## Abstract

A stationary model of three-dimensional magnetic reconnection in the absence of a null point is presented, with a non-ideal region that is localized in space. Analytical solutions to the resistive magnetohydrodynamic equations are obtained, with the momentum equation included so that the model is fully dynamic, and thus extends the previous kinematic solutions. A splitting of variables allows solutions to be written in terms of a particular non-ideal solution, on which ideal solutions may be superposed. For the non-ideal solution alone, it is shown that only the field lines linking the diffusion region are affected by the reconnection process, and counter-rotating flows above and below the diffusion region are present. It is only the dimensions of the diffusion region along the reconnection line that are important for the reconnection rate. Many features of the previous stationary kinematic model are also observed here.

## 1. Introduction

Magnetic reconnection is a fundamental process in a plasma whereby magnetic energy is rapidly converted into other forms. It is responsible for a wide range of dynamic processes in astrophysical and space plasmas, such as solar flares and geomagnetic substorms. Recently, much work on magnetic reconnection has centred on understanding the three-dimensional problem (for an overview, see Priest & Forbes 2000). The discovery that, while two-dimensional reconnection is fairly well understood, many aspects of the full three-dimensional problem are not present in two-dimensions has provided motivation for this work.

Our understanding of the nature of three-dimensional reconnection has been advanced both by numerical simulations and analytical work. Recent three-dimensional numerical experiments (e.g. those by Linton & Antiochos 2005) have provided important clues to our understanding of the behaviour of such magnetic fields. Analytical work has mainly focused on the kinematic problem (i.e. examining the implications of Ohm's law alone without the effects of the equation of motion) in a finite region. The localization of the non-ideal term in the induction equation is an important typical feature of reconnection in astrophysical plasmas. In these plasmas, localized regions of intense current concentration are the main cause of such localization, which may be reinforced when the resistivity is enhanced by current-driven microinstabilities. ‘Magnetic annihilation’ (Sonnerup & Priest 1975) and ‘reconnective annihilation’ (Craig & Henton 1995; Priest *et al*. 2000), solutions to the magnetohydrodynamic (MHD) equations, have also been studied by purely analytical means, although these models have non-localized diffusion regions.

In two dimensions, reconnection can only occur at X-type null points of the field. In three dimensions, however, reconnection can occur either at a null-point, or at locations where the field does not vanish (Schindler *et al*. 1988; Lau & Finn 1990). When null points are present, reconnection can take place by *spine*, *fan* or *separator* reconnection (Priest & Titov 1996). In the alternative situation where the field is everywhere non-zero, reconnection can still take place when a localized non-ideal region and an electric field component parallel to the magnetic field are present (Hesse & Schindler 1988). It is an example of this non-null reconnection process that we examine here.

In Priest *et al*. (2003), some of the striking differences between two- and three-dimensional reconnection were identified, examining, in particular, magnetic flux velocities and the rejoining of reconnecting flux tubes. Hornig & Priest (2003) set up a three-dimensional kinematic problem for the evolution of a non-null field in a finite diffusion region. The present work builds on their model, by including also the equation of motion in the analysis so that it is a fully ‘dynamic’ model rather than a kinematic one. We summarize here some aspects of the setup and findings of Hornig & Priest (2003), to allow a comparison with our model.

Hornig & Priest (2003) examined a stationary kinematic solution of the incompressible MHD equations with a localized diffusion region. An X-type magnetic field in the *x*–*y* plane (resulting in a uniform current in the *z*-direction) was superimposed on a uniform field in the *z*-direction. Thus, to obtain a localized non-ideal region, the resistivity *η* was assumed to be localized.

In order to satisfy the non-ideal Ohm's law, a particular electric field and non-ideal velocity were needed; these simplest solutions were termed ‘pure solutions’. We shall also use this term to refer to particular solutions. The purely diffusive velocity field was counter-rotational above and below the *x*–*y* plane, and confined to the hyperbolic flux tube (HFT), threading the diffusion region. The existence of such a counter-rotating flow was shown to be a consequence of the magnetic field structure and localization of the non-ideal region, and thus proved to be a more general property of three-dimensional reconnection. A cartoon illustrating the counter-rotating flows is shown in figure 1. Such flows have also been found in numerical experiments (Pontin *et al*. 2005). It was then observed that superimposed on this non-ideal flow could be any ideal velocity field and corresponding electric field, satisfying the equivalent ideal Ohm's law. Such solutions were termed ‘composite solutions’. Imposing, for example, an ideal stagnation point flow allowed for magnetic flux to be transported into the non-ideal region and subsequently removed after reconnection, with important consequences for the behaviour of the magnetic flux velocity, and so for the reconnecting flux surfaces.

This freedom to impose an ideal stagnation flow distinguishes the three-dimensional from the two-dimensional problem. In the latter, the external flow determines the reconnection rate and largely governs the solution in the diffusion region. In the three-dimensional kinematic case, we may impose an arbitrary ideal solution that is independent of the particular non-ideal solution in the diffusion region. In these papers, we wish to analyse whether this additional freedom arises in the kinematic solution through the neglect of the momentum equation. Thus, we present several elementary solutions of the full set of MHD equations.

A perturbation expansion of the stationary resistive MHD equations is made. Analytical solutions are obtained, and a splitting of variables is made in the limit of slow flow that allows for direct comparison with the ‘pure’ and ‘composite’ solutions of Hornig & Priest (2003). In this paper, the scheme is introduced and particular non-ideal solutions, termed ‘pure’ solutions, are examined. The paper is organized as follows. In §2, we introduce the model. In §3, we obtain two different pure solutions and discuss their implications, before concluding in §4.

## 2. Introducing the model

We take the stationary incompressible resistive MHD equations and non-dimensionalize by settingwhere all the dashed quantities are of order 1, and and are the typical magnetic field strength, length-scale and plasma velocity. Thus, Ohm's law becomes(2.1)where is the typical Alfvén speed of the plasma, andis the inverse Lundquist number.

The equation of motion is(2.2)where is the Alfvén Mach number. For simplicity, we choose to neglect here the effects that viscosity and any external forces (such as gravity) might have on the solutions. The remaining MHD equations are given by(2.3)

We assume , so that the inertial term is small, and look for a three-dimensional solution to these equations by expanding the variables in powers of the Alfvén Mach number of the flow as follows:where all the quantities are dimensionless.

This expansion assumes that the flow is much smaller than the Alfvén speed, since , where and are of order unity, so we have labelled the first term in the expansion of with the index 1. We have also taken , so that the lowest-order magnetic field is potential, an assumption that is crucial in allowing us to find analytical solutions to the equations.

Substituting these expansions into both Ohm's law and the equation of motion and comparing powers of , we find that at zeroth order the equation of motion is satisfied with a constant, while Ohm's law is given by(2.4)At first order, we obtain(2.5)(2.6)At second order, the equations become(2.7)(2.8)while at third order, we have(2.9)

(2.10)

It is clear that a natural coupling exists not between the same ordered equations for Ohm's law and the equation of motion, but rather between Ohm's law at a given order, and the equation of motion at the next order. Thus, to solve the system we will have to consider, for example, equations (2.4) and (2.6) together, and (2.5) and (2.8) together.

We set(2.11)where *k*>0. Thus, our basic state is an X-type current-free equilibrium in the *x*–*y* plane, superimposed on a uniform field in the *z*-direction. The field structure is illustrated in figure 2. The field is assumed to be reconnecting slowly , and is similar to that taken by Hornig & Priest (2003), although the separatrices were not inclined at right-angles, so allowing for a current. Previously, this field has been considered in the context of reconnection by Priest & Forbes (1989), and was termed a ‘hyperbolic flux tube’ (HFT) by Titov *et al*. (2002).

The equations, , of the field line passing through the point are given by(2.12)with the inverse mapping , given by(2.13)

As a further simplification we take , so that any zeroth-order flow is ideal. This assumption is not necessary if we are to obtain a complete solution to the system, but rather permits us to obtain ideal and non-ideal parts to Ohm's law in the zeroth- and first-order equations, respectively, with a corresponding equation of motion for both solutions (at first- and second-order, respectively). Thus, the construction of this model allows for a direct comparison of our solutions with the kinematic ones of Hornig & Priest (2003), where a similar decomposition resulted in pure solutions, satisfying the non-ideal Ohm's law, and composite solutions, in which an ideal solution was superposed on this basic state. We have here in addition an equation of motion for both the pure and composite solutions, and so can consider also how this affects the results.

Throughout this paper we consider the implications of the first-order solution alone, by assuming , and so satisfying Ohm's law at zeroth-order trivially. This is the equivalent of the pure solutions in Hornig & Priest (2003). Here it is at fourth order that the inertial term first appears, and thus the dynamic effects in this pure solution are primarily the Lorentz force and the pressure gradients. The implications of two different non-ideal terms, , are considered in §3, with particular reference to the resulting plasma flows and rate of reconnected flux.

## 3. Pure reconnection solutions

In this section, we examine the pure solutions obtained by setting . Ohm's law at zeroth order becomes , while the equation of motion is satisfied at zeroth and first order with and constants. Thus, for this pure solution it is first necessary to consider equations (2.5) and (2.8) together. With the assumptions taken thus far, these are now(3.1)(3.2)Localization of the non-ideal term can be achieved through a localization in three dimensions of either , or of , or, in the physically most realistic situation, through a localization of both terms. The important quantity in determining the main results presented here is , which is dependent only on the localization of the product , and not on how the localization is realized. As a simplification and in order to allow for analytical solutions, we here choose to prescribe a localization of the resistivity . This assumption was also taken by Hornig & Priest (2003), where a hyperbolic field similar to that given by (2.11) resulted in a uniform current in the -direction. By taking the curl of (3.2), we obtainwhich, assuming , gives , i.e. as constant along field lines of :(3.3)There are a number of ways to choose , two of which we examine here. In §3*a*, we take *f* to be uniform, as was the case in Hornig & Priest (2003). In §3*b*, we instead assume a form such that the current is localized along separatrices of , which is motivated by the numerical experiments of Pontin *et al*. (2005), where such a current was observed.

### (a) Uniform current

The simplest choice of is to takewhere is constant. Such a current can be obtained by taking, for example, the magnetic fieldwhich can be expressed aswhereThis is a particular solution for . Other particular solutions exist, to each of which we are free to add any potential vector field . With this perturbation, the field retains its X-type structure in the *x*–*y* plane, but now has separatrices inclined at a different angle. The sign of determines whether the greater angle between separatrices is across the *x*-axis (for ) or the *y*-axis (for ). In this section, we assume, without loss of generality, that .

Now (3.2) allows us to deduce aswhere is constant.

Considering next Ohm's law, (3.1), we seek a solution such that the non-ideal term is localized. Since the current is uniform, we must localize the resistivity, . To achieve this, together with an analytic form for the remaining terms, we prescribe a localized form for , then taking the scalar product of (3.1) with determine as

One suitable form is to impose(3.4)where . This expression is a function of the coordinates of the field lines, , and *s*, but, setting , we may use the inverse field line mappings (2.13) to find an equivalent expression in terms of *x*, *y* and *z*. Thus, we obtain the function as(3.5)Provided and have the same sign, this is a positive function. The parameter *L* gives the length of the diffusion region in the -direction, while *l* represents the width of the diffusion region in the *z*=0 plane. The hyperbolic nature of the field may render it necessary to decrease *l* with increasing *L* to ensure that the diffusion region remains localized. An example of such a diffusion region is shown in figure 3, where the surface is shown. The maximum value of occurs at the origin, where .

It remains to find and . We have , so, now that is given, we may integrate along the field lines to deduce :(3.6)Taking the gradient of this expression gives an analytical form for . Writingandwe find

The vector product of (3.1) with gives the component of perpendicular to asWe are free to add a velocity component parallel to , and choose to do so in such a way that the -component of is zero:This also ensures that the resulting velocity is divergence-free.

Thus, is given by(3.7)Figure 4 illustrates in two particular planes above and below the *z*=0 plane. The flow is counter-rotational above and below the *z*=0 plane, where it vanishes. Non-zero flow is limited to the region within the HFT, which consists of the field lines passing through the non-ideal region. Near to the origin the velocity field is almost circular, but becomes distorted by the magnetic field on moving away from the plane *z*=0, as shown in figure 4. The pure solutions of Hornig & Priest (2003) are very similar, themselves being counter-rotational flows within the HFT, distorted by the magnetic field.

We are left to consider the remaining second-order equation, (2.7), which becomes(3.8)This may be satisfied by taking and . We then may solve Ohm's law at all even orders, and the equation of motion at all odd orders, by taking

The equation of motion at all subsequent odd orders and Ohm's law at all subsequent even orders may also be solved, at least numerically, to determine completely all higher-order quantities. Here we outline a scheme for Ohm's law at third order and the equation of motion at fourth order:(3.9)(3.10)Since the components of both and parallel to are known, we may use (3.10) to calculate by integrating along the field lines, starting from in the plane *z*=0:Using the inverse field line mappings, this expression can be rewritten in terms of *x*, *y* and *z* and then deduced. In turn, this allows us to find the perpendicular component of the current :The freedom to add a component parallel to may then be used to ensure is divergence-free.

Turning to (3.9), it is left to determine and . The equation has essentially the same structure as (3.10), and so may be solved in the same way by again integrating along the field lines to findThe component of the flow perpendicular to is given byLettingensures is divergence-free, so that the continuity equation is satisfied. Note that this scheme would be effective even without the assumption , which has been used to allow a direct comparison with the kinematic case.

We now have sufficient information to determine the rate of reconnected flux, which is given by the integral of the parallel electric field along the reconnection lineWe identify the *z*-axis with the reconnection line, and so(3.11)The parameter *l* does not appear in this expression, and so we conclude that the extent of the diffusion region in the *x*–*y* plane does not affect the reconnection rate. This agrees with a similar finding in Hornig & Priest (2003) that the diameter of their non-ideal region did not affect the reconnection rate.

### (b) Localized current

In this section, we assume an alternative form for the current . We examine its effect on the remaining first- and second-order terms and compare the solutions with those found in §3*a*.

We have seen is constant along field lines of , i.e. satisfies (3.3). Another obvious choice for is one which produces an enhanced current at the origin, which, in turn, requires to be localized along the separatrices of . A suitable example is(3.12)A motivation for this choice is given by the numerical simulation of Pontin *et al*. (2005), who observed the evolution of magnetic flux in an isolated diffusion region within a HFT, and thus have a reconnection process similar in many respects to the one we are studying. The current concentration was found to grow throughout the run, and the final profile, as shown in figure 5, has a ‘bow-tie’ structure. The choice of current given by (3.12) results in a similar current density profile close to the origin.

Substituting (3.12) into (3.2) gives the pressure as(3.13)Whereas in the previous example (§3*a*) the pressure gradient had a stagnation structure, the localization of the current now gives the pressure gradient that is localized along the separatrices of . It is dependent on the sign of , taken to be negative here, although at this stage the choice is arbitrary. An example of the resulting pressure is shown in figure 6. The saddle-point pressure profile is a direct consequence of the hyperbolic nature of the field, since there is no inertial term in equation (3.2). Such saddle-point profile would persist in the presence of inertial terms of a magnitude similar to, or less than, the Lorenz force.

Setting , we may find a divergence-free field that produces the current given by (3.12). We are unable to use the method of infinite space Green's functions, since this would require the contribution of the ‘boundary’ terms of at infinity to vanish. Instead, we use the method described in appendix A to solvein the region with on the boundary. We obtain(3.14)where the coefficients are given byThe change of variables , , allows the integrand to be expressed in a form independent of *λ*, and we obtain the equivalent expression for the coefficients :(3.15)We find that each as and that as each tends to a limiting value. Thus, we use (3.14), with the coefficients (3.15) evaluated numerically, to find a form for .

is a smooth function with the opposite sign from that of , with the maximum of occurring at . The contours of , which are field lines for , are shown in figure 7. Superimposed is an outline of the current . The X-type structure of the field becomes flattened by the perturbation ; toward the *y*-axis when (which is assumed to be the case here) and toward the *x*-axis when . This is shown in figure 8 where the coefficient has been taken as to illustrate the effect.

Following the method used in §3*a*, we now prescribe a localized form for , and determine asby taking the scalar product of (3.1) with . Here we assume(3.16)with . This is similar to (3.4), with an extra factor to later ensure is sufficiently localized. Using the inverse field line mappings (2.13) to find an equivalent expression in terms of *x*, *y* and *z*, we obtain as(3.17)which is again a positive function, provided and are of the same sign.

Figure 9*a* shows the diffusion region in this example; it is seen to be very similar to that of §3*a*. Although the diffusion region given by (3.5) was circular in the *x*–*y* plane and elliptical for non-zero *z*-values, as illustrated by the cross-sections of figure 9*b*, in this case it is distorted from that shape by the current now lying along the separatrices of the field in the *x*–*y* plane.

We deduce an analytical form for using (3.6), which, in turn, allows us to find , the component of parallel to . This is given byto which we add a velocity component parallel to to set its -component to zero and ensure it is divergence-free:Settingthe resulting flow is given by(3.18)The additional factor , introduced in (3.16) (and not present in (3.4)), has had the effect of narrowing the HFT away from the -axis. The factor therefore has the same effect on the counter-rotational flow , as clearly shown in figure 10, although the qualitative structure remains largely the same.

The rate of reconnected flux can again be determined. The *z*-axis remains the reconnection line,(3.19)Note that this equation is precisely the same as that of the previous example, given by (3.11). The shape of the diffusion region in the *x*–*y* plane, which is different in both our examples, in turn, affects the shape of the HFT and therefore the structure of the plasma velocity . However, in the above expression these dimensions are unimportant, since it is the length of the diffusion region along the -axis which is key in determining the reconnection rate. In principal, any decaying function could have been used to determine this length.

The model used does not allow for a simple scaling of the reconnection rate with respect to the resistivity or Lundquist number, and so we cannot yet determine the maximum rate of reconnection. This is a consequence of three-dimensional reconnection being more complex and having a greater variety of solutions than the two-dimensional case. Consider the values of the variables at a height above the non-ideal region. There the ratio of the plasma velocity to the Alfvén velocity is given by(3.20)where is the global magnetic Reynolds number, and is a factor relating to the geometry of the magnetic field. The ordering of parameters has been assumed. The parameters *l* and *L*, which relate to the structure of the non-ideal region, and *g*, which relates to the field geometry, would not in a general three-dimensional reconnection event be arbitrary, but rather determined by the evolution of the magnetic field before the onset of a stationary phase. Therefore, we warn that the expression (3.20) should be interpreted with particular care. Although at first sight it appears to scale as , each of the other factors on the right-hand side of (3.20) can be much larger than unity and also depend on . Determining how scales with and so whether or not the reconnection is fast is therefore outside the scope of our simple stationary model.

## 4. Conclusion

We have examined a perturbation expansion of the three-dimensional stationary MHD equations. The basic state of an X-type equilibrium magnetic field has been assumed, and localized non-ideal region has been obtained by a localization of the resistivity .

Using this decomposition we obtained an ideal and non-ideal Ohm's law in the zeroth- and first-order Ohm's law, respectively, together with accompanying equations of motion. Assuming the trivial solution for the zeroth-order terms (excluding the magnetic field), we were able to examine separately the non-ideal solution alone. This directly corresponds to the pure solutions examined by Hornig & Priest (2003). The kinematic model of this analysis has here been generalized by setting up a model that also satisfies the momentum equation in our approximation scheme.

Two different magnetic fields have been considered, corresponding to a uniform current and to a current localized along the separatrices of the basic magnetic field. In the pure solutions obtained here, counter-rotating flows above and below the non-ideal region are observed that are limited to within the HFT, which threads the diffusion region. The same reconnection rate is observed in both examples, since the parallel electric field has been shown to be to some extent independent from the choice of current. Further, the dimensions of the diffusion region have been shown to be important only along the reconnection line (which is identified with the *z*-axis), with its extent and structure in the *x*–*y* plane being unimportant.

The structure of the plasma flow in the pure solution means that the reconnection is limited to effect the field lines within the HFT only. The inclusion of non-trivial solutions to the zeroth-order equations equates to the case of ‘composite solutions’ in Hornig & Priest (2003). We have some freedom to impose an ideal flow (e.g. a stagnation-point flow) on the solutions found in this paper. Such a flow could bring field lines into the diffusion region, and so change field line connectivity further away from this region. A certain degree of coupling between the ideal and non-ideal solutions will be required by the equation of motion, although this coupling may be weak. These questions will be addressed in future, where some various flows will be included in the analysis.

This system has been solved explicitly up to third order and a scheme outlined to allow a solution at all higher orders. Each new order includes enough free parameters to make the magnitude of the new contributions essentially free, and independent of the lower-order solutions. We are hopeful, therefore, that solutions may converge (provided ), and therefore that the solutions found are indeed good approximations to the full exact solutions.

## Acknowledgments

The authors would like to thank Piet Martens, Dick Canfield, Dana Longcope and Dibyendu Nandy for their support, and are grateful to the UK PPARC for funding.

## Footnotes

↵† Present address: Division of Mathematics, University of Dundee, Perth Road, Dundee DD1 4HN, UK.

- Received September 30, 2005.
- Accepted February 16, 2006.

- © 2006 The Royal Society