Vous êtes sur la page 1sur 8

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

Use of COMSOL Multiphysics in the Modeling of Ion Source Extraction


Marco Cavenago Laboratori Nazionali di Legnaro, Istituto Nazionale di Fisica Nucleare (INFN-LNL) viale dellUniversita n. 2, I-35020, Legnaro (PD) Italy, cavenago@lnl.infn.it
Abstract: A two dimensional description of plasma charged sheath and ion extraction, including the complete plasma sheath equation, is proposed and reduced to a system of coupled fields, the electric potential u, a front current jf, and moving boundaries, which are implemented with use of the COMSOL ALE (arbitrary Lagrangian Eulerian) mode and weak form. Ray tracing and crossing is also included. A separate application of weak form to the Hamilton-Jacobi and Poisson equation is also described, using five coupled fields. Keywords: ion sources, electromagnetism, accelerators, ray crossing, integral coupling The front plasma includes all the structures called presheath, sheath, and transition layers in literature on 1D models [4,5,6]. Result for V and ion current j are discussed, as a function of the extraction voltage VE and of the axial magnetic field B0; with increasing VE, the volume of plasma contributing to the extracted current is seen to increase. With decreasing B0 the ion trajectories deviation from parallel direction increases. Here a special script routine capable of treating ray crossing is also described. Most ion sources have a cylindrical symmetry so that (r,z) coordinates are appropriate, while some have a planar geometry (x,y). For the sake of simplicity, we here consider planar geometry. Ideas to use the Hamilton-Jacobi equation wave fronts in place of ray tracing and their relation to the weak form of PDE implementation in Comsol3.2/3 are discussed in a separate section. Work to include realistic negative ion formation rate is well in progress. In that case, note the complexity of having two extracted species, namely the electrons and the negative ions; moreover the sheath net charge changes sign near the extraction electrode edge.

1. Introduction
The extraction of a beam from a plasma is a complicate problem of non-linear physics, involving several coupled variables, as the ion density ni, the electron density ne, the selfconsistent electrostatic field V, and the particle trajectories xj (traditionally called rays) [1,2]. It is of great practical importance, since the design of high current particle accelerators critically depends on the beam parameters at extraction. Existing codes are usually based on an empirical simplification of the underlying physics, which makes extrapolation to multimegawatt beams difficult. Here we discuss a family of application scripts for Comsol3.2/3 where accurate expressions for total space charge na and for ion speed vi are used. The main application presented here is a simple positive ion source (see Figure 1), characterized by sizes comparable to the ionization length L (30 mm) and by a tenuous plasma (Debye length lD from 0.1 to 1 mm) [3,4]. We divide the plasma in two regions: the core, where the ion speed is small compared to a typical speed cs (Bohm speed) and can be treated analytically; and the front plasma, where vi sharply increase and a numerical two dimensional (2D) simulation is needed; we define their boundary by vi/cs = 2-1/2. Since position of this boundary is unknown, the ALE (arbitrary Lagrangian-Eulerian) mode is used. The same mode represents the boundary between vacuum and the ion beam.

2. The charge sheath


In figure 1 note the electrode (with V=0 by convention) which contains the plasma, and the hole or the slit (of aperture 2rh) where the ion beam exits. Since plasma electrons have a thermal energy (with typical temperature Te =5

Figure 1. A scheme of an ion source extraction showing also half of the source plasma

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

eV in energy units) and this usually produces a small maximum Vp of electric potential at the plasma center zp, it is convenient to use the adimensional potential u= - e V/Te, with q = ie where q and e are the ion and the proton charge, respectively. For positive ions the applied voltage at the extraction electrode E voltage is VE<0, with a modulus large compared to Te/e, so that u E = eVE / Te >> 1 . The electron density is well approximated by a Maxwellian distribution [5] ne=ne0 exp(up u) up=-e Vp/Te (1)

number much smaller than one; ua=1/4 in the following. Since for u ( z i ) u p < u a a series solution for u(zi; ) is known, the first contribution can be evaluated. The resulting expression [7] of N1 as a local function of u(z) of the form
N1 = N10 (u ) + ( 2 / ) N12 (u ) + O( 4 )

(5)

can be calculated by a subdomain expression in Comsol3 (see figure 2). The second integral can be approximated by N 2 = N f + N 2H + N 2L , (6)
Nf = jf u u p uc

where ne0 is the density at plasma center and up<0 is a scalar unknown, to be determined. The Poisson equation is written for u in the form
n n n e2 u = e i with = e 0 n Te 0 e 0 ne 0

and j f dzi
za

g ( zi ) L

(2)

where uc+ up is a sort of central or average value between u(za) and u, that is
uc = u a , b + u up
u 0.592,

(7)

where the Debye length lD=l-1/2 is the scale of sharp spatial variations of u. Ion density is here computed assuming that ions are generated at rest (by electron ionization of gas) at a position zi with a rate G=nione0g, and that they approximately move along a direction z following the magnetic field lines; here nio ne0 is the generation rate at plasma center and g(u)=exp(g(up-u)) is a form factor with g=1. Moreover u is increasing with z, so ions are accelerated by u and friction from collision is neglected (free fall ion model). Remembering that density=current/speed, the normalized ion density Ni=ni/ne0 is
Ni ( z) = dz i g (u ( z i )) | B( z ) | | B( z ) | u ( z ) u ( z ) L zp i i
z

with a = 2(u u a ) 2 and b = 2u 3u a . Here N2H and N2L are small corrections, due integration range ends, and can be estimated by local power expansions of u(zi). For example 1 3 H H N 2H N 21 (u ) (u , z L ) N 22 (u ) u , zz (u , z ) L1 (8) and the function NH21(u), NH22(u) were separately computed and loaded into fem.functions as interpolation functions, while a rough estimate is used for u,zz .Thus the normalized ion density is
N u N10 (u ) + ( 2 / ) N12 (u ) + N 2H + N 2L

Ni=Nf+Nu

(9)

(3)

where L= 21/2cs /nio has the physical meaning of the length that is needed to produce an ion density comparable to electron density and cs=(i Te/mi)1/2 is the Bohm speed. A uniform applied magnetic field B = B0 x is assumed in the following. The parameter compares this length to Debye length = 2 D / L . Let us show that Ni(z) is a mainly a function of u(z), while the complicated functional dependence on u(zi) can be approximately integrated and simplified. Let us divide Ni into two terms
N1 = 1 za dz i g (u ( z i )) 1 z dz i g (u ( zi )) u( z) u( z ) , N 2 = L u( z) u( z ) L zp za i i

where Nu is a simple local function of u. Finally, when ion motion is not completely magnetized, let y=yf(x,ys) be the trajectory (or

(4) where za satisfies u(za)-up=ua with ua a fixed

Figure 2. Plot of N10(u) vs asin([ua/(u-up)]1/2) for ua=0.25 and comparison with NH12, NH22 and uc.

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

its estimate) starting from the point (xs,ys) and passing from the point (x,y). For comparison, previously used empirical approximations are
N i = ki / u u p

(9)

3. Plasma boundary model


The region u(zi)-up<ua=1/4 will be excluded from the simulation domain, which becomes z z a , and u(za) =ua+up is one natural boundary condition. Moreover, u,z should be continuous in the plasma, so that it should match the analytical solution used in z z a : u,z(za)=u,z(za; ) (10)

Figure 3. In black, the model geometry in the reference plane (X,Y). Domain 1 is excluded from u and jf support and x is a symmetry axis. The extension u(x,-y)=u(x,y) is used in ray tracing. Note the boundaries AMD and GEH and line DNT. Some moving boundaries (x,y) are visible in red.

which is a second boundary condition. This implies a moving boundary, or in other words, two boundary conditions imply two fields, u and dz. In the application script, the spatial coordinates are called (x,y) instead of z, as summarized in table 1. Even if ALE will allow arbitrary deformation of boundary xa(Y), it seemed convenient, at least as beginning option, to restrain its deformation as xa(Y)-X= dx = a0 + a1 exp[-(Y/ac)2] (11)

b( x, y ) = y (dy / dx) ys

(13)

expresses the fact that, if trajectory converges, current density must increase to keep total current constant. A boundary value on line GEH can not be assigned for jf , since all the arriving current is acceptable. Indeed eq. 12 is parabolic, and if we neglect the diffusion term, since typically 1/Ld=0.001 mm-1, it is hyperbolic. With the weak density
W = g jf f T ( j, fy ) + b j f a j, fy j, x T ( j f ) L La Ld j, fy

where a0, a1 and ac are in millimeters. On the other side, the extractor is simply represented by a static flat plate GEH (figure 3), with the boundary condition u(xE,y)=uE . On symmetry axis x, we have dy=0 and u,y=0.

(14) where T is the test operator, the field momentum G is G=(0,G y)


y = j, fy dW = f dT ( j, y ) Ld

4. Use of a weak formulation mode for j

(15)

Note that jf is an adimensional quantity, equal the ion current density produced in the front plasma divided by j0=21/2 e ne0 cs. The definition of jf can be easily converted to a PDE as jf,z=g(u)/L. Since the front ions move slowly than core ions, they are more sensible to scattering, neutralization or adsorption, and deflection; these effects are here added to their governing equation: j ,x + a(x,y) j ,y = - b(x,y) j + S S=(g/L) (jf/La)+ (jf,yy/Ld) with a = (dy / dx) ys . Here S includes the source or sink terms, La is the neutralization length and Ld is the transverse diffusion length. Moreover
f f f

so that - n = 0 is an identity on GEH, where ny=0. Moreover on axis - n = 0 means j, fy = 0 that is the desired boundary condition.
Table 1: Summary of modes (cells in fem.appl )

Cell index Type Independent variables Field Variables Active domains

1 ALE X,Y x,y 1:5

2 General x,y

3 Weak x,y

(12)

u
2:5

jf
2:4

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

Only the boundary AMD needs the constraint j f = 0 to be set.

5. Iteration and particle tracing


As a first iteration step (labeled j=1), or when the magnetic field B0 is very strong, we can assume that yf=ys. In particular, the beam boundary is constant y=rh. Then table 1 model can be solved for x, y, u and jf. To avoid global couplings; parameter a0, a1 and up are adjusted in an internal iteration loop for i=1:Mi in a script, here named planar2init for reference. Equation for up is the condition that boundary integral of electron and ion current is zero:
Le I inj imi exp(u p ) + = Li g o + 4me j0
DT

where l = cs t and kA= (i e B0 )2/(miTe). Let ym be the maximum of the ys for which the trajectory yf(x,ys) arrives at line GEH; so that for all ys ym , the ion hits the V=0 electrode. This ym is found by iteration of ray tracing and ye(x)= max YD |yf(x,ys)| , YD = {y | 0 y s y m }

1 DT
f

(18) defines the beam envelope ye(x) on the set X D = {x | xa x xh } (figure 5). For postprocessing purposes, we have defined domain 1 as the x-axis reflected image of X D YD and define ys=-Y there; so that yf=yf(x,-Y). We want also to know the ion density in the ion beam, considering that ray crossing at (x,y), that is y= yf(x, ys1)=yf(x,ys2) and y 1 y s2 , (19) s is possible. The ion core density Nuc, where c means corrected to include ray convergence and crossing, is the sum of Nu(u(x,y)) |dys/dyf| on all rays (1 or 3 typically) arriving at (x,y). It is elegant, even if computationally expensive, to define the projection variable Q from subdomain 1 to its boundary XD and note
Q( x; y p ) = dy s ( y p | y f ( x, y s ) |) N u ( x, y f ( x, y s )) N uc ( x, y p ) = Q( x, y p ) / y p

(16)

where Le, Li are the lengths of the boundary regions where electrons (respectively ions) can impinge, g 0 0.270205 + 1 2 represents the 4 core ion current contribution, and Iinj is the externally injected ion current, if any. When we have a strong magnetic file, electrons motion along y is completely frozen and ion motion is almost frozen, so Le = NT and Li DT . Adjustments for a0, a1 are computed so that eq. 10 is approximately satisfied as shown in figure 4. Usually Mi=4 internal loop iterations suffice. After saving the obtained solution (and some intermediate result) into a file, the script planar3plot computes the trajectories of ions which come from plasma core, so that xs=xa and with several ys. Motion equations are
2x = u, x l 2

(20)

Ni=Nf+Nu

2 y = u, y + k A ( y s y ) l 2

(17)

where yp is a parameter and is the Heaviside function. Data for ye, yf, and Nuc can be stored in a file raydat2.ws. A comparison of Nu and Nuc is shown in figure 6. Then, at the j=2 iteration step, planar2init loads this file and: 1) it moves the boundaries MN and BE to the line y=ye(X); 2) for y<ye, that is in domains 2 and 4, it uses Nuc instead of Nu in ion density; 3) it computes values of a

Figure 4. The scaled electric field ux (in plasma units) at plasma core-front boundary: theoretical value (triangles), solution value ux (circles) for a self consistent boundary displacement dxale (in mm; squares); for graphical reason dx1=dxale - 0.1 is plotted. Here B0=1.2 T and uE=1.4 uch.

Figure 5. Ray tracing for B0=0.4 T, uE=1.4 uch and a=0.08. The green rays are produced iteratively to search for the beam envelope, while the other rays regularly spaced, provides helps interpolation of yf . Equipotentials of u shown in the plasma area

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

Figure 6. Result for the normalized total charge, with several models (B0=0.3 T). Eq. 9 should be compared with empirical model 9, with gt=0.371 or with the Bohm value kB=(2e)-1/2=0.429

and b for the jf equation 12, approximately, using ye(x) and the induced mesh formation. This completes the model physics. A new solution for x, y, u, jf and up, a0, a1 is obtained. Then planar3plot can compute new ray tracing data. Usually the j=3 iteration suffices for a full solution, since the solution for u was fairly good also at step j=1.

Figure 7. Ray tracing for B0=1.2 T.

6. Results
Typical results for beam trajectories and space charge are shown in figures 7, 8 and 9 for several B0= 1.2 T. In simulation we keep L=30 mm, comparable to the source size, and use a rather modest plasma density, so that lD is not smaller than the typical mesh size 1 mm. The ion current density is jEj0 with j E = ( g 0 y m + ND dy j f ) / rh shown in figure 10. We compare it with one dimensional analytical results for Te 0 and B0 , that are: first, the whole plasma emits an ion current density gt j0 where g t 0.371 weakly depends on ; second (Child law), a parallel beam can carry a current density
j= 4 0 9
3 2ie VE3 / 2 4 j0 u E / 2 = 2 mi d 9 d 2

Figure 8. The full space charge na=(ni-ne)/ne0 for B0=1.2 T, uE=1.4 uchild and a=0.08.

(21)

These limitations gives an expected current density jTj0 with jT=gtmin(1,(uE/uch)3/2) plotted in figure 10 where uch=(9gtld2/4)2/3 is the Child voltage. So, the 2D results show a smoother transition and a more stable behaviour than 1D ones. In figure 6 we also compare eq. 9 with ki=gt to our equation 9. As evident in figure 5, the extractor edge perturbs ion orbit, giving some beam envelope oscillation. Moreover, a higher magnetic field does not suppress this oscillation, but it makes them faster, see figure 7. The space charge maximum gives an idea of the sheath position

Figure 9. The space charge contribution N B0=1.2 T.

for

and faces the extraction hole, see figure 8; on the contrary the jf contribution peaks before, where front plasma ions are still slow, see figures 9 and 11.

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

Figure 10 shows that simulation results jE have a behaviour smoother than the expected jT for uE<uch. Some explanation is visible in figures 12 and 13; first, the plasma current reduces by a factor ym/ys at extraction; second, ion beam initially diverges, reducing space charge further.

7. Alternative models
As an alternative to ray tracing, we explore the possibility of using the Hamilton-Jacobi (HJ) equation in practice. While numerical solution of this equation for the action S are usually obtained with ray tracing based methods, here we use the Comsol solvers, that are based on finite elements in spatial coordinates. There also exist works related to solving HJ with finite difference methods [8,9], or with help of higher dimensional spaces [10]. An artificial viscosity e2DS is often added to HJ to have solutions, and the limit for 2 0 of these solutions is named viscosity solution [11]. Here we use only a transverse artificial diffusion e2S,yy, as we did for jf in eq. 12, since we want to perturb the solution as small as possible, and to preserve the physical fact that we have a particle extraction problem. Note some general HJ features when e2=0: the ion velocity is v=grad S/m, so that a streamline plot for S is a trajectory plot for the particle. Equipotentials of S are called wavefronts [12]. Moreover a ray crossing is a singular point for S, and after that, multivalued S(x,y) can be considered. In this study, we assume a simple fixed geometry (see figure 14) and that the particles with only one charge q and mass m originates from an emitting surface (the cathode DB), where V=0 with a number density n0c and a speed v0, normally directed. The HJ equation is
(2m) 1 ( S ,2x + S ,2y ) + qV = E

Figure 11. The space charge term N f for B0=0.3 T

Figure 12. Equipotentials u=-3.5:0.5:3.5 near the extractor edge and beam envelope for B0=0.6 T and uE=uch. Here up=-3.95.

(22)
Figure 13. Equipotentials u=-3.5:0.5:3.5 near extractor edge and beam envelope for B0=0.6 T and uE=0.8 uch. Here up=-3.95.

with E=mv02/2 the total energy. By defining Te=E now, and scaling the variables S= m v0 a and u=-qV/Te , we get (a,x)2+(a,y)2=1+u (23)

Figure 10. Extracted current jE versus uE/uch for several B0; also some comparison with jT.

The support of a and u in general differs, as explained in figure 14 caption, since the a=0 boundary line must have a continuous tangent.

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

Figure 14. The model geometry in the reference plane (X,Y). The condition u=0 holds on lines DB (cathode) and BL (focus electrode). The condition S=a=0 is imposed on line DBF and domains 2 and 4 were added, so that line BF is exactly tangent to DB in B, and its normal (parallel to initial speed) is continuous. No perfectly matched layers were necessary, even if a suitable domain subdivision was prepared.

Figure 15. The ion density c for uE=2, e2=e3=0.005 and k4=0.01 . Note that density increases at spot for convergence, but is reduced by acceleration. Note also streamlines of a (red lines) and a few particle track from cathode (green)

while the total current

I tot = dy a, x c

(a

subdomain to boundary projection variable) is plotted in figure 17 and compared x to I leak ( x) = 0 dxi a, y c( x i ) , the current crossing the x axis. Note that current leak appears after a singular point, near C. The reflection symmetry of the problem allows us to start a second HJ wavefront b from the x axis, carrying a particle density d, equal to the loss of c:
2 b, yy + 0.7 b, x + b, y 2 + 2u (b, x b, y ) 2 = 0 3 d , yy + b, x d , x + b, y d , y + d b = 0 (25)

We assume c=3y2/rh3 on cathode and c=0 on line BF, to generate a hollow beam. The special case u=0 and the concept of ray may be of help in the design the cathode. When u=0, the field a is the distance from the DBF curve and rays are the normal lines to this curve. The cathode DB is implemented as a Bezier curve (an arc of circle, parabola, or hyperbola); the circle gives the worst singularity, since all rays converge to its center C. We choose hyperbolic cathodes, so that rays would cross the axis in a segment CQ. In simulation here, DC=18.5 mm. The other equations are div(c grad a)=0 and the Poisson equation Du=k4c with k4=q2n0/Tee0,. The coupled PDE system has the nice form
2 a, yy + a, x 1 + u (a, y ) 2 = 0

In other words, after the singular point, we describe the multivalued S with two action fields. Note that boundary conditions for b , d are coupled to a, c ones on the x axis (26) b,y=-a,y d=c for a,y<0 and this requires caution. After some trial and error we set the constrained expressions to be (-a,y-b,y)*(a,y<-0.02) and c*(a,y<-0.02)-d (27) on the axis and d on source line DBF.

3 c, yy + a, x c, x + a, y c, y + c a = 0
u + k 4 ( c + d ) = 0

(24)

with the convention that the square root is defined to be zero for negative arguments; d is the scaled density of other particles, if any. The weak formulation is used for all fields. On line GEH, where ny=0, a zero weak term is suitable for a and c, as explained in section 4, and the constrain expression uE-u is entered for u. On the symmetry axis x, we should take in account the possibility of particle crossing, so that a,y can be negative. Since Gy=e2a,y, the boundary term -nye2a,yT(a) is entered. Similarly for c we set -nye2c,y T(c) . On the contrary u,y=0 holds on axis for symmetry, so that no weak boundary term is needed. Some solution is plotted in fig 15 and 16,

Figure 16. The potential u for uE=2, e2=e3=0.005 and k4=0.01 .

Excerpt from the Proceedings of the COMSOL Users Conference 2006 Milano

Figure 17. The balance of currents versus x. Note Ileak + Itot is about one; leak of current a,yc is reflected back as b,yd ; the x axis condition a,y+b,y=0 is satisfied when a,y<0, as imposed.

Figure 18. The current density jx=a,xc+b,xd for uE=2, e2=e3=0.005 and k4=0.01 . Note streamlines of a (red) and some particle tracks (blue) and hollow beam which converges, and then diverges.

Moreover we set e2 ny b,y T(b) and e3 ny d,y T(d) as weak terms on all the boundaries. Coherence of the b wawefront requires also b,y=cos(EBF)=0.163 on the line BF, since a,y=-cos(EBF) there. Solution is computed in three stages: 1) only a, c, u are solved for; 2) b and d are solved for, with a, c, u fixed; 3) a, b, c, d and u solution is completed. Some final solution is plotted in figure 18, and the current balance is compared in figure 17, with I refl = dy b, x d . Also the streamlines of a are compared to actual particle tracing. Comparison is fairly good, with possible difference due to e2, e3 values.

9. References
1. P. Spadtke, C. Muhle, Simulation of ion extraction and beam transport, Rev. Sci. Insrum., 71, 820-825 (2000) 2. R. Becker, W. B. Herrmannsfeldt, IgunA program for the simulation of positive ion extraction including magnetic fields, Rev. Sci. Insrum., 63, 2756-2758 (1992) 3. A.T. Forrester, Large Ion Beams, 45-76, John Wiley, NY (1996) 4. K.U. Riemann et al., The plasma-sheath matching problem, Plasma Phys. Control. Fusion , 47, 1949-1970 (2005) 5. H. Schmitz, K.U. Riemann, T. Daube, Theory of the collisional presheath in a magnetic field parallel to the wall, Phys. Plasma, 7, 2486-2495 (1996) 6. S.A. Self, Exact Solution of the Collisionless Plasma-Sheath Equation, Phys. Fluids, 6, 1762-1768 (1963) 7. M. Cavenago, V. Antoni, F. Sattin, Issues in modelling of negative ion extraction, Proceedings of Epac2006, 2218-2220 (2006) 8. Y. Tsai, L-T. Cheng, S. Osher, H. Zhao, Fast Sweeping Algorithms for a Class of Hamilton-Jacobi Equations, SIAM Jour. Numer. Analysis, 41, 673-694 (2003) 9. C.T. Lin, E. Tadmor, High-resolution Nonoscillatory Central Schemes for Hamiltion-Jacobi Equations, SIAM J. Sci. Comput., 21, 2153-2185 (2000) 10. T. Cecil, J. Qian, S. Osher, Numerical methods for high dimensional Hamilton Jacobi equations using radial basis functions, Jour. Computational Phys., 196, 327-347 (2004) 11. M. G. Crandall, P-L. Lions, Viscosity Solutions of the Hamilton-Jacobi Equations, Trans. Amer. Math. Society, 277, 1-42 (1983) 12. J. Ehlers, E. T. Newman, The theory of caustics and wave front singularities with physical applications, Jour. Math. Physics, 41, 3344-3378 (2000)

8. Conclusions
A description and solution of a model including plasma charged sheath and the beam extraction was here presented, at least with some simplification related to the simple magnetic field assumed. The critical point in the model is the integral coupling due to volume ion production, which is here carefully and approximately reduced to locally coupled expressions for u and a suitably defined jf. Ray tracing is modeled with interpolation from a few computed rays and a satisfactory principle solutions for ray crossing was found. From the practical point of view, not only the sheath meniscus, but also the whole front plasma provides some regulation of the extracted current to the acceleration voltage. The achieved accurate description of extractor edge effects on plasma is basilar to technological optimization of ion sources. As regards to alternative to particle tracing, a simple application of the HJ equation was described and solved. This is promising, also in view of higher dimensional methods, including Vlasov equation.

Vous aimerez peut-être aussi