Sei sulla pagina 1di 31

MAGIC course, spring 2010

Topological uid mechanics


Andrew D. Gilbert and Mitchell Berger
Mathematics Research Institute, School of Engineering, Mathematics and Physical Sciences, University of Exeter, U.K.

Why topological uid mechanics?

To the authors knowledge, the term was rst coined for an IUTAM conference in Cambridge, U.K. in 1989, by Keith Moatt and Arkady Tsinober [13]. The aim of the conference was to bring together scientists interested in uid mechanics and the sister subject of magnetohydrodynamics (MHD) who were wrestling with problems of understanding the complexity of uid ows and magnetic elds. Overall the aim is to approach such problems from a general viewpoint that complements the more classical approaches. So for example a classical question would be to understand the instability of plane parallel ows, introducing problems involving eigenmodes and eigenfunctions for specic proles, and some general results such as Rayleighs inection point theorem. On the other hand the type of question that topological uid mechanics would aim to address is: what is the general structure of a steady inviscid uid ow? Can stability be proven for whole classes of two- or three-dimensional ows in arbitrary shaped regions? Can we construct ows with general streamline topology? If a magnetic eld is tangled and knotted, how much energy may be realised if it is allowed to reconnect? Such questions have relevance in understanding a whole range of phenomena, from the structure of turbulent ow, to superuid ow (in which vortices form a tangle of lines), to the magnetic elds of the solar corona and generation of eld deep in the Earth or Sun. In each case a range of methods of looking at general elds and their structure have proved useful and the aim of this course is to set out some of the theory and applications. In a sense the term topological uid mechanics is something of an umbrella

Preprint submitted to Elsevier

8 February 2010

term, and some of the methods and ideas could perhaps be better described as geometrical. Nonetheless it is a useful title to bring together a whole range of topics that take a general view of whole classes of uid ows and magnetic elds and their properties. This includes many works with a distinctly pure mathematical avour, and some areas of study rapidly move into studying general inviscid ows on manifolds of arbitrary dimension, or to quantifying how big groups of dieomorphisms are. However our approach will be from the angle of applied mathematics, with a close eye on applications and not being concerned about technicalities. We have benetted from a range of sources of information, and in particular have made use of material in the unpublished lecture notes of Steve Childress [3] for the beginning of the course, with his permission.

Topics

The course will be given by Andrew Gilbert and Mitchell Berger and cover the following topics: Part I (AG): basics, helicity and relaxation Background and motivation, hydrodynamics and magnetohydrodynamics. Revision of Kelvins theorem and magnetic analogies. Fluid, magnetic and cross helicity, geometrical interpretation. Magnetic relaxation. Part II (MB): knots, tangles, braids and applications Link, twist and writhe of ux and vortex tubes. Braiding of ux and vortex tubes. Vortex tangles in quantum uids and vortex tubes in turbulence, crossing numbers. Chaotic mixing, stirrer protocols, pA maps and topological entropy. Part III (AG): dynamics of point and line vortices Point vortex dynamics, invariants, integrability. Vortex tube dynamics, local induction approximation, invariants, solitons. If time: introduction to Lagrangian uid mechanics. These Latex notes cover the part I, which is the rst three lectures, and there are exercises and bibliography at the end. The notes are pretty informal and I have not cross referenced equations, preferring to repeat them where necessary. This is because these notes are a basis for the lectures I will give: the lectures will also have an informal style but be more bullet point orientated and include pictures and sketches. Finally there may be (will be!) errors and slips in the notes and lectures: please do not hesitate to bring these to my attention. 2

A few general notes

The summation convention is in force in these notes unless otherwise stated. We sum over repeated (dummy) indices in each product. Single (free) indices are not summed and label dierent equations. We will refer to position with vector r or x and use either (x, y, z) or (x1 , x2 , x3 ) interchangeably. Note that r = |r|. We may use components of a vector u as (u1 , u2 , u3 ), especially in general theory, but also (u, v, w), particularly in examples. We use the convenient abbreviation t for /t and i for /xi . For a vector b we often write b2 for b b = |b|2 . We will often write something like u A. This can be thought of as (u )A where u = ui i is an operator acting on the vector eld A. Or, we can think of A as a matrix whose ijth component is i Aj . In this case we are simply dotting u into the rst index by taking ui i Aj . In the latter case we consider A as dening a matrix and we dene | A|2 = (i Aj )(i Aj ), that is the sum of the squares of the entries of the matrix. We assume a good knowledge of vector calculus identities: many are given below.

Vector calculus identities

Here is an incomplete list of useful identities: () = + (u) = u + u (u) = u + u (u v) = u v + v u + u ( v) + v ( u ( u) = 1 u2 u u 2 (u v) = v u u v (u v) = u v v u + v u u v ( u) = ( u) 2 u =0 u=0

u)

We also make much use of Gauss theorem (the divergence theorem) u dV =


V S

n u dS

where V is a volume enclosed by a surface S with outward normal n. Or we 3

have it in a more general form i dV = ni dS

where i = /xi and ni is the ith component of the normal vector n. Here can be any quantity, so could have a set of indices itself. We also occasionally use Stokes theorem n
S

u dS =
C

u dr

where S is a surface with boundary C and the direction of the normal n to S is related to the direction of integration around C by the usual right-hand rule.

Governing equations for uid ow

We consider uid ow u(r, t) with constant density, governed by the Navier Stokes equation in the form t u + u u= p+ u=0 The latter condition, of a solenoidal or divergenceless vector eld, is used a great deal below. The viscosity will always be a constant. We will often make use of the material derivative Dt = for which the equation becomes Dt u = p +
2 2

D = t + u Dt

The ows may be in innite space, or in a bounded domain D with boundary S in which case we apply a no-slip condition u=0 (r S)

We will usually focus on inviscid or ideal uid ow, governed by the Euler equation, with zero viscosity (this is a highly singular limit!), namely, t u + u 4 u= p

This can also be written in the attractive form t u = u where the Bernoulli function is
1 P = p + 2 u2

and the vorticity is given by = u which is another solenoidal vector eld =0 Note that by taking the divergence, the pressure is given by
2

P =

(u )

Actually we have lapsed here: P is the Bernoulli function, but we often slip into calling the pressure. Whatever it is called, obtaining P is thus a nonlocal procedure: the ow and vorticity at all points in the domain aect all other points. In a numerical code, this Poisson equation has to be solved each time-step. Taking the curl eliminates the pressure and gives the vorticity equation t = or Dt = t + u = u This latter equation shows the is transported with the uid ow (left-hand side) but also undergoes stretching and rotation (right-hand side). Technically it is Lie-dragged in the ow and for this reason vorticity (rather than velocity) plays a central role in discussion of topological uid mechanics. In other words given a ow u, the above equation carries vectors as though they each join two innitesimally close uid particles. Here the Lie bracket of two vector elds is dened as [u, v] = u v v u It is antisymmetric [u, v] = [u, v] and satises the Jacobi identity [[u, v], w] + [[v, w], u] + [[w, u], v] = 0 For more information, see books on dierential geometry, for example that of Schutz [16]. In the ideal case we also have boundary conditions: in a nite domain D we impose a no normal ow condition nu=0 5 (r S) (u )

where n is a normal vector to S pointing outwards from D. As setting = 0 reduces the order of the equation, we are forced to reduce the number of boundary conditions and can no longer require that the ow be no-slip at the boundary. Note that we will use D for the domain, the entire volume, in which we are working, and V if we are using a subvolume of this. We will use S for the surface of either D or V: which will be clear from the context. Sometimes it is convenient to work in unbounded space D = R3 , for which we typically assume that the vorticity eld is localised, so falling o with r faster than any negative power of r so that for any n, |(r, t)| = o(rn ) (r )

For working in innite space we start with a nite sphere SR of radius R enclosing a volume VR . We then derive an exact result valid in VR and then allow R to tend to innity. A condition such as that the vorticity eld be localised will typically mean that the surface integral over SR vanishes as R , leaving us with a result valid over all space D = R3 . Sometimes it is instead convenient to work in periodic space, for example we set D to be the periodic cube T3 dened as [0, 2]3 with periodic boundary conditions so that surface integrals sum to zero on opposite faces by periodicity (as the normal vector n changes sign, all other quantites remaining the same). One may have to take care dening potentials in this geometry, for example magnetic vector potential A below. Note that in a nite domain D with the boundary condition u n = 0, the divergence-free eld u is orthogonal to any gradient in the sense that u
D

dV =
D

(u) dV =
S

u n dS = 0

using Gauss theorem (the divergence theorem).

Cauchy solution

A particle or uid element will be carried by a uid ow. It will start at say a at time t = 0 and at each moment it will have position x(t) and velocity u(x, t). Its position x(t) is then given at later times by the dierential equation dx = u(x(t), t) dt and initial condition x(0) = a It is convenient often to use a label a to label the initial condition, so we essentially follow the whole family of trajectories x(a, t) from all the initial 6

positions a and write x t and initial condition x(a, 0) = a Thinking about uid dynamics in terms of these Lagrangian coordinates a gives the Lagrangian picture of the ow, which has some advantages, though many disadvantages (in that the pressure transforms in a complicated way in Lagrangian coordinates). Within the Lagrangian picture one can solve for the vorticity eld if the velocity eld is known. Of course the two are linked in a complex way for ideal uid ow! But life is a bit easier in the kinematic dynamo problem: see later. What about an innitesimal vector that joins x and x + x? We get the equation for this by linearising the above ODE to yield x t = x
a

= u(x(a, t), t)
a

u(x(a, t), t)

We have Taylor expanded in x and dropped quadratic and higher order terms. This essentially denes an innitesimal vector: it is always so short that we can neglect these higher order terms (one can imagine rescaling it). This is the vector used for example for measuring Liapunov exponents (and technically it is in the tangent bundle of the domain in question). This is essentially the same equation as that for vorticity in the form Dt = u

conrming that vorticity vectors are carried like innitesimal vectors in the ow. We can also obtain a similar equation for the Jacobian: given x(a, t) we may set xi J (a, t) = x/a or Jij = aj Dierentiating the equation for particle motion with respect to a gives J t = ( u)T J
a

or

Jij t

= k ui (x(a, t), t) Jkj ,


a

J (a, 0) = I

(with I as the identity matrix). Here u is the matrix of derivatives of u, with ( u)ij = i uj and T gives its matrix transpose. 7

This gives us solutions to transport of vectors and scalars in the ow. First consider a scalar eld that is carried in the ow passively according to Dt = t + u =0

with initial conditon (a, 0) = 0 (a). The solution is given by (x(a, t), t) = 0 (a) as can be checked by dierentiating with respect to time at constant a. Similarly suppose the initial vorticity is given by (a, 0) = 0 (a) and the vorticity (x, t) obeys Dt = t + u = u

Then the solution to this, the Cauchy solution, is given by (x(a, t), t) = J (a, t) 0 (a) or i (x(a, t), t) = Jij (a, t)0j (a) or, taking x = x(a, t) as read, we have (x, t) = J (a, t) 0 (a) or i (x, t) = Jij (a, t)0j (a) This natural link between the motion of particles, innitesimal vectors and Jacobians, and the equations governing vorticity and (later) magnetic elds is at the heart of the notion of Lie-dragging. It is also natural to think of vortex lines, that is integral curves of the vorticity eld given by dx2 dx2 dx1 = = 1 2 3 at a xed time t, being carried materially by the uid ow. If we mark vectors on these lines, joining material points, then as the points move apart or rotate the vectors stretch or rotate accordingly. Given that we can solve the vorticity equation by means of the Cauchy solution it may seem attractive now to use this to solve the Euler equation. The problem though is that it is very hard to invert the vorticity solution to give the ow eld (this being a non-local procedure) and it is only in very special cases that these Lagrangian methods can be used to solve uid problems. Note that for similar reasons there is no obvious means to transport ow vectors u in the uid. Whereas the equations for vorticity permit a local calculation of the evolution of vectors, those for u involve the non-local pressure term, which requires a global calculation. 8

Transport of gradients or one-forms

There is another natural means of transporting vectors in a ow. Suppose we dierentiate the equation Dt = t + u =0 . This

for a passive scalar to obtain the equation for its gradient, say g = satises Dt g = t g + u g = ( u) g Here u is the matrix of derivatives of u, with ( u)ij = i uj

Thus gradients are transported dierently from vorticity or magnetic eld vectors. This is natural when you think how they are dened: they point in the direction of greatest increase of at a point: if there is stretching in that direction then the gradient is reduced! Technically we ought to distinguish vectors that are stretched like vorticity, obeying the Lie dragging equation Dt b = t b + u b=b u

and one-forms or covectors which obey a dierent Lie-dragging equation Dt g = t g + u g = ( u) g

Books on dierential geometry (also modern books on relativity) make this important distinction. We do not want to get too sidelined by the manifolds and dierential geometry, but it is useful to know that there are these two natural kinds of vectors, with dierent stretching properties in the ow. When the properties are important we will refer to one-forms as distinct from vectors, but we may be a little lax sometimes. Furthermore if we take a scalar product we can check that Dt (b g) = b ( u) g b ( u) g = 0 so the scalar product of a vector eld and one-form eld is a scalar eld with Dt (b g) = 0 If we consider 3 innitesimal vectors transported in a uid ow, b, c and d, then these demarcate an innitesimal area element, whose volume is preserved as u = 0. Thus we have Dt (b c d) = 0 9

Now, this is true for all choices of vector elds b, c and d and so it can be shown that bc gives a one-form which may be thought of as an area element, say dS = n dS. We obtain the result that area elements are transported like gradients, according to Dt dS = ( u) dS Again, thinking of how an area element is transported under stretching or shearing gives some intuition of what is going on, and why the transport equation is dierent from that for vectors. Wavevectors are also transported in the same way, in other words as one-forms. We can also write for innitesimal vectors again and for innitesimal volumes, Dt dr = dr Dt dV = 0 Finally we have the Cauchy solution for vector elds, and something tells us that there should be something analogous for one-form elds. Suppose g is a one-form eld and b is an arbitrary vector eld, then from the Cauchy solution for a vector eld, say b(x, t) = J (a, t)b0 (a) or bi (x, t) = Jij (a, t)b0j (a) and the fact that Dt (b g) = 0 gives (b g)(x, t) = (b0 g 0 )(a) or (bi gi )(x, t) = (b0i g0i )(a) one can check that g(x, t)J (a, t) = g 0 (a) or gi (x, t)Jij (a, t) = g0j (a) Or, if we dene the matrix K(a, t) to be the inverse of J (a, t), J K = KJ = I then one can check that g(x, t) = g 0 (a, t)K(a, t) or gi (x, t) = g0j (a)Kji (a, t) An alternative route is to obtain the equation for K as K t = K( u)T
a

or Jij Kjk = Kij Jjk = ik

or

Kij t

= Kik j uk (x(a, t), t),


a

K(a, 0) = I

10

Kelvins theorem

Kelvins theorem states that in ideal ow, the circulation of the uid ow around any material curve C(t) is conserved in time. C = u dr
C(t)

By virtue of Stokes theorem this may be written as an integral over a surface S(t) that spans C(t), with an appropriate choice of outward normal C = dS
S(t)

The invariance of these follows easily from the discussion above. We know that since is a vector and dS is a one-form Dt ( dS) = 0 and thus dC = Dt ( dS) = 0 dt S(t) Note that we had to take a material derivative inside the integral. To appreciate this or visualise it, it is perhaps most natural to break up the surface S(t) into many innitesimal surface elements dS and apply the innitesimal version to each.

Energy and helicity invariants for uid ow

We have conservation of (kinetic) energy EK =


1 2 u 2

dV

with the no normal ow boundary condition, and (kinetic) helicity HK = with the boundary condition that n=0 (r S) u dV
D

This boundary condition states that vortex lines must be parallel to the boundary. In a sense this is an unnatural boundary condition, and it would not be preserved if viscosity were present. However in ideal uid ow it is is preserved 11

if it is the case initially: to understand it we need to know more about the interpretation of the helicity integral. To check the conservation of helicity, we take dHK = dt = =
D

D D

[(t u) + u (t )] dV [(u u P) + u (u )] dV

(u ) dV

with the n = 0 condition used to remove the P integral (recall the orthogonality property discussed earlier). We now use = u, the identity (a b) = b and apply Gauss theorem to give dHK = dt = (u ) u n dS
S

aa

[u2 (u )u] n dS = 0
S

This vanishes provided the natural boundary condition un = 0 holds and the condition that vortex lines do not penetrate the boundary, n = 0. Under these conditions HK is constant, independent of time. It is interesting to look at the evolution of HK (V) where now V(t) is a material volume, carried in the uid ow. In this case we calculate as follows dHK (V) = dt = =
V

V V

[(Dt u) + u (Dt )] dV [ p + u ( )u] dV

(p + 1 u2 ) dV 2 (p + 1 u2 ) n dS = 0 2

=
S

provided that n = 0 on the boundary S(t) of the volume V(t). Suppose that initially the volume V(0) is bounded by vortex lines, lying on the surface S(0). Then, as vortex lines are carried by the uid ow the material surface S(t) will continue to be composed of vortex lines and so the boundary condition n = 0 will continue to hold. Thus for a material volume V(t) bounded by vortex lines HK (V) is conserved. When we discuss the topological interpretation of HK this will become clear. Here we just note that there could be whole families of surfaces, for example nested tori, giving a continuous family of invariants, labelled by the surface. 12

It would be nice to illustrate this with some exact solutions for uid ows satisfying the unsteady Euler equations, but such solutions are few and far between! Instead we will content ourselves with the steady Euler equations and a family of ows (a special case of the family of ABC ows we will meet soon), u = A(0, sin x, cos x) + B(cos y, 0, sin y) There are two parameters A and B (though only the ratio is important as we can rescale u and time t) and the ows may be thought of as dened in a 2 periodic cube, i.e. a 3-torus T3 . This ow has the Beltrami property that = u u, vorticity and velocity are everywhere parallel, and so it is a solution of the steady Euler equation. The ow may also be written as u = (y , x , ), = A cos x + B sin y

and so we see that in the (x, y)-plane the motion is along curves of constant , while the vertical motion, that is in the z-direction, is constant on each curve of constant . In full 3-dimensional space, helical streamlines lie on cylinders of constant and these are also vorticity surfaces. The helicity inside each surface is constant, as it must be trivially for a steady ow! But in a more general context we have to imagine such a ow evolving with time and the helicity inside each surface of constant would be conserved.

10

Ideal MHD and gauges

Before going on to study the topological interpretation of uid helicity, we set up the analogous quantity of magnetic helicity. We consider the equations of magnetohydrodynamics (MHD), which in suitable units take the form of the NavierStokes equation t u = u + j b P +
2

with the addition of the Lorentz force j b and the magnetic induction equation t b = (u b) ( b) Here we take the viscosity and the magnetic diusivity to be constants (otherwise the above forms need modifying). The limit 0 is called the perfectly conducting limit as 1 is proportional to the conductivity of the uid, and so is large in plasmas and liquid metals. The two elds are divergenceless u= b=0

The rst of these is an approximation (no uid is entirely incompressible) while the second is a fundamental law of physics (no magnetic monopoles). 13

The electric current is given by j= b

Note that for divergenceless elds the two forms of the nal term are equivalent as a result of the vector identity ( u) = ( u)
2

One can swap between these as needed. There are a number of ways of rewriting the equations, for example as Dt u = t u + u u=b b PM +
2 2

Dt b = t b + u where

b=b

u+

1 PM = p + 2 b 2 now includes the magnetic pressure.

In the ideal case we set = = 0 and consider the perfectly conducting and no normal ow boundary conditions, namely nu=nb=0 (r S)

for a nite domain D with boundary S. This requirement gives no ow through the boundary and also no magnetic eld lines penetrating the boundary. The latter is natural for theoretical discussion but not for many applications, for example the solar magnetic eld which penetrates the photosphere, into the magnetosphere; see later. Alternatively we may impose that the vorticity and magnetic elds fall o rapidly (faster than any power of radius) in an innite domain, so that (at any time t) for any n, |(r, t)|, |b(r, t)| = o(rn ) We now focus on the ideal case = = 0. t u = u + j b t b = (u b) or Dt u = t u + u u = b b Dt b = t b + u b = b u PM P (r )

14

We note rst that given b = 0 initially, by taking the divergence of the induction equation it is preserved in time, t b=0

This is why there is no pressure needed in the magnetic induction equation. For u pressure is dened to constrain u to remain divergenceless. Also, the induction equation is completely analogous to the vorticity equation, except that the magnetic eld is not directly linked to the ow eld u, whereas for vorticity = u. Of course we did not use that link earlier in our discussion of Lie-dragging and so we still have the Cauchy solution for the magnetic induction equation b(x, t) = J (a, t)b0 (a) or bi (x, t) = Jij (a, t)b0j (a) In fact in kinematic dynamo theory we assume the eld is so weak that the Lorentz force term may be neglected in the NavierStokes equation, and the ow eld u may be specied independently of the magnetic eld. In this case the Cauchy solution can be really useful if the ow is not too complicated. We also have the analogue of Kelvins theorem, namely Alfvns theorem that e the ux S of magnetic eld through any material surface S(t) is conserved, S = b dS
S(t)

We have the usual conservation of energy with EK + EM =


D 1 2 u 2

dV +
D

1 2 b 2

dV = constant

What of helicity? Just as =

u, plainly we need to obtain a eld A with b= A

The existence of such a vector potential follows from b = 0, at least locally. This denition of A can be done globally in innite space or in a bounded domain provided that it is simply connected. However there can be problems in dening A (as a single-valued function) if the domain D is multiply connected. For example if we work in three-dimensional periodic space D = T3 we cannot uncurl a periodic b to obtain a periodic A unless it b has a zero average over D: no mean eld. We will not get side-tracked by these considerations (another way in which topology intrudes into MHD) here, but they may reemerge later on. Instead we focus on a dierent issue, that of gauge freedom: we can always add a gradient to A without aecting b: AA+ 15 (r, t)

We uncurl the ideal induction equation in the form t A = u b + =u( A) +

where the presence of corresponds to the freedom to choose the gauge: whatever we put here will disappear when we take the curl to obtain the induction equation for b. There are a number of choices for : one is to make A = 0, in which case we require this to hold for the initial condition for A and then evolve A with determined by
2

(u b)

This is a non-local equation, rather like that for the pressure in the Euler equation.

11

Magnetic helicity

Given A with some choice of gauge we can dene the magnetic helicity over the domain D by A b dV HM =
D

We then have t HM = [(u b +


D

) b + A

(u b)] dV

Now the integral of ( ) b vanishes as usual with n b = 0 on the boundary and so the integral is reassuringly independent of the choice of gauge. Also u b b = 0, removing one term above, and with this we have ((u b) A) = A from which after applying Gauss theorem t HM = n (u b) A dS =
S S

(u b)

n [(u A)b (b A)u] dS = 0

by virtue of the boundary conditions: n b = 0 so no eld lines leave through the boundary, and n u = 0 so there is no ow to transport helicity density b A out of the domain. For ideal MHD, the kinetic helicity HK is no longer conserved but a quantity called the cross-helicity is: HX = u b dV
D

What of our magnetic vector potential A: is it a vector or a one-form? As it stands it is neither, in the sense that it is generally not transported locally: 16

for example if one imposes the gauge A = 0 then one has to solve a global, Poisson equation for the gauge . However we can write t A = u ( A) + = u A + ( A) u +

With the choice = u A we have = ( u) A ( A) u and so Dt A = t A + u A = ( u) A which is the equation for the Lie-dragging of a one-form. A is a one-form eld with this particular choice of gauge and then we not only have conservation of the integral of helicity but also of its value, the helicity density hM = A b advected in the uid ow Dt hM = t hM + u hM = 0

Although this is attractive, having a materially conserved quantity, it does depend on following the eld A in this gauge, and as it evolves it may become very ne-scaled. Typically people would rather follow the magnetic eld b and then compute A in a suitable gauge such as A = 0: in this case we can only use the integrated form HM of the helicity as a conserved quantity. As in the case of kinetic helicity we can establish that HM is conserved if taken over a material volume V(t) contained in the domain D, HM (V) = We can use the form Dt b = b u, Dt A = ( u) A + A b dV
V(t)

for an arbitrary gauge and obtain (with Dt dV = 0) dHM (V) = dt [b (Dt A) + (Dt b) A] dV = b
V(t)

dV = 0

V(t)

using Gauss theorem with n b = 0 on the boundary of V. As for vorticity, if the surface S volume is composed of magnetic eld lines, then the helicity within is conserved. Historical note: the conservation of magnetic helicity was noted by Woltjer in 1958: the attractive name of helicity was chosen because HM or HK change sign under reection, being pseudo-scalars, by Moatt in 1969. For example consider the ow u = (y, x, x2 + y 2 ) = (0, 0, 2) 17 u = 2(x2 + y 2 )

and its mirror image under inversion r r u = (y, x, x2 y 2 ) = (0, 0, 2) u = 2(x2 + y 2 )

12

Topological interpretation of helicity as linking number

We now consider the interpretation of helicity. We can work with either the uid helicity HK = u dV
D

or the magnetic helicity HM = A b dV


D

in the whole domain D where the ow or eld is dened. We shall choose to work with the magnetic helicity. Consider two thin linked tubes Tj in D, of magnetic eld as depicted, with centre lines Cj for j = 1, 2. Here b = 0 outside each one, and each carries a ux j > 0: the direction of traversal along Cj is in the direction of the eld. The ux is the integral j = b dS
Aj

where Aj is a surface spanning Tj with normal in the direction of the eld, i.e. a cross sectional area of the tube. We also assume that, individually, each tube is unknotted. Focussing on the fact that b is non-zero only in the tubes, we have two contributions to the helicity integral, HM = H1 + H2 , where Hj = A b dV
Tj

We can approximate these as integrals along the centrelines Cj of the tubes, replacing b dV with j dr, as Hj = j A dr
Cj

We may now use Stokes theorem: the integral of A around Cj is the integral of the eld b = A through a surface Sj spanning Cj Hj = j b dS
Sj

and so is plus-or-minus the ux of the other tube, depending on orientation. For the picture shown and j = 1 we have a plus sign and so H1 = 1 2 . Likewise H2 = 1 2 and the total helicity is HM = 21 2 . If the tubes are 18

not linked the value is zero, and if one set of arrows is reversed (or the whole system is reected) then HM changes sign. In general we have the result for a whole collection of tubes (each individually unknotted) labelled by j that HM =
i j

i j ij

where ij are integers that label the signed number of times that the centre line Cj of the jth ux tube crosses the surface Si spanning the centre line Ci of the ith ux tube. Here the sign is +1 when the direction of Cj agrees with the outward normal of Si , obtained from a right-hand rule from the direction of the centre line Ci Note that it is important that the ux tubes individually are not knotted (in knot theory these are called unknots) and we set ij = 0 for i = j. If a tube is knotted then we may break it up into a collection of unknotted tubes (see pictures) by inserting links and then compute the helicity this way. This denes the value of ij for i = j in this more complicated case. A number of notes are in order. First, the topological nature of the invariant, in terms of the linkage of magnetic eld lines or of vortex lines, explains why it is invariant under ideal, frozen eld evolution: linkages cannot be broken if magnetic eld or vortex lines move materially with the uid. It also explains why the condition that the eld, or vortex, lines must satisfy n = b n = 0, either on the surface of the domain D for the total helicity, or on the surface of a moving material volume V(t). We cannot dene linkage (easily) if the eld or vortex lines do not close up. Note, though, that some important applications involve eld lines that do intersect surfaces, for example in the solar corona, and ways of measuring linkage have been adapted to this: see later. There is a connection between helicity and an invariant known as the Gauss linking number. Consider a magnetic eld b in unbounded space that occupies tubes Tj with centrelines Cj and has uxes j . The corresponding A (in the gauge A = 0) may be obtained at any time by a BiotSavart integral (as can the velocity from the vorticity) namely, A(r) = 1 4 (r r ) b(r ) dV |r r |3

If we replace b(r ) dV in the jth tube by j dr then we obtain A(r) =


j

j 4

Cj

(r r ) dr |r r |3

We now plug this into H=


i

A dr
Ci

19

from above to obtain H=


i j

i j ij

with ij = 1 4
Ci Cj

(r r ) dr dr |r r |3

This integral gives an integer, which is the Gauss linking number of curve i with curve j for i = j. Note that if the linking number ij between any two curves is non-zero then they are linked, and cannot be smoothly separated. However one can obtain congurations such as the Borromean rings, which are 3 linked rings: they cannot be pulled apart but have the property 12 = 23 = 31 = 0. Furthermore we have assumed there is no linkage associated with the eld lines inside a given ux tube: this is not always correct and leads on to how one measures the twist of eld in a tube (which may itself be in a complicated knot) and the related topic of writhe. This will be taken up later in the course. A nal issue here is that we have focussed on eld that is concentrated in discrete ux tubes (or vorticity in discrete vortex tubes). But what of eld in a continuous distribution, for example take the vorticity eld = u for members of the ABC family above. In this case very few eld lines actually close. The work above has been generalised by Arnold to cover this case, and he refers to this as the asymptotic Hopf invariant. We will not reproduce the proof here, but the idea is to take magnetic eld lines starting at points a and integrate trajectories x(a, t) along them for a time T , say. One then uses a set of paths to link the end points x(a, T ) to the initial points a: these paths are to be short and straight. This then gives a collection of closed curves whose Gauss linking numbers may be computed and integrated. An appropriate limiting procedure taking T yields the helicity or Arnolds asymptotic Hopf invariant as the limit of the Gauss linking number, for a continuous magnetic eld.

13

Beltrami ows and the ABC ows

We now return to uid ows: our aim is to look at the structure of some steady Euler ows and how this is constrained. Consider an ideal uid in a nite region D with helicity HK = 0. What is the minimum value of the enstrophy K = 2 dV
D

20

for this given value of helicity? By the CauchySchwartz inequality we have


2 HK = 2

u dV
D

u2 dV
D

2 dV

= 2EK K

We also have a Poincar inequality that for a given nite region D there is a e constant L such that K = 2 dV L2
D D

u2 dV = 2L2 EK

for any u with n u = 0 on the boundary. The number L1 is the smallest value such that = u = L1 u has a non-trivial solution in D subject to the no normal ow boundary condition u n = 0. From these two inequalities we have
2 HK 2EK K L2 2 K

and so can write a lower bound for the enstrophy K L1 |HK | in terms of the helicity. Plainly equality occurs when = u everywhere in D, where = L1 is a constant. Such ows with the vorticity and velocity everywhere parallel are called Beltrami ows and satisfy the steady Euler equation since u = 0 identically. Here the ow and vorticity are proportional by a constant, namely . The condition u = u gives an eigenvalue problem, and the least eigenvalue gives L1 . We have already met one such solution which is a special case of the ABC (or ArnoldBeltramiChildress) family of ows, taking the form u = A(0, sin x, cos x) + B(cos y, 0, sin y) + C(sin z, cos z, 0) (in periodic space T3 with no mean ow). If one of A, B or C is zero, we have a ow with regular streamsurfaces, seen earlier. In a certain plane there will be a repeated pattern of cats eyes. However if all three are non-zero, then typically we see a complex mixture of regular vortices, with stream surfaces that take the form of tori, interlaced with regions of chaos. To make life dicult the ow is highly three-dimensional. It has been studied by means of Poincar e sections, in which a trajectory x(a, t) is follows and points plotted whenever this intersects one of a family of planes, for example a plane of constant x. 21

Amazingly, this is a steady Euler ow, and raises the question: just how complicated can steady Euler ows be? Are there any constraints on their structure? An answer to this question (and pretty much all that is known) is provided in a rigorous result of Arnold, which we set out informally as follows. We worm in a nite domain D which could be a sphere, or could be periodic space T3 . A steady Euler ow obeys u = P and so u P = P =0 Now, provided P does not vanish identically in a 3-dimensional region of space, there will exist a family of 2-dimensional surfaces S(P0 ) on which P = P0 is constant. The above result says that streamlines and vortex lines will be tangent to such surfaces. To understand what form the surfaces can take we need to be careful about points where P = 0. We assume there are only nitely many such points in the domain D under consideration. We then choose a value P = P0 = constant, to give a surface S(P0 ) which does not intersect any points with P = 0 (in other words we avoid the nite number of exceptional values of P0 where this occurs). We also assume the surface does not intersect the boundary of D. The vorticity and velocity vectors are tangent to this surface S(P0 ) and can vanish nowhere on it. It can be shown that the only such surface is topologically a torus, and that the velocity and vorticity elds wind around it. As we vary the value of P0 we obtain nested tori S(P0 ), provided we do not encounter a point where P = 0: at such a point surfaces may intersect. Note that the reason the surfaces of constant P must be tori is linked to the non-vanishing of the ow eld u (or ) on the surfaces. For example the surfaces cannot be spheres as it is impossible to dene a non-vanishing vector eld: this is the famous result that one cannot comb a hairy ball, or that at any time the horizontal wind velocity must vanish at at least one location on the earths spherical surface. (For more information, in particular the link to the Euler characteristic of a surface, see the book of Arnold and Khesin [1].) This analysis assumes that P varies in space and P = 0 at only nitely many points, so that nested two-dimensional surfaces P = P0 = constant may be dened. But what if P is constant and so P = 0 over a whole 3-dimensional volume V, a subdomain of D? In such a case we must have that u = 0 everywhere in V, so that = u for some function (r): the ow must be Beltrami. But given that u = = 0, we can take the divergence of or u and rapidly show that u = =0

Now the argument repeats itself: if (r) denes two-dimensional surfaces S() and with = 0 assumed only at a nite number of points (which we avoid), 22

then the velocity and vorticity vectors must be tangent to the surfaces and non-vanishing. The surfaces are again tori. The nal possibility then is that (r) is in fact a constant, say (r) = , over a whole 3-dimensional region V and in this case we have

= u

there, which is the eigenvalue problem giving us the ABC family of ows we had earlier in the particular case when V is all of periodic space. In other domains there will be other families of ows: for example there are analogues of ABC ows in spherical geometry. In conclusion, ow lines and vorticity lines lie on nested tori except in the special case of constant Bernoulli function P = p + 1 u2 and = u with 2 = constant. The ABC ows provide an example of this: for example the values A = B = C = 1 show a complex topology of uid trajectories. These include invariant surfaces on which some stream lines sit, but interleaved with these are thin bands of chaotic trajectories. There is no mathematical reason why there should be any surfaces at all, and in fact the amount of chaos in the ow varies with the parameters. At another extreme, if C = 0 the ow shows a cats eye structure as mentioned before. Here there are two distinct families of tori on which streamlines (or vortex lines sit). The reader may wonder why we cannot have regions where there are twodimensional surfaces P = constant interleaved with three-dimensional regions in which P is identically constant. Or one could have a sequence of points where P = 0 that accumulate at some point in space. There are a lot of possibilities in three dimensions! In fact the theorem of Arnold excludes this by requiring that the eld u be complex analytic: this means that critical points P = 0 (or = 0) cannot accumulate, and that if P (or ) is constant in any three-dimensional volume then it is constant everywhere. Thus either there are P = P0 surfaces with P = 0 at only nitely many points, or P is constant with P = 0 everywhere in D: there is no messy mixture allowed. In the latter case either = 0 surfaces exist with = 0 at nitely many points, or = 0 everywhere in D. The assumption of analyticity gives clear and clean alternatives at each point in the argument. More general situations, in which the elds are only required to be innitely dierentiable, or arise from magnetic relaxation (see below) do not appear to have been studied, and are likely to be complicated. 23

14

Magnetic relaxation

Consider the following problem: we consider evolution of magnetic eld with zero magnetic diusion = 0, under the full MHD equations including nonzero viscosity > 0. These are t u = u + j b P +
2

t b = (u b) u = b = 0, j = b or equivalent. We work in a domain D which is nite with u = 0, n b = 0 on the boundary S. We have kinetic and magnetic energies EK = Then it may be shown that d (EK + EM ) = dt | u|2 dV
D D 1 2 u 2

dV,

EM =

1 2 b 2

dV

and so the total energy E = EK + EM 0 of the system decreases monotonically as t and must tend to a limit, positive or zero. Recall that | u|2 = (i uj )(i uj ) 0. Can the limit be zero? Suppose we start with a given magnetic eld b0 and zero ow u = 0: then the Lorentz force j b will tend to drive a ow (eld lines are elastic and tend to contract, reducing magnetic energy). But the eld lines are frozen in the ow and so any linkages and any helicity integrals (over the whole domain or over subvolumes) must be conserved: this constrains how small the magnetic energy may be. In fact we have from our discussion of Beltrami ows that K = 2 dV L1
D D

u dV = L1 |HK |

where L is the constant appearing in the Poincar inequality. In magnetic e terms this tells us that 2EM = B 2 dV L1
D D

A b dV = L1 |HM |

and so the helicity, if non-zero initially, gives a lower bound on the magnetic energy. (In fact a little care must be taken with the denition of L here as it involves boundary conditions on A: see Moatt [9] for more information.) The conclusion is that non-zero helicity must lead to a non-trivial nal state of this relaxation process, and in fact it has been shown that any linkage of eld lines (for example Borromean rings) must do the same (by Freedman [5]). 24

In the limit t we expect the kinetic energy to go to zero and leave us with a magnetostatic equilibrium satisfying jb= P, j= b

But this is analogous to the steady Euler equation 0 = uE E under the replacement j E , b uE , P PE PE , E = uE

and so we nd a steady Euler ow from the relaxation process and then applying this analogy. We use an E subscript to distinguish the analogous ow from the original one in the MHD system To recap: we choose a magnetic eld with whatever magnetic eld line, that is choose b-line topology, we like but with some non-trivial linkage. This could be total helicity, or the helicity restricted to the eld inside magnetic surfaces, perhaps a whole family or magnetic surfaces. Or any helicities may be zero and there could be higher order linkages analogous to the Borromean rings. We then do the relaxation process and let t , in which case the ow u 0 and so we are left with the relaxed eld with the same topology as initially and non-zero magnetic energy. This corresponds to a steady Euler ow uE in which the stream lines (not the vortex lines) have the same topology as the eld lines b specied initially. Moatt [9; 11; 12] gives appealing pictures of relaxing magnetic elds and corresponding Euler ows. For example we can start by a set of nested surfaces S() in which the helicity will be conserved, being a parameter labelling the surface. If we then relax this magnetic eld we obtain a corresponding Euler ow, and the helicities will be preserved: this helicity function gives the signature of the vortex. Nonetheless, despite this appealing picture, there are many unknowns (and probably unknown unknowns, particularly for general three-dimensional elds). Does u tend to zero pointwise? Can singularities occur in the elds as they relax in the limit t ? Or rather, how bad can singularities be? How can a sensible limit then be dened? For example we have the situation described by Bajer (see [12]) where an initial X-point topology leads to current sheet formation in 2-dimensional relaxation. Here discontinuities in the b eld emerge in the innite time limit, corresponding to a singular current distribution. Suppose a sensible limit can be taken. Then how does the resulting Euler ow uE t into Arnolds classication: presumably the requirement of analyticity 25

will no longer hold and will allow all sorts of complex mixtures of surfaces and chaotic regions, with discontinuities and maybe devils staircases for functions such as P , as suggested by Bajer [2]. Also these questions are very hard to investigate numerically in the dicult limit of zero magnetic diusion and t .

15

A variational formulation

To investigate a little more, we temporarily forget about magnetic relaxation and consider any velocity eld u and vorticity eld = u. This has kinetic energy 1 2 u dV EK = 2
D

Suppose some ctitious, divergenceless, innitesimal ow eld acts and moves the vortex lines, so we have a new vorticity eld + and velocity eld u u + u given by, at leading order = ( )

u = + Here the scalar eld is xed to make the eld u divergenceless. This is called a frozen eld or isovortical displacement. The corresponding change in energy is EK = u u dV = u ( + ) dV
D D

With u n = 0 in the domain where we are working, we obtain EK = ( u) dV


D

Now if u is the gradient of some scalar eld then we obtain zero by the divergence theorem: = 0 means that is orthogonal to gradients provided n = 0 on the boundary of the domain. Conversely it can be shown that if EK vanishes for all divergenceless then u must be the gradient of a scalar eld, say u = P This is Eulers equation for steady ow and we obtain the variation result: steady Euler ows are stationary points of EK with respect to general isovortical displacements. Further study looks at the second variation of the energy and if it is a maximum or a minimum then stability can be shown [10]. This has led to powerful stability theorems in two dimensions [1]. In terms of relaxation, we move the magnetic eld lines b with the ow u so as to minimise the magnetic energy EM , so in particular we are making it 26

stationary and tending to a stable magnetic equilibrium. When we identify b with uE we have a stationary point of the corresponding kinetic energy EK of uE . But this need not be a minimum, and in general there is no reason why the uid ow uE should be stable.

27

16

Exercises

These vary in diculty: if you cant do one, then do move on to the next. Most involve vector calculus identities and the use of Gauss theorem. Questions 6, 8, 9 are very easy, and question 5 is potentially quite tricky. Q 1. Show that in the presence of viscosity , for a ow (no magnetic eld) in a bounded domain D with u = 0 on the boundary S, dEK = dt where | u|2 = (i uj )(i uj ). Q 2. Show that for a ow in innite space with a localised vorticity distribution dHK = 2 dt
D

2 dV =
D D

| u|2 dV

dV

Q 3. A divergence free eld takes the form u = + where , and are smooth functions such u = 0. Show that if these functions are localised and so decay suciently rapidly at innity, then the ow eld has zero total helicity HK . Q 4. Prove the Jacobi identity of Lie brackets for 3 general vector elds, [[u, v], w] + [[v, w], u] + [[w, u], v] = 0

Q 5. Show that given arbitrary vector elds b, c and d transported in the usual way, namely Dt b = b u, Dt c = c u, Dt d = d u

then for incompressible ow

u = 0 we have Dt (b c d) = 0

Hence show that b c is transported as a one-form, namely Dt (b c) = ( u) b c 28

[Hint for rst part: after a short time t, the vectors b, c and d are moved to b (I + t u), c (I + t u) and d (I + t u) (I being the identity matrix). The change in the volume is given by the determinant of (I +t u) which can be linked to the trace of u through the identity det M = exp(tr log M ).] Q 6. Show that for two dimensional ows u = (u(x, y, t), v(x, y, t), 0) with vorticity = (0, 0, ) the vorticity equation may be written in the form Dt = t + u = t J(, ) = 0

where u is written in terms of a stream function u = (y , x , 0) and J(a, b) = (x a)(y b) (y a)(x b) is a Jacobian. From t + u = 0 show that the enstrophy K = 2 dV
D

is conserved for ideal ow and a localised vorticity distribution in the innite plane. Show also that the moments of the vorticity distribution Kn = are also conserved. Q 7. Show that for ideal MHD the total energy EK + EM =
D 1 2 u 2

n dV
D

dV +
D

1 2 b 2

dV = constant

in a nite domain with no normal ow, perfectly conducting boundary conditions nu=nb=0 Show that the cross helicity HX = u b dV
D

is conserved in time with n u = n b = 0 on the boundary, but that the kinetic helicity obeys dHK = 2 j b dV dt D provided n j = 0 on the boundary. Interpret the cross helicity in terms of linkage by considering magnetic eld b conned to tubes Tj . Q 8. Show that ideal MHD may be written in terms of Elsasser variables + = u + b, 29 = u b

as t + + + = PM t + + = PM + = = 0 Hence show that any solution of the form u = b or u = b satises the ideal MHD equations. Q 9. Consider magnetic relaxation under the following equation for the uid ow in a porous medium, with constant > 0 t u = u + j b P u

Show that EM + EK decreases in time, for u non-zero.

30

References [1] V.I. Arnold & B. Khesin 1998 Topological methods in hydrodynamics. Springer. [2] K. Bajer 2005 Abundant singularities. Fluid Dyn. Res. 36, 301317. [3] S. Childress 2004 Topological uid dynamics for uid dynamicists. http://www.math.nyu.edu/faculty/childres/tfd.pdf [4] T. Dombre et al. 1986 Chaotic streamlines in the ABC ows J. Fluid Mech. 167, 353391. [5] M.H. Freedman 1988 A note on topology and magnetic energy in incompressible perfectly conducting uids. J. Fluid Mech. 194, 549551. [6] A.D. Gilbert 2003 Dynamo theory. In Handbook of mathematical uid dynamics (ed. S.J. Friedlander, J.P. Serre), volume 2, pp. 355441. North Holland. [7] H.K. Moatt 1969 The degree of knottedness of tangled vortex lines. J. Fluid Mech. 35, 117129. [8] H.K. Moatt 1978 Magnetic eld generation in electrically conducting uids. Cambridge University Press. [9] H.K. Moatt 1985 Magnetostatic equilibria and analogous Euler ows of arbitrary complex topology. Part 1. Fundamentals. J. Fluid Mech. 159, 359378. [10] H.K. Moatt 1986a Magnetostatic equilibria and analogous Euler ows of arbitrary complex topology. Part 2. Stability considerations. J. Fluid Mech. 166, 359378. [11] H.K. Moatt 1986b On the existence of localized rotational disturbances which propagate without change of structure in an inviscid uid. J. Fluid Mech. 173, 289302. [12] H.K. Moatt 1990 Structure and stability of solutions of the Euler equations: a lagrangian approach. Phil. Trans. R. Soc. Lond. A 333, 321342. [13] H.K. Moatt & A. Tsinober (eds.) 1989 Topological uid mechanics. Cambridge University Press. [14] H.K. Moatt et al. (eds.) 1992 Topological aspects of the dynamics of luids and plasmas. Kluwer. [15] R.L. Ricca (ed.) 2001 An introduction to the geometry and topology of uid ows. Kluwer. [16] B. Schutz 1980 Geometrical methods of mathematical physics. Cambridge University Press.

31

Potrebbero piacerti anche