The Astrophysical MUltiscale Software Environment (AMUSE)

mewstennisSoftware and s/w Development

Nov 4, 2013 (3 years and 7 months ago)

63 views

The Astrophysical MUltiscale
Software Environment (AMUSE)

P
-
I: Portegies Zwart

Co
-
Is: Nelemans, Pols, O’Nuallain, Spaans

Adv.: Langer, Tolstoy, Hut, Ercolano, de Grijs,
Mellema, Spurzem, Bischof, Quillen

AMUSE

The objectives of AMUSE


More science with existing software


Combine existing astrophysical codes


This is a technical problem


It is technically possible


Impression of how it works

Existing codes


Excellent single
-
physics codes exist


hydro


gravity


radiation


stellar evolution


All written in different languages, different
format, different architecture....


Need a homogeneous environment for utilizing
these resources

More science with existing code


Universe is multi
-
physics ...


Scientific objectives:


dense stellar systems (hydro+gravity+stellar evo.)


evolution of galactic environments, star formation,
AGN, ... (hydro+gravity+radiation)


planet formation (hydro+gravity+radiation)


galaxy formation and interaction
(gravity+hydro+radiation+stellar evo.)


Single physics software solutions exist, try to
combine existing codes

This is a technical problem


No new physics needed


Combining requires understanding of how
software and computer hardware interacts


Development to a usefull toolbox requires
professional engineering


Requires substantial manpower

It is technically feasible


Developing new code not optimal because


it is a time consuming task


large codes tend to become unmanageable


initial assumptions tend to require redesign at a late
stage in the development process


Combining existing code via wrapper has been
tried, and works


Propose homogeneous software framework,
à

la Numerical Recipes

Flow control layer (scripting language)

Gas dynamics

Radiative

transport

Stellar evolution

Stellar dynamics

Interface layer (scripting and high level languages)

Smoothed

particles

hydrodynamics

Metropolis

Hastings

Monte Carlo

Henyey

multi
-
shell

stellar evolution

4
th

order Hermite

block timestep

N
-
body

AMUSE

Limitations and Merits

-

Only problems whose physics are expressible
through module coupling (different time scales)

-

Low and high level use possible

-

Radiative transfer (and stellar evolution)
module links to VO (through eg. ‘spiegel’ and
‘partiview’): dust and stellar continuum, atomic
and molecular lines;
ELT, JWST, ALMA, Herschel

Impression of how it works

A)

install

B)

suite of test applications

C)

design your own multi
-
physics problem

D)

write script

E)

run

F)

analyze data

G)

download package from website

H)

write Nature paper

Design/Performance


AMUSE module must be written in language
with Foreign Function Interface (C, C++, Fortran
as well as high level languages like C#, Java,
Haskell. Low level applications optimized.


Top level uses a scripting language. These are
slow, but do just I/O, GUI, call sequence.


Top level can run in parallel (using MPI, GRID
technology); data exchange through HDF


Low level can run in parallel or on dedicated
hardware (eg GRAPE or GPU for direct N
-
body)

Initial Applications


Young and dense star cluster


Evolution of gas and stars near a black hole in
a galactic nucleus


Dynamics of embryonic planets in a debris disk

Relation to other projects


Different concept but with similar scientific
objectives/physics:


FLASH


Gadget


Starlab


Comparable in setup but with different scientific
objectives:


Atmosphere/Ocean/Tectonic
simulations by NASA


Molecular dynamics


QUESTIONS?

management/development plan


programmers under daily supervision of
software engineer and PI


regular interaction with
postdoc, who protects
scientific objectives

The cost


6
-
year of programming effort (3x2years?)


2 years of software engineering


2 years of postdoc


travel, webservices, hardware, etc.


total cost: 640Keuro


NOVA request: 500kEuro