This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON NUCLEAR SCIENCE 1

Radioactive Decays in Geant4

Steffen Hauf, Markus Kuster, Matej Batič, Zane W. Bell, Dieter H. H. Hoffmann, Philipp M. Lang, Stephan Neff,
Maria Grazia Pia, Georg Weidenspointner, and Andreas Zoglauer

Abstract—The simulation of radioactive decays is a common [2], see Section IV-A ) or the statistical outcome of many decays
task in Monte-Carlo systems such as Geant4. Usually, a system (e.g., MCNP [3], [4] and FLUKA [5]).
either uses an approach focusing on the simulations of every Whereas the first approach may be inefficient if the individual
individual decay or an approach which simulates a large number
of decays with a focus on correct overall statistics. The radioac- decay is not of interest, the latter approach does not allow for
tive decay package presented in this work permits, for the first the physically correct simulation of an individual decay and its
time, the use of both methods within the same simulation frame- associated effects. General purpose Monte Carlo codes would
work—Geant4. The accuracy of the statistical approach in our benefit from the capability of providing both approaches in the
new package, RDM-extended, and that of the existing Geant4 same environment, in response to the simulation requirements
per-decay implementation (original RDM), which has also been
refactored, are verified against the ENSDF database. The new of different experimental scenarios.
verified package is beneficial for a wide range of experimental In the following a software package for the simulation of ra-
scenarios, as it enables researchers to choose the most appropriate dioactive decay, which realizes both approaches for Geant4, is
approach for their Geant4-based application. presented. This package includes a refactored implementation
Index Terms—ENSDF, Geant4, Monte-Carlo Simulation, ra- of the existing Geant4 per-decay approach [6], and extends the
dioactive decay, validation. functionality of Geant4 radioactive decay simulation by a novel
implementation based on a statistical approach. It is based on
the ENSDF (Evaluated Nuclear Structure Data File) data library
I. INTRODUCTION [7], which was chosen due to its widespread usage in the nuclear
science community.

R ADIOACTIVE decays and the resulting radiation play an This paper reports on the verification of both implemented
important role for many experiments, either as an observ- approaches against a large set of evaluated data. To the best of
able, as a background source, or even as a potential hazard when the authors’ knowledge, such a thorough verification of Geant4
they are a source of radiation-induced damage for hardware and radioactive decay simulation has not yet been documented in the
human beings. Detailed knowledge of the radiation inside and literature. The experimental validation of the software for both
around an experiment and its detectors is thus required for a suc- approaches is reported on in a separate paper [8].
cessful outcome of the experiment and to guarantee the safety
II. RADIOACTIVE DECAY PHYSICS
of the operator. The increasing complexity of experiments often
makes it prohibitively expensive, if not impossible, to com- Radioactive decay is a physical process where an atomic nu-
pletely determine the radiation characteristics and response of cleus of an unstable atom transmutes into a lower energy state
an experiment from measurements alone. In order to circum- by spontaneous emission of ionizing radiation. The process does
vent these limitations, it has become increasingly important to not require external interactions to occur. It results from either
estimate an experiment’s radiation and response characteristics nucleus-internal processes or interactions of the nucleus with
with the help of computer simulations. (inner) shell electrons. A brief overview of the main physics
General-purpose Monte-Carlo simulation codes either focus of radioactive decay is summarized here to facilitate the com-
on the correct simulation of individual decays (e.g., Geant4 [1], prehension of the functionality of the software described in this
paper.
Different types of radioactive decay are commonly identified
Manuscript received January 31, 2013; revised June 03, 2013; accepted June
13, 2013. This work has been supported by Deutsches Zentrum für Luft- und according to the type of emitted particles.
Raumfahrt e.V. (DLR) under grants 50 QR 0902 and 50 QR 1102. • During an -decay a He–nucleus is emitted from the parent
This work was supported by the Deutsche Zentrum fuer Luft– und Raumfahrt
nucleus. This results in a daughter nucleus with two fewer
(DLR) under Grant number 50QR902 and 50Q1102.
S. Hauf and M. Kuster are with European XFEL GmbH, Hamburg, Germany protons and two fewer neutrons than the parent nucleus.
(e-mail: steffen.hauf@xfel.eu). • The -decay is a weak process during which a neutron
D. H. H. Hoffmann, P. M. Lang, and S. Neff are with Institute for Nuclear
is converted into a proton. An electron and anti-neutrino
Sciences, TU Darmstadt, Darmstadt, Germany.
M. Batič and M. G. Pia are with the INFN Genova, Genova, Italy. are emitted by the parent nucleus: consequently, the atomic
Z. W. Bell is with the Oak Ridge National Laboratory, Oak Ridge, TN, USA. number of the daughter nucleus increases by one and the
G. Weidenspointner is with the Max-Planck Halbleiter Labor, Munich, Ger-
atomic mass number stays constant. The electron and anti-
many and the Max-Planck Institut für extraterrestrische Physik, Garching,
Germany. neutrino share the energy released during the decay. Since
A. Zoglauer is with the Space Science Laboratory, University of California, both particles are not bound in their final state, their energy
Berkeley, CA, USA.
distribution follows a continuous spectrum.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. • During a -decay a bound proton of a nucleus is con-
Digital Object Identifier 10.1109/TNS.2013.2270894 verted into a neutron. A positron and a neutrino are emitted

0018-9499/$31.00 © 2013 IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

by the parent nucleus; the atomic number decreases by material properties after irradiation: for instance, of a sample ir-
one and the atomic mass number stays constant. Similar radiated by a neutron beam.
to -decays, both particles are not bound in their final The study of activation and build-up of radioactive nuclei is
state, and accordingly, their energy distribution follows a relevant to various applicative domains: nuclear reactors (fis-
continuous spectrum. sion and fusion), particle accelerators or intense light sources,
• If a daughter nucleus is left in an excited state, after a where the statistical effect of many decays and activations is
transmutation by the previously mentioned decay types, it relevant. It also concerns space-borne X-ray and -ray instru-
can deexcite by emitting -radiation. In case the excited ments: in these scenarios meta-stable states usually must be ac-
daughter state is a long-lived (metastable) state, its deex- counted for, and the statistical effect of many decays and acti-
citation is called isomeric transition, which will also result vations is relevant. Additionally, individual decays contribute
in -radiation. In both cases the atomic number and atomic to the prompt instrument background. The estimation of the
mass number remain unchanged. in-orbit cosmic-ray induced background is important for space-
• During an electron capture, the parent nucleus absorbs based detectors to distinguish individual radioactive emission
an inner shell electron (usually K- and L-shell electrons) from intended observables.
and simultaneously emits a neutrino. During this process, Low background astro-particle physics experiments are
which is also called inverse -decay, a proton is trans- concerned with the influence of the cosmic-ray induced back-
muted into a neutron, thus the atomic number decreases ground and natural radioactivity: here accurate simulation of
by one and the atomic mass number stays constant. In the individual decays’ spatial and temporal distribution can be
contrast to a -decay, an electron capture is a two-body important.
decay, resulting in a discrete neutrino energy.
• As an alternative process to -emission, an excited nu-
IV. FOUNDATIONS FOR A RADIOACTIVE DECAY SIMULATION
cleus can return to its ground state by transferring its ex-
citation energy to one of the lower shell electrons of the
atom. This process is called internal conversion, and re- A. Requirements
sults in the emission of an electron by the atom, leaving The simulation of radioactive decays consists of the task
the atom in an excited state. The electron carries a dis- of decaying an unstable nucleus and generating the resulting
crete fraction of the decay energy, and by this is distinct products.
from -particles with continuous energy spectra. As with The decay of the parent nucleus should proceed according to
-decays, no transmutation of the nucleus takes place, and the physical parameters governing it: decay type, initial excita-
both the atomic number and atomic mass number remain tion and half-life time. A daughter nucleus should be produced,
unchanged. with the physical properties resulting from the decay: atomic
Radioactive decay is a stochastic process. The time at which number and mass determined by the decay type, excitation en-
a given unstable atom decays is not predetermined; instead de- ergy and kinetics determined by parent kinetics and decay ki-
cays occur with a certain probability. In consequence, experi- netics. Secondary particles and radiation associated with the
ments will measure statistical observables such as the amount decay (e.g., -, - and -emission, neutrinos and conversion
of ionizing radiation of a certain type and energy emitted within electrons) should be generated.
a given time period. The software should handle the deexcitation of the daughter
Due to practical impossibility of calculating all relevant atom, involving the production of - and X-rays, and of Auger-
parameters from theory, the simulation of radioactive decay electrons.
physics for a large number of decays or decay chains relies on Theoretical calculations of the required parameters are not
the usage of empirical or precalculated data. practically feasible in the course of the simulation; therefore the
algorithm must use empirical or precalculated data.
Radioactive decays of nuclei are often associated with prior
III. EXPERIMENTAL SCENARIOS
activation of stable nuclei of a given material; examples for
Many experiments measure the time-accumulated statistical such applications include shielding analysis and radiation safety
distributions of observables such as energy, type, momentum analysis. This scenario requires the ability of dealing with acti-
and timing of radiation resulting from radioactive decays. Often vation, and thus replenishment of nuclei, within the simulation.
the radioactive decay products are not the intended observable Meta-stable states and isomeric transitions should be taken into
but contribute to the experiment’s or application’s measured account.
data as background. In addition to these functional [9], physics-induced require-
Radioactive decay physics modeling plays an important role ments, the software should take into account nonfunctional
in various fields; the overview summarized here is not intended ones, which derive from the experimental context of the simu-
to be exhaustive. lation: the possibility to efficiently simulate a large number of
Measurement and analysis of the properties of decay chains decays, when the physical accuracy of the individual decay is
of naturally abundant radioactive isotopes is commonly per- of less importance, but overall statistics are relevant; the possi-
formed in material sciences, radiation safety and nuclear pro- bility of efficiently simulating decay chains, when intermediate
liferation monitoring. Radioactive decay modeling is also of in- products in a chain are of lesser interest; the provision of user
terest to experimental scenarios involving the measurement of input.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 3

If radioactive decay simulation occurs in the context of a MCNP(X) [3], [4] does not include a full radioactive decay
more general Monte Carlo simulation system, the software re- simulation by default except for the generation of delayed
sponsible for the radioactive decay process should interact with -rays resulting from decays sampled from MCNP’s photon
other components of the system. data library (phtlib). Instead MCNP can be linked via scripts to
specialized codes such as ORIGEN2 [17] or CINDER90 [18],
B. Problem Domain Analysis [19]. Both of these use their own data libraries. These codes
Software objects with specialized responsibilities collaborate are generally used to model reactor fuel cycles and accelerator
to satisfy the requirements mentioned in Section IV-A. For the induced transmutations, but also provide functionality for
code presented in this work the division of responsibilities is as simulating radioactive decays and decay chains. Due to the
follows: codes’ nature, they include replenishment through activation.
• data management; The - and -emission needs to be generated from tabulated
• sampling of the (discrete) emission resulting from the in- user input. As a result, MCNP is specialized on the simulation
dividual decay and generation of the daughter nucleus; of many decays and the resulting statistics.
• calculation of the -emission spectrum; FLUKA [5] generates and transports - and -radiation,
• calculation of the number of nuclei within a decay but only started including -radiation in its latest release.
chain—which may include activation; Decay chains are possible and include replenishment through
• the user interface; activation. FLUKA uses its own data libraries, largely based on
• the interface with the Monte-Carlo code. NNDC (National Nuclear Data Center) data and thus ENSDF.
Similar to MCNP, the emphasis lies on the simulation of a large
V. ENSDF DATA number of decays.
The sampling of radioactive decays will usually rely on the In Geant4 [1], [2] radioactive decays are treated on a per-
use of empirical or precalculated data. The Evaluated Nuclear decay level, based on data taken from ENSDF [7]. Decay chains
Structure Datafile (ENSDF) [7] is one such collection of data, including activation are possible and produce the associated
which for instance contains information on half-life times, decay emission. The - and -emission are sampled from the
decay types, branching ratios, emission energies and transition decay database. Deexcitation radiation of the daughter nucleus
types. It is maintained by the National Nuclear Data Center is not produced by the radioactive decay simulation itself, but
(NNDC) and distributed, amongst others, by the International by other physics processes included in Geant4, which use their
Atomic Energy Agency (IAEA). respective databases. The emphasis lies on the per-decay simu-
ENSDF is an evaluated library, i.e., it contains data from ex- lation, not on the sampling of a large number of decays.
periments and theoretical calculations which are recommended None of the above mentioned Monte-Carlo codes allows the
for use as a reference after a critical analysis of uncertainties, simulation of either individual decays or a statistical treatment
inter- and extrapolation methods and underlying models has of many decays in the same software environment.
been performed. IAEA defines it as the master library for evalu-
ated experimental structure and decay data [10]. Other special- VII. RADIOACTIVE DECAY IN GEANT4
ized libraries and bibliographic databases, such as NuDAT [11], A package for the simulation of radioactive decays [6], [20]
CINDA [12] and NSR [13], exist as well, but are commonly de- has been available in Geant4 since version 2.0, where it was
rived from or related to ENSDF. Therefore ENSDF is frequently named radiative_decay. Since Geant4 version 6.0 it has been
considered the de-facto standard for nuclear structure and radi- named radioactive_decay, although it is conventionally known
ation data. as the Geant4 RDM (Radioactive Decay Module). This code
For -ray intensities and energies the ENSDF data usually was originally developed by P. Truscott and F. Lei; it imple-
consists of evaluated measurements. Conversion electron inten- ments per-decay sampling.
sities and energies are derived from theory. The following discussion is based on the radioactive decay
However, the data present in ENSDF are not sufficient to code of Geant4 9.4p04, but is also pertinent to subsequent ver-
calculate all atomic deexcitation emissions. Specifically, data sions 9.5 and 9.6, the latter one being current at the time of
on electron binding energies, fluorescence and Auger-electron submission of this paper. In these latter versions the problem
yields are necessary to calculate the respective intensities. of not producing fluorescence emission in case of decays other
To mitigate this problem, analysis programs distributed with than electron capture has been addressed and the handling of
ENSDF, such as RADLST [14], are supplied with the necessary forbidden -decays has been added, but all other features and
information taken from data of Bearden and Burr [15] and problems of the implementation mentioned in this work are still
Bambynek et al. [16]. In order to stay consistent with these valid.
ENSDF-related programs, the aforementioned data are also A unified modeling language (UML) [21] class diagram of
used to derive quantities for the radioactive decay database of the RDM code in Geant4 9.4p04 is shown in Fig. 1. It shows the
the RDM-extended package and the verification data used in cooperation between the different classes in the code. This dia-
this work. gram is supplemented by the activity diagram shown in Fig. 2.
The two diagrams highlight problems [22] inherent in the code’s
VI. RADIOACTIVE DECAY IN MONTE CARLO CODES design.
Models for the simulation of radioactive decays exist in most • Since each decay type is defined by distinctive physics,
general-purpose Monte-Carlo codes. it would make sense to implement the individual types
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

Fig. 1. UML class diagram of the original RDM, which implements per-decay sampling, showing the relations between the individual classes. The colors (grey
shades) reflect the responsibilities and are matched to the activity diagram in Fig. 2: physics simulation (blue/medium grey) and decay chain processing (purple/
dark grey). Yellow/light grey objects can not be attributed to a single responsibility. Data retrieval, initialization and management is provided by the main class
(G4RadioactiveDecay). The individual decay channels initialize a G4NuclearDecayChannel object which includes the physics of all decay types in one class.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 5

Fig. 2. UML activity diagram of the original RDM, which implements per-decay sampling showing the workflow of the code when decaying a nucleus. The
color (grey shade) scheme is matched to the class diagram in Fig. 1 and reflects different activity types: physics simulation within the radioactive decay code
(blue/medium grey), physics simulation by other processes (red/darkest grey) and decay chain calculations (purple/dark grey). Yellow/light grey objects can not
be attributed to a single responsibility.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

as separate objects. Instead, in the original Geant4 RDM G4RadioactiveController (data-management) and G4Decay-
package all decay types are implemented together in the ChainSolver (decay chains).
G4NuclearDecayChannel class. The decay type classes The RDM-extended package also respects encapsulation
merely provide an interface to this class with decay type- rules. Furthermore, physics functionality implemented in the
dependent initialization parameters. This complicates unit different classes can be combined as needed, thus allowing
tests of individual components. Additionally, a much larger both sampling approaches to use a common code base for
amount of code has to be checked if, for instance, an error functionality required by both.
is found for one decay type. The addition of a statistical sampling approach is reflected in
• Whereas the decay physics for each type is distinct, the in- the activity diagram: the initial configuration and library data ac-
terface to each type is similar: all decay type objects should cess are common for both design approaches. They differ after
have a method to produce decay emission and a daughter a decay channel is chosen: in the per-decay approach fluores-
nucleus. Such an interface could be provided by a common cence emission is sampled for electron capture decays by del-
base class. In the existing design the G4RadioactiveDecay egating responsibility to the G4AtomicDeexcitation class. At a
class needs to know the implementation and interface de- later point the deexcitation emission of the nucleus is sampled
tails of each individual decay type. by using the G4PhotonEvaporation class. If the statistical ap-
• Objects should have one specific responsibility, e.g., the proach is chosen instead, all photon emission and discrete elec-
interface to the data library. This is not the case: the tron emission processes are sampled simultaneously at the end
G4RadioactiveDecay class is responsible for initializing of handling decay physics.
and loading values from the data libraries, initializing the
decay types and the variance reduction and decay chain
handling. Such a design again enlarges the fraction of VIII. PER-DECAY SAMPLING
code which needs to be checked for a specific error or
maintained for the update of a specific responsibility. Ad- Per-decay sampling is already present in Geant4. This orig-
ditionally, due to inadequate domain decomposition, two inal RDM code provides the required functionality: therefore
distinct responsibilities—the simulation of radioactive the general approaches to physical treatment and data handling
decay and event biasing—are mixed in the same class. were preserved, but refactored before their inclusion into the
• In case of the -Fermi-function implementation the inter- RDM-extended in order to conform to the design discussed in
face is implementation-dependent. Changing the algorithm the previous section. This refactored per-decay sampling is con-
may thus also involve changing the interface—and as a re- sistent with the original RDM. Additionally, small errors in the
sult all other code parts depending on this class. physical treatment were addressed. As a result the structure of
A new package, named RDM-extended, has been developed the code is different from the original one, but the inherent func-
to address existing issues of Geant4 RDM, and to extend and tionality is conserved.
improve the capabilities of radioactive decay simulation in a Per-decay sampling is based on reprocessed ENSDF data
Geant4-based environment. both in the original RDM and the RDM-extended package.
The software design of the RDM-extended package follows For the RDM-extended we reprocessed the data using a parser,
an object-oriented programming approach with clear responsi- which implements the ENSDF format as given in the ENSDF
bility definitions. For this design the relevant entities and re- manual [7]. The Geant4 classes handling nuclear and atomic
quirements for radioactive decay physics have been identified. deexcitation in the per-decay approach use their own libraries.
Decay type-dependent approaches have been considered along- For the G4PhotonEvaporation class (nuclear deexcitation) this
side common tasks for all decay types. The design also takes into
is ENSDF-based, with conversion electron probabilities com-
account that two sampling methods are to be handled within a
piled from [23]–[25]. The G4AtomicDeexcitation class (atomic
common framework, and that the external classes the code de-
deexcitation), which has been previously validated in [26] and
pends on may be subject to interface changes.
[27], uses EADL [28] data. Concerns about the accuracy of the
The UML diagrams in Figs. 3 and 4 document that the respon-
EADL data have been reported in [29].
sibilities of objects are clearly defined and that functionality is
neither duplicated nor aggregated into nonspecialized classes. The electron capture probabilities included in both radioac-
An example for implementing functionality only once tive decay libraries are not given in ENSDF directly. They need
is the class G4RandomDirection, which provides function- to be calculated using an additional data source, which gives
ality for sampling random particle momentum vectors to the fluorescence yields and binding energy information. For con-
classes modeling the individual decay types and the statistical sistency reasons the atomic data file distributed with ENSDF is
sampling. used for the RDM-extended package, which is based on data by
The G4RadioactiveDecay class of the original RDM (Fig. 1) Bearden and Burr and Bambynek et al. [15], [16]. This also fa-
is an example of a nonspecialized class: it is responsible cilitates comparisons with ENSDF-based online-databases such
for physics simulation, data preparation and decay chain as NuDat, which use the same data.
handling. In contrast the G4RadioactiveDecay class of the Further refactoring is foreseen to use a package for atomic
RDM-extended is a pure management class, which coordinates data management exploiting the results of [30] and an improved
the interaction of specialized classes for the aforementioned package for the simulation of atomic relaxation, which are cur-
tasks: the different emission classes (physics simulation), rently under development. These packages are intended to
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 7

Fig. 3. UML class diagram for the RDM-extended package implementing a refactored per-decay sampling and a statistical sampling method. The colors (grey
shades) reflect the responsibilities and are matched to the activity diagram in Fig. 4: physics simulation (blue/medium grey), data management and preparation
(orange/medium dark grey) and decay chain processing (purple/dark grey). Each decay type class is responsible for simulating the type’s characteristic physics.
The deexcitation and fluorescence emission for the statistical simulation approach is produced by the G4VDecayEmission base class.

satisfy requirements common to the simulation of radioactive A. Energy Conservation

decay and of electromagnetic interactions.
The radioactive decay database of both codes is supplied in For a physically correct outcome energy needs to be con-
plain text on a per isotope basis. served for the decay, all decay emissions and when deexciting
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

Fig. 4. UML activity diagram for the RDM-extended package implementing a refactored per-decay sampling and a statistical sampling method, showing the
workflow of the code when decaying a nucleus (top) and handling of decay chains (bottom). The color (grey shade) scheme is matched to the class diagram
in Fig. 3 and reflects different activity types: physics simulation within the radioactive decay code(blue/medium grey), physics simulation by other processes
(red/darkest grey), data management and preparation (orange/medium dark grey) and decay chain calculations (purple/dark grey). The emission produced by the
statistical decay approach is highlighted in green/medium light grey.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 9

the daughter nucleus. Nuclear deexcitation is handled by the Similarly, the momenta of any fluorescence and Auger-emis-
G4PhotonEvaporation class, which uses a reprocessed ENSDF- sion are sampled by the G4AtomicDeexcitation class and should
based data library different from the radioactive decay class. also be taken into account as vectorial quantities. This is not the
Since it is not guaranteed that the level-energies in the two data case in the original Geant4 RDM code: here the difference of all
libraries are exactly the same, a possible deviation has to be produced emissions’ summed kinetic energies and the binding
taken into account. This is done in the original Geant4 RDM energy of the innermost vacated shell are added to the kinetic
on three levels. energy of the daughter nucleus. Whereas the computation of
• If the level energy passed by the radioactive decay code scalar values might result in a marginal performance increase, it
does not correspond to a tabulated level in the G4Pho- neglects that the emission is not unidirectional, but isotropic. A
tonEvaporation library, the nearest level present in the full vectorial treatment is the physically accurate solution and is
G4PhotonEvaporation database is used to retrieve the thus implemented in the RDM-extended package.
possible transitions. The initial deexcitation step occurs
from the energy passed by the decay code and will transi- IX. STATISTICAL SAMPLING
tion to an energy tabulated in the deexcitation library. The The statistical sampling approach discussed in this section
energy of the first -ray will thus not be in accordance is a novelty for Geant4-based radioactive decay simulation.
with the tabulated evaporation data, but deviate by the The necessity for such an approach has become evident as a
difference between the tabulated radioactive decay and consequence of the experimental requirements identified in
evaporation library values. Further transitions will then Section III. Statistical sampling in this context means that the
result in emission at energies in accordance with the full simulation of an individual decay is considered of lesser
G4PhotonEvaporation data. importance as long as the emission and nuclei produced by
• For all transitions along a deexcitation chain it is checked if many decays on average lead to a physically correct result.
the level resulting from the transition is within a tolerance Because all decay emission is treated as independent of each
other and the intensity of each emission is known, sampling
of 1 keV of the ground state. If this is the case, the photon
generally reduces to the problem of repeated random number
energy is set to the energy of the excited level and the
generation within the intensity range of . Additionally,
nucleus is deexcited to the ground state. Again energy is
one has to take into account that intensities may exist,
conserved, but the energy of the emitted photon may not
if both nuclear and atomic deexcitation result in radiation at the
be in accordance with the tabulated data.
same energy. In the latter case at least emissions
• The final state energy is passed back to the radioactive will be generated for every decay.
decay code. In case this energy is not 0, but below 1 keV, This approach does not require to take allowed or forbidden
the daughter’s excitation energy is set to 0 keV. Here en- level transitions into account during nuclear deexcitation since
ergy is not conserved. Otherwise the decay code outputs an the order of occurrence of transitions is considered irrelevant,
excited (possibly meta-stable) daughter nucleus as defined as long as for a large enough number of transitions, each oc-
by the data library. curs with the correct probability, in result yielding the correct
The first two conservation treatments are necessary if slightly intensity.
divergent data libraries exist. They are handled internally by the These intensities are included in ENSDF alongside normal-
G4PhotonEvaporation class for all processes which delegate to ization information and as such all required decay information
it, and have thus not been altered as part of the radioactive decay may be retrieved from a single consistent library. Accordingly,
code development. Rectifying these divergencies would require energy mismatches between libraries do not occur and no inter-
a consolidation of the Geant4 data libraries, which exceeds the dependencies between different physics classes exist. As a con-
scope of this work. The last approach to energy conservation is sequence, an alteration of tabulated library values will manifest
physically unjustifiable and accordingly has been corrected in itself in a straight-forward and immediate fashion in the simu-
the refactored implementation. In case of an energy mismatch, lated decay intensities and energies.
the energy of an excited nucleus is deposited in the geometrical The conservation of energy and momentum is simplified, as
volume it is located in. Whereas this does not adress the library the information of momenta and energy of all decay emissions
inconsistency, which leads to the energy mismatch, it does con- is known to the radioactive decay simulation. Accordingly, it
serve energy and is accordingly seen as the preferable solution. can be taken into account in a physically meaningful fashion.
In consequence, statistical sampling is a more efficient,
straight-forward, ENSDF-consistent and performant approach
B. Momentum Conservation and Recoil of Nucleus for simulating radioactive decays for the majority of ex-
perimental applications which do not require knowledge of
Like energy, momentum should be conserved during the com- individual decays.
plete decay simulation. In the per-decay approach nuclear deex-
citation is delegated to G4PhotonEvaporation, which also gen- X. SAMPLING-METHOD-INDEPENDENT FUNCTIONALITY IN
erates the momenta of the -rays and conversion-electrons in a THE RDM-EXTENDED RADIOACTIVE DECAY CODE
deexcitation cascade. The radioactive decay simulation is then In the previous two sections a distinction between the per-
responsible to account for the resulting recoil of the daughter decay and statistical sampling approach was made. For a com-
nucleus. plete solution to radioactive decay physics, in accordance with
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

the requirements, sampling schemes for -emission and decay set is computed for the first time, a tabulated energy distribu-
chains are necessary. These are approach-independent as they tion is calculated, from which the -particle energy is drawn.
occur in both sampling scenarios. Future occurrences of the same parameter set will draw particle
energies directly from this distribution, thereby minimizing pro-
A. -Fermi-Function – Sampling of the Continuous cessing time. Venkataramaiah et al. found this approximation
-Spectrum to be accurate within a one-percent margin of error, when com-
-decay is an unbound three-body decay and the emitted radi- pared to the tabulated values of Rose [32]. Because the interface
ation will thus have a continuous spectrum. This spectrum can of the -Fermi-function class is independent of the model im-
be sampled using the -Fermi-function, which also takes cor- plemented therein, it is easily possible for the user to substitute
the RDM-extended’s -Fermi-function approximation, by one
rections resulting from the interaction of the charged nucleus
which is has better accuracy for a certain isotope.
with the -particle into account.
The parameters passed to the -Fermi function should be
physics-relevant, but code-independent. In particular this means B. Decay Chains and Activation
that, instead of passing a binning scheme dependent energy, as
is done in the original Geant4 RDM implementation, the phys- Isotopes resulting from a radioactive decay are often unstable
ical parameters and endpoint energy , as well as the decay themselves. This leads to chains of subsequent decays until a
type ( or ) and optionally forbiddenness, are passed in the stable daughter product is reached. Often it is desirable to only
RDM-extended implementation. This parameter set is sufficient fully simulate those isotopes in a chain which are of interest as
for many Fermi-function approximations, like to those summa- observables, either due to their half life or due to the radiation
rized by Venkataramaiah et al. in [31]. Currently, the compu- emitted when they decay. In such a case a full Monte-Carlo sim-
tationally most performant approximation given therein is used ulation of every decay is very inefficient. The sampling of inter-
in the RDM-extended package for calculating the Fermi correc- esting isotopes only requires determining the number of nuclei
tion factor for an isotope with atomic number with of a given species present in a chain at a specific time.
the total energy (given in MeV): Additionally, radioactive isotopes may be created by nuclear
activation, often as a result of proton or neutron collisions with a
(1) nucleus. While the collisions and resulting activation is handled
by the hadronic processes of Geant4 [33], bookkeeping of the
activation buildup and the decay of the created unstable nuclei
The constants and were determined by Venkataramaiah et are tasks for the radioactive decay simulation. In the RDM-ex-
al. through linear regression using the data from Rose [32], and tended the activation simulation is also capable of altering the
were found to satisfy: material composition on a per-volume basis to the isotope com-
position present at discrete user-defined time steps.
(2a) For a system of nuclei, where the th nuclei decays into
(2b) the ( )th nuclei of the chain, this calculation can be done
by solving a system of coupled differential equations with the
with general form:

and
..
(3) .
(5)
with

for with being the quantity of the th nucleus at time and

for containing information on the decay constant and a
for constant particle number dependent activation rate .
for . A general approach for solving (5) for any number of prod-
ucts was first derived by Bateman [34] and the above equations
The correction factor is then input into the -Fermi have as such become known as the Bateman equations.
function One of the main disadvantages of the original Bateman
solution is that it does not consider branching, i.e., if a parent
(4) nucleus can decay to different daughter nuclei via different
decay types. Many computational algorithms using Bateman’s
where is the end-point energy of the -spectrum, obtained approach or a derivative thereof exist, with the one imple-
from the tabulated radioactive decay data library, is the total mented in the original radioactive decay code [6], being just
energy of a -particle with momentum . If a given parameter one example which overcomes the branching limitation and
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 11

includes activation. These algorithms are generally compu- as described by Onega. In (13) is a square matrix with the
tationally expensive, in the sense that they require recursive th column consisting of the th eigenvector of , so that
loops for each nucleus in a decay chain, but may achieve good . is its inverse and a diagonal matrix
overall performance, if the calculations can be reused. with the elements being the th eigenvalue of .
An alternative is the algebraic approach derived by M. M. Amaku et al. then derive an algorithmic approach for cal-
Amaku et al. in [35], building upon work by R. J. Onega [36], culating the matrices , and , which is computationally
D. Pressyanov [37], L. Moral and A. F. Pacheco [38], as well less expensive and more accurate than a general approach of
as T. M. Senkov [39] and D. Yuan and W. Kernan [40]. In this numerically calculating the matrix elements. The elements of
approach the properties governing the decay chain are written can be calculated with the recurrence expression
into a matrix in Hessenberg-form

(14)
.. .. (6)
. .
for , and . Similarly the
elements of are given by
and a vector with

(15)
.. (7)
.
for and . Using and , is given by

The system of (5) may then be written as (16)

(8) resulting in a general form for 15 of

This original algebraic approach was introduced by Onega in

1969 and extended by Yuan and Kernan as well as Semkow to ..
include branching by modifying to .

(17)
In the RDM-extended code the replenishment and generation
.. .. ..
. . . of unstable isotopes through activation is calculated using the
(9) above equations on a per-volume level. In a first simulation
.. .. .. . . run, a bookkeeping class G4ActivationBookkeeping keeps track
. . . . of all activations occurring in a given volume and calculates
the new material and isotope composition of the volume at a
with given time. In a second run, the material properties are then up-
dated within the geometry, and the volumes are set as radioac-
(10) tive background sources, with the calculated activity via the
G4RadDecayVolumeBookkeeping class. This class samples the
for and radioactive background emission, which should be generated
alongside a primary particle and pushes it on the event stack.
(11) In this way minimal user intervention is required.
In (10) denotes the branching ratio from the th to the th
component of the decay chain with . In compu- XI. VERIFICATION OF SAMPLING METHODS
tational practice can be easily constructed by iterating through
The radioactive code implementations have been verified
the decay chain and taking activation rates into account, if nec-
for consistency with ENSDF data. Since ENSDF is established
essary. Because is independent of the actual nuclei numbers, it
must only be constructed once for each set of nuclei and activa- in the experimental community as an authoritative reference,
tions characterizing a given chain. Calculations using different comparison of Monte Carlo models against it provides valuable
nuclei numbers can reuse these initial matrices as needed. Using information for the users of these codes. The results presented
the above definitions, the number of nuclei of each species in the here are chosen to highlight the physics-performance improve-
decay chain is then given by ments of the statistical sampling in the RDM-extended package
when compared to the per-decay sampling of the original
(12) Geant4 RDM code. Accordingly, the refactored per-decay
which can be rewritten to code, which has been verified to produce equivalent results to
the approach used in the original Geant4 RDM, is not detailed
(13) further.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

In addition to the -ray and conversion electron emission di-

rectly given in ENSDF, the verification of intensities and en-
ergies of fluorescence and Auger-emission with the Bambynek
et al. data [16] has been included.

A. Simulated Data Production

The Geant4 simulations consisted of decays of an un-
stable nucleus in an otherwise empty geometry. Any resulting
unstable daughter nuclei were not decayed further. The kinetic
energy of radiation and particles resulting from these decays
was recorded separately for each radiation type ( , , , non-
electrons) into two histograms: the first histogram ranging from
0—30 keV with a binning size of 0.05 keV, the second his-
Fig. 5. Percentage of simulated emission within a given deviation from the
togram ranging from 0—30 MeV with a binning size of 0.2 keV. simulation input.
The two binning schemes were chosen in order to properly dis-
tinguish discrete radiation in the X-ray and –energy regime.
Using the above approach, data for 2910 parent excitation
levels of isotopes were simulated using the original RDM code
with radioactive decay database version 3.3 and the per-decay
simulation. The statistical approach simulations included 3040
parent isotopes and excitation levels, i.e., 130 more than the
sample used for the production with per-decay sampling. The
numbers differ due the different versions of ENSDF data and
different parsers used for reprocessing. For both sets of simula-
tions -emission was distinguished from Auger- and conversion
electron emission.

B. Evaluated Data Preparation

The verification data were extracted from ENSDF into a tabu-
lated form suitable for further analysis using the RADLIST [14]
program. By using this ENSDF-supplied parser it was assured
that no errors were introduced into the evaluated data during the
automated extraction. Fig. 6. Distribution of the relative error of the observed deviations between
simulated and evaluated radiation intensities with respect to the level intensity.
C. Data Analysis The contour levels correspond to the percentile of values at given deviation error
with respect to the number of values at a given level intensity. At higher inten-
The automated, comparative analysis takes the level and par- sities values are more sparse resulting in the observed ”single box” contours.
ticle energies from the evaluated data as an initial energy esti-
mate or for where emission may be present
in the simulated data. The simulated data histograms are then calculated from the total number of the events in these bins. The
scanned in an interval of at energies below 30 evaluated data can indicate that emissions at a different energy,
keV and at higher energies for emission events. but in proximity to the energy currently being processed may
Data outside these windows was considered not to belong to the have also contributed to the number of simulated events. In this
currently compared emission energy. case the summed intensities of all emissions in neighboring
In case of discrete emission contained in a single bin the in- bins are used for further comparisons. The energy position of
tensity of the emission is given by these emissions is then set to the median energy of the events,
accordingly its uncertainty is the median deviation.
These steps provide energy and intensities of the radiation
(18)
from the evaluated and simulated data. Absolute and relative
energy and intensity deviations can then be calculated. In this
The energy uncertainty is determined by the bin size. The inten- way overall comparisons (see Section XII) and comparisons of
sity uncertainty is with the event number individual isotopes, such as those measured in [8], are possible.
per bin and the simulated number of decays . In order to assess the consistency of the simulation model, a
Should the emission be distributed into (multiple) neigh- complete set of simulations using the statistical approach was
boring bins, which is possible due to recoil from previous compared to the simulations’ input data. Ideally, one would not
emissions transferred onto the emitting nucleus, the intensity is expect any energy deviations at all and the intensity deviations
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 13

Fig. 7. Nuclide charts showing the median relative intensity deviations per isotope for - (top), conversion-electron (middle) and -emission (bottom). Simula-
tions using the per-decay approach of the original Geant4 RDM are shown on the left; simulations using the RDM-extended statistical sampling method are shown
on the right.

should be compatible with the statistical error of the simula- XII. RESULTS: CONSISTENCY OF THE RADIOACTIVE DECAY
tion. In practice the simulations’ output is binned spectra, in CODES WITH ENSDF
order to keep total data amounts at a manageable level. Due to In the following we focus upon identifying global trends and
this, in rare cases the emission at nearby energies may not be exploring the regions in which the radioactive decay simulations
distinguished properly, which in turn leads to intensity devia- gives reliable results. A validation with measured data can be
tions. As is apparent from Fig. 5, 99% of all simulated emis- found in [8].
sions are within an intensity deviation of 3% of the simulation For the comparisons presented in the following, knowledge
input. of the comparisons’ uncertainties is important. Fig. 6 shows a
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

14 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

Fig. 8. Nuclide charts showing the median relative intensity deviations per isotope for X-ray (top) and Auger-electron emission (bottom). Simulations using the
per-decay approach of the original Geant4 RDM are shown on the left; simulations using the RDM-extended statistical sampling method are shown on the right.

compilation of the intensity deviation uncertainties, which de- amount to for -rays and for
pend on the uncertainty of the evaluated intensity for a given -emission when using the per-decay code.
level and the statistical uncertainty of the simulated data. As is For the statistical approach the deviations are minimal, with
apparent from the figure, even at low intensities the uncertain- for -rays and for -emis-
ties are in a range below 10% for 99% of the data points. sion. The outliers ( discrepancy) in the simulation using
the RDM-extended package can be explained by the way the
A. Intensity Deviations ENSDF data are parsed into the radioactive data library. In cases
where multiple datasets describing the same emission exist, the
In order to help readers quickly identify the simulations’ in- implemented parser is tuned to pick experimentally determined
tensity discrepancy for isotopes occurring in their simulation, data or, if such a distinction is not possible, use the first data
it was chosen to display the results as nuclide charts (a colored set. The RADLIST program preferably uses theoretical data ac-
online version is available). cording to the documentation [14].
Fig. 7 shows a comparison of the intensity discrepancy of It is further apparent from Fig. 7 that the original per-decay
the original Geant4 RDM per-decay sampling alongside the dis- code does not reproduce the ENSDF intensities of conversion
crepancy of the statistical approach of the RDM-extended when electrons well. This manifests itself in a mean deviation of
compared to ENSDF data. As is apparent from the figures, both compared to for the statistical
sampling methods reproduce the - and -emission intensities approach.
given in ENSDF within a few percent deviation for the majority For the original Geant4 RDM per-decay code the - and
of isotopes. Specifically, the mean deviations ( conversion electron intensity deviations must be attributed
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 15

Fig. 9. Distribution of absolute energy deviations with respect to the radiation energy for all level energies. The contour levels correspond to the percentile of
values at a given energy deviation with respect to the number of values at a given radiation energy. At higher energies evaluated values are more sparse resulting in
fractured contours. Simulations using the per-decay approach of the original Geant4 RDM are shown on the left; simulations using the RDM-extended statistical
approach are shown on the right.

Fig. 10. Distribution of absolute energy deviation with respect to the radiation energy for level energies below 25 keV. The contour levels correspond to the
percentile of values at a given energy deviation with respect to the number of values at a given radiation energy. Simulations using the per-decay approach of the
original Geant4 RDM are shown on the left; simulations using the RDM-extended statistical approach are shown on the right.

to the G4PhotonEvaporation model or more specifically its ciated EADL data library. It is interesting to note that the devi-
underlying data library [26]. A comprehensive verification and ations are largest for the isotope on the left of the nuclide chart,
validation of this process would be beyond the scope of this i.e., isotopes which decay via electron capture. This substanti-
paper and was thus not undertaken. ates the conclusion that the EADL data library is the source of
For fluorescence and Auger emission the deviation between the deviations, as it is only called by the per-decay approach
evaluated ENSDF intensities and those produced by the original if electron capture decays need to be sampled (in Geant4 9.4,
RDM per-decay approach are even larger, as is apparent from changed in Geant4 9.5). Again the extended-RDM’s statistical
Fig. 8. For X-ray emissions the deviation amounts to method yields results with smaller deviations, which amount to
; for Auger-electron to . Again not the for X-ray emission and for
radioactive decay code alone is responsible for these offsets but Auger electron emission. This is a more than 10-fold improve-
its interplay with the G4AtomicDeexcitation class and its asso- ment in intensity consistency with respect to the per-decay ap-
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

16 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

proach of the original Geant4 RDM. A similar magnitude of

deviations is reported in [29] for EADL data, in which dif-
ferent binding energy data libraries were compared with refer-
ence data. In this work EADL consistently showed the largest
deviations.
It should be stressed that the observed intensity deviations
for the per-decay approach result mainly from the incoherence
of the data libraries involved in the sampling of nuclear deex-
citation and fluorescence emission with the ENSDF-database,
from which the reference data were derived. For the statistical
approach the deviations are naturally smaller, as all radioactive
decay data are derived from a single ENSDF-library and sup-
plementary atomic data files.

B. Energy Deviations
Most application scenarios will depend on the correct sam- Fig. 11. decay chain simulated for performance testing. The initial iso-
tope of the chain was varied from to to test different chain lengths.
pling of the energy of the - or X-ray radiation and that par-
ticles as well as Auger- and conversion-electrons are emitted at
the experimentally determined energies. From the comparisons
shown in Figs. 9 and 10, one can conclude that at energies in
the X-ray and Auger-electron range less than 30 keV, a devia-
tion of less than 0.2 keV with respect to ENSDF data is to be
expected for all radiation, when using the original Geant4 RDM
per-decay method. Statistical sampling is again more consis-
tent with ENSDF data. Here the observed energy deviation is
less than 0.1 keV for the majority of emissions. For -radiation,
-particles and conversion electrons at higher energies, devia-
tions of less than 0.5 keV are observed with the per-decay ap-
proach. The statistical approach shows deviations at or less than
the bin size of 0.2 keV.
Fig. 12. Computational performance change when using the RDM-extended
XIII. COMPUTATIONAL PERFORMANCE package with the statistical approach in comparison to the original Geant4 RDM
Computational performance is a significant aspect for large and the per-decay approach. Each simulation consisted of 100 000 decays and
was repeated 200 times. Three different sampling techniques are shown: sta-
scale Monte-Carlo simulations. The RDM-extended package tistical approach, classical approach and classical approach without postdecay
was designed keeping this in mind (Section X-B). The imple- fluorescence production. This last approach resembles the original Geant4 RDM
mentation reflects this performance optimization: for instance, implementation.
it uses hash-maps (std::unordered_map) from the C++11-stan-
dard [41], which allow element access times which scale with
the number of elements in , instead of , as would decay chain shown in Fig. 11 was simulated. In order
be the case for traditional std::map containers (see e.g., [42]). In to estimate the performance for different decay chain lengths,
order to estimate the performance gain achievable using this im- different initial nuclei in this chain were chosen.
plementation in comparison to the original RDM, the following All performance tests were done on a 12-core XEON ma-
performance tests were undertaken: chine at 2.93 GHz running Ubuntu 10.10 Maverick Meerkat
• decay an isotope 100 000 times using the refactored per- and an identical application based on Geant4 9.4p04, which
decay and the statistical approach; was unaltered except for the decay code. The gcc 4.4.5 com-
• decay a chain of isotopes and retrieve all emission which piler with Geant4 standard compiler flags and the extensions
has occurred from all isotopes in the chain within a sam- of the C++11-standard enabled was used for compilation. The
pled time period of , which could be, for instance, an radioactive decay was the only physics simulation process
experiment’s measurement duration, 10 000 times; included in the test environment, thus guaranteeing that no
• decay a chain of isotopes and retrieve the emission from other processes, which the involved particles might be subject
a single isotope within the chain occurring within a time to during further tracking, would influence the results.
period 10 000 times. The relative performance of the RDM-extended statistical
Each of the above tests was repeated 200 times to ensure sampling method shown in Fig. 12 asserts that this method is –
that temporary CPU load from the operating system or other in almost all cases more than 20% – faster than the per-decay ap-
processes would not bias the results. For the first test the proach implemented in the original Geant4 RDM. An exception
isotopes , , and are isotopes which have a large number of deexcitation emis-
were decayed, constituting an example for each of the particle sions such as . Here the linear increase of sampling time
emitting decay types. As an example of a full decay chain, the with the number of emissions results in a performance penalty.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

HAUF et al.: RADIOACTIVE DECAYS IN GEANT4 17

simulated. In the original RDM the user has to filter for the
emission of interest after it has been produced by the decay
code and has been passed to tracking. This is inefficient and
computing time intensive, as can be seen from Fig. 13. An ex-
treme case for such a scenario is when only the emission of the
final isotope in a chain is of interest. This case is documented
in Fig. 13 by the two ”end of chain” data values.

XIV. CONCLUSION
Experimental requirements concerning radioactive decay
simulation have been analyzed and evaluated against ra-
dioactive decay models and functionality available in present
Monte-Carlo codes. It was found that none of the available
codes offers the possibility to correctly simulate physics on
Fig. 13. Computational performance of the RDM-extended package and the the per-decay level and to correctly simulate the statistical
original RDM when decaying the decay chain. The chain length was outcome of many decays without unnecessary overhead within
varied by setting different initial nuclei. The two data points labeled end of chain one framework.
are the performance values for when only the emission of the last isotope in the
chain is of interest. A software package, which addresses these requirements and
allows simulations using both approaches, has been designed,
implemented and verified with respect to the established refer-
When comparing the refactored per-decay approach decay ence of the ENSDF evaluated data library. The RDM-extended
emission production, i.e., delegation to G4PhotonEvapora- package described in this paper reproduces the functionality
tion and G4AtomicDeexcitation, two scenarios have been of the original Geant4 RDM package implementing per-decay
distinguished. In the first case, labeled ”per-decay” in Fig. 12, sampling, although with improved software design and gener-
the vacant shell index, which may be output by the G4Pho- ally faster computational performance. In addition, it encom-
tonEvaporation class at the end of deexcitation, is passed to passes functionality for statistical sampling of radioactive de-
G4AtomicDeexcitation for X-ray and Auger-electron produc- cays: with respect to the per-decay approach of the preexisting
tion. The original Geant4 RDM code does not pass the atomic Geant4 RDM, this approach has been verified to achieve better
shell index, a scenario also simulated with the RDM-extended consistency with ENSDF data, and better computational perfor-
package and labeled ”per-decay w/o fluor.” in Fig. 12. It is mance. Significant consistency improvements have been veri-
apparent from the figure, especially for , that the inclu- fied especially in the X-ray regime.
sion of G4AtomicDeexcitation results in a severe performance The RDM-extended package can be used transparently in
penalty. If a full treatment of X-ray and Auger-electron emis- Geant4-based simulation applications. Its experimental valida-
sion is needed and simulation performance is critical, it is thus tion is documented in a distinct dedicated paper [8].
recommended to use statistical sampling, if the applications
scenario permits it. REFERENCES
For the decay chain performance comparison shown in
[1] S. Agostinelli et al., “Geant4 – A simulation toolkit,” Nucl. Instrum.
Fig. 13 one has to consider that the per-decay approach of Methods., vol. 506, no. 3, pp. 250–303, 2003.
the original Geant4 RDM does not take the duration of the [2] J. Allison et al., “Geant4 developments and applications,” IEEE Trans.
time period to be sampled into account. Instead it is up to Nucl. Sci., vol. 53, pp. 270–278, 2006.
[3] J. F. Briesmeister, MCNP : A General Monte Carlo Code for Neutron
the user to filter for the relevant emission according to the and Photon Transport Los Alamos National Laboratory, Los Alamos,
application scenario. This explains the increase in computing 1986, 3.
time needed by the RDM-extended between as the initial [4] J. S. Hendricks, MCNPX Version 2.5.c. : , 2003, American Nuclear
Society.
isotope and . As is shown in Fig. 11, thorium has a [5] A. Fassò, A. Ferrari, P. R. Sala, and J. Ranft, “FLUKA: Status and
much shorter half life time than uranium. Regardless of which prospects for hadronic applications,” in Advanced Monte Carlo for Ra-
isotope is taken as the initial one in the chain, the simulated diation Physics, Particle Transport Simulation and Applications, F. B.
A. Kling, Ed. : , 2001, p. 955.
time duration for which decays are sampled stays the same at [6] P. Truscott, “Treatment of Radioactive Decay in Geant4,” Tech. Rep.
. During this time, much less uranium Qinetiq, 2002.
than thorium will have decayed. Accordingly, when uranium is [7] J. Tuli, Evaluated Nuclear Structure Data File (ENSDF) 2010 [Online].
Available: http://www.nndc.bnl.gov/ensdf/
chosen as an initial isotope, less emission has to be sampled, [8] S. Hauf et al., “Validation of Geant4-based radioactive decay simula-
than is the case for thorium. This reflects itself in the perfor- tion,” IEEE Trans. Nucl. Sci., 2013, to be published.
mance increase of . [9] Systems and Software Engineering – Life Cycle Processes –Require-
ments Engineering, , 2011, ISO/IEC/IEEE 29148:2011(E).
Similarly, the statistical sampling in the RDM-extended [10] O. Schwerer, Nucl. Data Libraries and Services Provided by the IAEA,
package is faster, if only the emission from selected isotopes Sep. 2001.
in the chain is of interest. In the RDM-extended code only the [11] A. A. Sonzogni, “NuDat 2.0: Nuclear structure and decay data on the
internet,” in AIP Conf. Proc., 2005, vol. 769, pp. 574–577, 1.
emissions from these isotopes are actually sampled and passed [12] CINDA 2000, The Index to Literature and Computer Files on Micro-
to tracking, reducing the number of particles which need to be scopic Neutron Data IAEA, 2000.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

18 IEEE TRANSACTIONS ON NUCLEAR SCIENCE

[13] B. Pritychenko, E. Běták, M. Kellett, B. Singh, and J. Totans, “The Nu- [28] S. T. Perkins, M. H. Chen, D. E. Cullen, and J. H. Hubbell, Tables
clear Science References (NSR) database and Web Retrieval System,” and Graphs of Atomic Subshell and Relaxation Data Derived From
Nucl. Instrum. Methods., vol. 640, no. 1, pp. 213–218, 2011. the LLNL Evaluated Atomic Data Library (EADL),
[14] T. W. Burrows, The Program RADLST Brookhaven National Labora- Lawrence Livermore Nat. Lab., Livermore, CA, 1991.
tory, 1988, Tech. Rep. BNL-NSC-52142. [29] M. G. Pia, H. Seo, M. Batic, M. Begalli, C. H. Kim, L. Quintieri, and
[15] J. A. Bearden and A. F. Burr, “Reevaluation of X-ray atomic energy P. Saracco, “Evaluation of atomic electron binding energies for Monte
levels,” Rev. Mod. Phys., vol. 39, pp. 125–142, Jan. 1967. Carlo particle transport,” Nucl. Sci., IEEE Trans., vol. 58, no. 6, pp.
[16] W. Bambynek et al., “X-ray fluorescence yields, Auger, and Coster- 3246–3268, 2011.
Kronig transition probabilities,” Rev. Mod. Phys., vol. 44, pp. 716–813, [30] M. Batič et al., “Algorithms and parameters for improved accuracy in
1972. physics data libraries,” in J. Physics Conf. Series, Dec. 2012, vol. 396,
[17] A. G. Croff, “ORIGEN2: A versatile computer code for calculating the p. 022039, 2.
nuclide compositions and characteristics of nuclear materials,” Nucl. [31] P. Venkataramaiah, K. Gopala, A. Basavaraju, S. S. Suryanarayana,
Technol., vol. 62, no. 3, 1983. and H. Sanjeeviah, “A simple relation for the Fermi function,” J. Phys.
[18] W. B. Wilson, T. R. England, D. C. George, D. W. Muir, and P. G. G, vol. 11, no. 3, p. 359, 1985.
Young, in Recent Development of the CINDER’90 Transmutation [32] M. E. Rose, Beta-and Gamma-Ray Spectroscopy, K. Sieghahn, Ed.
Code and Data Library for Actinide Transmutation Studies, Versailles, Amsterdam: North Holland Publishing Company, 1955.
France, 1995, Global ’95. [33] D. H. Wright et al., S. R. Magill and R. Yoshida, Eds., “Recent devel-
[19] M.-L. Giacri-Mauborgne et al., “Status of the photonuclear data library opments and validations in Geant4 hadronic physics,” in Calorimetry
for CINDER’90,” in Int. Conf. on Nucl. Data for Sci. and Technol., in High Energy Physics: XII, Oct. 2006, vol. 867, American Institute
2005, vol. 769, American Institute of Physics Conference Series, pp. of Physics Conference Series, pp. 479–486.
195–198. [34] H. Bateman, “The solution of a system of differential equations occur-
[20] P. Truscott et al., “Geant4-a new Monte Carlo toolkit for simulating ring in the theory of radioactive transformations,” in Proc. Cambridge
space radiation shielding and effects,” in Radiation Effects Data Work- Philos. Soc., 1910, vol. 15, p. 423.
shop, 2000, 2000, pp. 147–152. [35] M. Amaku, P. R. Pascholati, and V. R. Vanin, “Decay chain differential
[21] J. Rumbaugh, I. Jacobson, and G. Booch, Unified Modeling Language equations: Solution through matrix algebra,” Comp. Phys. Comm., vol.
Reference Manual, 2nd ed. : Pearson Higher Education, 2004. 181, pp. 21–23, 2010.
[22] M. Fowler, Refactoring: Improving the Design of Existing Code. [36] R. J. Onega, “Radioactivity calculations,” Am. J. Phys., vol. 37, no. 10,
Reading, MA: Addison-Wesley, 1999, Addison-Wesley Object Tech- pp. 1019–1022, 1969.
nology Series. [37] D. S. Pressyanov, “Short solution of the radioactive decay chain equa-
[23] I. Band, M. Trzhaskovskaya, and M. Listengarten, “Internal conversion tions,” Am. J. Phys., vol. 70, pp. 444–445, 2002.
coefficients for atomic numbers ,” Atomic Data and Nucl. Data [38] L. Moral and A. F. Pacheco, “Algebraic approach to the radioactive
Tables, vol. 18, no. 5, pp. 433–457, 1976. decay equations,” Am. J. Phys., vol. 71, pp. 684–686, 2003.
[24] F. Rösel, H. M. Fries, K. Alder, and H. C. Pauli, “Internal conversion [39] T. M. Semkow, “Comment on “Algebraic approach to the radioactive
coefficients for all atomic shells,” Atomic Data and Nucl. Data Tables, decay equations,” by L. Moral and A. F. Pacheco [Am. J. Phys. 71 (7),
vol. 21, no. 2–3, pp. 91–289, 1978. 684-686 (2003)],” Am. J. Phys., vol. 72, pp. 410–411, 2004.
[25] R. Hager and E. Seltzer, “Internal conversion tables Part I: K-, L-, [40] D. Yuan and W. Kernan, “Explicit solutions for exit-only radioactive
M-shell conversion coefficients for to ,” Nucl. Data decay chains,” J. App. Phys., vol. 101, no. 9, p. 094907, 2007.
Sheets. Section A, vol. 4, no. 1–2, pp. 1–11, 1968. [41] SC22/WG21. ISO/IEC 14882: 2011, Information Technology, Pro-
[26] S. Guatelli, A. Mantero, B. Mascialino, M. Pia, and V. Zampichelli, gramming Languages, C++, , 20122, ISO JTC.
“Validation of Geant4 atomic relaxation against the NIST physical ref- [42] M. Han, C.-H. Kim, L. Moneta, M. G. Pia, and H. Seo, “Physics data
erence data,” Nucl. Sci., IEEE Trans., vol. 54, no. 3, pp. 594–603, Jun. management tools: Computational evolutions and benchmarks,” ArXiv
2007. e-Prints, Dec. 2010.
[27] M. G. Pia, P. Saracco, and M. Sudhakar, “Validation of k and l shell
radiative transition probability calculations,” IEEE Trans. Nucl. Sci.,
vol. 56, pp. 3650–3661, 2009.