1. Introduction
Magnetic reconnection is a fundamental mechanism for energy dissipation in plasmas. It enables the rapid conversion of magnetic energy into kinetic and thermal energy, and non-thermal particle populations through the reconfiguration of magnetic field topology (Zweibel & Yamada Reference Zweibel and Yamada2009). As such, it plays a central role across laboratory, space and astrophysical environments (Yamada et al. Reference Yamada, Ji, Hsu, Carter, Kulsrud, Bretz, Jobes, Ono and Perkins1997; Priest & Forbes Reference Priest and Forbes2002; Paschmann, Øieroset & Phan Reference Paschmann, Øieroset and Phan2013). Reconnection is widely invoked to explain a range of explosive phenomena, from solar and stellar flares and coronal mass ejections to magnetospheric substorms and high-energy particle acceleration (Angelopoulos et al. Reference Angelopoulos2008; Shibata & Magara Reference Shibata and Magara2011; Sironi, Uzdensky & Giannios Reference Sironi, Uzdensky and Giannios2025). It is important in large-scale astrophysical systems such as galaxy clusters, where reconnection has been proposed as a source of diffuse radio emission (Ghosh & Bhat Reference Ghosh and Bhat2025).
Magnetic reconnection occurs in localised regions with high current density, where the magnetic field undergoes sharp reversals. In the classical Sweet–Parker model, this process is described as occurring within an elongated current sheet of length
$L$
and thickness
$\delta$
, where oppositely directed magnetic fields with strength
$B$
are brought together and reconnect due to finite, non-zero resistivity
$\eta$
. The resulting reconnection rate is constrained by the aspect ratio
$({\sim} \delta /L)$
of the diffusion region and scales as
$S^{-1/2}$
, where
$S = BL/\eta \sqrt \rho = V_AL/\eta$
is the Lundquist number, written here in units where the vacuum permeability is unity, and
$\rho$
is the density of the plasma (Parker Reference Parker1957; Sweet Reference Sweet and Lehnert1958).
Typical astrophysical systems have very large Lundquist numbers, owing to their large spatial scales and extremely small resistivities. In this regime, the Sweet–Parker model predicts exceedingly high-aspect-ratio current sheets and correspondingly slow reconnection rates, rendering it incapable of explaining the rapid energy release observed in many astrophysical phenomena.
This limitation has led to the exploration of several mechanisms for fast reconnection. One such mechanism is the plasmoid instability: sufficiently long and thin current sheets become unstable and fragment into multiple magnetic islands and secondary current sheets, forming a hierarchical structure. This fragmentation enables multiple simultaneous reconnection sites and substantially enhances the overall reconnection rate, making it only weakly dependent on the Lundquist number (Loureiro, Schekochihin & Cowley Reference Loureiro, Schekochihin and Cowley2007; Bhattacharjee et al. Reference Bhattacharjee, Huang, Yang and Rogers2009; Huang & Bhattacharjee Reference Huang and Bhattacharjee2010; Uzdensky, Loureiro & Schekochihin Reference Uzdensky, Loureiro and Schekochihin2010; Loureiro et al. Reference Loureiro, Samtaney, Schekochihin and Uzdensky2012; Comisso et al. Reference Comisso, Lingam, Huang and Bhattacharjee2016). A similar result was obtained by Galsgaard & Nordlund (Reference Galsgaard and Nordlund1996) in the context of flux-tube braiding at low plasma beta, where the formation of a hierarchy of smaller current sheets causes the resulting heating rate to be independent of resistivity. A complementary perspective was proposed by Pucci & Velli (Reference Pucci and Velli2014), who argued that at sufficiently large Lundquist numbers, Sweet–Parker current sheets are super-Alfvénically unstable and therefore cannot form, with the tearing instability disrupting the current sheet before the classical Sweet–Parker state is established.
Fast reconnection can also arise in turbulent environments, where stochastic field-line wandering broadens the effective outflow region and enhances magnetic flux transport across the reconnection layer (Lazarian & Vishniac Reference Lazarian and Vishniac1999; Kowal et al. Reference Kowal, Lazarian, Vishniac and Otmianowska-Mazur2009, Reference Kowal, Lazarian, Vishniac and Otmianowska-Mazur2012; Vicentin et al. Reference Vicentin, Kowal, de Gouveia Dal Pino and Lazarian2025). In addition, kinetic-scale processes may further modify the structure of the diffusion region, enabling fast reconnection in weakly collisional or collisionless plasmas (Shay et al. Reference Shay, Drake, Rogers and Denton1999; Bessho & Bhattacharjee Reference Bessho and Bhattacharjee2005; Drake, Shay & Swisdak Reference Drake, Shay and Swisdak2008; Bhat & Loureiro Reference Bhat and Loureiro2018).
While these developments have largely focused on reconnection in isolated or idealised current sheets, many astrophysical plasmas are instead in a turbulent state, where reconnection may occur ubiquitously across a broad range of scales. In this context, Galishnikova, Kunz & Schekochihin (Reference Galishnikova, Kunz and Schekochihin2022) identified the parameter regime in which tearing-mediated reconnection operates across a range of scales in forced MHD turbulence. Reconnection is also believed to operate in unforced, non-helical MHD turbulence (Bhat, Zhou & Loureiro Reference Bhat, Zhou and Loureiro2021), where it may additionally regulate the decay time scale of the turbulence, in conjunction with the conservation of some dynamically important quantity.
Two quantities have been proposed as candidates for this role. The first is the so-called Hosking integral (Hosking & Schekochihin Reference Hosking and Schekochihin2021, Reference Hosking and Schekochihin2023; Zhou, Sharma & Brandenburg Reference Zhou, Sharma and Brandenburg2022; Brandenburg Reference Brandenburg2023; Brandenburg et al. Reference Brandenburg, Kamada and Schober2023, Reference Brandenburg, Neronov and Vazza2024), whose conservation and gauge invariance were recently examined under broad classes of gauge choices by Hew et al. (Reference Hew, Hosking, Federrath, Beattie and Kriel2026). The second is the mean-squared vector potential, or anastrophy (Zhou et al. Reference Zhou, Bhat, Loureiro and Uzdensky2019; Bhat et al. Reference Bhat, Zhou and Loureiro2021; Dwivedi, Anandavijayan & Bhat Reference Dwivedi, Anandavijayan and Bhat2024; Anandavijayan & Bhat Reference Anandavijayan and Bhat2026). To address the gauge dependence of anastrophy, Brandenburg, Kahniashvili & Tevzadze (Reference Brandenburg, Kahniashvili and Tevzadze2015) defined it instead as
$\langle A_{\mathrm{2\text{-}D}}^2\rangle$
, where
$A_{\mathrm{2\text{-}D}}$
is the component of the magnetic vector potential along the intermediate eigenvector of the rate-of-strain tensor. This choice minimises the anastrophy, but does not render it fully gauge-independent; see table 5 of the supplemental material of Brandenburg et al. (Reference Brandenburg, Kahniashvili and Tevzadze2015).
If reconnection indeed governs the decay of magnetic energy in such systems, its rate should imprint directly on the global decay law. However, numerical evidence presents a more nuanced picture. In particular, Brandenburg, Neronov & Vazza (Reference Brandenburg, Neronov and Vazza2024) reported a decay time scale at large Lundquist numbers that appears largely independent of the magnetic Prandtl number,
$\textit{Pr}_{\mathrm{M}}$
– the ratio of the plasma viscosity,
$\nu$
, to the resistivity,
$\eta$
. This finding is in tension with expectations from plasmoid-mediated reconnection theory. For instance, Comisso, Grasso & Waelbroeck (Reference Comisso, Grasso and Waelbroeck2015) obtain a reconnection rate scaling as
${\sim} {\textit{Pr}_{\mathrm{M}}}^{-1/2}$
in the high-
$S$
regime, at high
$\textit{Pr}_{\mathrm{M}}$
. This scaling was obtained from analyses of the boundary-driven Taylor problem, where a current sheet is externally forced to form.
The discrepancy raises a key question: does reconnection, as realised in dynamically evolving turbulent systems, actually follow the same scaling laws as in externally driven configurations? The Taylor problem represents a highly idealised set-up, in which the current sheet is both formed and driven by imposed boundary perturbations. In such studies (e.g. Comisso et al. Reference Comisso, Grasso and Waelbroeck2015, hereafter CGW15), the system typically begins from a current sheet that is initially stable to tearing, and reconnection is triggered by externally imposed perturbations. In contrast, current sheets in decaying turbulence arise self-consistently from the nonlinear interaction and coalescence of magnetic structures, which may be spontaneously unstable to the plasmoid instability.
Motivated by this distinction, we revisit the
$\textit{Pr}_{\mathrm{M}}$
dependence of the reconnection rate using a minimal, self-consistent set-up that resembles the magnetic structures observed in simulations of (at least two-dimensional) decaying MHD turbulence. Specifically, we study reconnection in the current sheet that sits at the interface of two coalescing magnetic islands. This configuration captures the spontaneous generation of plasmoids within a pre-existing current sheet during island coalescence, as opposed to externally driven set-ups.
Our study complements previous theoretical and numerical investigations of viscous magnetic reconnection. These include the visco-resistive Sweet–Parker model of Park, Monticello & White (Reference Park, Monticello and White1984), the plasmoid-mediated reconnection theory of Comisso et al. (Reference Comisso, Grasso and Waelbroeck2015), simulations incorporating viscosity along with thermal conduction (Minoshima, Miyoshi & Imada Reference Minoshima, Miyoshi and Imada2016) and simulations of tearing-mode evolution with anisotropic (Braginskii) viscosity (Tenerani et al. Reference Tenerani, Rappazzo, Velli and Pucci2015).
This paper is organised as follows. Section 2 presents the numerical set-up, including the governing equations, the initial configuration representing the two coalescing magnetic islands and the diagnostics used to measure the reconnection rate. In § 3, we present results from our direct numerical simulation (DNS) studies, which support only a weak
$\textit{Pr}_{\mathrm{M}}$
dependence of the reconnection rate in the plasmoid-mediated regime. In § 4, we compare these results with simulations of the boundary-driven Taylor problem. We also discuss the key physical differences between the two set-ups that may underlie the different
$\textit{Pr}_{\mathrm{M}}$
scalings obtained. Finally, in § 5, we summarise our findings and discuss their possible implications.
2. Numerical set-up
We perform two-dimensional (2-D) direct numerical simulations of a system of two coalescing islands spanning a wide range of Lundquist and magnetic Prandtl numbers.
2.1. Basic equations
We use the Pencil Code (Pencil Code Collaboration et al. 2021) to simulate the compressible isothermal visco-resistive equations of magnetohydrodynamics,
where
$\textrm {D} {}/\textrm {D} {} t=\partial /\partial t+\boldsymbol{U}\boldsymbol{\cdot }{\boldsymbol{\nabla }}$
is the advective derivative,
$\boldsymbol{B}={\boldsymbol{\nabla }}\times \boldsymbol{A}$
is the magnetic field (again in units where the vacuum permeability is unity),
$\boldsymbol{J}={\boldsymbol{\nabla }}\times \boldsymbol{B}$
is the current density,
$\boldsymbol{U}$
is the velocity field,
$c_s$
is the sound speed,
$\nu$
is the viscosity,
$\eta$
is the resistivity and
${\mathsf{S}}_{\textit{ij}}=(\partial _i U_j+\partial _j U_i)/2-\delta _{\textit{ij}}{\boldsymbol{\nabla }}\boldsymbol{\cdot }\boldsymbol{U}/3$
are the components of the traceless rate-of-strain tensor
$\boldsymbol{\mathsf{S}} {}$
. For a subset of runs, we use a GPU-accelerated version of the Pencil Code coupled with the Astaroth library (Pekkilä et al. Reference Pekkilä, Väisälä, Käpylä, Rheinhardt and Lappi2022), enabling efficient simulations at higher resolutions.
The computational domain is a square box with sides
$L_x = L_y = 1$
, spanning
$-({L_x - \Delta x})/{2} \leqslant x \leqslant ({L_x + \Delta x})/{2}$
and
$-({L_y - \Delta y})/{2} \leqslant y \leqslant ({L_y + \Delta y})/{2}$
, where
$\Delta x = L_x/(N-1)$
and
$\Delta y = L_y/(N-1)$
are the grid spacings in the
$x$
and
$y$
directions, respectively, and
$N$
is the number of grid points in each direction. This choice ensures that there exist grid points that lie exactly at
$x=0$
and
$y=0$
. Conducting, free-slip boundary conditions are imposed on all boundaries.
2.2. Initial conditions
We adopt the initial configuration of two coalescing islands as described by Huang & Bhattacharjee (Reference Huang and Bhattacharjee2010) (HB10 hereafter) for our numerical experiments. We initialise the system with no flows and a magnetic vector potential such that only the
$z$
-component is non-zero,
where
$A_{\mathrm{amp}}$
sets the amplitude and
$a$
is a measure of the current sheet half-width. The magnetic field is obtained from
$\boldsymbol{B} = \boldsymbol{\nabla }\times \boldsymbol{A}$
.
We choose
$A_{\mathrm{amp}} = 1/{2\pi }$
so that the resulting reconnecting field has an amplitude
$B_{\mathrm{amp}} = 1$
. The streamlines in figure 1 show the magnetic field structure. The resulting current sheet is extended along the
$y$
-direction with the current density pointing along the negative
$z$
direction.
Initial configuration showing the two coalescing islands at
$t=0$
. The colour map represents the density, while the magnetic field is visualised using streamlines of
$\boldsymbol{B}$
. This particular snapshot is from a run with
$S=40\,000$
and the corresponding current sheet half-width,
$a=0.005$
.

This configuration is a smooth regularisation of
where the discontinuity at
$x=0$
is replaced by a tanh profile, yielding a current sheet of half-width
$a$
. We choose the current sheet half-width to have a classical Sweet–Parker scaling such that
$a = L_yS^{-1/2}$
.
The density is initialised to ensure an approximate force balanced equilibrium,
where
$\rho _0=1$
, and we set
$\ln \rho$
accordingly. This density profile is shown in figure 1. Here,
$B_y = \boldsymbol{B} \boldsymbol{\cdot }\hat {\boldsymbol y}$
and
$B_{y,0} = (\boldsymbol{\nabla }\times \boldsymbol{A}_0)\boldsymbol{\cdot }\hat {\boldsymbol y}$
. Note here that this is not simply a pressure equilibrium, but the non-uniform density approximately balances the full Lorentz force, the sum total of the magnetic tension and the magnetic pressure. This is an important subtlety that we discuss in Appendix A.
The important non-dimensional number that we characterise the system with is the Lundquist number
$S=V_AL/\eta$
. The characteristic length scale for us is the length of the current sheet,
$L_y=1$
, the Alfvén speed,
$V_A = B_{\mathrm{amp}}/\sqrt {\rho _0} = 1$
, and so the Lundquist number is simply
$S=1/\eta$
. Further, we minimise compressibility effects by setting the sound speed to
$10 V_A$
, which implies a plasma beta of
$\beta = 2\rho _0c_s^2/B^2= 200$
. The Alfvén time is defined as the ratio of the length of the current sheet to the Alfvén speed,
$\tau _A = L/V_A = 1$
.
The other relevant dimensionless parameters are the magnetic Reynolds number,
$Re_{\mathrm{M}} = UL/\eta$
, the kinetic Reynolds number,
$Re = UL/\nu$
, and the magnetic Prandtl number,
$\textit{Pr}_{\mathrm{M}} = \nu /\eta$
. Since reconnection outflows are expected to be Alfvénic, we take the characteristic velocity to be the Alfvén speed,
$U = V_A$
. With the current-sheet length
$L=L_y=1$
, this choice gives
$Re_{\mathrm{M}} = V_AL/\eta = S$
, so the magnetic Reynolds and Lundquist numbers are identical in our set-up.
2.3. Diagnostics
The primary quantity we aim to estimate is the reconnection rate. We adopt a measure similar to that proposed by HB10. The instantaneous non-dimensional reconnection rate is defined through the rate of depletion of magnetic flux,
Here, the quantity inside the brackets measures the magnetic flux contained within an island,
$\varPhi _{\mathrm{island}}$
(see Appendix B for further details). As reconnection proceeds, this flux decreases and its time derivative therefore provides a direct measure of the reconnection rate. The definition in (2.7) is identical to that used by HB10 (see also Comisso et al. Reference Comisso, Grasso and Waelbroeck2015 for a closely related diagnostic), except that HB10 assume
which is strictly valid only in the ideal (
$\eta =0$
) limit. At finite resistivity,
$\max _x A_z(x,0,t)$
evolves solely owing to resistive diffusion, whereas
$\max _y A_z(0,y,t)$
evolves owing to both resistive diffusion and magnetic reconnection. Neglecting the evolution of
$\max _x A_z(x,0,t)$
therefore attributes not only the reconnection-driven flux depletion but also the diffusive component of the flux depletion to reconnection, leading to a slight overestimate of the reconnection rate. Retaining this term removes the contribution from resistive diffusion, thereby isolating the reconnection-driven flux depletion and providing a more accurate estimate of the reconnection rate. An alternative measure of the reconnection rate used by Vicentin et al. (Reference Vicentin, Kowal, de Gouveia Dal Pino and Lazarian2026) is described in Appendix B, and gives consistent results.
3. Results from direct numerical simulations
In this section, we present results from a suite of direct numerical simulations performed over a range of Lundquist numbers and magnetic Prandtl numbers.
3.1. Runs with
$\textit{Pr}_{\mathrm{M}}=1$
We first summarise results from
$\textit{Pr}_{\mathrm{M}}=1$
simulations to check consistency with existing literature and recover known scaling relations. Table 1 summarises various run parameters for all the runs.
Summary of the simulation parameters. The columns give the numerical resolution
$N$
, Lundquist number
$S$
, magnetic Prandtl number
$\textit{Pr}_{\mathrm{M}}$
, and the initial current-sheet half-width
$a$
. Runs marked with an asterisk (*) denote the high-resolution simulations used for the convergence study (see Appendix C).

Table 1. Long description
A table with four columns and twelve rows. The columns are labeled ‘S.No’, ‘Resolution, N’, ‘S x 10^3’, ‘PrM’, and ‘Current sheet half-width, a’. The rows provide data for different simulation runs, including numerical resolution, Lundquist number, magnetic Prandtl number, and initial current-sheet half-width. Runs marked with an asterisk denote high-resolution simulations used for a convergence study. Row 1: 1, 1024, 1, 1, 10, 20, 40, 0.0316. Row 2: 2, 1024, 2, 1, 0.0224. Row 3: 3, 1024, 4, 1, 10, 20, 40, 0.0158. Row 4: 4, 4096, 8, 1, 0.0112. Row 5: 5, 4096, 10, 1, 0.0100. Row 6: 6, 4096, 20, 1, 0.0071. Row 7: 7, 4096, 40, 1, 10, 20, 40, 100, 0.0050. Row 8: 8, 4096, 100, 1, 0.0032. Row 9: 9*, 6144, 100, 1, 0.0032. Row 10: 10, 16384, 200, 1, 0.0022. Row 11: 11, 16384, 500, 1, 10, 20, 40, 50, 0.0014. Row 12: 12*, 32768, 500, 1, 0.0014.
Figure 2 shows the dependence of this non-dimensionalised reconnection rate on the Lundquist number,
$S$
. Consistent with existing results, we find that the non-dimensionalised reconnection rate follows the Sweet–Parker scaling
$V_{\mathrm{rec}} \propto S^{-1/2}$
up to a critical Lundquist number,
$S_c \approx 10^5$
. Close to and beyond
$S_c$
, the plasmoid instability sets in, giving rise to a Lundquist number independent regime. Unlike recent work by Vicentin et al. (Reference Vicentin, Kowal, de Gouveia Dal Pino and Lazarian2026) with a similar set-up, we do not observe an intermediate
$V_{\mathrm{rec}}\propto S^{-1/3}$
scaling. In Appendix A, we show that this scaling can nevertheless be recovered using their initial condition, suggesting that it is sensitive to the initial force balance rather than representing a universal tearing-mediated reconnection regime.
Scaling of the reconnection rate versus Lundquist number,
$S$
, at
$\textit{Pr}_{\textrm {M}} = 1$
. The orange circles show the reconnection rates obtained from simulations. The black dashed line shows the Sweet–Parker scaling,
$V_{\mathrm{rec}} \propto S^{-1/2}$
.

3.2.
$\textit{Pr}_{\mathrm{M}}$
dependence for
$S\lt S_c$
In the Sweet–Parker regime, the dependence of the reconnection rate on the magnetic Prandtl number arises from the influence of viscosity on the outflow from the current sheet. Park et al. (Reference Park, Monticello and White1984) first showed that viscous stresses reduce the outflow velocity, thereby broadening the current sheet through mass conservation and reducing the inflow speed; see also the modern discussion by Baty (Reference Baty2022). We briefly summarise this scaling argument before comparing it with the reconnection rates measured in our simulations.
3.2.1. Visco-resistive Sweet–Parker scaling
For a steady Sweet–Parker current sheet of length
$L_{\textrm {SP}}$
and thickness
$\delta _{\textrm {SP}}$
, with inflow speed
$V_{\textrm {in}}$
and outflow speed
$V_{\textrm {out}}$
, mass conservation gives
Balancing advection and diffusion in the steady induction equation,
yields
Combining these relations gives
The outflow speed follows from an energy balance. The magnetic energy carried into the current sheet per unit time,
is converted into kinetic energy of the exhaust together with viscous dissipation inside the current sheet.
The kinetic energy flux leaving the two exhausts scales as
while the viscous dissipation rate is
where
$C_\nu$
is a dimensionless coefficient that depends on the detailed velocity profile across the current sheet. Using (3.1) to eliminate
$V_{\textrm {in}}$
, the energy balance,
becomes
where
$V_A=B/\sqrt {\rho }$
. Using (3.4) gives
or
Substituting this into (3.4) yields
and therefore,
Factoring out the coefficient
$C_\nu$
,
defining
$\textit{Pr}_{\textrm {M}0}\equiv C_\nu ^{-1}$
and absorbing the constant factor
$C_\nu ^{-1/4}$
into the overall proportionality coefficient gives
which reduces to
3.2.2. Comparison with simulations
We now compare the above-mentioned scaling with the reconnection rates measured in our simulations in the Sweet–Parker (
$S\lt S_c$
) regime.
Figure 3 compares the measured reconnection rates with the visco-resistive Sweet–Parker scaling for three Lundquist numbers,
$S=1000$
,
$4000$
and
$40\,000$
. Figure 3(a) shows that, for each value of
$S$
, the quasi-steady reconnection rate decreases with increasing magnetic Prandtl number. At the same time, the reconnection rate decreases with increasing Lundquist number, consistent with the Sweet–Parker scaling
$V_{\textrm { rec}}\propto S^{-1/2}$
.
Dependence of the reconnection rate on the magnetic Prandtl number,
$\textit{Pr}_{\textrm {M}}$
, for three Lundquist numbers,
$S=1000,\ 4000$
and
$40\,000$
in the Sweet–Parker regime. (a) Measured quasi-steady reconnection rate,
$V_{\textrm {rec}}$
, as a function of
$\textit{Pr}_{\textrm {M}}$
. For all three Lundquist numbers, the reconnection rate decreases with increasing magnetic Prandtl number. (b) Reconnection rates rescaled by
$S^{1/2}$
, showing an approximate collapse of the data onto a single curve. The error bars denote an assumed uncertainty of
$5\,\%$
. The solid curve shows the best-fitting model,
$S^{1/2}V_{\textrm {rec}} = 1.97(1.0+\textit{Pr}_{\textrm {M}}/8.0)^{-0.23}$
, indicating that the rescaled reconnection rate is well described by the theoretical prediction in (3.15).

To isolate the dependence on
$\textit{Pr}_{\textrm {M}}$
, figure 3(b) shows the reconnection rates rescaled by
$S^{1/2}$
. The rescaled data exhibit an approximate collapse onto a single curve, particularly for the higher Lundquist numbers, consistent with the expected
$S^{-1/2}$
scaling over the range of
$S$
considered. We fit the rescaled reconnection rates with the shifted power-law
obtaining
$V_0 = 1.97$
,
$\textit{Pr}_{\textrm {M}0}=8.0$
and
$\alpha =-0.23$
, in good agreement with the theoretical prediction (
$\alpha = -0.25$
) of (3.15). No plasmoid formation is observed in any of these simulations, confirming that the current sheet remains in the laminar Sweet–Parker regime throughout the measurement interval. This is illustrated in figure 4, which shows representative current-density maps for the
$S=40\,000$
runs spanning the range of
$\textit{Pr}_{\mathrm{M}}$
considered. In all cases, the current sheet remains smooth and elongated, with no evidence of secondary islands or plasmoid formation.
Representative current-density maps (
$|J_z|$
) for the Sweet–Parker regime at
$S=40\,000$
, and magnetic Prandtl numbers
$\textit{Pr}_{\mathrm{M}}=1$
, 10, 20, 40 and 100. (a) Evolution of the magnetic flux contained within the islands, with the vertical dashed line indicating the time at which the current-density snapshots are shown in panel (b).

Figure 4. Long description
Panel A: A line graph shows the evolution of magnetic flux contained within islands over time. The x-axis represents time (t) and the y-axis represents the difference in maximum magnetic flux. Different colored lines represent various magnetic Prandtl numbers (PrM) with values of 1, 10, 20, 40, and 100. The vertical dashed line indicates the time at which the current-density snapshots are shown in Panel B. Panel B: Five heatmaps display current-density maps for different magnetic Prandtl numbers (PrM) at a specific time. Each heatmap has the z-axis on the vertical axis and the y-axis on the horizontal axis. The color scale at the top indicates the magnitude of current density (|Jz|), ranging from 0 to 300. The heatmaps show how current density varies spatially for PrM values of 1, 10, 20, 40, and 100.
3.3.
$\textit{Pr}_{\mathrm{M}}$
dependence for the plasmoid regime
We now turn to the regime in which the current sheet becomes unstable to the plasmoid instability and fragments into multiple secondary islands. In our simulations, this transition is clearly observed for
$S \gtrsim 10^5$
. Previous studies have shown that viscosity influences not only the Sweet–Parker reconnection rate, but also the linear tearing instability itself by reducing its growth rate and increasing the characteristic tearing time, with the precise scaling depending on the tearing regime (e.g. Ofman et al. Reference Ofman, Chen, Morrison and Steinolfson1991). In the present work, however, we are primarily interested in the reconnection rate after the current sheet has fully entered the plasmoid-mediated regime. The reconnection rates reported herein are therefore measured well after the initial onset of the tearing instability, when the current sheet has fragmented into multiple plasmoids, and the dynamics are dominated by their nonlinear evolution and interactions.
We focus on runs with
$S = 5\times 10^5$
and magnetic Prandtl numbers in the range
$1 \leqslant \textit{Pr}_{\mathrm{M}} \leqslant 50$
. All simulations were performed at a numerical resolution of
$16\,384^2$
. To assess numerical convergence, the
$\textit{Pr}_{\textrm {M}}=1$
case was additionally repeated at twice the resolution,
$32\,768^2$
. The evolution of the magnetic flux contained within the islands, together with the corresponding reconnection rates, is shown in figure 5.
Time evolution of the magnetic flux and the corresponding reconnection rate for different magnetic Prandtl numbers at
$S=5\times 10^5$
. (a) Quantity
$\max _x A_z(x,0)-\max _y A_z(0,y)$
, which measures the magnetic flux contained within the magnetic island. (b) Corresponding reconnection rate, computed from the time derivative of this quantity as defined in (2.7). The dashed lines indicate a reconnection rate of
$0.008$
, obtained from a linear fit to the flux evolution in the
$\textit{Pr}_{\mathrm{M}}=1$
run. The paired markers indicate the time interval used to compute the mean reconnection rate and its
$1\sigma$
uncertainty for each simulation. The start of the averaging interval was selected by visual inspection as the point where the reconnection rate increases following the initial transient, while the end coincides with the end of the simulation.

We find that the reconnection rate exhibits only a weak dependence on
$\textit{Pr}_{\mathrm{M}}$
across this range, remaining broadly consistent with a
$\textit{Pr}_{\mathrm{M}}$
-independent scaling. In particular, the instantaneous reconnection rate in all runs reaches, and often exceeds, values of
${\approx} 0.008$
, consistent with the characteristic rates expected for plasmoid-mediated reconnection (Bhattacharjee et al. Reference Bhattacharjee, Huang, Yang and Rogers2009; Huang & Bhattacharjee Reference Huang and Bhattacharjee2010; Loureiro et al. Reference Loureiro, Samtaney, Schekochihin and Uzdensky2012).
In particular, let us focus on the run with
$\textit{Pr}_{\mathrm{M}}=40$
. As can be seen from figure 5, the reconnection proceeds in a quasi-cyclic manner, alternating between phases of relatively fast and slow reconnection. We investigate this behaviour in greater detail in the following and identify a plausible physical explanation for these oscillations.
We reproduce in figure 6 the flux evolution curve from figure 5 for the run with
$\textit{Pr}_{\mathrm{M}} = 40$
. Alongside this, we show contours of the vector potential at three representative times. Two of these snapshots (
$t=3.25$
and
$t=6.25$
) correspond to phases of relatively fast reconnection (
$V_{\mathrm{rec}}\gtrsim 0.008$
), while the intermediate snapshot (
$t=5.00$
) corresponds to a slower reconnection phase. The contour plots reveal a marked qualitative difference between these states. During the fast-reconnection phases, the current sheet contains multiple plasmoids of appreciable size, many of which appear to be interacting and undergoing mergers. In contrast, during the quiescent phase, the layer is dominated by a single comparatively small plasmoid and exhibits much less dynamical activity. This suggests that interactions and coalescence between plasmoids may play an important role in sustaining enhanced reconnection rates. A similar conclusion was reached by Vicentin et al. (Reference Vicentin, Kowal, de Gouveia Dal Pino and Lazarian2026), who found that sustained fast reconnection is associated with a strongly nonlinear regime characterised by vigorous inter-plasmoid interactions and mergers, rather than merely by the onset of plasmoid formation.
(a) Time evolution of the magnetic flux
$\varPhi _{\mathrm{island}}$
and (b) corresponding reconnection rate,
$V_{\mathrm{rec}}$
for the
$S=5\times 10^5$
,
$\textit{Pr}_{\mathrm{M}}=40$
run in the plasmoid-unstable regime. The vertical dashed lines mark the three times at which contours of the vector potential
$A_z$
are shown in panel (c). These snapshots correspond to distinct phases of the reconnection dynamics: a period of rapid reconnection associated with plasmoid interactions and mergers (
$t=3.25$
), a relatively quiescent phase with slower reconnection (
$t=5.00$
) and a later episode of enhanced reconnection rate accompanied by plasmoid activity (
$t=6.25$
). The cyclic alternation between faster and slower reconnection phases is reflected in the temporal variability of
$V_{\mathrm{rec}}$
.

Figure 6. Long description
Panel A: A line graph shows the time evolution of magnetic flux (phi_island) on the y-axis against time (t) on the x-axis. The magnetic flux decreases over time with notable drops around t = 3.25, t = 5.00, and t = 6.25. Panel B: Another line graph displays the reconnection rate (V_rec) on the y-axis against time (t) on the x-axis. The reconnection rate shows peaks around t = 3.25, t = 5.00, and t = 6.25, indicating periods of rapid reconnection. Panel C: Three contour plots of the vector potential (A_z) at different times (t = 3.25, t = 5.00, and t = 6.25) show the spatial distribution of the vector potential in the x-y plane. The color scale ranges from 0.000 to 0.040, with yellow indicating higher values and black indicating lower values. The contours illustrate the structure and evolution of the vector potential during different phases of reconnection dynamics.
Moreover, in the plasmoid-mediated regime, the number of plasmoids is expected to increase with the Lundquist number (see e.g. HB10). We therefore hypothesise that, at astrophysically large Lundquist numbers, the current sheet may host a sufficiently large population of interacting plasmoids such that prolonged quiescent intervals become increasingly unlikely. If so, reconnection could remain persistently in the nonlinear, interaction-dominated state, leading to a fast,
$\textit{Pr}_{\mathrm{M}}$
-independent reconnection rate. Testing this conjecture, however, requires simulations at substantially higher Lundquist numbers than those considered here.
To better quantify the dependence of the reconnection rate on the magnetic Prandtl number, we compute the time-averaged reconnection rate for each simulation over the intervals indicated by the paired markers in figure 5. The start time (
$t_{\textrm {start}}$
) of each interval is chosen by visual inspection as the point where the reconnection rate begins to increase following the initial transient, while the end corresponds to the end of the simulation. Empirically, we find that the inverse onset time exhibits the same magnetic-Prandtl-number dependence as the visco-resistive plasmoid growth rate derived by Comisso et al. (Reference Comisso, Grasso and Waelbroeck2015),
as shown in figure 7. This can equivalently be written as
where
Inverse start time,
$1/t_{\textrm {start}}$
, as a function of the magnetic Prandtl number,
$\textit{Pr}_{\textrm { M}}$
. The solid curve shows the best-fitting shifted power law,
$1/t_{\textrm {start}}=4.41\,(2.07+\textit{Pr}_{\textrm { M}})^{-0.51}$
, consistent with the visco-resistive scaling of Comisso et al. (Reference Comisso, Grasso and Waelbroeck2015).

At first sight, the fact that
$t_0\lt \tau _A$
may appear surprising, since reconnection is often expected to occur on time scales longer than the Alfvén time. However, the relevant Alfvén time scale governing the tearing instability is not the global crossing time, but the Alfvén time across the initial current-sheet half-thickness (
$a$
), given by
For the present simulations,
$S=5\times 10^5$
, giving
Consequently,
so that the ordering
is entirely consistent with the expected hierarchy of time scales.
To estimate the uncertainty in the time-averaged reconnection rate, we approximately account for temporal correlations in the instantaneous reconnection rate by assuming a correlation time of one Alfvén time,
$\tau _{\textrm {A}}$
, corresponding to the characteristic dynamical time scale of the system. Using the standard expression for the correlated time series (Sokal Reference Sokal, DeWitt-Morette, Cartier and Folacci1997; Wolff Reference Wolff2004), we estimate the effective number of statistically independent samples as
where
$T$
is the duration of the averaging interval. The uncertainty in the mean reconnection rate is then estimated as
where
$\sigma$
is the standard deviation of the instantaneous reconnection rate over the averaging interval. The measured mean reconnection rates are then fitted using the functional form
where both the normalisation
$V_0$
and the exponent
$\alpha$
are treated as free parameters.Footnote 1 For each fixed value of
$\alpha$
, the corresponding best-fitting normalisation
$V_0$
is determined by a weighted least-squares fit using the measured uncertainties, and the resulting
$\chi ^2$
is computed. Figure 8 shows the corresponding
$\Delta \chi ^2=\chi ^2-\chi ^2_{\textrm {min}}$
as a function of
$\alpha$
, together with representative fits to the measured reconnection rates in the inset. The minimum occurs at
$\alpha =-0.107$
, with a
$68\,\%$
confidence interval of
$-0.241\lt \alpha \lt 0.018$
(
$\Delta \chi ^2\lt 1$
for one fitted parameter), indicating only a weak dependence of the reconnection rate on
$\textit{Pr}_{\textrm {M}}$
. In particular, a
$\textit{Pr}_{\textrm {M}}$
-independent reconnection rate (
$\alpha =0$
) is fully consistent with the measurements, whereas the commonly assumed scalings
$\alpha =-1/4$
and especially
$\alpha =-1/2$
provide progressively poorer fits. The latter yields
$\Delta \chi ^2=5.68$
, making it significantly less consistent with the data than the nearly
$\textit{Pr}_{\textrm { M}}$
-independent scaling favoured by the simulations. This statistical analysis quantitatively supports the conclusion drawn from figure 5 that, once the current sheet enters the fully plasmoid-mediated regime, the reconnection rate depends at most only weakly on the magnetic Prandtl number.
Constraints on the magnetic Prandtl number scaling of the time-averaged reconnection rate. The main panel shows
$\Delta \chi ^2$
for fits of
$V_{\textrm {rec}}=V_0(1+\textit{Pr}_{\textrm {M}})^\alpha$
, with the dashed line and shaded region indicating the
$68\,\%$
confidence interval. The inset compares the measured rates (circular markers with
$1\sigma _{\textrm {mean}}$
error bars) with fits for
$\alpha =-1/2$
,
$-1/4$
,
$0$
, and the best-fit exponent,
$\alpha =-0.107$
.

Having characterised the behaviour in our set-up, we now turn to understanding why the resulting
$\textit{Pr}_{\mathrm{M}}$
dependence differs from that reported in studies of the Taylor problem. To this end, we perform a set of simulations using a configuration closely following CGW15.
4. Comparison to the Taylor problem
We now compare our results with the driven Taylor problem studied by Comisso et al. (Reference Comisso, Grasso and Waelbroeck2015) (hereafter CGW15). Their analysis predicts a strong magnetic Prandtl number dependence,
$V_{\textrm {rec}}\propto (1 + \textit{Pr}_{\textrm {M}})^{-1/2}$
for
$\textit{Pr}_{\textrm {M}}\gg 1$
in the plasmoid-mediated regime. This is a steeper dependence than the scaling derived by Park et al. (Reference Park, Monticello and White1984) for the Sweet–Parker regime.
CGW15 analysed the plasmoid-mediated regime of the forced Taylor problem. In their model, the global reconnection layer fragments into a hierarchy of secondary current sheets that remain close to marginal stability to the plasmoid instability. The dimensions of these local sheets are therefore not determined by the global Sweet–Parker balance, but by the condition that their effective Lundquist number remains close to the viscosity-dependent critical value for plasmoid onset. Combining this marginal-stability condition with the visco-resistive Sweet–Parker scaling of Park et al. (Reference Park, Monticello and White1984) for each secondary sheet yields the stronger dependence
$V_{\textrm {rec}}\propto \textit{Pr}_{\textrm {M}}^{-1/2}$
for
$\textit{Pr}_{\textrm {M}}\gg 1$
. Thus, the difference between the
$\textit{Pr}_{\textrm { M}}^{-1/4}$
and
$\textit{Pr}_{\textrm {M}}^{-1/2}$
scalings reflects the different current-sheet structures being assumed: a single steady laminar sheet in the former case and a hierarchy of marginal secondary sheets in the latter.
Further details of the Taylor problem can be found from Comisso et al. (Reference Comisso, Grasso and Waelbroeck2015). Here, we summarise only those aspects of the set-up that are relevant for comparison with the present work.
The equilibrium magnetic field is given by
$B_y=x$
, with perfectly conducting boundaries at
$x=\pm L_x/2=\pm 1$
. This equilibrium is stable to the tearing instability, and reconnection is instead driven externally through a boundary perturbation of the form
where
$\varXi (t) = \varXi _0(1-e^{-t/\tau } - (t/\tau )e^{-t/\tau })$
with
$\varXi _0=0.04$
and
$\tau =2$
, and
$x_w$
represents the instantaneous
$x$
-coordinate of the displaced conducting wall. Following CGW15, we choose a domain with
$L_y=16\pi$
and perturb the fundamental mode,
We use the Pencil Code to perform only two simulations in the region of parameter space identified by CGW15 as leading to plasmoid-dominated reconnection. Our numerical grid has a uniform resolution with
$\Delta x=L_x/2048 \approx 9.76\times 10^{-4}$
and
$\Delta y = L_y/8192 = 6.13\times 10^{-3}$
. Both simulations use the same resistivity,
$\eta = 4\times 10^{-7}$
, while the viscosity is varied such that the magnetic Prandtl numbers are
$\textit{Pr}_{\mathrm{M}}=10$
and
$40$
. The reconnection rate is measured as
following equation 10 of Comisso, Grasso & Waelbroeck (Reference Comisso, Grasso and Waelbroeck2014).
Figure 9 shows the reconnection-rate time series for the two driven reconnection simulations. Similar to CGW15, the system first develops an elongated Sweet–Parker-like current sheet, which subsequently becomes unstable to plasmoid formation.
Time evolution of the reconnection rate for the two Taylor-problem simulations, following the set-up of CGW15. In the main panel, the reconnection rate is rescaled by
$\textit{Pr}_{\mathrm{M}}^{1/2}$
, while in the inset, it is rescaled by
$\textit{Pr}_{\mathrm{M}}^{1/4}$
. The former produces similar amplitudes in the plasmoid-dominated phase, whereas the latter gives a collapse during the Sweet–Parker-like phase. The horizontal reference lines (pink dashed in the main panel and black dotted in the inset) are shown only as visual guides.

To examine the
$\textit{Pr}_{\mathrm{M}}$
dependence in the different stages of the evolution, we rescale the reconnection rate by different powers of
$\textit{Pr}_{\mathrm{M}}$
. In the inset, where the reconnection rate is multiplied by
$\textit{Pr}_{\mathrm{M}}^{1/4}$
, the curves show a reasonable collapse during the Sweet–Parker-like phase, consistent with the visco-resistive Sweet–Parker scaling obtained by Park et al. (Reference Park, Monticello and White1984) and CGW15. In contrast, during the plasmoid-dominated phase, a collapse of the reconnection rate is obtained when rescaled by
$\textit{Pr}_{\mathrm{M}}^{1/2}$
, as shown in the main panel. This behaviour is again consistent with the prediction of CGW15 and Loureiro, Schekochihin & Uzdensky (Reference Loureiro, Schekochihin and Uzdensky2013) in the plasmoid-mediated regime.
Both our simulations and those presented in CGW15 encounter numerical difficulties shortly after entering the plasmoid-mediated regime. This typically occurs once the plasmoids begin to coalesce, leading to the development of increasingly sharp gradients also in the
$y$
-direction. As a result, neither set of simulations is able to sustain a long statistically steady plasmoid-dominated state with ongoing plasmoid interactions and mergers. We therefore identify two potentially important differences between our coalescing islands set-up and the Taylor problem considered in CGW15.
First, in the Taylor problem, the equilibrium current sheet is stable to the tearing instability. Reconnection is initiated externally through boundary perturbations and therefore corresponds to a case of forced reconnection. In contrast, the coalescing islands set-up undergoes spontaneous reconnection arising self-consistently from the nonlinear evolution of the system.
Second, in CGW15 (as well as in our corresponding simulations reproducing their setup), the plasmoid-mediated reconnection rate is measured during the early plasmoid phase, before substantial mergers between individual plasmoids occur. By contrast, in our coalescing-islands set-up, as discussed in § 3.3, interactions and mergers between multiple plasmoids appear to be essential for attaining the large reconnection rates that become nearly independent of
$\textit{Pr}_{\textrm {M}}$
. This distinction may also help reconcile our results with the magnetic-Prandtl-number dependence of the plasmoid onset time shown in figure 7. Prior to the onset of the fast-reconnection phase, the current sheet is expected to undergo the initial tearing instability, producing plasmoids that remain largely isolated and evolve without significant mutual interaction. One possible interpretation is that the time required for these initially isolated plasmoids to grow to sizes at which they begin to interact is governed by the same visco-resistive physics that controls the early plasmoid-mediated reconnection rate measured by CGW15. If so, this would naturally account for the fact that the onset time follows the
$\textit{Pr}_{\textrm {M}}$
scaling predicted by CGW15. Once the plasmoids have grown sufficiently to interact and merge, however, the dynamics enter a strongly nonlinear regime in which repeated plasmoid mergers substantially enhance the reconnection rate. Our results suggest that it is this merger-dominated regime that gives rise to the large reconnection rates exhibiting only a weak dependence on
$\textit{Pr}_{\textrm {M}}$
.
5. Discussion
In this work, we have investigated the dependence of the reconnection rate on the magnetic Prandtl number in a two-dimensional set-up consisting of two coalescing magnetic islands. Our primary motivation was to understand whether the reconnection scalings obtained in externally driven configurations, such as the Taylor problem, remain applicable in more self-consistent reconnecting structures resembling those observed in decaying MHD turbulence.
For the Lundquist numbers below the onset of the plasmoid instability, our simulations recover the expected visco-resistive Sweet–Parker scaling, with the reconnection rate decreasing approximately as
$\textit{Pr}_{\mathrm{M}}^{-1/4}$
. Once the current sheet becomes plasmoid unstable, however, the behaviour changes qualitatively. In the fully plasmoid-mediated regime (
$S \gtrsim 10^5$
), we find that the reconnection rate depends only weakly on
$\textit{Pr}_{\mathrm{M}}$
over the explored parameter range, in contrast to the
$\textit{Pr}_{\mathrm{M}}^{-1/2}$
scaling predicted for the plasmoid-mediated Taylor problem by CGW15.
A closer examination of the high-
$S$
runs suggests that this difference may be connected to the dynamical state of the plasmoid population. In our coalescing islands set-up, the largest reconnection rates are attained during phases in which multiple plasmoids interact, merge and undergo continual nonlinear evolution. By contrast, phases dominated by a single isolated plasmoid are associated with slower reconnection. This suggests that the reconnection dynamics in the strongly nonlinear plasmoid-interaction regime may no longer be controlled by the same local Sweet–Parker balances that underlie the analytic
$\textit{Pr}_{\mathrm{M}}^{-1/2}$
prediction.
The comparison with the Taylor problem further supports this interpretation. In the boundary-driven set-up, the equilibrium current sheet is initially stable and reconnection is triggered externally through imposed perturbations. Moreover, both our reproductions of the CGW15 set-up and the original simulations themselves terminate shortly after the onset of plasmoid coalescence, owing to the development of increasingly sharp gradients. As a result, the reconnection rate is effectively measured during the early plasmoid phase, before a sustained nonlinear state with strong plasmoid interactions is established. In contrast, the coalescing islands set-up naturally evolves into a state with repeated plasmoid formation, interaction and merger. The differing
$\textit{Pr}_{\mathrm{M}}$
scalings may therefore reflect a genuine physical distinction between these regimes, rather than a disagreement between numerical results.
Our results may also have implications for reconnection-mediated decay in magnetically dominated turbulence. Recent studies of decaying MHD turbulence have reported decay time scales that appear to be largely independent of
$\textit{Pr}_{\mathrm{M}}$
at sufficiently high Lundquist numbers (Brandenburg et al. Reference Brandenburg, Neronov and Vazza2024). If reconnection in such systems predominantly occurs through strongly nonlinear plasmoid interactions similar to those observed here, then the weak
$\textit{Pr}_{\mathrm{M}}$
dependence found in our simulations may help explain these observations.
From a cosmological perspective, these results may also have implications for the evolution of primordial magnetic fields. For example, Hosking & Schekochihin (Reference Hosking and Schekochihin2023) estimated turbulent decay rates using the commonly assumed
$\textit{Pr}_{\mathrm{M}}^{-1/2}$
scaling of the reconnection rate when assessing the survivability of primordial magnetic fields in cosmic voids. If the weak
$\textit{Pr}_{\mathrm{M}}$
dependence found here persists in fully turbulent systems and if magnetic reconnection is indeed the process governing the decay of primordial magnetic fields, then such estimates may need to be revisited, potentially modifying the resulting constraints on primordial magnetogenesis scenarios; see Neronov et al. (Reference Neronov, Vazza, Mtchedlidze and Carretti2024). Indeed, Brandenburg et al. (Reference Brandenburg, Neronov and Vazza2024) found, in the context of decaying MHD turbulence, that the decay-time prefactor exhibits little or no additional dependence on the magnetic Prandtl number, although it retains a dependence on the Lundquist number over the range accessible to their three-dimensional (3-D) simulations. Replacing the
$\textit{Pr}_{\mathrm{M}}^{1/2}$
enhancement of the decay time assumed by Hosking & Schekochihin (Reference Hosking and Schekochihin2023) with this approximately
$\textit{Pr}_{\mathrm{M}}$
-independent behaviour, they estimated that, for
$\textit{Pr}_{\mathrm{M}}\sim 10^7$
, the magnetic field strength at recombination would shift from
${\sim} 10^{-9}\,\mathrm{G}\,(\xi _M/1\,\mathrm{pc})$
to
${\sim} 10^{-12.8}\,\mathrm{G}\,(\xi _M/1\,\mathrm{pc})$
, where
$\xi _M$
is the comoving magnetic integral scale. This substantially alters the inferred constraints on primordial magnetic fields, although the resulting field strengths remain above the lower limits inferred from blazar observations (Neronov & Vovk Reference Neronov and Vovk2010). We emphasise, however, that our simulations probe only
$1\leqslant \textit{Pr}_{\mathrm{M}}\leqslant 50$
and, therefore, any extrapolation of the weak
$\textit{Pr}_{\mathrm{M}}$
dependence found here to the cosmologically relevant regime of
$\textit{Pr}_{\mathrm{M}}\sim 10^7$
should be regarded as tentative.
Several important caveats remain. First, all simulations presented here are two-dimensional. In fully 3-D systems, reconnection may proceed through additional pathways involving kink-like distortions (Oishi et al. Reference Oishi, Mac Low, Collins and Tamura2015), turbulence and stochastic field-line wandering (Lazarian & Vishniac Reference Lazarian and Vishniac1999; Kowal et al. Reference Kowal, Lazarian, Vishniac and Otmianowska-Mazur2009, Reference Kowal, Lazarian, Vishniac and Otmianowska-Mazur2012; Russell Reference Russell2025; Vicentin et al. Reference Vicentin, Kowal, de Gouveia Dal Pino and Lazarian2025), and reconnection away from simple X-points. Recent work has shown that, at least at low Lundquist numbers, tearing-like instabilities can nevertheless persist in genuinely 3-D flux-tube-like magnetic configurations while still retaining several similarities with classical 2-D tearing dynamics (Kumar & Bhat Reference Kumar and Bhat2025). Determining whether the weak
$\textit{Pr}_{\mathrm{M}}$
dependence observed here survives in fully nonlinear 3-D plasmoid-mediated reconnection remains an important direction for future work.
Second, although our simulations reach Lundquist numbers well into the plasmoid-mediated regime, the explored parameter range remains limited compared with many astrophysical systems. Establishing the true asymptotic behaviour at simultaneously large
$S$
and large
$\textit{Pr}_{\mathrm{M}}$
will require substantially higher-resolution simulations, likely in three dimensions.
Overall, our results suggest that the reconnection rate may become only weakly dependent on the magnetic Prandtl number once the plasmoid dynamics enter the strongly nonlinear regime. Whether this behaviour persists at significantly larger Lundquist numbers and in fully 3-D turbulent systems remains an important question for future work.
Acknowledgements
The simulations were carried out on the Dardel supercomputing cluster under allocations granted by the Swedish National Infrastructure for Computing through the Center for Parallel Computers at the Royal Institute of Technology (KTH), Stockholm. We thank the reviewers for their careful reading of the manuscript and for their constructive suggestions, which have strengthened the manuscript and improved the clarity of its presentation. We also thank M. Rheinhardt (Aalto, Helsinki) for helpful comments on the manuscript. V.K. thanks P. Bhat (ICTS, Bengaluru), A. Bhattacharjee (Princeton) and A. Lazarian (Wisconsin–Madison) for insightful discussions. V.K. also thanks S. Kapoor (ICTS, Bengaluru), C. Rajarshi (ICTS, Bengaluru) and Chandranathan A. (ICTS, Bengaluru) for constructive discussions during the revision process that helped shape several improvements incorporated into the revised manuscript.
Editor Steve Tobias thanks the referees for their advice in evaluating this article.
Funding
V.K. gratefully acknowledges the hospitality and support of Nordita through the Nordita Visiting PhD Fellowship, as well as support from the Department of Atomic Energy, Government of India, under Project Nos. RTI4013 and RTI4019. A.B. was supported in part by the European Research Council through the ERC Synergy Grant COSMOMAG under grant No. 101224803, the Swedish Research Council (Vetenskapsrådet) under grant No. 2025-05957, the National Science Foundation under grant Nos. NSF AST-2307698, AST-2408411, and NASA Award 80NSSC22K0825.
Data availability statement
The data that support the findings of this study are openly available on Zenodo at doi:10.5281/zenodo.20268567 (v2026.05.18) or, for easier access, at http://norlx65.nordita.org/∼brandenb/projects/plasmoid-mediated/. All calculations have been performed with the Pencil Code (Pencil Code Collaboration et al. 2021); doi:10.5281/zenodo.3961647.
Declaration of interests
The authors report no conflict of interest.
Appendix A. Pressure equilibrium versus complete equilibrium
In § 2.2, we mentioned an important subtlety regarding pressure equilibrium of the initial condition. Vicentin et al. (Reference Vicentin, Kowal, de Gouveia Dal Pino and Lazarian2026) (hereafter VKDL26) performed 2-D simulations with a set-up very similar to ours, using the same magnetic field profiles, to investigate the onset of fast, Lundquist-number-independent reconnection mediated by plasmoids. They reported the existence of an intermediate Lundquist number regime,
$10^4 \lesssim S \lesssim 10^5$
, in which the reconnection rate scales as
$S^{-1/3}$
. We do not observe such a regime in our simulations.
We argue that this discrepancy is most likely due to two subtle but important differences between their set-up and ours. In VKDL26, the initial density profile is chosen such that it balances only the magnetic pressure gradient. Figure 10 shows the corresponding density profile, which visually matches the profile used in VKDL26. However, because the magnetic field lines are curved, the field also exerts magnetic tension forces. As a result, pressure balance alone does not correspond to a complete force equilibrium. The resulting force imbalance can become dynamically important in the low-
$\beta$
limit. In VKDL26,
$\beta \approx 2$
, placing their simulations firmly in the low-
$\beta$
regime where the residual force imbalance associated with magnetic tension can become dynamically important and potentially influence the observed reconnection-rate scaling. In contrast, our simulations have
$\beta \approx 200$
, for which these effects are expected to be much weaker. We show in figure 11 the normalised magnitude of the difference between the total Lorentz force and the pressure gradient – a measure of the residual force imbalance – for both set-ups. The pressure-equilibrium-only configuration used in VKDL26 exhibits an order-unity imbalance over a substantial fraction of the domain, whereas in our approximately full-equilibrium set-up, this imbalance remains below
$10^{-3}$
throughout almost the entire domain.
Density isocontours overlaid with magnetic field streamlines for the pressure equilibrium case. As in figure 1, this snapshot is taken from the run with
$S=40\,000$
to enable a direct comparison.

Comparison between (a) the full equilibrium and (b) pressure-equilibrium-only cases. The colours show the normalised net force,
$|\boldsymbol{J}\times \boldsymbol{B}-\boldsymbol{\nabla }p| / (\rho c_s^2)$
, at
$t=0$
for both cases.

To investigate whether this residual force imbalance could account for the discrepancy, we also performed simulations using the pressure-equilibrium-only set-up of VKDL26 (albeit without a guide field) with
$\beta \approx 4.5$
. This was achieved by setting the sound speed to
$1.5$
in Alfvén units. We carried out simulations at
$\textit{Pr}_{\mathrm{M}}=1$
over a range of Lundquist numbers and extracted the corresponding reconnection rates. These results are shown in figure 12, together with the results from the full-equilibrium set-up (reproduced from figure 2).
Comparison between reconnection rates as a function of Lundquist number with different initial density configurations at
$\textit{Pr}_{\textrm {M}} = 1$
. The configuration with full equilibrium shows a clean
$S^{-1/2}$
scaling up to
$S=10^5$
, whereas the simulations with only pressure equilibrium show a different,
$S^{-1/3}$
scaling for
$S\gtrsim 10^4$
, as obtained by VKDL26.

One can clearly see that the intermediate
$S^{-1/3}$
scaling regime appears only in the pressure-equilibrium configuration and in precisely the same Lundquist number range reported by VKDL26. While this strongly suggests that the residual force imbalance associated with the pressure-equilibrium initial condition is responsible for the observed intermediate scaling, our comparison does not isolate this effect completely, as the pressure-equilibrium runs also differ from our standard set-up in plasma
$\beta$
and the absence of a guide field.
Whether the pressure-equilibrium-only configuration or the approximately full-equilibrium configuration more accurately captures the conditions relevant to reconnecting structures in MHD turbulence and astrophysical plasmas is beyond the scope of the present work. Nonetheless, we expect our results in the fully plasmoid-mediated regime (
$S \gtrsim 10^5$
), where plasmoids strongly interact and merge with one another, to remain largely unaffected by the differences that give rise to the intermediate scaling regime over
$10^4 \lesssim S \lesssim 10^5$
.
Appendix B. Measures of reconnection rate
In § 2.3, we discussed different measures of the reconnection rate. In this appendix, we explain our diagnostic for measuring it and compare with an alternative diagnostic used in the literature. We also provide some physical intuition for the diagnostic adopted in this work.
We measure the rate of flux depletion at the current sheet as the reconnection rate. In two dimensions, with no out-of-plane magnetic field component, the magnetic field can be written in terms of the
$z$
-component of the vector potential as
so that
$A_z$
acts as a flux function. Since we impose perfectly conducting boundary conditions and initialise
$A_z = 0$
at the domain boundaries,
$A_z$
remains zero there for all times.
Using
$\partial _x A_z = -B_y$
, we write
so that, along the current sheet (
$x=0$
), we have
Using
$\partial _y A_z = B_x$
, we may equivalently write
and, across the current sheet (
$y=0$
), we have
The two expressions for
$A_z(x,y)$
, (B2) and (B4), are consistent since
$\boldsymbol{\nabla }\boldsymbol{\cdot }\boldsymbol{B} = 0$
. The magnetic flux contained within an island,
$\varPhi _{\mathrm{island}}$
, is proportional to the difference between the extrema of
$A_z(x,0)$
and
$A_z(0,y)$
. The instantaneous non-dimensional reconnection rate is therefore defined as
This definition is identical to that used by Bhattacharjee et al. (Reference Bhattacharjee, Huang, Yang and Rogers2009) and HB10, except that those works assumed
which is well satisfied, particularly in high-
$S$
runs. Under this approximation, the reconnection rate reduces to
Our diagnostic is also closely related to that used by VKDL26, where the total unsigned magnetic flux is computed across a line perpendicular to the current sheet spanning the full domain (the
$y=0$
line for an orientation similar to ours). This quantity is approximately equal to
$4\varPhi _{\mathrm{island}}$
, whose time derivative, when appropriately normalised, gives the reconnection rate.
Figure 13 compares the two flux measures as functions of time. Their appropriately normalised time derivatives yield the reconnection rate, and the close agreement between the two measures demonstrates the consistency of the diagnostics.
Comparison between the diagnostic quantity used by us and by VKDL26.

Appendix C. Convergence of our simulations
To verify that our measured reconnection rates are not influenced by numerical resolution, we repeated the
$\textit{Pr}_{\textrm {M}}=1$
simulation at the highest Lundquist number considered in this work,
$S=5\times 10^5$
, using resolutions of
$16\,384^2$
and
$32\,768^2$
. Figure 14 compares the evolution of the reconnected flux and the corresponding reconnection rate. The two simulations exhibit very similar flux evolution and reconnection rates. The small differences that develop are attributed to the different realisations of the initial perturbations at the two resolutions.
Flux evolution and corresponding reconnection rate for the runs at
$S=5\times 10^5$
and
$\textit{Pr}_{\textrm { M}}=1$
at numerical resolutions of
$16\,384^2$
and
$32\,768^2$
.

Owing to the substantially increased computational cost, we did not perform equivalent convergence tests for the higher-
$\textit{Pr}_{\textrm {M}}$
simulations. However, increasing
$\textit{Pr}_{\textrm {M}}$
is not expected to introduce finer spatial structures than in the
$\textit{Pr}_{\textrm {M}}=1$
case. The higher-
$\textit{Pr}_{\textrm {M}}$
runs should therefore be at least as well resolved as the converged
$\textit{Pr}_{\textrm {M}}=1$
simulation. This expectation is further supported by figure 15, where the zoomed views show that the peak-current region is resolved by sufficiently many grid cells for all values of
$\textit{Pr}_{\textrm {M}}$
considered, with no evidence of Gibbs ringing or other grid-scale numerical artefacts.
Current-density structure during plasmoid-mediated reconnection for different magnetic Prandtl numbers. (a) Snapshots of the out-of-plane current density,
$J_z$
, for
$\textit{Pr}_{\textrm {M}}=1$
,
$10$
,
$20$
,
$40$
and
$50$
, shown at representative times during the nonlinear plasmoid regime. The dashed rectangles indicate the regions enlarged in panel (b), each centred on the location of the maximum current density magnitude,
$|J_z|$
. The zoomed views demonstrate that the current sheets remain well resolved across the entire range of magnetic Prandtl numbers considered. The white vertical and horizontal lines in each part of panel (b) correspond to a length of 15 grid cells.

We therefore conclude that the reconnection rates reported in this work are numerically converged and are not significantly affected by the adopted grid resolution.
Appendix D. Three-parameter fits to the plasmoid-mediated reconnection rate
In addition to the two-parameter model adopted in § 3.3,
we also explored the more general three-parameter form
where
$V_0$
,
$\alpha$
and
$\textit{Pr}_{\textrm {M0}}$
were all treated as free parameters.
As it turns out, the three-parameter fit does not provide a meaningful constraint on
$\textit{Pr}_{\textrm {M0}}$
. Regardless of the initial guess, the optimisation consistently converges to the largest value of
$\textit{Pr}_{\textrm {M0}}$
permitted by the fitting bounds, indicating that the available data do not contain sufficient information to determine this additional parameter. The preference for large values of
$\textit{Pr}_{\textrm {M0}}$
is also consistent with the weak magnetic-Prandtl-number dependence inferred in the main text: for
$\textit{Pr}_{\textrm {M0}}\gg \textit{Pr}_{\textrm {M}}$
, the model
$V_{\textrm {rec}}=V_0(\textit{Pr}_{\textrm {M0}}+\textit{Pr}_{\textrm {M}})^\alpha$
approaches an approximately
$\textit{Pr}_{\textrm {M}}$
-independent form over the range of
$\textit{Pr}_{\textrm {M}}$
considered here. The present data therefore do not justify the introduction of the additional parameter
$\textit{Pr}_{\textrm {M0}}$
.
To illustrate this point, we also performed fits with
$\textit{Pr}_{\textrm {M0}}$
fixed to a set of representative values. Table 2 lists the resulting best-fit values of
$V_0$
and
$\alpha$
, together with the corresponding
$68\,\%$
confidence interval for
$\alpha$
, for
$\textit{Pr}_{\textrm {M0}}=0$
,
$1$
and
$50$
. The fits with
$\textit{Pr}_{\textrm {M0}}=0$
and
$1$
yield very similar values of the scaling exponent, while adopting the much larger value
$\textit{Pr}_{\textrm {M0}}=50$
produces a steeper best-fit exponent but with a substantially broader confidence interval that still encompasses a weak or negligible
$\textit{Pr}_{\textrm {M}}$
dependence. In fact, fixing
$\textit{Pr}_{\textrm {M0}}=50$
requires a best-fit exponent of
$\alpha =-0.649$
, which is even steeper than the
$\alpha =-1/2$
scaling predicted by Comisso et al. (Reference Comisso, Grasso and Waelbroeck2015) and substantially steeper than the
$\alpha =-1/4$
scaling of Park et al. (Reference Park, Monticello and White1984). The increasingly steep best-fit exponent obtained as larger values of
$\textit{Pr}_{\textrm {M0}}$
are imposed illustrates the strong degeneracy between
$\textit{Pr}_{\textrm {M0}}$
and
$\alpha$
, indicating that the available data do not provide sufficient leverage to determine a characteristic
$\textit{Pr}_{\textrm {M0}}$
independently.
Best-fit parameters for the model
$V_{\textrm {rec}}=V_0(\textit{Pr}_{\textrm {M0}}+\textit{Pr}_{\textrm {M}})^\alpha$
with fixed values of
$\textit{Pr}_{\textrm {M0}}=0$
,
$1$
and
$50$
. The quoted limits correspond to the lower and upper bounds of the 68 % (
$1\sigma$
) confidence interval for
$\alpha$
.

Table 2. Long description
The table presents best-fit values for three different models with fixed values of PrM. It has four columns and three rows. The columns are labeled Model, V0, alpha, alpha_min, and alpha_max. The row labels are V0PrM^alpha, V0(1 + PrM)^alpha, and V0(50 + PrM)^alpha. The values in the table are as follows: Row 1: Model, V0PrM^alpha; V0, 7.833 x 10^-3; alpha, -0.085; alpha_min, -0.193; alpha_max, 0.017. Row 2: Model, V0(1 + PrM)^alpha; V0, 8.458 x 10^-3; alpha, -0.107; alpha_min, -0.241; alpha_max, 0.018. Row 3: Model, V0(50 + PrM)^alpha; V0, 1.016 x 10^-1; alpha, -0.649; alpha_min, -1.416; alpha_max, 0.021.

t=0
B
S=40000
a=0.005
N
S
PrM
a
S
PrM=1
Vrec∝S−1/2
PrM
S=1000, 4000
40000
Vrec
PrM
S1/2
5%
S1/2Vrec=1.97(1.0+PrM/8.0)−0.23
|Jz|
S=40000
PrM=1
S=5×105
maxxAz(x,0)−maxyAz(0,y)
0.008
PrM=1
1σ
Φisland
Vrec
S=5×105
PrM=40
Az
t=3.25
t=5.00
t=6.25
Vrec
1/tstart
Pr M
1/tstart=4.41(2.07+Pr M)−0.51
Δχ2
Vrec=V0(1+PrM)α
68%
1σmean
α=−1/2
−1/4
0
α=−0.107
PrM1/2
PrM1/4
S=40000
|J×B−∇p|/(ρcs2)
t=0
PrM=1
S−1/2
S=105
S−1/3
S≳104
S=5×105
Pr M=1
163842
327682
Jz
PrM=1
10
20
40
50
|Jz|
Vrec=V0(PrM0+PrM)α
PrM0=0
1
50
1σ
α