Scribd
Upload
35 views24 pages

8-Stability Analysis - Diffusion On Unstructured Meshes

Uploaded by

alagarg137691
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
35 views24 pages

8-Stability Analysis - Diffusion On Unstructured Meshes

Uploaded by

alagarg137691
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Lecture 8: Stability Analysis

Diffusion on Unstructured Meshes


Last Time…

z Completed unsteady scheme discussion


» Properties of implicit scheme
» Crank Nicholson scheme and its properties
z Truncation error analysis
» Mean value approximations are second-order
accurate
» Implicit scheme is first-order accurate
This Time…

z Von Neumann stability analysis


» Apply to explicit scheme and establish stability
limits
z Diffusion on orthogonal unstructured meshes
» What changes with respect to Cartesian?
Von Neumann Stability Analysis

z For steady problems:


» Can we obtain a solution using an iterative
scheme?
z For unsteady problems
» Will the numerical solution be bounded in time if the
original problem yields a bounded solution in time?
z Note analogy between iteration and time-stepping!
Stability of Explicit Scheme

z Explicit scheme in 1D:


Error Definition

z Consider “exact solution” of the discrete equation, Φ.


» This is the solution to infinite precision
z Numerical solution to machine precision is φ
z Error:
z Substitute into discrete equation:

z But Φ satisfies discrete equation exactly.


Error Equation

z Therefore error satisfies:

z Now expand error in Fourier series


Amplification Factor

z σm may be complex
z Amplification factor:

z Since problem is linear, let’s look at the behavior of a


single term:
Amplification Factor (cont’d)

z Substitute into discrete equation

z Divide through by

to yield
Amplification Factor (cont’d)

z Substitute for the coefficients aP , a0P and anb


(assuming uniform mesh) to yield:

z Define

so that
Amplification Factor (cont’d)

z Thus

Γ∆t
r=
ρ∆x 2

z For a stable scheme:


Amplification Factor (cont’d)

z For stability
1 − 4r sin 2 β ≤ 1
Or
−1 ≤ 1 − 4r sin 2 β ≤ 1
i.e,
−2 ≤ −4r sin 2 β ≤ 0 or 0 ≤ 4r sin 2 β ≤ 2

z Since
1
0 ≤ sin 2 β ≤ 1 0 ≤ 4r ≤ 2 or 0 ≤ r ≤
2
Stability Limits

z In 1D, the stability limit therefore is:

z In 2D, expand the error as:

z Assuming ∆x = ∆y
Summary

z Von Neumann stability analysis yields the same limit


heuristically derived by requiring all-positive
coefficients for the explicit scheme
z However, the stability limits derived from stability
analysis are more rigorous
z Similar analysis for implicit scheme reveals that it is
unconditionally stable
z The Crank-Nicholson scheme can be shown to be
unconditionally stable
» Can yield oscillatory solutions because of the
negative coefficient, but wiggles die down
Diffusion on Unstructured Meshes

z Recall unstructured mesh:


» Each node connected to
arbitrary number of
neighbors
z First we will look at
“orthogonal” meshes made up
of convex polyhedra
z Then we will look at “non-
orthogonal” unstructured
meshes
Diffusion on Orthogonal Meshes

z Consider mesh of convex polyhedra


z Mesh is considered orthogonal if line joining cell
centroids is perpendicular to face
z This admittedly constrains how “unstructured” a mesh
can get
Orthogonal Mesh
2D Steady Diffusion

z Governing equation:
z Here

z Integrate over control volume


around C0

z Apply Divergence theorem


Finite Volume Scheme

z Sum over faces of cell


z Linearize source term
z Here

z Instead of using global (x,y) system


use local (ξ,η) orthogonal coordinates
Profile Assumption

z Outward-pointing face area vector

z Face flux

z Note how only ξ direction component survives


z Make linear profile assumption between centroids to
find
Final Discrete Equation

z Cell C0 is connected
to face neighbors (nb)
z Did not worry about
shape of cell
z Need convex polyedra
or centroid may fall
outside cell
z Exactly the same
derivation in 3D
Discussion

z Scheme is fully conservative, just like on Cartesian


meshes, regardless of mesh size
z Think of Jf as belonging to the face f rather than to
either cell. Flux leaving C0 enters C1.
z Since in the absence of a source term, we
are guaranteed boundedness by face neighbors
z Scarborough criterion satisfied in the equality in the
absence of source terms – must look to Dirichlet and
mixed bc to satisfy inequality
Discussion (cont’d)

z All other procedures carry over from Cartesian


meshes
» Source linearization
» Interpolation of Γf
» Under-relaxation
z Can use Gauss-Seidel to solve since there is no
assumption of structure in it
Closure

In this lecture, we
z Looked at how to do a von Neumann stability analysis
» Applied it to the explicit time-stepping scheme
» Found that the limit is the same as the one we
derived heuristically
z Started looking at diffusion on unstructured meshes
» Started looking first at orthogonal meshes
» Found that structure of discrete equation is very
similar to the Cartesian case

You might also like