4. Conclusions and future work

After more than one year of operations with our algorithms, including the one for attributions presented in this paper and the one for orbit identifications presented in [Milani et al. 2000a], we think the results can be rated satisfactory. Recovering more than 2,500 asteroids using telescopes would have consumed significant resources and so, obtaining the same result by pure computation frees the telescope and observer resources for the work which can only be done in that way. The comparison of our results with those of other groups is not straightforward. The largest number of identifications is found by the MPC staff, as is natural since they have access to the new observations before anyone else. They disseminate the data with an average delay of 2-3 weeks and they efficiently use this time to scan for identifications. The very fact that we find some identifications implies that our algorithms are capable of finding nontrivial identifications that could escape the MPC scrutiny. Another large fraction of the published identifications are credited to the so called DANEOPS consortium (A. Doppler, A. Gnädig, G. Hahn and others). We understand that their methods belong to the attribution class, but it is not easy to compare the efficiency of the two algorithms; probably their good results also depend upon a very efficient data flow and a rather advanced automation. Their purpose is to obtain the largest number of identifications in the shortest possible time after the data are available, and they are indeed very successful in this. Although we also try to be as efficient as possible and to conclude our computations in just a few days after the monthly update, the speed is not our main concern. Our main goal is to define new algorithms, which are published in the scientific literature and therefore can be used by others. Some improvements, which could result in a faster selection of the identifications to be proposed, are indeed suggested by the analysis of our results contained in this paper. As an example, the most computationally intensive steps of the identification search procedure should be performed in an incremental way, rather than by testing all the possible pairs each month. This way of acting could apply to the first filter, as pointed out in Section 2.1, and also to the third filter, which is even more time consuming, especially for short arc orbits. We are indeed experimenting with ``cleanup routines'' that remove from the list of pairs passed by the second filter the ones having already been submitted to the differential correction test in the previous months. More in general, the entire procedure needs to move towards a ``steady regime'' in which only the orbits and attributables either being new or having changed are processed. However, such a regime cannot be adopted until the algorithm control values and parameters, as well as the algorithms themselves, are finalized and frozen. The level of automation of the entire procedure we are using to propose identification is good, but could be improved. The problem is that a fully automated procedure removes the possibility of applying human judgment and we are reluctant to give up the possibility of human intervention, at least in the final stage where the selection of the pairs to be submitted takes place. However, the procedure in the previous steps, including the three filtering stages and even the recursive procedure to search for additional attributions to newly identified orbits, can be automated. In the final selection of the pairs to be submitted, the number of cases to be examined is comparatively small and some human intervention is affordable. The year of operations described in this paper has been the first and, of course, we have had to learn from experience. Nevertheless, in this time span our group has been the third largest contributor for identifications, and with a significant fraction of the total number. This fact implies that our work is not only of theoretical interest, but is a practical contribution to the problem of finding identifications, which can in turn play an important role in other issues such as the prediction of close approaches [Milani and Valsecchi 1999], and the recovery. On the other hand, the number of identifications that is currently found by all groups is by far insufficient. Every month the number of new designations greatly exceeds the number of new identifications. The conclusion is not that so many new asteroids are actually discovered, but only that the data archives are more and more full of short observational arcs, with corresponding very poor orbits (if any), and to which one should add the existing one night stands that are not published. Most of these short arcs should be identified, but they remain unidentified. We believe that in the long run this problem will be the main one: how to clean up the data archives from all these essentially unused data. It is very difficult to estimate the number of identifications remaining to be found, but they must be many tens of thousands and hence our contribution of more than 2,500 appears inadequate. It follows that more advanced algorithms need to be found, and this will be an important goal of our continuing research in this field.



Acknowledgements: The OrbFit free software is maintained by a consortium led by A. Milani, M. Carpino, Z. Knezevic and G.B. Valsecchi; it is available at http://newton.dm.unipi.it/asteroid/orbfit/. This research has been supported: by the Italian Space Agency under grants ASI-ARS-98-240 and ASI-ARS-99-81; and by the North Atlantic Treaty Organization under a grant awarded in 1998.