Документ взят из кэша поисковой машины. Адрес оригинального документа : http://star.arm.ac.uk/~csj/papers/conference/2002_ATMOS_2.ps
Дата изменения: Thu Aug 15 12:33:13 2002
Дата индексирования: Tue Oct 2 10:12:41 2012
Кодировка:

Поисковые слова: arp 220
**TITLE**
ASP Conference Series, Vol. **VOLUME**, **PUBLICATION YEAR**
**EDITORS**
Computational Astrophysics Tools for the GRID
C. Simon Je ery
Armagh Observatory, College Hill, Armagh BT61 9DG, Northern
Ireland
Abstract. The newest generation of telescopes and detectors and facil-
ities like the Virtual Observatory (VO) will deliver enormous volumes of
astronomical data. The analysis of these data relies heavily on computer-
generated models of growing sophistication and realism. To carry out sim-
ulations at increasingly high spatial and temporal resolution and physical
dimension and in a growing parameter space will require a new approach
to data modelling. A proposal to develop toolkits to faciliate such mod-
elling and to integrate them within a heterogeneous network (such as the
VO) is described.
1. Inroduction
The proposals outlined in this poster have been developed by a team compris-
ing the author, Dr Robert J. Allan (CLRC), Dr Martin A. Barstow (Univ. of
Leicester), Prof Tom J. Millar (UMIST), Dr Juliet C. Pickering (ICSTM) and
Dr Jeremy Yates (UCL), together with other associates, involved in a range of
science projects:
 measuring the properties and dynamics of plasma, elds and particles in the
near-Sun heliosphere,
 investigating the ne-scale structure and dynamics of the Sun's magnetised
atmosphere,
 the structure and evolution of stellar atmospheres other than the Sun,
 probing the atmospheres of stellar remnants including white dwarfs, hot sub-
dwarfs, helium stars and RCrB stars,
 the structure and chemistry of hot molecular cores,
 shock chemistry associated with protostellar jets,
 the evolution of asymptotic giant branch stars and planetary nebula,
 photochemistry in the envelopes of AGB stars and proto-planetary disks,
 the processing of dust from AGB stars through the ISM and protostellar cores
to planets,
 the hydrodynamics of planetary atmospheres,
 and the excitation of cosmic masers.
Our goal is to develop software tools and a computational framework that will
facilitate modelling astronomical data in large parameter spaces using widely
distributed and heterogeneous computer systems. It builds on experience and re-
sources acquired during the lifetime of the Collaborative Computational Project
No. 7 for the Analysis of Astronomical Spectra (Je ery 1996).
1

2 Je ery
spectral databases
image databases
Database
exploration
exploration
Database
Parameter space
exploration
Visualisation
structure modelling
spectral modelling
atomic databases
exploration
Database
AstroGrid
Data modelling
model libraries
Figure 1. Schematic view of relation between theoretical models,
atomic data and spectroscopic and imaging data in a computational
grid.
2. Interpreting Data
The interpretation of nearly all astronomical phenomena involves tting a pre-
dictive model to quantitative data by some process {  2 minimization or data
inversion, for example. This could involve combining new observations with data
from established archives or new databases. A goal-seeking algorithm uses this
data to constrain a sequence of theoretical calculations. In turn these simula-
tions may need to retrieve data from other sources, atomic data, pre-computed
model atmospheres and so on. Already, data resources are distributed and com-
putations are frequently done remotely from the scientist.
The principal elements of the problem to be solved are the tting of multi-
parameter theoretical models to multi-element observational data. A key ap-
plication is the modelling of astronomical spectra. These may contain several
thousand elements (in frequency, time and space). Spectra are produced in me-
dia in which the temperature, density, velocity and chemical structure must be
determined empirically { these are the `free' parameters in the model. Other ex-
amples include the structure of astrophysical jets, or modelling the light curves
of active galactic nuclei.
The general modelling problem will have many of the following properties.
The parameter space may be very large, but only speci c volumes will be as-
trophysically interesting. The models may be linear in some parameters, but
non-linear in others. Theoretical models require ever increasing cpu, RAM and
data resources at geographically remote locations. The physical sizes of indi-
vidual models or grids of models are increasing, exceeding tens of gigabytes in
many cases. Distributed computer systems on a cluster or intranet allow di er-
ent regions of parameter space to be searched at the same time, but monitoring
and guidance require overall control systems and, possibly, human intervention.
In the classical approach, a single task may combine model construction,
data organisation and goal-seeking algorithms. The increasing sophistication
of the model construction process and the need to outsource data organisation
(e.g. atomic data, satellite archives) suggests that this will not be suфcient.

Computational Tools 3
The overall procedure may need to be broken down into several tasks (Fig. 1),
and distributed across a computational grid.
3. Automated Analysis
A decade ago, classical methods for the ne analysis of a high-resolution stel-
lar spectrum would take several months of e ort; even the process of spectral
classi cation could be time consuming. Multi-object, 2D and echelle spectro-
graphs, high-speed detectors, high spectral resolutions, heterodyne receivers,
and long-running survey programmes are generating ever-increasing volumes of
high-quality astronomical spectra. Gains in S/N provided by large-aperture
ground-based telescopes are complemented in the UV and X-ray by a variety of
space missions (e.g. HST, FUSE, Chandra, XMM-Newton). Recent technologi-
cal developments promise spectral resolutions at X-ray and EUV wavelengths to
rival those routinely provided by UV and optical observatories. Array detectors
and interferometers being developed for the submillimeter will enable the 3-D
exploration of interstellar clouds. Datasets may contain up to tens of thousands
of frequency points, a few hundred positions in 1 or 2 spatial dimensions, tens to
thousands of snapshots in time, or spectra of a few thousand objects from a single
night's observing. Many scienti c goals (e.g. identifying solar analogues in the
Galaxy, measuring lithium abundances in young low-mass stars, distinguishing
brown dwarfs from free- oating planets) require the analysis of large samples.
Establishing precise astronomical dimensions from the study of time-dependent
phenomena (e.g. planets, active galactic nuclei, pulsating stars, binaries) may
require the analysis of long series of spectra with wavelength coverage from the
infrared to the X-ray. Under these circumstances, extracting anything but triv-
ial information on a competitive timescale now demands a revolution in the
methods of spectral modelling.
We have piloted a scheme to automatically extract e ective temperatures,
surface gravity, helium abundance, radial velocity and interstellar reddening
from a series of moderate-resolution spectra of a variable star (Je ery, Woolf
& Pollacco 2001). Extending this to extract chemical information for more
than one spectrum was prohibitive with the resources available. The classical
solution would be to move the problem onto a faster computer. That o ered by
grid technology would be to distribute speci c tasks across a network, allocating
tasks to optimized architectures and available resources. In the face of imminent
data inundation, the scienti c motivation for automating the data modelling
process is compelling.
4. Dynamic Modelling
In addition to observational data, theoretical simulations regularly use libraries
of atomic data and of pre-calculated grids of models. Such data libraries are char-
acteristically di erent to observational data and lie outside the interest of most
contemporary projects. Their signi cance is that in a computational grid, ap-
plications may be designed to access or save data in remote libraries as required,
thereby automating some of the most laborious steps in a scienti c investigation.

4 Je ery
4.1. Atomic and Molecular Databases.
Most astrophysical models require atomic and molecular data. There is a con-
fusing array of experimental and theoretical data of varying quality and com-
pleteness. This costs astronomers dearly in terms of wasted time and resources
and may often result in a awed analysis of an astronomical spectrum which may
have been obtained at great expense. The solution is to develop methods using
metadata to allow an astronomer or, better, an application to automatically ac-
cess the most appropriate databases around the world in real time. These would
select the best available data for a particular problem and take advantage of the
very latest, state-of-the-art measurements or calculations. New tools would be
required to access databases of varying formats and to handle large quantities
of data eфciently. Standards for nomenclature, units and le formats would
be required. Where search tools fail to nd speci c atomic data, an automatic
request (e.g. from an R-matrix calculation) could be generated and used to up-
date relevant databases. Thus astrophysical requirements could lead to directly
to the generation of required atomic data `on demand'.
4.2. Model libraries.
In general, astrophysical models have a much shorter lifetime than astronomical
observations; obsolescence is an inevitable result of progress. However, the com-
putational cost of producing many models remains large compared with that of
data storage and remote retrieval. When computed for a suфciently general vol-
ume of parameter space, precomputed grids of models provide a vital resource
for interpreting data. A simple example is the Kurucz database of model struc-
tures, ux distributions and synthetic colours for stellar atmospheres (Kurucz
1991 and subsequent), which have found applications throughout astronomy.
These might be supplemented by databases of high-resolution ux and intensity
spectra for normal stars. Similar databases might subsequently be generated
with non-standard compositions or improved atomic data.
Applications in a computational grid need to know about such model li-
braries and have tools to explore them and to introduce new data, especially
where these extend the volume of parameter space represented by existing mod-
els.
5. Project Realisation
The ideas outlined in this poster are challenging, but they are widely recognised
as important to the future of astronomical research. The project team are
exploring means and resources by which their goals may be realised.
References
Je ery C.S. 1996, QJRAS, 37, 39
Je ery C.S., Woolf V.M., & Pollacco D. 2001, A&A 376, 497
Kurucz R.L. 1991, Precision Photometry: Astrophysics of the Galaxy, ed. A.G.Davis
Philip, A.R.Upgren, & K.A.Janes (L.Davis Press, Schenectady)