2d cold-ion equations
TRANSCRIPT
2D Cold-Ion EquationsDiscontinuities in its Solutions
James Cheung
Department of Scientific ComputingFlorida State University
Tallahassee, Florida 32306
December 4, 2013
The Cold-Ion Equations
∂n
∂t+∇ · (n~v) = 0
∂~v
∂t+ ~v · ∇~v = −∇φ
∇2φ = eφ − n
I n denotes the plasma density
I ~v denotes the flow velocity
I φ denotes the electrostatic potential
An Explicit Form
∂n
∂t+
∂
∂x(nu) +
∂
∂y(nv) = 0
∂u
∂t+
∂
∂x(
u2
2+ φ) + v
∂u
∂y= 0
∂v
∂t+
∂
∂y(
v 2
2+ φ) + u
∂v
∂x= 0
∂2φ
∂x2+∂2φ
∂y 2= eφ − n
What are plasmas?
I A plasma is a gas that has enough energy to seperateelectrons from its respective atoms.
I A plasma is a mixture of electron gas and positive ions.
I Plasmas are often considered collisionless, which led to thedevelopment of the Vlasov-Maxwell and Vlasov-Poissonequations.
I Plasmas are often super-critical fluids.
I Cold plasmas are not super-critical, and the electrontemperature is negligible.
I Cold plasmas can be modelled using the Cold-Ion Equations.
What do these equations model?
I These equations are representative of the cold-ion limit ofcollisionless plasma modelled by the Vlasov-Poisson equations.
I Instead of determining a distrubution of ion-densities at everypossible velocity and space coordinate, there is only 1 uniquedensity associated to 1 velocity and 1 spatial coordinate.
I Instead of using Maxwell’s equations to model the behavior ofthe charged nature, the Poisson equation is used instead asthe zero magnetic field limit. This assumption is valid becausein a cold-ion the charge separation between the electron andits ion is very small.
I This model can be viewed as the Euler equations coupled withelectrostatic phenomena.
Why are these equations interesting? (i.e. Focus of myresearch)
With initial conditions in density that contain step-discontinuities,an infinte spike forms at approximately x ≈ t. This has beenshown both numerically and analytically using the stationary phasemethod for the one dimensional case. (See Perego, et. al.)
Numerical Methods
I The elliptic PDE was solved using finite differences withNewton’s method to deal with the nonlinearity
I The hyperbolic system was solved using an explicit staggeredLax-Friedrichs finite difference scheme.
Staggered Lax-Friedrichs
The scheme is a two-step scheme. For the transport equation, it isgiven as:
Un+ 1
2
i+ 12,j+ 1
2
=1
4
(Un
i ,j + Uni+1,j + Un
i ,j+1 + Uni+1,j+1
)− ∆t
2∆x
(Un
i+1,j − Uni ,j
)− ∆t
2∆y
(Un
i ,j+1 − Uni ,j
)
Un+1i ,j =
1
4
(U
n+ 12
i− 12,j− 1
2
+ Un+ 1
2
i− 12,j+ 1
2
+ Un+ 1
2
i+ 12,j− 1
2
+ Un+ 1
2
i+ 12,j+ 1
2
)− ∆t
4∆x
(U
n+ 12
i+ 12,j− 1
2
− Un+ 1
2
i− 12,j− 1
2
+ Un+ 1
2
i+ 12,j+ 1
2
− Un+ 1
2
i− 12,j+ 1
2
)− ∆t
4∆x
(U
n+ 12
i− 12,j+ 1
2
− Un+ 1
2
i− 12,j− 1
2
+ Un+ 1
2
i+ 12,j+ 1
2
− Un+ 1
2
i+ 12,j− 1
2
)
Numerical Error
The theoretical truncation errors:
I Second Centered Difference: O(∆x2 + ∆y 2
)I Two-Step Lax-Friedrichs: O
(∆x2+∆y2
∆t + ∆x + ∆y + ∆t)
In addition to truncation error, we have phase errors for theTwo-Step Lax-Friedrichs scheme. The scheme is dissipative anddispersive.
Why do I want to use such a “terrible” method?
I I am not particularly interested in achieving an extremelyprecise solution
I The numerical dissipation allows the algorithm not to blow upwhen the infinite spike forms.
I The dipersion error is smoothed out by using a staggered grid.
I The scheme perserves monotonicity, whereas higher orderschemes, i.e. Lax-Wendroff does not. (Godunov’s Theorem)
I The phase errors can be reduced by taking timesteps close tothe theoretical limit of stability.
Verification
Because the Poisson equation convergence rates were studied todeath in the lab, I will focus my presentation on verifying that mycode is consistent with the properties of the staggeredLax-Friedrichs method. The test problem is the 2D transportequation.
∂u
∂t+∂u
∂x+∂u
∂y= 0
u (x , y , 0) =
{1 if |x | < 5 and |y | < 5
0 otherwise
Verification (continued)
The numerical rates of convergence
Numerical Error and Rates of Convergence
h ||u(xi , yj , tn)− Un
i ,j ||2 rate
1/4 0.29521/8 0.2321 0.3469
1/16 0.1856 0.32251/32 0.1501 0.30631/64 0.1224 0.2943
The numerical rate of convergence is low due to the phase errorcaused by the numerical method. ∆t was taken to be h
2
Visual Verification
These pictures represent the numerical solution at t = 5.
Figure: Solutions of the test equation. ∆t = 0.1h, ∆t = 0.5h, and ExactSolution
The numerical solutions verify that my algorithm is dissipative andthat it does not create spurious oscillations.
Some Cool Pictures
Three initial conditions for the PDE will be considered. Thesecases model the expansion of cold-ion plasma of a higher densityinto a cold-ion plasma of a lower density.
I The 2D analogue of the initial conditions studied in Perego,et. al.
I A square of high density
I A circle of high density
It will be assumed that an infinite domain is adequatelyrepresented using homogeneous Neumann boundary conditions.
A Square of High Density
n (x , y , 0) =
{1 if ||x || < 5 and ||y || < 5
0.25 otherwise
~v (x , y , 0) = 〈0, 0〉
Algorithm
While final time has not been reached:
1. Solve for φ on base grid.
2. Solve for n and ~v on staggered grid.
3. Solve for φ on staggered grid.
4. Solve for n and ~v on base grid.
5. Timestep
This algorithm is expensive. Newton’s method must be used tosolve the nonlinear Poisson equation for φ twice per time step.Remember, ∆t ≤ h
2 . Thus, in these simulations a gridsize h = 0.1and ∆t = 0.4 was used. In my 1D simulations, it was found that agridsize of h = 0.02 was ideal for modelling the phenomena. Thiswould be cost-prohibitive for my current 2D algorithm.