this discussion appeared on the forum tonight.
first is the question, then the answer
"Hi all,
I was wondering what science is behind the new WUs.
I have found only very limited info in this forum:
CSS in the names of the WUs means Catalina Sky Survey, 5y, 10y, 20y denotes how many years the simulation takes into account.
However, does anybody know more detailed info? I suppose that we are of course simulating trajectories of asteroids, but (questions just for illustration):
- one asteroid for one WU?
- are general relativity effects taken into account?
- is reliability of the input data somehow taken into account? (more observations means more precision)
Simply, I would highly appreciate any scientific info about the new WUs - I like knowing what I am crunching (at least roughly).
Maybe this would be announced/explained at the end of July after the conference, but at least some basic preliminary info I think would be useful for many crunchers.."
--------------------------------------
Now Pasquale's answer..
"Simply put, each WU simulates 512 NEOs and one or more telescopes, over a given period.
NEOs belong to two populations: "real" and "synthetic". Both populations are simulated, related only statistically to the population of known NEOs in the Solar System.
Most WUs (between few hundred and two thousand units) simulate "synthetic" NEOs, while only a few (usually 20) simulate "real" NEOs. This means that in each batch we simulate up to one million synthetic NEOs and about 10,000 real NEOs.
In all WUs generated during the same batch, the behavior of the telescope is identical. The scheduling at this point is very simple.
WUs with "real" NEOs keep track of the real NEOs discovered during the simulated period, while WUs with "synthetic" NEOs keep track of ... well, this is the tricky part. Every time a synthetic NEO is observed, its likelihood to represent a real asteroid gets lower and lower. A synthetic NEO that never gets observed is likely to represent a real NEO that has not been discovered yet. Otherwise, a synthetic NEO observed multiple times is quite unlikely to be related to a real NEO still undiscovered: if that was the case, we should have discovered already the real NEO.
After the completion of all the WUs in the batch, we have:
- a list of real NEOs discovered over the simulated period,
- a list of real NEOs that have not been discovered yet,
- a list of synthetic NEOs with associated their likelihood to represent a real NEO.
Now it's time to test the "improved" scheduling for the same telescope.
In a new, single simulation that includes ALL the NEOs in the batch, and over an extended period that is after the period in the batch (not overlapping), I simulate on the server the telescope that now points exclusively at regions of the sky where the synthetic NEOs concentrate, focusing on those that are more likely to represent real NEOs. As simple as that. As a comparison, I also perform the same simulation where the telescope scheduling is as in the batch.
By pointing at regions with high density of "good" synthetic NEOs, the telescope performance seems to improve very significantly, at least according to our simulations. The idea behind it as you can see is pretty simple: use a "control" population to track the past performance, and use it for future decisions.
After the summer, I believe we'll be at a good point to stop working on simulated NEOs, and start using real telescope data and real NEOs, and see how this method performs in the real world. But it's going to take some time... we're probably going to see initial results from real world testing not before spring 2009. Before then, the screensaver will help making things clear on how this works."
first is the question, then the answer
"Hi all,
I was wondering what science is behind the new WUs.
I have found only very limited info in this forum:
CSS in the names of the WUs means Catalina Sky Survey, 5y, 10y, 20y denotes how many years the simulation takes into account.
However, does anybody know more detailed info? I suppose that we are of course simulating trajectories of asteroids, but (questions just for illustration):
- one asteroid for one WU?
- are general relativity effects taken into account?
- is reliability of the input data somehow taken into account? (more observations means more precision)
Simply, I would highly appreciate any scientific info about the new WUs - I like knowing what I am crunching (at least roughly).
Maybe this would be announced/explained at the end of July after the conference, but at least some basic preliminary info I think would be useful for many crunchers.."
--------------------------------------
Now Pasquale's answer..
"Simply put, each WU simulates 512 NEOs and one or more telescopes, over a given period.
NEOs belong to two populations: "real" and "synthetic". Both populations are simulated, related only statistically to the population of known NEOs in the Solar System.
Most WUs (between few hundred and two thousand units) simulate "synthetic" NEOs, while only a few (usually 20) simulate "real" NEOs. This means that in each batch we simulate up to one million synthetic NEOs and about 10,000 real NEOs.
In all WUs generated during the same batch, the behavior of the telescope is identical. The scheduling at this point is very simple.
WUs with "real" NEOs keep track of the real NEOs discovered during the simulated period, while WUs with "synthetic" NEOs keep track of ... well, this is the tricky part. Every time a synthetic NEO is observed, its likelihood to represent a real asteroid gets lower and lower. A synthetic NEO that never gets observed is likely to represent a real NEO that has not been discovered yet. Otherwise, a synthetic NEO observed multiple times is quite unlikely to be related to a real NEO still undiscovered: if that was the case, we should have discovered already the real NEO.
After the completion of all the WUs in the batch, we have:
- a list of real NEOs discovered over the simulated period,
- a list of real NEOs that have not been discovered yet,
- a list of synthetic NEOs with associated their likelihood to represent a real NEO.
Now it's time to test the "improved" scheduling for the same telescope.
In a new, single simulation that includes ALL the NEOs in the batch, and over an extended period that is after the period in the batch (not overlapping), I simulate on the server the telescope that now points exclusively at regions of the sky where the synthetic NEOs concentrate, focusing on those that are more likely to represent real NEOs. As simple as that. As a comparison, I also perform the same simulation where the telescope scheduling is as in the batch.
By pointing at regions with high density of "good" synthetic NEOs, the telescope performance seems to improve very significantly, at least according to our simulations. The idea behind it as you can see is pretty simple: use a "control" population to track the past performance, and use it for future decisions.
After the summer, I believe we'll be at a good point to stop working on simulated NEOs, and start using real telescope data and real NEOs, and see how this method performs in the real world. But it's going to take some time... we're probably going to see initial results from real world testing not before spring 2009. Before then, the screensaver will help making things clear on how this works."