the upper ring and ending at the blue + in the centre of the lower ring (also in inset).

Frame 3b quantifies the separation of the rings spawned by the original trefoil by showing the temporal evolution of vertical profiles of the enstrophy Z(z,t)Z(z,t), also seen experimentally [12], and how the position of ωm\omega_m jumps between the rings over time. With 3c following Δzsplit(t)\Delta z_{\mathrm{split}}(t), the distance between the two vertical Z(z,t)Z(z,t) maxima after the vertical enstrophy profile splits in 3d. This growth rate is preserved from 2tx≈102t_x \approx10 to the end of the calculation at t=43.2t = 43.2.

Large-scale lengths and the Loitsyanskii invariant I\mathcal{I}

Besides the energy spectra, additional decaying turbulence statistics given by Matsuzawa et al. [15] include E(t)∼t−10/7E(t) \sim t^{-10/7} energy decay and the evolution of three large-scale lengths in Figure 4c. The integral time-scale ℓ(t)\ell(t), plus Ru,rmsR_{u,\mathrm{rms}} and Rω,rmsR_{\omega,\mathrm{rms}} for turbulent blob expansion, with

ℓ(t)≡(3π/4)∫0∞k−1E(k,t) dk∫0∞E(k,t) dk,(8)\ell(t) \equiv(3\pi/4) \frac{\int_{0}^{\infty} k^{-1}E(k,t)\,\mathrm{d}k}{\int_{0}^{\infty} E(k,t)\,\mathrm{d}k}, \tag*{(8)}
Ru,rms2(t)=∫E(x,t)r2 dx∫E(x,t) dx; Rω,rms2(t)=∫E(x,t)ω2 dx∫E(x,t) dx.(9)R_{u,\mathrm{rms}}^{2}(t) = \frac{\int E(x,t)r^{2}\,\mathrm{d}x}{\int E(x,t)\,\mathrm{d}x};\ R_{\omega,\mathrm{rms}}^{2}(t) = \frac{\int E(x,t)\omega^{2}\,\mathrm{d}x}{\int E(x,t)\,\mathrm{d}x}. \tag*{(9)}

A thorough discussion of the two theories for scaling decaying turbulence, based upon the Saffman L\mathcal{L} and Loitsyanskii I\mathcal{I} integrals, are given by Ishida et al. [6] and Krogstad & Davidson [13], respectively in the context of initial k→0k \to0, knk^n simulations and grid turbulence. One

way to describe the origin of L\mathcal{L} and I\mathcal{I} is to transform the two-point velocity correlation E(k)=1π∫0∞⟨u⋅u′⟩krsin⁡(kr) drE(k)=\frac{1}{\pi}\int_0^\infty\langle\mathbf{u}\cdot\mathbf{u}'\rangle kr\sin(kr)\,d\mathbf{r}, then expand the result in kk

E(k→0)=Lk24π2+Ik424π2+…withL=∫⟨u⋅u′⟩dr and I=∫r2⟨u⋅u′⟩dr,(10)E(k\to0)=\frac{\mathcal{L}k^2}{4\pi^2}+\frac{\mathcal{I}k^4}{24\pi^2}+\ldots\quad\text{with}\quad\mathcal{L}=\int\langle\mathbf{u}\cdot\mathbf{u}'\rangle d\mathbf{r}\ \text{and}\ \mathcal{I}=\int r^2\langle\mathbf{u}\cdot\mathbf{u}'\rangle d\mathbf{r}, \tag*{(10)}

For grid turbulence there is a well-defined streamwise momentum that gives L≠0\mathcal{L}\ne0, which scales as uˉ2ρ3\bar{u}^2\rho^3 [18] and leads to a prediction with E(t)∼t−6/5E(t)\sim t^{-6/5}. In contrast, without a well-defined linear impulse such that L=0\mathcal{L}=0 (here the average ⟨uz⟩x,y=0\langle u_z\rangle_{x,y}=0), then I∼uˉ2ℓ5\mathcal{I}\sim\bar{u}^2\ell^5 can be used, which leads to E(t)∼t−10/7E(t)\sim t^{-10/7}. On dimensional grounds, I\mathcal{I} could go as the angular momentum squared ⟨H2⟩/V\langle H^2\rangle/V [14], which in a finite periodic domain is difficult to define. So instead the circulation Γ\Gamma about the vortex knot, with the same dimensions, is considered by integrating on Γ\Gamma-loops like those in Figure 3a.

Second and fourth-order velocity structures

Figure 5. Second and fourth-order velocity structures for t=10.8t=10.8 to 36. (a) For DLL(r,t)=δu2(δr,t)D_{LL}(r,t)=\delta u^2(\delta r,t) and scaled by the Kolmogorov prediction δu2(r)/(0.8δrε)2/3\delta u^2(r)/(0.8\delta r\varepsilon)^{2/3}. The scaled value is consistent with the experimental value Matsuzawa et al. [15]. (b) Exponent ζ4\zeta_4 of the Dlog⁡δrD_{\log\delta r} derivative of the fourth-order longitudinal structure function. Note that for δr/η8≲100\delta r/\eta_8 \lesssim100 the plateau with ζ4∼1.2\zeta_4 \sim1.2 is growing.

Figure 4 shows two versions of the energy decay E(t)E(t) in 4a, with the corresponding integral scales in 4b and further length scales in 4c. The energy decay for the full fields in shades of blue all obey ET(t)∼(t+70)−10/7E_T(t)\sim(t+70)^{-10/7} and for ν=8.35×10−5\nu=8.35\times10^{-5}, the integral scale grows as ℓ=(t+70)2/7±0.03\ell=(t+70)^{2/7\pm0.03} in 4b When the E(k∼0)E(k\sim0) shell from figure 1b is removed, the decay of the remaining E−0(t)E_{-0}(t) (red) still goes as (t+70)−10/7(t+70)^{-10/7}, however its integral scale ℓ(−0)\ell_{(-0)} (red) roughly obeys ℓ(−0)∼(t+70)0.11\ell_{(-0)}\sim(t+70)^{0.11}, the red-dashed in 4b.

Figure 4c compares the growth of Δzsplit\Delta z_{\mathrm{split}} to the root-mean-square radii Ru,ω-rmsR_{u,\omega\text{-rms}} (2.3) about the vortex knot, with 4d showing the ratio ℓ/Rω,rms\ell/R_{\omega,\mathrm{rms}}. The inset shows that ℓ\ell and Rω-rmsR_{\omega\text{-rms}} are of the same magnitude and converging in time, with the magnitude and growth of both much less than that of Δzsplit\Delta z_{\mathrm{split}}.

Note that the experimental ℓX∼t0.16±0.06\ell_X\sim t^{0.16\pm0.06} (Matsuzawa et al. [15]) is closer to the ℓ(−0)\ell_{(-0)} curve than to the full ℓ=(t+70)2/7±0.03\ell=(t+70)^{2/7\pm0.03} above, suggesting that the experimental spectral analysis might be missing wavenumbers at the scale of their apparatus.

As a check on their decay results, Matsuzawa et al. [15] compares a modelled dissipation rate ε~=E3/2/ℓ\tilde{\varepsilon}=E^{3/2}/\ell to the true dissipation rate ε=dE/dt\varepsilon=dE/dt. Assuming that E∼t−10/7E\sim t^{-10/7}, then dE/dt∼t−(17/7≈2.43)dE/dt\sim t^{-(17/7\approx2.43)} and ε~∼t−2.30\tilde{\varepsilon}\sim t^{-2.30}. A difference in exponents that would be consistent with the curves in their figure 3a. Figure 2b shows an equivalent ε\varepsilon comparison for these simulations, with ε(t)=dE/dt(t)\varepsilon(t)=dE/dt(t) and ε~=0.5E3/2/ℓ\tilde{\varepsilon}=0.5E^{3/2}/\ell converging for t≥25t\geq25.

Figure 4c includes Δzsplit(t)\Delta z_{\mathrm{split}}(t), the separation of the rings, with the two versions of Rrms(t)R_{\mathrm{rms}}(t) (2.3). For t>10t>10, the two rms radii grow together and more slowly than Δzsplit(t)\Delta z_{\mathrm{split}}(t). The inset, figure 4d, shows that Rω,rms(t)R_{\omega,\mathrm{rms}}(t) grows faster whilst converging to ℓ\ell, the integral scale in figure 4b. And all are significantly larger than the Taylor microscale λ\lambda in 4c.

Structure functions Finally, a few longitudinal structure functions δup(δr)\delta u^p(\delta r) have been determined, with scaled second-order DLL=δu2D_{LL}=\delta u^2 in figure 5a and in 5b the exponent of its fourth-order Dlog⁡δrD_{\log\delta r} derivative, ζ4=d(log⁡(δu4)/dlog⁡(δr)\zeta_4=d(\log(\delta u^4)/d\log(\delta r). DLLD_{LL} is consistent with the equivalent values from Matsuzawa et al. [15] and for ζ4\zeta_4 for δr/η8≲100\delta r/\eta_8\lesssim100, a plateau with ζ4∼1.2\zeta_4\sim1.2, similar to forced isotropic turbulence (Gotoh & Fukayama [5]) is developing. Not shown: All third-order D~LLL=DLLL/(δrε)\tilde{D}_{LLL}=D_{LLL}/(\delta r\varepsilon) have negative slopes for small δr\delta r, with widely varying slopes, except for t=30.4t=30.4 which has of subrange with relatively steady D~LLL(δr)∼−1\tilde{D}_{LLL}(\delta r)\sim-1. Sixth-order, has a short ζ6≈1\zeta_6\approx1 regime, plus very large oscillations.

Concluding thoughts

The three connected threads from this decaying turbulence calculation, with caveats, are:

  • It has been shown how an unadorned trefoil vortex knot can, immediately after self-reconnection, generate a k−5/3k^{-5/3} energy spectrum that persists over several turbulent time-scales whilst obeying Kolmogorov scaling. Further results include an E(t)∼t−10/7E(t) \sim t^{-10/7} energy decay rate and an ℓ(t)∼t2/7\ell(t) \sim t^{2/7} integral length scale growth rate that are consistent with assuming that an invariant exists that goes as I∼uˉ2ℓ5I \sim\bar{u}^{2}\ell^{5}.

    • Figure 4b shows us that to understand the scaling for ℓ\ell will require a better understanding of the role of the ηk∼0.01\eta k \sim0.01 wavenumbers along with their role in determining the circulation Γ\Gamma.

  • The persistence of the -5/3 spectra and the energy decay rate are consistent with those observed by a recent experiment following the decay of isotropic turbulence. With the differences in the growth rates of the respective integral scales ℓ\ell possibly due to the effect of the largest scales, or k∼0.01k \sim0.01 wavenumbers.

    • For times between t=5.2t = 5.2 and 30.4, a careful examination of how braids are shed by one ring, then wrap around the other, might shed light upon how the Kolmogorov spectra are maintained over these times.

  • That H(t)/2E(t)Z(t)\mathcal{H}(t)/\sqrt{2E(t)Z(t)} is relatively constant in Figure 2b would support the existence of an invariant II that could be related to the helicity H(t)\mathcal{H}(t), or something with its dimensions. With one possible example coming from the streamlines shown in the inset of Figure 3a, whose associated circulations up to t=10t = 10 are Γ∼2\Gamma\sim2.

    • Additional measures for the circulation Γ\Gamma need to be considered.

  • In addition to complementing the Matsuzawa et al. [15] experiment, these results should encourage further simulations using a collection of colliding vortex rings that would resemble the experiments more closely.

Acknowledgements The author is grateful for discussions with P.A. Davidson and J.D. Gibbon. This work was partially supported by EPSRC small grant no. R.MAAA.3656.STB with computing resources provided by the Scientific Computing Research Technology Platform at the University of Warwick.

References

  1. [1]Brachet, M.E., Meiron, D.I., Orszag, S. A.,Nickel, B. G., Morf, R.H., & Frisch, U. 1983 Small-scale structure of the Taylor-Green vortex. J. Fluid Mech. 130, 411–452.DOI
  2. [2]Bodenschatz, E., Pesch, W., & Ahlers, G. 2000 Recent developments in Rayleigh-Bénard convection. Annu. Rev. Fluid Mech 32, 709778.DOI
  3. [3]Bustamante, MD, & Kerr, RM 2008 3D Euler about a 2D symmetry plane,. Physica D 237, 1912–1920.DOI
  4. [4]Gallay, T., & Smets, D. 2020 Spectral stability of inviscid columnar vortices. Anal. PDE 13, 17771832.DOI
  5. [5]Gotoh, T., & Fukayama, D. 2002 Velocity field statistics in homogeneous steady turbulence obtained using a high-resolution direct numerical simulation. Phys. Fluids 14, 10651081.DOI
  6. [6]Ishida, T., Davidson, P.A., & Kaneda, Y. 2006 On the decay of isotropic turbulence. J. Fluid Mech. 564, 455–475.
  7. [7]Kerr, R.M. 1985 Higher order derivative correlations and the alignment of small-scale structures in isotropic numerical turbulence.. J. Fluid Mech. 153, 31.
  8. [8]Kerr, R.M. 2018a Enstrophy and circulation scaling for Navier-Stokes reconnection,. J. Fluid Mech. 839, R2.DOI
  9. [10]Kerr, R.M. 2023 Sensitivity of trefoil vortex knot reconnection to the initial vorticity profile. Phys. Rev. Fluids 8, 074701.DOI
  10. [11]Kerr, R.M. 2025 Compact Navier–Stokes trefoils in large domains with finite dissipation. J. Fluid Mech. 1025, A3.DOI
  11. [12]Kleckner, D., & Irvine, W.T.M 2013 Creation and dynamics of knotted vortices. Nature Phys. 9, 253–258.
  12. [13]Krogstad, P.A., & Davidson, P.A. 2010 Is grid turbulence Saffman turbulence?. J. Fluid Mech. 642, 373–394.DOI
  13. [14]Landau, L., & Lifshitz, L. Fluid Dynamics (1959).
  14. [15]Matsuzawa, T., Zhu, M., Goldenfeld, N., & Irvine, W. 2026 Nonlinear diffusion and decay of a blob of turbulence spreading into a quiescent fluid. Proc. Nat. Acad. Sci. 123, e2526858123.DOI
  15. [18]Saffman, P. 1967 Note on decay of homogeneous turbulence. Phys. Fluids 10, 1349–1349.DOI
  16. [19]Scheeler, M. W., Kleckner, D., Proment, D., Kindlmann, G. L., & Irvine, W.T.M. 2014 Helicity conservation by flow across scales in reconnecting vortex links and knots. Proc. Nat. Acad. Sci. 111, 15350–15355.DOI
  17. [20]Scheeler, M. W., van Rees, W., Kedia, H., Kleckner, D., & Irvine, W. 2017 Complete measurement of helicity and its dynamics in vortex tubes. Science 357, 487.DOI
  18. [21]Sinhuber, M., Bodenschatz, E., & Bewley, G.P. 2015 Decay of turbulence at high reynolds numbers. Phys Rev Lett. 114, 034501.pubmed.ncbi.nlm.nih.gov/25659002
  19. [22]Smits, A.J., McKeon, B.J., & Marusic, I. 2011 High-Reynolds number wall turbulence. Annu. Rev. Fluid Mech. 43 (1), 353–375.DOI
  20. [23]Yeung, P.K., Ravikumar, K., Uma-Vaideswaran, R., Dotson, D.L., Sreenivasan, K.R., Pope, S.P., Meneveau, C., & Nichols, S. 2025 Small-scale properties from exascale computations of turbulence on a 32768³ periodic cube. J. Fluid Mech. 1019, R2.

Paper details

vlnc:W_y89qqx3s
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents