MHD flow in an insulating rectangular duct under a non-uniform magnetic field
© Moreau et al 2010
Received: 23 June 2010
Accepted: 1 October 2010
Published: 1 October 2010
Skip to main content
© Moreau et al 2010
Received: 23 June 2010
Accepted: 1 October 2010
Published: 1 October 2010
Followed by a review of previous studies of magnetohydrodynamic (MHD) duct flows in a non-uniform magnetic field at the entry into a magnet (fringing magnetic field), the associated MHD problem is revisited for a particular case of a nonconducting rectangular duct of a small aspect ratio ε = b/a (here, b is the duct half-width in the magnetic field direction, and a is the half-height). The suggested model includes a realistic three-component div- and curl-free fringing magnetic field as well as inertia terms and takes into account the mechanism of electric current exchange between the core of the flow and the Hartmann layers. The original three-dimensional flow equations are reduced to a quasi-two-dimensional (Q2D) form for three basic scalar quantities: the vorticity, the streamfunction and the electric potential. This Q2 D formulation implies that the velocity field in the core region between the two Hartmann layers does not change in the magnetic field direction and thus is two-dimensional, while the induced electric current forms both cross-sectional and axial circuits and is essentially three-dimensional. A new parameter R = ε 2 Re/Ha has been identified to characterize the role of inertia in duct flows with insulating walls (Re and Ha stand for the Reynolds and Hartmann numbers). Computations and analytical studies are performed for inertialess (R ≪ 1) and inertial (R ≫ 1) flows at ε = 0.2 for Re up to 300,000 resulting in new scaling laws for typical lengths, velocities, electric current densities and pressure drops, which provide a new theoretical basis for potential applications.
PACS Codes: 47.65.-d, 47, 47.11.-j
Magnetohydrodynamic (MHD) flows in a non-uniform magnetic field, or in ducts of a varying cross-sectional area, have been extensively studied since the seminal paper of Hartmann , where the formation of the axial current loops and the origin of the pressure losses associated with the fringing magnetic field either at the entry or exit of a magnet are qualitatively described. Later, Shercliff, in his two books [2, 3] and educational movie , showed that the presence of the axial currents promotes the formation of the so-called "M-shaped" velocity profile and associated vorticity. A few important experiments were made in Riga [5–8] and in Saint Petersburg , which revealed the spectacular behavior of such non-uniform flows. Incidentally, new phenomena of turbulence suppression in a fringing magnetic field was recently discovered experimentally in Ilmenau, Germany [10, 11]. Attempts to build a theory of those complex flows started with the papers by Kulikovskii [12, 13], where the concept of characteristic surfaces was first introduced. Although this concept is limited to the asymptotic case of a perfectly conducting fluid, it allows for qualitative predictions of the main features of many MHD flows, including those in a fringing field. Subsequently, different research groups, around Walker [14–23] and more recently Molokov [24–27], investigated such flows, using simplifying assumptions, e.g. limiting the analysis to inertialess flows. These authors obtained velocity and electric current distributions, and suggested first explanations of how these flows are organized. In particular, numerical results based on the inertialess flow model  have demonstrated a fair agreement with the experimental data for both pipe and rectangular duct flows in a strong magnetic field. However, other experiments for flows in a slotted duct at high Reynolds (Re ≈ 105) and moderate Hartmann numbers (Ha ≈ 102) have demonstrated strong inertial effects , which cannot be explained using inertialess models. As a matter of fact, except for Holroyd  and recent attempts [29, 30] to model the Ilmenau experiment [10, 11] most of the theoretical efforts are lacking direct comparisons with the experimental data, which often do not comply with the theoretical predictions. Quite recently, Alboussière reexamined the theory, paying a special attention to the full role of the Hartmann layers . His considerations are, however, still based on a few striking assumptions that limit the model applicability.
Most of previous theoretical investigations are based on limiting assumptions, which are revisited here. First, since our magnetic field is div- and curl-free, we do not restrict our considerations to the so-called straight magnetic field approximation, in which only the predominant field component B z (x) is retained (see, for instance, [16, 24, 31]). As a result, the magnetic field derived in the present paper (see Section 2) decays slowly at a large distance from the magnet as x -1. As a consequence of this slow decay, a well-known flow structure with the Hartmann and side layers and internal core similar to that in fully developed MHD flows is established at large upstream distances from the magnet.
The exchange of current between the core and the Hartmann layers is another fundamental mechanism that we intend to readdress. Unlike previous theories, where this mechanism is not introduced or included in an approximate way, we derive the three current components directly based on the properties of the Hartmann layers, the vorticity equation and by satisfying the charge conservation equation. This is different, for example, from the approach in Ref.  by Alboussière, where the exchange of electric currents between the core flow and the Hartmann layers is included indirectly by integrating (averaging) the current density distribution. Such an approach seems not to be valid in the flow region where the M-shaped velocity profile is formed and the local current density can differ significantly from its averaged value. Despite this limitation, Alboussière's analysis has the merit to renew the basic scaling law in this problem, introducing a new important length scale of the order of Ha -1/4 in the case of a pipe flow.
A third strong common assumption is the inertialess approximation  in which inertia in the core flow is neglected in comparison with the Lorentz force and the pressure gradient. This implies that, except the side layers where viscosity is significant, the Lorentz force is purely irrotational and thus cannot generate any vorticity. The rationale for this approximation is the assumption that the ratio of the Lorentz to inertia forces scales as the interaction parameter N = Ha 2/Re, which in fusion blanket conditions is much larger than unity. However, in electrically insulating ducts, this scaling law appears not to be correct any longer providing the exchange of current between the core and the Hartmann layers is taken into account. As a consequence, the Lorentz force is in fact Ha times smaller, and the relevant parameter is Ha/Re. As applied to the fusion blanket conditions, this suggests that in the blanket flows inertial effects are not necessarily negligible.
Notice that among various designs of a liquid-metal blanket, the EU Helium-Cooled Lead-Lithium (HCLL) concept requires low velocities (a few mm/s in the blanket ducts and a few cm/s in the feeding duct), whereas in the US Dual-Coolant Lead-Lithium (DCLL) concept much larger velocities are required for heat and tritium removal and proper cooling (of the order of 10 cm/s in the poloidal ducts and about 50 cm/s in the access ducts). Even higher velocities, of the order of 1 m/s or higher, may be needed in a self-cooled blanket. Therefore, to cover the whole velocity range in the blanket applications, our goal is to consider the flows in a fringing magnetic field starting from regimes with zero or weak inertia to highly inertial regimes.
The structure of the paper is the following. Following this introduction, a div- and curl-free magnetic field is derived in Section 2. This magnetic field is used in Section 3 in the derivations of the basic quasi-two-dimensional (Q2D) equations governing the reference flow in a fringing magnetic field. These equations are obtained from the original three-dimensional equations in terms of three scalar variables, vorticity, stream function and electric potential assuming that the flow variations in the core in the magnetic field direction are negligible, whereas the induced electric current is essentially three-dimensional. Then, in Section 4, typical numerical results are shown. The derivation of explicit asymptotic solutions is the purpose of Section 5, with a special emphasis on regimes with strong inertia and new scaling laws for typical lengths, velocities, electric current densities and pressure drops as compared to previous theories. Finally, a discussion and concluding remarks are given in Section 6.
only if the z-dependent component of the electromotive field u × B in equation (2c) is taken into account. The term ∂ z j z is of order unity even though j z is much smaller than j x and j y . Its contribution appears to be essential as it expresses the leakage of current from the axial loop toward the Hartmann layers.
The aim of this section is to get a correct expression for the function B(x, y), which is in fact one of the main parameters of this problem. Note first that B(x, y) should vary as x -3 or y -3 at large distances from the gap, if the magnetic field were a three-dimensional dipole. If this dipole were two-dimensional, as in the case of infinite magnet poles in the y-direction, the far field should vary as x -2. In the reference case, when the poles are semi-infinite in the x-direction and infinite in the y-direction, the far potential is logarithmic and the far field must vary as x -1. If so, the Hartmann layers still exist at a long distance from the magnetic poles. Indeed, considering that their thickness is proportional to , the overall length over which the flow exhibits Hartmann layers is b m Ha in the case of a two-dimensional very long magnet. It is b m Ha 1/2 in the case of a two-dimensional short magnet, and b m Ha 1/3 in the case of a three-dimensional short magnet.
where K is a complex constant which has to be determined. This mapping transforms the half-plane Im(t) > 0 into the region between the symmetry axis and the yoke of the semi-infinite magnet in the r-plane. Due to the exponent 1/2, rotation of the vector representing the complex number t - 1 by π between the values t = 1 + η and t = 1 - η (η ≪ 1 is a real number), corresponds to rotation of the vectors (t - 1)1/2 and dr by π/2. Therefore, the suggested transformation can reproduce the sharp corner of the yoke, around which the potential lines of the magnetic field turn in the fringing field region.
Consider the flow of an incompressible and electrically conducting fluid in the rectangular duct shown in figure 1. Since the duct width is constant (2b), and since ε m ≪ 1, the characteristic surfaces coincide with the cross-sections x = const. As shown by Kulikovskii [12, 13] and Holroyd & Walker , neither the streamlines, nor the electric current lines can cross them within the core flow.
Where Φ is the magnetic flux, implies that the electric current crossing the characteristic surfaces is determined by the Ohmic losses (locally j/σ). As a consequence, the whole Hartmann layers (not only their portions overlapping the side layers), where these Ohmic losses have the highest values, must be taken into account on both sides of the core. This differs from the Holroyd & Walker  ideas, according to which the only possibility for the fluid to cross the characteristic surfaces is within the side layers.
Note the presence in (16) of , which represents the exchange of current between the core and the Hartmann layers, and which, like the other terms, is only a function of x and y.
At the same time, providing that ω 0,ψ 0, φ 0 and φ 1 are known, the current density given by (14), remains three-dimensional since it contains the z-dependent terms and j 0z ≠ 0.
Equation (32) could also be used to derive a higher order equation for ω 1(x, y) by collecting terms proportional to z 2. In this investigation, however, we limit ourselves to a quasi-two-dimensional approximation based on equations (35) for ω 0(x, y), (29) for φ 0(x, y), (30) for φ 1(x, y), and the condition of incompressibility Δ⊥ ψ 0 = -ω 0 for ψ 0(x, y). In this closed set of equations, the vorticity production due to the non-zero derivatives ∂ x B and ∂ y B is clearly exhibited by the first and second terms on the right-hand-side of equation (35). The Hartmann damping of vorticity is represented by the third term . The additional viscous dissipation is represented by the last term, which is significant only in the side layers. This basic equation may be seen as a generalization of the Sommeria & Moreau  model equation to the case of non-uniform magnetic fields. It is also worth mentioning that, after an integration of equation (32) between the two Hartmann walls, a more complete model might be obtained, which would additionally take into account the contribution of the parabolic z-dependent terms. The derivation of the complete form of the vorticity transport equation would, however, require the introduction of terms proportional to z 4 in the basic expression (10) for the electric potential. Such a model, would be a generalization of the approach presented in , allowing for high ε values.
Notice that in the particular case B = B(x), equation (35) can also be interpreted as an expression for the axial part of the current density component j 0x (x, y, z = 0) = -σ(∂xφ0 - Bv 0), parallel to ∇⊥ B, which is expressed through other unknowns (velocity and vorticity). In other words, this equation explicitly demonstrates that, within the core, electric current lines must cross the characteristic surfaces due, essentially, to two main causes: inertia and Hartmann current sheet. As a consequence, a full three-dimensional electric current distribution can accurately be analyzed (see Section 5), in which j 0x and j 0z are derived from equations (35) and (31), while j 0y can be easily obtained from the charge conservation equation.
The above expressions for J X and J Y are, of course, valid to the leading order in terms of ε.
We now restrict the analysis to the case of a fringing field of the type studied in Section 2.2 where the magnetic field is y-independent and given by the function β(X) (see equation (8)).
One should notice that equations (46) and (47) show explicitly the exchange of current between the core and the Hartmann and side layers. It is also clear that in the limit of negligible inertia (R = 0), the three current density components in the core have the same scaling law, all being proportional to Ha -1. Equations (46) have also the merit of exhibiting the contributions of inertia and viscosity on the axial current loop.
which is the basic equation for the vorticity in the fringing field. This equation expresses all the relevant contributions to the transport of vorticity, which are, respectively, the advection of vorticity, Hartmann damping, damping in the side layers, torque and exchange of current. Note that the advective term, Δ(RΓ/β'), involves the dimensionless parameter R = ε 2 Re/Ha, which results as the suitable measure of the importance of inertia, in contrast with previous theories where the relevant scaling for inertia was N -1 = ε 2 Re/Ha 2.
Although the equations presented in previous sections are quasi-two-dimensional, their accurate numerical solution requires serious efforts, in particular if the Hartmann and Reynolds numbers are high, since computations become more time and memory demanding. Unlike a fully three-dimensional approach, the present one has the advantage of not computing the flow in the Hartmann layers. However, resolution of high-gradient flow regions associated with the side layers of thickness at the walls y = ± a is still needed to reproduce correctly the axial and cross-sectional current loops and the current exchange between the bulk flow and the boundary layers. Another important feature of the flow, which should be properly treated in the computations, is the formation of a thin transverse layer at the entry to the magnet (see Section 5). It should also be taken into consideration that the reference flow exhibits high velocity jets once the liquid enters the magnet zone. Such velocity profiles present inflection points, it is thus important that the computed results can indicate whether or not the flow loses its stability and develops a time-dependent behavior. Then, computing time-dependent flows needs small time increments and special approximations of the convective terms to minimize the schematic viscosity and to keep stable computations at the same time.
In the past, many computations of MHD flows demonstrated severe restrictions on the magnitude of the Hartmann number, even for relatively simple and well-understood MHD flows, e.g. fully developed flows in a rectangular duct under a uniform magnetic field. Typically, the Hartmann number in such studies was limited to Ha ≈ 102 [35, 36]. The limitations were to some extent related to the computational time, which increases roughly exponentially with Ha. However, the most severe limitations are caused by a catastrophic loss of accuracy at high Ha numbers, which may occur in the course of computations due to accumulation of round-off errors. A potential source of computational errors can be related to the use of the electric potential as the electromagnetic variable. As discussed by Smolentsev et al. , the loss of accuracy occurs when an induced electric current of small magnitude (O(Ha -1)), is calculated from the Ohm's law as a difference between two large numbers, associated with -∇Φ and U × B 0, which are both O(1). In the present formulation, the nature of this problem can be seen directly from the equation for the potential (37). Two terms on the right-hand-side of this equation, βΩ and Ω/Ha, are different by a factor Ha, once the magnetic field is almost uniform. However, ignoring the term Ω/Ha, the whole physical mechanism responsible for the current exchange between the bulk flow and the Hartmann layers would be missed. First numerical tests showed that using the electric potential in the computations for the reference problem is reasonable only for moderate values of Ha, limited to a few tens, but the above mentioned accuracy problem becomes extremely critical at higher Ha.
Usually the accuracy problem associated with the electric potential does not appear if the alternative formulation based on either the induced magnetic field or electric current is used. For example, when the induced magnetic field is used to calculate the electric current from Ampere's law, the continuity equation for the electric current is automatically satisfied based on the vector identity div(rot) ≡ 0. Therefore, the computed electric current always satisfies the charge conservation equation. However, when applying the approach based on the induced magnetic field, the boundary conditions should be formulated far enough from the flow domain, where the induced magnetic field vanishes. It is therefore more convenient to use the electric current, namely its J X -component, which obeys equation (48), as an electromagnetic variable, since the boundary conditions can be formulated at the interface between the liquid and the wall (insulating duct) or at the external surface of the wall (conducting duct).
The boundary condition at the walls Y = ± 1, which are assumed to be non-conducting, is ∂ Y J X = 0, as follows directly from the expression for the Z-component of ∇ × J and taking into account that J Y = 0 at the walls. The J X current density component is then used to calculate the Lorentz force term on the right-hand-side of the vorticity equation. In the post-processing phase, the other current component, J Y , can be derived from (50). After computing the current density components J X and J Y , the whole three-dimensional current distribution, including J Z , can be reconstructed. This formulation was intensively tested recently for both steady and unsteady flows . The tests demonstrated very good match with analogous computations based on the traditional induced magnetic field formulation.
In the code, a finite-difference technique using a non-uniform grid in the Y -direction that clusters grid points in the side layers is applied. The whole problem is solved by advancing in time, until a steady-state solution is achieved. The vorticity equation at each time step is solved using the TDMA (Tridiagonal Matrix Algorithm) method in both coordinate directions. The convective terms in the vorticity equation are approximated with the central-difference scheme, which is known to be dissipation free. Using convective term approximations free of numerical dissipation is particularly important when the flow is essentially time-dependent. However, in the chosen range of parameters, including high values of Re (up to 300,000), the numerical solution always converges to a steady-state regime without showing any tendency to instability. The elliptic equations for the streamfunction (39) and for the electric current (48) are both solved with the cos fast Fourier transform (FFT) in X and the tridiagonal algorithm in the other direction. The use of the FFT technique limits the computational grid in X to a uniform spacing, but is more effective compared with much slower relaxation techniques since solution of these elliptic equations may take over 90% of the total computational time. Using the cos FFT in X also requires the inlet/outlet boundaries to be positioned far enough from the fringing field region, so that Neuman boundary conditions ∂ X = 0 are allowed. Typical calculation domain is 40-60 units. Satisfactory resolution is achieved by using 512 or 1024 grid points in X-direction and 201-501 points in Y -direction. The numerical tests included grid sensitivity tests, and comparisons with the analytical solution for a fully developed flow and unsteady vortical flows [37, 38]. The comparisons demonstrated coincidence in local values between the analytical and the numerical results in four digits.
In the asymptotic case of large Ha and small R, these computed results suggest that the whole flow domain can be subdivided into five characteristic regions: i) the far upstream developing Shercliff-like flow, where the velocity profile is uniform with a moderate over-velocity to compensate the missing flow rate in the side layers; ii) the upstream region where the M-shape starts to form, whose position is close to abscissa X ≈ -15 and varies very slowly with Ha; iii) the jet-region, where most of the flow rate is transported by two jets located on both sides of the almost motionless core, and whose length seems to be weakly Ha-dependent; iv) a short transverse layer, whose length seems significantly smaller than the duct half width, suggesting that the X-derivatives should be locally larger than the Y -derivatives; and finally v) the fully-established Shercliff flow with its familiar plateau-type velocity between the side layers. In Section 5, a theoretical attempt is taken to derive the most important characteristics for the three intermediate regions where the flow is strongly affected by the variation of the magnetic field. The fact that the jet formation occurs over a distance of a few units, which agrees with the present asymptotic theory (see Section 5), seems to be one of the newest ideas arising from this investigation.
The electric current distribution, also shown on figure 3, shows a transition from a three-dimensional structure in the jet region to a purely cross-sectional current distribution in the fully-established Shercliff flow. In turn, in the jet region the three-dimensional current density demonstrates two characteristic families of current loops: an internal one where the current lines are almost quasi-two-dimensional and closed in the core, and an external one where the current lines passing through the jets and side layers are strongly three-dimensional and close themselves within the Hartmann layers. Additional numerical results are presented in Section 5, within the framework of the theoretical ideas that allow the interpretation of these results in terms of scaling laws, specific to each sub-domain.
Figures 6 and 7 show that, when inertia is significant, the flow is quasi-parallel all over the fringing region. When inertia is weak, the numerical results suggest that this may still be true, but only in the far upstream region (β 2 ≪ 1) and in the jet region, where the streamlines exhibit a very small curvature (see figure 3). Therefore, let us pay attention to conditions where it can be assumed that in the Laplacians of equations (39), (41), (48) and (49). Of course, there are situations, for instance when inertia is weak, where this approximation is not valid and consequently deviate from the coming analysis. Inertiales regime is studied in Section 5.3.3 where an attempt is presented to analyze the short transition region toward the downstream Shercliff flow. In this case, the flow is far from being quasi-parallel and hence, becomes predominant upon (see figure 3).
Its maximum (ΔP)2 = (πHa/10ε)(3/5)3/2 ≈ 0.146(Ha/ε) is reached for .
The terms of this linear equation express all major effects: from left to right, we have advection of vorticity , Hartmann damping , damping in the side layers , torque (Haβ' 2Ω) and exchange of current (β'2∂βΩ).
Note that an additional term of the order of U 1 δ SL Y might be introduced in the expression for Ψ, to take into account the missing flow rate in the side layers. However, as suggested by (61), this term is small enough to be neglected in this asymptotic approach.
In particular, in most experimental conditions, these velocities deviate from their asymptotic limit as Nβ 3/ε, where N = Ha/R is the interaction parameter. This shows that the typical length of the upstream region where the M-shape profile is formed, scales as L up ≈ (ε 2 Ha/R)1/3 and becomes shorter the higher the R values. This can be interpreted as the squeezing of the upstream region against the fringing field by the advection of vorticity, clearly observed in figures 6 and 7.
The correction for the U 0 value is small since t/6 ≪ 1, but the correction for U 1 is important and implies a reduction of its maximum value by a factor of the order Ha/εR. Nevertheless, the typical length necessary to reach the Shercliff profile remains controlled by e-X/R and scales as L down ≈ R. This downstream transport of the M-shape over a long distance is also clear on figures 6 and 7.
This is a heat-like equation with the non-uniform diffusivity β/β', the minus sign implying that diffusion is directed upstream. Remarkably, the Hartmann number has disappeared from equation (92), what implies that it does not affect the geometry, neither the typical length scale over which the M-shape profile is formed, nor the jet thickness. The Hartmann number influences only the local value of Ω(β) within this region through the amplitude coefficient e Haβ . It is worth noticing that the two terms involving ∂ β Ω in (67) illustrate the competition between the downstream advection of vorticity, predominant in the case analyzed in Section 5.2 and the upstream diffusion due to the exchange of current, predominant in the present analyzed case.
and shows that α deviates from its zero value in the far upstream region when β 3 ≈ (ε 2/Ha)(1+3πR/2ε). Therefore, the length over which the M-shape profile develops is L M-sh ≈ (εHa)1/3 in the inertialess limit. In the case of flows with weak inertia such that R » 1, we just recover the previous estimate obtained in Section 5.2: L up ≈ (ε 2 Ha/R)1/3It is worth noticing that this estimate combines two effects: the upstream diffusion in a geometry independent of H a, and the vorticity distribution, which is itself H a-dependent.
The components of the current density in this region immediately follow from equation (46) for the axial loop and (47) for the cross-sectional loop, while expression (64) yields the stream function h a . Similarly, the pressure variation follows from equations (56a) and (56b), whereas equations (63) yield the head loss. We do not give details of these expressions since, as will be shown below, the predominant contribution does not come from the upstream region, but from the short transition toward the downstream Shercliff flow.
which requires U 0 = 0. Note that the numerical results suggest values significantly smaller than unity, but nevertheless positive, even for Ha = 103 or larger. This small discrepancy is certainly a consequence of the quasi-two-dimensional approximation that neglects the exchange of current and derivatives in the Laplacians. Notice that these quantities, neglected in both the upstream and jet regions, are certainly signficant between these regions, as suggested by figure 3.
This is a well-known equation in MHD duct flows in the presence of a uniform magnetic field although, usually, a coefficient Ha 2 appears instead of Ha. This equation characterizes all kinds of free shear layers parallel to the applied magnetic field, often named inertialess Ludford layers, located between cores having different properties. The main novelty introduced by (106) is the thickness of the transverse layer, of the order Ha -1/4 in ξ-units, instead of Ha -1/2 in the case of Ludford layers. This layer has a clear similarity with the free shear layer discovered by Alboussière  along a singular characteristic surface "which splits" in ducts having a non-uniform width in the magnetic field direction. We now find that this layer appears along one of the characteristic surfaces whose shape is not at all singular, since all surfaces have the same shape (plane cross-sections of the rectangular duct). Indeed, it appears that this layer exists as soon as the slowly X-dependent solution exhibits a singularity. In our case, the origin of the singularity lies in the solution for the M-shape profile region, when β →1 and β' →0. Hence, the transverse layer provides the smooth transition between the M-shape flow region and the downstream fully developed Shercliff flow.
Several solutions of equations similar to (106) for flows in a uniform magnetic field have been found, for instance by Moffatt  (see also Moreau , pp. 138-150; Müller & Bühler , pp. 85-89). They have a remarkable structure determined by the superposition of two layers, one issuing from Y = -1 with a zero initial thickness, propagating in the positive Y-direction and diffusing in the ± X-directions, and the other issuing from Y = +1 with a zero thickness, propagating in the negative Y-direction and diffusing also in the ± X-directions. Here the term "propagating" reminds that these layers can be interpreted as diffusive Alfvén waves propagating in the ± B directions. The analogy between the previously studied free shear layers and the present transverse layer that appears in rectangular ducts at the entry of a strong magnetic field seems to stop at this point, since the analogous propagation is not any more in the direction of the magnetic field. This transverse layer seems to be the only possibility to feed the downstream fully developed flow, complying with the requirement that the characteristic surfaces cannot be crossed by the streamlines.
This prediction of a Ha-independent thickness in X-units is a priori surprising. However, it has been confirmed through additional numerical results (not shown in this paper), obtained when the parameters used in figure 3 are modified within the range of small R-values to maintain inertia negligible. This actually demonstrates that the properties of this flow in a fringing field are not generic enough to be analyzed without making the choice of a specific β(X) distribution. For instance, would we use the distribution β' = (πν/ε) n , the above derivation would yield δ X ≈ (Ha n-1 ε 2n )1/(4-2n). This clearly illustrates that the choice n = 1, corresponding to the magnetic field distribution (8), is the one which makes the transverse layer thickness independent of Ha.
It can be verified that the global current is Y-independent and takes the remarkably simple value . As shown in figure 9, this explains that the axial current issuing from the jets, completes its trajectory through the transverse layer.
This a priori surprising result for the inertial regime is a consequence of the required balance between inertia (proportional to ) and Lorentz force (proportional to ).
In this paper, the problem of the flow in a non-uniform magnetic field is revisited by paying special attention to accurate representation of the applied fringing magnetic field, electric current exchange between the core and the Hartmann layers, and the role of inertia. The starting point is the derivation of a consistent div- and curl-free magnetic field which models the inhomogeneities of the actual field at the entrance of a large yoke. A second important point is the explicit consideration of the Hartmann layers and the electric current exchange mechanism between them and the core flow. One of the key consequences of this mechanism is that the relevant scaling for the ratio of inertial and Lorentz forces in the case of insulating walls is R =ε 2 Re/Ha instead of N -1 = ε 2 Re/Ha 2. Another important consequence is that the three-dimensional electric current distribution is formed by a pair of two-dimensional current loops, namely, one in the axial (x, y), and another in the cross-sectional (y, z) planes. In the observed three-dimensional electric current distribution, in contrast with previous theories, a part of the induced current can cross the Kulikovskii characteristic surfaces within the core flow, in spite of the constraint imposed by the Alfvén theorem. As a result, the core flow can exhibit a non-zero vorticity generated by the rotational Lorentz force, which can be balanced by inertia, local Hartmann damping, and friction in the side layers.
The quantification of inertia in the reference flows is another novelty. After adapting the asymptotic theory based on the quasi-parallel flow assumption ( ) and Oseen's approximation to flows with strong inertia we were able to construct scaling laws for the length over which the M-shape profile develops, which is found to be (ε 2 Ha/Re)1/3, and for the velocity maximum in the form U 1 - U 0 ≈ Ha/εR. These scaling laws are in qualitative agreement with the numerical predictions. For comparison, in the inertialess regime this length scale was found as (εHa)1/3 and the maximum velocity of the M-shape profile as U 1 - U 0 ≈ 1. Besides, the predicted downstream advection of the M-shape profile over a distance of order of R into the region where the magnetic field is uniform, also seems to be a distinguished feature of inertial flows. We have also observed that the electric current circuits in the fringing field region become less and less three-dimensional, with more currents closing in the axial planes, as inertia increases. Hence, for the inertial regime the exchange of current between the core and the Hartmann layers seems to be much less important than for inertialess flows.
Another remarkable property of the regimes with strong inertia is that the computational results remain quite stable, even at large Reynolds numbers (Re ≈ 3 ×105). The stability of the computations is in fact in agreement with recent observations of turbulence suppression by Andreev et al. [10, 11] in the experiments with a short magnet. This striking feature can be considered as one of the most noticeable properties of the flows in fringing fields. While in a strong uniform field quasi-two-dimensional flow disturbances are promoted and damped slowly for a typical Hartmann time (see [33, 43]), in a region of non-uniform field, disturbances seem to be much more rapidly damped. As shown elsewhere , the main damping mechanism in the fringing field zone is due to electric currents crossing the Kulikovski's characteristic surfaces which create extra pressure losses in the mean flow and extra Joule dissipation affecting turbulent pulsations. These currents are induced by any vortical disturbance with a significant velocity component in the x-direction. In the non-uniform field zone where β' and β are of order unity, the additional damping is Ha times larger than the Hartmann damping. This has major consequences. In the first place, the extra damping overcomes the Hartmann braking effect, as well as both the non-linear interactions between vortices and the mean flow, and the instability mechanisms based on inflectional velocity profiles. Secondly, the quasi-two-dimensional vortices typical of MHD turbulence, which travel across the characteristic surfaces, are more rapidly damped than vortices elongated in the y-direction, parallel to these surfaces. This explains that, in spite of its simplicity, the linear model presented in  provides predictions in fair agreement with measurements reported in [10, 11]. In short, the characteristic surfaces progressively laminarize the turbulent upstream disturbances and stabilize the flow more rapidly and more efficiently than any other mechanism.
Let us emphasize that the choice of a particular div- and curl-free two-dimensional magnetic field is particularly important, instead of assumptions on the length scale over which this field varies. Indeed, many crucial results, such as the thickness of the transverse layer and the final head losses, would be different with another magnetic field distribution. There remains some questions where the asymptotic analysis does not provide any clear answer, such as the dependence of the maximum velocity U 1max reached in the jets on the parameters Ha and ε. Additional important points, such as the flow properties when ε ≥ 1, or when C ≠ 0 would deserve to be re-examined. The general guideline of this investigation might easily be adapted to such additions, by adjusting the general formulation to new assumptions.
Finally, the results presented in this paper for flows with a strong or moderate inertia agree rather well with Tananaev's experiments performed in a similar geometry. The orders of magnitude of typical velocities and pressures agree fairly well. It is however difficult to quantitatively benchmark these data, because the exact values of the parameters (aspect ratio, definitions of the Re and Ha numbers) are not given with enough care in Tananaev's book .
Consider now the opposite case where inertia is strong: N ≪1. In the expression of U 1 - 1 for β ≫ 1, N ≪1, substitute to the varying variable t its limit when β = 1: t max = πHa/15εR.
It is noticeable that, if N ≫ 1 the total head loss is controlled by the initial asymptotic branch where U 1 increases toward its maximum, whereas it is controlled by the final asymptotic branch where U 1 decreases from its maximum if N ≪1.
We are indebted to Mohamed Abdou, Thierry Alboussière, Neil Morley, Mingjiu Ni, Ramakanth Munipali and the group at Hypercomp Inc., for fruitful discussions. RM is indebted to the UCLA fusion group for supporting his visits during the accomplishment of this work. SS acknowledges support from the U.S. Department of Energy via grant DE-FG02-86ER52123-A040. SC thankfully acknowledges support from CONACYT under project 59977.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.