harald paganetti (geant4) monte carlo benchmarking
Post on 21-Dec-2015
229 Views
Preview:
TRANSCRIPT
Harald Paganetti
(Geant4) Monte Carlo benchmarking(Geant4) Monte Carlo benchmarking
Why benchmarking ?
1. We are in Medical Physics !2. Monte Carlo is considered to be the gold standard3. There are various competing codes/algorithms
G4N
AM
UM
arch
200
6
4. We want confidence in our code
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory• MC intercomparison
B Benchmarking based on ‘complex’ data• Comparison with experimental data• Comparison with theory• MC intercomparison
G4N
AM
UM
arch
200
6
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory
Benchmarking done bydevelopers (and users)
G4N
AM
UM
arch
200
6
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory• MC intercomparison
cross section data need to be accessible (MCNPX ?)
G4N
AM
UM
arch
200
6
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory• MC intercomparison
B Benchmarking based on ‘complex’ data• Comparison with experimental data• Comparison with theory• MC intercomparison
Benchmarking done by users only (!)
We can all contribute (e.g., Poon, Paganetti)
G4N
AM
UM
arch
200
6
G4N
AM
UM
arch
200
6
G4N
AM
UM
arch
200
6
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory• MC intercomparison
B Benchmarking based on ‘complex’ data• Comparison with experimental data• Comparison with theory• MC intercomparison
G4N
AM
UM
arch
200
6
Often notsufficient !
Often toouser specific !
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory• MC intercomparison
B Benchmarking based on standard (‘complex’) data• Comparison with experimental data• Comparison with theory• MC intercomparison
Standard benchmarking problems should be simple (used by multiple users in various codes; easy error analysis)
G4N
AM
UM
arch
200
6
Cu Cu Cu….. …..
A B C
Multi-Layer Faraday Cup (MLFC)
p+
p+
char
ge
e.g., (p,pn) charge : 0 (+p; -recoil) 1 (+p)
p+
p+
n
80 %
Clean benchmark for nuclear models:
separated nuclear buildup region device has 100% acceptance for
charged secondaries technique measures charge, not
dose (no problems of dosimeter linearity and response to particle types)
7.6 cm x 7.6 cm
G4N
AM
UM
arch
200
6
Multi-Layer Faraday Cup (MLFC)
p+
Cu Cu Cu….. …..ch
arge
p+
p+
n
A B C
CH2 CH2CH2….. …..p+
p+
n
A B
e.g., (p,pn)charge : 0 (+p -recoil) 1 (+p)
can be used for high and low-Z !
e.g., (p,pn) charge : 0 (+p; -recoil) 1 (+p)
G4N
AM
UM
arch
200
6
Types of Monte Carlo benchmarking
A Benchmarking based on basic data• Comparison with experimental data• Comparison with theory• MC intercomparison
B Benchmarking based on standard ‘complex’ data• Comparison with experimental data• MC intercomparison
We need standard benchmarking problems
G4N
AM
UM
arch
200
6
The Computational Medical Physics Working Group (CMPWG) is an international group dedicated to the pursuit of better computational tools in medical and health physics applications. CMPWG consists of individuals from the American Nuclear Society (ANS), American Association of Physicists in Medicine (AAPM), Society of Nuclear Medicine (SNM) , Health Physics Society (HPS) among others.CMPWG is hosted by two divisions within the American Nuclear Society.
• Mathematics and Computations Division (MCD) • Biology and Medicine
Upcoming Conferences • April 2006 - ANS RPSD 2006 - Carlsbad, New Mexico• June 2006 - ANS Annual Meeting - Reno, Nevada• July 2006 - AAPM 48th Annual Meeting - Orlando, Florida• October 2006 - First European Workshop on Monte Carlo Treatment Planning
G4N
AM
UM
arch
200
6
Benchmarks
Contribute a benchmark problemSubmitted problems
• Radiation TherapyBrachytherapyExternal Beam TherapyProton Therapy
• Imaging• Nuclear Medicine• Health Physics
Too complex !
G4N
AM
UM
arch
200
6
G4N
AM
UM
arch
200
6Submit only successful simulations to
CMPWG and report unsuccessful ones to Geant4-developers !
Discussion:
• We should compare cross section data used by various codes
WHO ?
• We need to design a set of (simple) benchmarking problems
WHO ?
G4N
AM
UM
arch
200
6
top related