3 ms·
I am out of the loop, but I have a few questions, from a scientific perspective. Inputs into system can be either boundary/initial conditions and constants. It
by noobermin 11y ago
I am out of the loop, but I have a few questions, from a scientific perspective.
Inputs into system can be either boundary/initial conditions and constants. It seems like here, you essentially have to design the number system to match the problem, is this correct?
How can someone solve odes with this? I can see this solving constraints (which is basically algebra) but how about dynamical systems, especially when you don't quite know the behavior of the system, specifically, the bounds of your solution (which allow you to define your interval boundaries)? For example, what about chaotic systems (which are difficult in normal math too, I know). How about min/max problems (finding the global minimum of a function)?
It sounds like one would have to iterate your number system until you find your solution, which sounds like just offloading the difficulty somewhere else. (For finding the minimum of a function, you iterate as well, essentially). I think it makes sense that this does some problems well, but I need help to buy that it does all problems well.
- jacobolus 11y agoA chaotic system is by definition also beyond the reach of existing methods. With unums you’ll presumably get numbers that are less and less precise as you step through the computation, whereas with standard floats you get a “precise” answer which you know must be substantially wrong. I’m not sure to what extent if any unums can help solving ODEs. I haven’t seen any ODE experts examine unums yet or write down their impressions.
- noobermin 11y ago>A chaotic system is by definition also beyond the reach of existing methods. Not quite, it mainly is a system that is sensitive to initial conditions. For that reason, for a "range" of initial conditions wouldn't be helpful for a chaotic system, as you would get wildly different final results over that range for a particular set of conditions (ie., constants in the ODE), unless the range is small...and bam, you have a floating point with a +/- on it and probably an unum with a large number of intervals. For the weather, for example, you can still tell someone what it will be (with some degree of accuracy) tomorrow, but the weather three weeks from now is unpredictable due to that sensitivity to the initial conditions (ie. now).