Week2 Slides
Week2 Slides
OVERVIEW
FINITE DIFFERENCE
ACCURACY
METHODS – II
Prof. Krish Sankaran DISPERSION
STABILITY
© Prof. K. Sankaran
OVERVIEW ACCURACY
Type – I Error
ACCURACY
DISPERSION
STABILITY
1
7/25/2016
ACCURACY ACCURACY
Type – II Error Type – III Error
ACCURACY ACCURACY
MODELLING ERROR – dependent
PHYSICAL MATHEMATICAL NUMERICAL
SYSTEM MODEL SOLUTION
Error
Mesh size
© Prof. K. Sankaran © Prof. K. Sankaran
2
7/25/2016
OVERVIEW DISPERSION
ACCURACY
STABILITY
where
© Prof. K. Sankaran © Prof. K. Sankaran
DISPERSION DISPERSION
FD solution Compare and
is numerical wavelength Take first 3 terms in the series expansion of
Substituting in FD equation above
where
3
7/25/2016
DISPERSION OVERVIEW
For
ACCURACY
DISPERSION
Error converges Error Caution
quadratically control by valid for
choice 1D only STABILITY
STABILITY STABILITY
LAX EQUIVALENCE THEOREM
4
7/25/2016
STABILITY STABILITY
A Numerical scheme is stable if a small error Define an error at time step as for one
at any stage produces a smaller cumulative independent variable
error at next stages.
Then at time step , this error amplifies as,
Otherwise it is unstable !!!
STABILITY STABILITY
For multivariable problem becomes a For a Finite Difference scheme to be stable,
matrix
or
5
7/25/2016
STABILITY STABILITY
Consider
ICs
STABILITY STABILITY
CFL condition
6
7/25/2016
STABILITY STABILITY
An FD scheme is stable if errors made at one von Neumann Stability Analysis
time step do not magnify as computation
continues
Decompose round off errors into Fourier
space and analyze their time evolution
ERROR ERROR ERR OR
STABILITY STABILITY
For FTCS to be stable,
7
7/25/2016
STABILITY STABILITY
To make scheme unconditionally stable all the
terms on RHS are replaced by average of
values at time-steps and
STABILITY STABILITY
Implicit scheme will take CD at
8
7/25/2016
STABILITY END
ACCURACY
© Prof. K. Sankaran