在计算的过程中有如下提示

1.     The following message from the master process (MYID= 0):
WARNING 33193: Negative vertex coordinates found in the geometry, STL export will not be conformal to the standard, one might consider moving the geometry to make all coordinates positive
         See also message in the output file A1_MLFMM.out
Checking the geometrical input data
2.    The following message from the master process (MYID= 0):
WARNING  1663: Inhomogeneous segmentation for triangles
         See also message in the output file  A1_MLFMM.out
   Checking the geometrical input data for triangles
3.    The following message from the master process (MYID= 0):
NOTE    33242: For the FMM there is the possibility to save memory by switching to single precision (see EG card), this is normally recommended
          See also message in the output file  A1_MLFMM.out
第一个问题 我认为是有警告是有负向的面元 但是我差了半天也没有（并不排除我出错误的原因）
第二个问题 是不是因为最大的边与最小边的比值超过100所以显示Inhomogeneous segmentation
第三个问题 用fMM会节省内存 这个怎么设置呢?

多看一下 PDF 手冊跟範例對你會有幫助....
第一個問題 我認爲是有警告是有負向的面元 但是我差了半天也沒有（并不排除我出錯誤的原因）
--> 方向性... file A1_MLFMM.out...裡會告訴您是那個面。
第二個問題 是不是因爲最大的邊與最小邊的比值超過100所以顯示Inhomogeneous segmentation
沒架構，思考方向有2：
1. inhomogeneous dielectric bodies
2.inhomogeneous meshing
從字面看來應該是2
WARNING  1663: Inhomogeneous segmentation for triangles .See also message in the output file  A1_MLFMM.out
Examples of BQ card usage:
The BQ card can be used to create the mesh shown in Figure 13-6, or also for inhomogeneous
meshing as shown in Figure 13-7.
Figure 13-6: Example of a BQ card from demo BQ1.pre
Figure 13-7: Example of a BQ card with an inhomogeneous segmentation from demo BQ3.pre
July 2008 FEKO User’s Manual - DESCRIPTION OF THE GEOMETRY CARDS 13-9
第三個問題 用fMM會節省内存 這個怎麽設置呢?
NOTE    33242: For the FMM there is the possibility to save memory by switching to single precision (see EG card), this is normally recommended
1. single precision (see EG card)
Solution accuracy: This parameter can be set to force FEKO to use single precision for the storage of some of the memory critical arrays. Single precision storage is the default behaviour, and as compared to double precision the memory requirement is then half. Using double precision is recommended when the FEKO kernel gives a warning to switch to double precision (this might happen for instance at low frequencies where an increased accuracy is required).
2.  MLFMM
21 Using the MLFMM for electrically large models
Keywords: MLFMM, large model, Radar cross section, trihedral
In this example we consider a single plane wave incident (from ϑ=60◦ and ϕ=0◦) on a
large trihedral. The size of the trihedral (13.5λ2 surface area) was chosen such that it
can still be solved incore on a PC using 519 MByte of RAM. Larger examples show a
proportionally larger resource saving by using the MLFMM, but the absolute increase in
solution time makes such an example impractical for explanation purposes. This example
is large enough to demonstrate the advantage of using the MLFMM.
22 Antenna coupling on an electrically large object
Keywords: electrically large, MLFMM, CFIE, coupling, antenna placement,
S-parameters
A Rooivalk helicopter mock-up model with 3 monopole antennas located near the front,
middle and back of the model respectively. S-parameters (coupling) are computed between
the 3 antennas over a frequency range.

FMM是他自己MLFMM自己改成的一种算法吧 而且用单精度会节省一半内存 那个反向面元还是不大好找 能不能再看看