How can I improve my residuals
 

I'm very new to CFD and Star-CCM+ and I'm not sure how to improve my residuals. The image below shows my simulation after 2000 iterations.

http://i.imgur.com/FGuNTDS.png

I would very much like to improve on my energy and continuity. What causes the residuals to barely drop below 0.01?

I have tried to lower the under-relaxation factors and create lower base size on my mesh, but no help so far.

The convergence criteria differs between simulations, although .01 is rather large. I would run reports to check mass flow at inlet and outlet and compare, also do you have reversed flow at the outlet? Additionally maybe your physics model or volume meshing model isn't optimal for the simulations that you're doing.

There are no massflow inlet or outlets. The velocities are purely because of natural convection. I'm using a polyhedral meshing model with a 2-layer prism layer.

Your meshing model should be fine. Maybe check if the flow is turbulent or laminar. If you run a turbulent model as a laminar model then it won't converge; I'm not sure if laminar flow under turbulent model would converge or not although I feel like it would.

Also if its turbulent try running implicit unsteady instead of steady.

Describe the mesh, please. From what i see - i would suppose this as a result of coarse mesh.

It worked out fine when I changed from turbulent to laminar model, so everything is all right now. Thanks.

Hi
No one can say that the results of the simulation are wrong because of high residuals. As Kylea&M told, you need to check mass conservation or energy conservation.
If you are simulating air movements in the room caused by a heat source, the pattern is substantially turbulent and simulating by laminar give wrong results.

With natural convection problems, I believe it is best to switch to a bouyancy driven "two-layer type" when using the k-epsilon turb. model.

This is under Continua->Physics->Models->Realizable K-Epsilon Two Layer

disregard if you are using k-omega.

The residuals fluctuate sometimes for the reason that the flow is an unsteady and we try to run it as a steady state problem. Take for example pressure at outlet may not be atmospheric but we usually give pressure outlet boundary condition that forces the simulation to run likewise.

Similarly the energy model can lead to such fluctuations. If the problem is independent of the change in density or 'incompressible' then a segregated flow model should be well able to solve the problem.

A compressible flow otherwise treated as an incompressible will again lead to fluctuation of residuals.

So in short we need to familiarize ourselves with the problem very well and apply the proper flow physics before solving the problem.

I would like to add that a residual that oscillates about some mean position may not necessarily yield a wrong solution. :)

I would agree with your first statement. However, as you said that segregated flow solver could resolve these fluctuations - I don't think it helps. Since, I am simulating the flow around a wind turbine blade to predict the location of transition using segregated flow but residuals have such fluctuations even they drop down to 1E-3 and 1E-4 whereas other monitoring parameters such as drag and moment coefficients become steady so I think there is no point to only rely on "making these residuals smooth".