Dear Users,

1. Routines for CSR were changed to reduce the memory usage, computing time, etc.

> Dear Users,
>
> 1. Routines for CSR were changed to reduce the memory usage, computing time, etc.
>
cfg1, cfg2, cfg3, cfg4の型は、INTEGERですが、REALで無くてよいのでしょうか？

以下の理由から INTEGER型が自然とは思えません
* 0.d0、cf/12.d0、cg/6.d0(DOUBLE PRECISION REAL)で初期化している
* 差分的には、cf+cg等の REAL型の式を置き換えとして導入されている

すみません。お粗末な間違いです。V1.0.10.4.6a20で直します.



> > Dear Users,
> >
> > 1. Routines for CSR were changed to reduce the memory usage, computing time, etc.
> >
> cfg1, cfg2, cfg3, cfg4の型は、INTEGERですが、REALで無くてよいのでしょうか？
>
> 以下の理由から INTEGER型が自然とは思えません
> * 0.d0、cf/12.d0、cg/6.d0(DOUBLE PRECISION REAL)で初期化している
> * 差分的には、cf+cg等の REAL型の式を置き換えとして導入されている

Dear Users,

1. Routines for CSR were changed to reduce the memory usage, computing time, etc. The 5 point stencil differentiation introduced in 10.4.6a10 was abandoned as it makes symmetry of the CSR matrix worse to generate more complex eigenvalues, which is unphysical.

Dear Users,

1. Now CSR allows non-square meshes for transverse differential matrix calculation. If MeshRatio -> {mx, my}, it means different mesh ratios for x and y. MeshRatio is the ratio (rho / k^2)^(1/3) / (dx, dy) where dx and dy are the mesh sizes, rho the bending radius, and k the wake number.

2. More tiny improvements were made for CSR routines.

Dear Users,

1. Some corrections were made on the boundary condition of CSR to have quadratic approximation for the normal component of the field. Actually It was tried in the previous version, but there was a mistake.

2. Parallel processing of CSR was modified to make the CPU load flatter in time.

2. Shared now accepts operations AddTo, IncrementTo, etc.

Dear Users,

1. Routines for CSR were changed to place the boundary separated by a half mesh size from the wall. It used to be a full mesh size, then caused slow convergence.

Dear Users,

1. A stupid mistake was found in the previous change (a60) in CSR, in the calculation of the beam field. After correcting it, the convergence has become better.

Dear Users,

1. I found that some of the changes above were not effective in some cases depending on the shape of beam pipe. I reverted solving eigensystem by perturbation and half-mesh horizontal boundary. It still does not reasonable results for instance a circular cross section, depending on the length and bending radius. Anyway I have no idea now, so I just check-in as it is now.