Context Navigation

Meeting Summary

The sources of asymmetry could be broken down into 4 categories…

Non-symmetric addition [ a+b-c-d /= (a-c)+(b-d) ]
- Diffusion algorithm's calculation of the divergence
- Characteristic limiters that do matrix multiplication from slowest to fastest wave speed
- Calculation of back forces on particles
Upwind type calculations that expect quantities to be < or > 0…
- Riemann fans that have an even number of states (odd number of waves) such as HLLC, HLLD, ExactRS
- MHD eigenvectors which depend on the sign of Bx or the 'polarization angle' [atan(Bz/By)] which break down when Bx = 0 or By=Bz=0
Bugs in the code
- Extrapolated MHD physical boundaries not updating cell centered b fields using aux
- Clump object's default behavior to persist in boundaries
- Wind object's failure to zero out other fields - miscellaneous tracers etc…
Machine optimization
- Apparently optimization can lead to a breakdown in accuracy (perhaps because it reorders additions/subtractions to be faster?). Fortunately this behavior can be corrected by the addition of -fp-model precise to the compiler options.

f_{\mbox{rotating}}=-2\vec{\Omega} \times \vec{v} - \vec{\Omega} \times (\vec{\Omega} \times \vec{r})

or if we assume that \vec{\Omega} = \Omega \hat{z} then we have

f_{\mbox{rotating}}=2\Omega \vec{v} \times \hat{z} + \Omega^2 \vec{x}_p where \vec{x}_p = \vec{x}-\vec{x} \cdot \hat{z}

No comments.