Solver Related

What does "floating point error" mean? How can I avoid it?

The floating point error has been reported many times and discussed a lot. Here are some of the
answers found in the Fluent Forum:

From numerical computation view point , the basic operations performed by computer are
represented inside computer in what is called floating point numbers. The errors that are either
because of invalid numeric computation initiated by user or limitation of machine that is used are
floating point errors.

1)Invalid Operations:- Simplest example is if one uses Newton Raphson root finding method to
solve f(x)=0 and for some Nth iteration if we get x = x(N) such that derivative of function f(x),
f'(x(N))=0 then formula for calculating next iterate x(N+1) = x(N) - f(x(N))/f'(x(N)) requires
division by f'(x(N)) which is zero. Here you get divide by zero type of floating point error.

2) Over or Underflow:- Another type is having data with either too large or too small magnitude
called 'overflow' or 'underflow' respectively.Such data cannot be physically represented on
computer for direct processing by arithmetic processing part of Processor.

3) Rounding off errors :- While rounding off a decimal number , some significant digits are lost
which cannot be recovered . e.g. if we round off 0.1 to integer (not greater than it called 'floor' of
the given no.) then it is zero. If this value if further used for computation then it may lead to
several errors.

SOLVER AND ITERATION -----I think if you set shorter time step, it may be good. Or
changing little Under-Relaxiation-Factors, it may be good. In my experience, I set 1/3 Under-
Relaxiation-Factors as default.� -----�also lower the values of under relaxation factor and use
the coupled implicit solver� -----�Try to change under-relaxation factors and if it is unsteady
problem maybe time step is to large.� -----�you can improve the ratio in the solve--control--
limits, maybe that can help.� -----�you will need to decrease the Courant number� -----�If
you still get the error, initialize the domain with nothing to 'Compute from...' Then click 'init'.
Again select the surface from which you want to compute the initial values & iterate. This should
work.� -----�Another reason could be a to high courant number - that means, that the steps
between two iterations are too large and the change in the results is too large as well (high
residuals)�

GRID PROBLEMS -----�this error comes when I start scaling grid. in gambit, all my
dimension is in mm, when in fluent i convert it in meter using buttone SCALE. after it, when i
iterate, about hundred iteration, this error appeared. but when i not scale my drawing to m...and
let it be as in gambit..then the iteration is success. -----�hi I think you should check your mesh
grid mesh is very high. your problem solve by selection a low mesh.� -----�Your mesh is so
heavy that your computers resources are not enough. try to use coarser mesh.�
BOUNDARY CONDITIONS -----�In my case I had set a wall boundary condition instead of
an axis boundary condition and then FLuent refuses to calculate telling me 'floating point
error'.� -----�Your Boudary Conditions do not represent real physis.� -----�wrong boundary
condition definition might cause the floating point error. For example setting an internal
boundary as interior� -----�Once I had the problem, simulating a 2D chamber with a
symmetry BC. I set the symmetry somewhere as �axe symmetric� and the floating point error
occur� -----�check the turbulence parameter you set. reduce the turbulence intensity to less
that one for first, say 50 iterations.

MULTI PROCESSOR ISSUES -----"I've had similar problems recently with floating point
errors on a multi processor simulation. The solution for my problem seems to be to run on a
single processor, where it runs fine....?�

WRONG INITIATION ----- Initiating the case with wrong conditions may lead to floating
point error when the iterations start.