# How to get correct results when baking a fluid simulation

I have a student who is trying to bake her fluid, but it won't do it correctly. Below are the steps she told me she takes when she does it. I have no idea on how to fix it and I have been trying for days. Any help would be appreciated. I also included screen shots of what she does.

"Before I bake, I assign the larger icosphere as the domain, and the cube to the fluid and the other objects as obstacles. When I bake it, the fluid moves up to its starting point at the square like it's supposed to. After that, however, when the fluid falls, it passes through the objects marked as obstacles and lands as if it were landing inside a large rectangular prism, or some of it gets suspended in midair. (the pictures show it)"

• Is it possible you could upload the .blend? Commented Jun 8, 2015 at 23:14
• "Before I bake, I assign the larger icosphere as the domain" Only rectangular solids can be domains. an icosphere will not work. Commented Jun 8, 2015 at 23:39
• I see that the obstacles are only surfaces (I assume), obstacles work better with mesh with thickness Commented Jun 9, 2015 at 8:26
• @gandalf3 pasteall.org/blend/36462 Commented Jun 9, 2015 at 12:31

There are several issues:

• The scene is upside down. The Z axis (blue) always points up in world space. In the file provided, the view is upside down and the blue arrow appears to be pointing down. When righted, it looks like this:

So in your scene the the fluid emitter is actually below the domain and collision objects.

• The fluid domain is always a box shape, as it uses the object's bounding box. When assigned to a non-box shape, this means that it will be a cuboid with the same dimensions as the shape.

If you replace the icosphere with a cube, the reasons for the odd behavior become clearer:

Note that I duplicated the domain object in this screenshot so that the domain bounds would be visible at the same time as the fluid

• The collision planes are non-manifold. The fluid simulator has no way of determining where the insides and outsides of the 2D planes are, so it will guess, usually resulting in strange behavior.

If the fluid emitter object is moved above the collision objects and the collision objects are given volume and correctly oriented normals, then it appears to work: