... | ... | @@ -94,7 +94,7 @@ clusters based on the Message Passing Interface (MPI). The underlying |
|
|
concepts of parallelism and load balancing are described in some more
|
|
|
detail .
|
|
|
|
|
|
#### Adaptive mesh refinement
|
|
|
#### Adaptive mesh simulations
|
|
|
|
|
|
NIRVANA features the possibility of multi-scale simulations by applying
|
|
|
the technique of Adaptive Mesh Refinement (AMR). AMR works in all
|
... | ... | @@ -388,7 +388,7 @@ cell-centered. |
|
|
|
|
|
| location | `g->` | definition: cylindrical | definition: spherical |
|
|
|
|:------------------------------|:-------------|:--------------------------------------------------------------------------------|:--------------------------------------------------------------------------------|
|
|
|
| cell-centroid x | `xc[ix]` | *z*<sub>`ix`+ 1/2</sub> | 3*Δ* <sub>`ix`</sub>*r*<sup>4</sup>/(4*Δ*<sub>`ix`</sub>*r*<sup>3</sup>) |
|
|
|
| cell-centroid x | `xc[ix]` | *z*<sub>`ix`+1/2</sub> | 3*Δ*<sub>`ix`</sub>*r*<sup>4</sup>/(4*Δ*<sub>`ix`</sub>*r*<sup>3</sup>) |
|
|
|
| cell-centroid y | `yc[iy]` | 2*Δ*<sub>`iy`</sub>*R*<sup>3</sup>/(3*Δ*<sub>`iy`</sub>*R*<sup>2</sup>) | *Δ*<sub>`iy`</sub>(*θ*cos *θ* − sin *θ*)/*Δ*<sub>`iy`</sub>cos *θ* |
|
|
|
| *xy*/*xz*-face-centroid x | `xf[ix]` | *z*<sub>`ix`+1/2</sub> | 2*Δ*<sub>`ix`</sub>*r*<sup>3</sup>/(3*Δ*<sub>`ix`</sub>*r*<sup>2</sup>) |
|
|
|
| *yx*-face-centroid y | (≡) `yc[iy]` | 2*Δ*<sub>`iy`</sub>*R*<sup>3</sup>/(3*Δ*<sub>`iy`</sub>*R*<sup>2</sup>) | *Δ*<sub>`iy`</sub>(*θ*cos *θ* − sin *θ*)/*Δ*<sub>`iy`</sub>cos *θ* |
|
... | ... | @@ -399,7 +399,7 @@ are identical to their cell/face-centered coordinates.* |
|
|
*Note 6: In Cartesian geometry all cell/face-centroid coordinates are
|
|
|
identical to their cell/face-centered counterparts.*
|
|
|
|
|
|
### Adaptive mesh refinement
|
|
|
## Adaptive mesh refinement
|
|
|
|
|
|
The mesh refinement algorithm relies on the oct-tree data structure
|
|
|
`_G0` (of type `*GRD`) which represents a hierarchy of nested generic
|
... | ... | @@ -487,7 +487,7 @@ criteria which are |
|
|
user-specific threshold with the reciprocal value giving the minimum
|
|
|
number of grid cells the Field length should be resolved.
|
|
|
|
|
|
### Parallelism and load balancing
|
|
|
## Parallelism and load balancing
|
|
|
|
|
|
The NIRVANA code is parallelized making use of the MPI library. The code
|
|
|
infrastructure provides routines for mesh partitioning in order to
|
... | ... | |