Physics simulator concept + problems

EXTREMELY ABRIDGED VERSION:
Look at how a bunch of small things behave together, make a medium thing that behaves like that bunch of small things; Look at how a bunch of medium things behave together, make a big thing that behaves like that bunch of medium things; and so on...

FULL VERSION:
The idea of this is a physics simulator that only gets linearly more demanding as the scale of it increases exponentially, while still doing subatomic scale simulation.
This is done by doing lots of small simulations, throwing them into a database, making larger simulations out of those smaller simulations, throwing the larger simulations into a database, etc.
Those "small simulations" are "level 1 regions", the "larger simulations" are level 2 regions, and so on.

So you have:
Database of level 1 regions that are made from hardcoded objects (Smallest known fundamental particles)
Database of level 2 regions that are constructed from level 1 regions
Database of level 3 regions that are constructed from level 2 regions
etc.

Each region subdivides into a number of smaller regions; areas with similar emissions and absorbtions are regarded as part of the same region when calculating what these smaller regions are (This is because doing this would cause the boundaries of regions to be what we regard as boundaries to distinct objects, or distinct parts of objects. For example, an ice cube would not be regarded as part of the water it is in, because the ice has far less thermal emission among other differences in emission and absorbtion.)

Each region is given a pattern of emissions and absorbtions; these can be quanta of energy, hardcoded objects, or smaller regions.

There are still lots of holes in this idea, but only because I don't know what to do for said holes.
!How should the creation, deletion, splitting, and merging of regions be managed, when there are any number of instances of each region being used by any number of larger regions?

!Emission and absorbtion data needs to be accurate in terms of spatial and chronological patterns, and needs to be factored in by the parent region simulation. How is this data best formatted and used? (Answer for this question likely depends on the answer for the next question, and the question after that)
^Note: Region overlap means that this data cannot be a 2D texture.
!Should regions be ballooned to completely pack the parent region, should a new region be created as the medium, or what? Because there's a lot of space between atoms.
^Solution: If empty space takes even quanta of energy into account when deciding subdivisions, then the space between regions subdivides itself in a manner that also solves the problem of managing the transmission
^!Movement of regions could create rapid and thus resource-intensive adjustment
^Note: This adjustment could be lessened by allowing one region to be completely overlapped by another.

!To increase accuracy and efficiency, how could regions that have varying output based on varying input be created, rather than regions that ignore alteration in conditions until it forces a change in the database?
<Proposed solution is held, but it will require review and explanation by someone who actually understands it.

!It's all very well saying how regions subdivide to give a very neat structure for the simulation, but how can one go from a blank simulation to that structure? (Answer for this question likely depends on the answer for the next question, and perhaps ties in with first question)
^Notes exist for solving this.
!Something like pure, completely still water would subdivide in one of these two ways: Whole volume of water is a single level 3 region and literally every molecule is part of that region as a level 2 region, or the whole volume of water is defined on every possible region level from 2 onwards. This is problematic.
^Possible solution: The ability for the formation of a large region to assimilate small regions should decrease (Get more sensitive to difference) as the number of small regions increases, so as to prevent crash-causingly huge regions from forming; and quantum effects should naturally deal with fracturing things that are perfectly consistent. The fact that low difference in output data causes large regions to form with this may or may not be a good thing.

!As a camera moves through the simulation, looking at small and large things, how should information change when the camera comes close, if at all?
^Solution: The region that the camera is in is continuously simulated, rather than simulating until it is decided that the region is properly documented. All other regions, including regions that are part of the one the camera is in and sibling regions of the one the camera is in are simulated as normal.

!What implementation of quantum physics interacts cleanly and cleverly with all of the above?

PERSONAL COMMENTS:
March 12th:
If textures applied to regions were used for calculating emissions, they would have to be 3D to compensate for overlapping.

March 12th:
Each region could be given a number of degrees of freedom to which would also have to be taken in consideration by behaviour interpolation.

March 19th:
The physics simulator should be able to form its hierarchy of subsimulations by reshaping the regions of an octree.
Octrees are traditionally cubical, but any mesh volume can become an octree.
The way they can be applied is by dividing the mesh volume into cubic shapes where it is possible to do using the vertices of the mesh, and then anything remaining can be filled with cubic shapes distorted into simpler shapes (For example, a cube can become a tetrahedron by compressing the top face into a line, and then doing the same to the bottom face but in the other direction). By having non-cubic shapes as distortions of cubes, traditional octree subdivision can be used.