Density Functional Theory Approach to Nuclear Fission
Extracted source text ยท arXiv 1212.3356
\documentclass{appolb}
\usepackage{graphicx}
\usepackage{amsmath}
% graphicx package included for placing figures in the text
%------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% %
% BEGINNING OF TEXT %
% %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\title{Density Functional Theory Approach to Nuclear Fission
\thanks{Presented at the XLVII${\text{th}}$ Zakopane Conference on Nuclear Physics}
}
\author{N. Schunck
\address{Physics Division, Lawrence Livermore National Laboratory, CA 94551, USA, }
}
\maketitle
\newcommand{\gras}[1]{\boldsymbol{#1}}
\begin{abstract}
The Skyrme nuclear energy density functional theory (DFT) is used to model
neutron-induced fission in actinides. This paper focuses on the numerical
implementation of the theory. In particular, it reports recent advances in
DFT code development on leadership class computers, and presents a detailed
analysis of the numerical accuracy of DFT solvers for near-scission calculations.
\end{abstract}
\PACS{21.60.Jz, 21.10.-k, 21.30.Fe, 21.65.Mn, 24.75.+i}
\section{Introduction}
In spite of successful applications in energy production and national security,
relatively little is known about the fundamental mechanisms of the nuclear fission
process. Even today, the most widely-used theories of fission rely on the
macroscopic-microscopic approach to nuclear structure and semi-classical dynamics
based on the Langevin equations. These methods are quite powerful but, in the
long term, cannot be expected to yield the predictive power needed to understand
the fission of very neutron-rich nuclei in the fission-recycling stage of the
formation of elements, or to give enough accuracy for precise simulations of new
generations of nuclear reactors.
Already in the 1980ies, promising attempts were made to understand fission in a
microscopic framework based on the self-consistent nuclear mean-field theory with
effective pseudo-potentials \cite{[Ber84],[Ber89]}. At the time, the computing
power was not sufficient for these approaches to compete with more empirical
models, but these pioneer works yielded a lot of insight on the quantum mechanics
of fission. In the recent years, the rapid development of leadership-class
computers scaling to hundreds of thousands, and soon millions, of processing units
has given us for the first time the computing power needed to successfully
implement the full microscopic theory of fission. In parallel, novel forms of
scientific collaborations gathering nuclear theorists, applied mathematicians and
computer scientists have considerably improved our ability to utilize such
large-scale machines \cite{unedf}.
The goal of this paper is to provide some tools to verify and validate DFT
simulations of nuclear fission. In particular, we pay special attention to the
problem of the numerical accuracy of DFT solvers at very large deformations.
After a brief reminder on the theoretical framework, we recall some of the
recent developments in DFT calculations on leadership class computers, then
give a detailed analysis of truncation errors arising in DFT implementations
using the one-center finite harmonic oscillator basis.
\section{Theoretical Framework}
The nucleus is described in the local density approximation at the Hartree-Fock
Bogoliubov (HFB) approximation. The particle-hole channel is modeled by effective
pseudo-potentials up to second-order derivatives in the density. In practice,
this is equivalent to using zero-range effective interactions of the Skyrme
type \cite{[Dob95],[Dob96b]}. The results presented below were thus obtained with
the SkM* parameterization of the Skyrme interaction \cite{[Bar82]}. The
particle-particle channel is characterized by a density-dependent contact
interaction with mixed volume and surface character \cite{[Dob02]}. An energy
cut-off of $E_{\text{cut}} = 60$ MeV is used to reduce the number of quasi-
particles in the definition of the densities. The HFB equations are solved in a
one-center harmonic oscillator basis.
In the DFT picture of nuclear fission, the HFB energy of the nucleus depends on
an ensemble of collective variables $\gras{q} = (q_{1}, \dots, q_{N})$. These can
be, for example, variables describing the nuclear shape, excitation or spin. In
this work, we considered as collective variables the expectation value (on the
HFB ground-state) of the multipole moments $\hat{Q}_{\lambda\mu}$. In practice,
the axial $\hat{Q}_{20}$ and triaxial $\hat{Q}_{22}$, as well as the mass octupole
$\hat{Q}_{30}$ and hexadecapole $\hat{Q}_{40}$ moments were considered. The
collective space is thus four-dimensional. Expectation values of $\hat{Q}_{\lambda\mu}$
will simply be denoted $Q_{\lambda\mu} \equiv \langle \hat{Q}_{\lambda\mu}\rangle$.
In the actinide region, the part of the potential energy surface relevant to
nuclear fission spans a rather large range in deformations. The axial quadrupole
moments runs typically from $\sim 30$ b in the ground-state to nearly 600 b at
scission for symmetric fission; the octupole moment from 0 to about 70 b$^{3/2}$
for very asymmetric fission (cluster radioactivity, see \cite{[War11],[War12]});
the hexadecapole moment from nearly $\sim 3$ b$^{2}$ near the ground-state to
typically more than $\sim 350$ b$^{2}$ for symmetric fission. Assuming for sake of
simplicity a uniform sampling of each degree of freedom, and a $1$ b$^{\lambda/2}$
mesh size, the size of the collective space is more than 1.4 millions points, only
for the axial collective variables. Adding triaxiality multiplies this estimate by
another 2 orders of magnitude.
\section{Large-Scale Potential Energy Surfaces}
In this section, we present the performance of our DFT solver, and provide a
detailed analysis of convergence properties of our Skyrme HFB calculations in
the case of $^{240}$Pu.
\subsection{DFT Solvers on Leadership Class Computers}
All calculations were performed with the DFT solvers HFBTHO \cite{[Sto12]}
and HFODD \cite{[Sch11]}. Both codes solve the HFB equations in the harmonic
oscillator (HO) basis. The program HFBTHO assumes axial and time-reversal
symmetry, while HFODD is fully symmetry-unrestricted. The two programs have
been benchmarked against one another and agree within a few eV for an axial
configuration \cite{[Sto12]} .
\begin{figure}[htb]
\begin{center}
\includegraphics[width=0.8\linewidth]{hfodd_scaling_Titan_defs.eps}
\caption{Performance of the DFT solver HFODD on the Titan supercomputer at the
Oak Ridge Leadership Computing Facility in Oak Ridge. Each experiment measures
the time of performing six full HFB iterations. The term 'deformation point'
refers to a set of constraints on multipole moments, i.e. a point in the
collective space.}
\label{fig:titan_scaling}
\end{center}
\end{figure}
Owing to the block-diagonal structure of the HFB matrix induced by axial
symmetry, the typical runtime of HFBTHO is of the order of the minute
(depending on the type of nucleus, size of the basis and 'difficulty' of
converging the HFB iterations). By contrast, it is of the order of several
hours for HFODD. In practice, HFBTHO is used as pre-conditioner for HFODD:
for any given point $\gras{q}$ in the collective space, the HFB iterations
are first solved with HFBTHO, and the densities at convergence are used to
initialize HFODD. If the point of the collective space is axial, no additional
iterations are therefore needed.
Clearly, the very large size of the collective space together with the current
runtime of the DFT solvers requires using today's most powerful supercomputers.
A lot of effort was, therefore, devoted to porting our codes to leadership
class computers, and ensuring that good scaling with the number of processing
units could be achieved. From a computational point of view, mapping the nuclear
collective space in DFT is a naturally parallel problem. The code HFODD has,
therefore, a hybrid MPI/OpenMP programming model, where points in the collective
space are distributed across the MPI grid, and on-node multi-threading enables
to take advantage of highly optimized linear algebra libraries. Figure
\ref{fig:titan_scaling} shows the performance of HFODD on the Titan supercomputer
at the Oak Ridge Leadership Computing Facility. In this experiment, up to 300,000
processors were used in parallel. The slight degradation of the computing time
is due to the original input/output backend, which has not been optimized and
taxes the operating system at large scale.
\subsection{Numerical Accuracy}
Modeling nuclear fission requires to explore regions of the collective
space with extreme deformations. The finite size of the HO basis may thus lead
to a significant dependence of the results on the basis parameters. In our
calculations, we only considered axially-deformed bases, characterized by
a quadrupole deformation $\beta$ and an oscillator frequency $\omega_{0}$.
The maximum number of shells is denoted by $N_{\text{max}}$ and the maximum
number of states by $N_{\text{states}}$. In the case of a full spherical HO
basis, the two are related through the well-known relation
$N_{\text{states}} = (N_{\text{max}}+1)(N_{\text{max}}+2)(N_{\text{max}}+3)/6$.
This relation is not valid anymore for a deformed basis. In practice, we
introduce two cut-offs, one on the number of shells and another on the number
of states. Note that the cut-off on the number of states is only the practical
consequence of using a symmetry-unrestricted solver, for which the size of
the matrices involved goes approximately as $2N_{\text{max}}^{3}$. By contrast,
the block structure induced by the built-in axial symmetry in HFBTHO would
enable to consider all full shells up to $N_{\text{max}}$.
Basis dependence of the calculations thus comes from 4 parameters (i) the
maximum number of shells $N_{\text{max}}$, (ii) the maximum number of states
$N_{\text{states}}$, (iii) the oscillator frequency $\omega$ and (iii) the
basis deformation $\beta_{2}$. In principle, at every point $\gras{q}$ in
the collective space, we should seek the HFB solution that is the minimum in
this 4-dimensional parameter space. Clearly, such a strategy is not sustainable
even on today's largest computers. Instead, one is bound to estimate truncation
errors by exploring the parameter space locally, and extracting asymptotic expressions.
\begin{figure}[!sht]
\center
\includegraphics[width=0.8\linewidth]{publi_basis_omega.eps}
\caption{(color online) Convergence of the HFB energy as function
of the oscillator frequency $\omega_{0}$, for two configurations characterized
by $Q_{20} = 30$ b (black squares) and $Q_{20}=300$ b and $Q_{40} = 120$
b$^{2}$ (red circles). The deformation $\beta$ is adjusted according to the
formula (\ref{eq:beta}).
}
\label{fig:HObasis1}
\end{figure}
As a first example, we show in figure \ref{fig:HObasis1} the dependence of
the HFB energy on the spherical-equivalent frequency of the harmonic oscillator
$\omega_{0}$, that is the frequency such that $\omega_{0}^{3} =
\omega_{x}\omega_{y}\omega_{z}$. The deformation of the basis is fixed at
each point according to the formula (\ref{eq:beta}). Two typical points in
the collective space are considered, one near the ground-state with deformation
$Q_{20} = 30$ b, one way past the second barrier on the descent to scission
at $Q_{20} = 300$ b and $Q_{40} = 120$ b$^{2}$. We note that the dependence
on $\omega_{0}$ is more marked at large deformations, and that the optimal
frequency shifts toward smaller values as the deformation increases, which
is consistent with the need to then include basis states with a larger
spatial extension. Importantly, it is possible to extend this analysis and
extract an empirical fit $\omega_{0}(Q_{20})$ giving the optimal basis frequency
as function of the quadrupole moment of the collective point. In our tests, we
found that the expression
\begin{equation}
\omega_{0} =
\left\{ \begin{array}{l}
0.1 \times Q_{20}e^{-0.02 Q_{20}} + 6.5\; \text{MeV}\ \text{if}\ |Q_{20}| \leq 30\; \text{b} \\
8.14\; \text{MeV} \ \text{if}\ |Q_{20}| > 30\; \text{b}
\end{array}
\right.
\label{eq:omega}
\end{equation}
gives a reasonably accurate fit of the frequency as function of the quadrupole
moment.
\begin{figure}[!sht]
\center
\includegraphics[width=0.8\linewidth]{publi_basis_deform.eps}
\caption{(color online) Convergence of the HFB energy as function of the basis
deformation $\beta$. The oscillator frequency is adjusted according to the
formula (\ref{eq:omega}). For the second configuration, an additional
constraint on $\hat{Q}_{60}$ to $Q_{60} = 150$ b$^{3}$ was added.
}
\label{fig:HObasis2}
\end{figure}
The dependence of the HFB energy on the deformation of the basis, for a fixed
$N_{\text{max}}$ and basis deformation, is illustrated in figure
\ref{fig:HObasis2}. Not surprisingly, the minimum is always obtained for basis
deformation that are 'close' to the requested value of the axial quadrupole
moment. Note that the dependence on deformation is rather marked. However, as for
the oscillator frequency, it is {\it a priori} possible to obtain a fit
$\beta(Q_{20})$ such that the optimal basis deformation is chosen at point in the
collective space. Our tests showed that the simple formula
\begin{equation}
\beta = 0.05\sqrt{Q_{20}}
\label{eq:beta}
\end{equation}
provides a reasonable expression that remains applicable up to the largest
values of $\hat{Q}_{20}$.
\begin{figure}[!ht]
\center
\includegraphics[width=0.8\linewidth]{publi_basis_states.eps}
\caption{(color online) Convergence of the HFB energy as function of the number
of states $N_{\text{states}}$.
}
\label{fig:HObasis3}
\end{figure}
Last but not least, we show in figure \ref{fig:HObasis3} the error induced
by the truncation of the number of states. Contrary to the previous two
parameters of the HO basis, this truncation is imposed by memory limitations,
and can not really be mitigated: for a given value of $N_{\text{max}}$, the
optimal number of states is always given by
$N_{\text{states}} = (N_{\text{max}}+1)(N_{\text{max}}+2)(N_{\text{max}}+3)/6$,
a number that can grow very large for large $N_{\text{max}}$ For example, at
$N_{\text{max}} = 30$, we have $N_{\text{states}} = 5456$. Taking into account
the spin degree of freedom, the total number of basis states is more than 10,000,
which implies that the size of the HFB matrix exceeds 20,000 $\times$ 20,000. At
this time, it is not possible to handle in a reasonable time frame iterative
processes involving dense, complex matrices in double precision of that size. As
can be seen from figure \ref{fig:HObasis3}, restricting $N_{\text{states}}$ to
manageable values around $N_{\text{states}}\approx 1000$-$1200$ may easily lead
to 2 to 3 MeV errors beyond the second fission barrier.
To finish this section, we would like to emphasize two important consequences of
basis truncation effects:
\begin{itemize}
\item At constant truncation, the error increases with deformation, albeit not
necessarily linearly. This is bound to have a very significant impact for, e.g.,
calculations of barrier penetrability, since the numerical precision is not the
same at the entry and exit points, and errors of the order of the MeV can lead to
orders of magnitude uncertainties on fission lifetimes;
\item Truncations magnify the impact of discontinuities in the potential energy
landscape. In practice, calculations with very different basis characteristics
initialized with similar density/wave-functions could converge to two
very different points of the multi-dimensional PES. This effect is the reason
why, in the lower panel of figure \ref{fig:HObasis2}, an additional constraint
on $\hat{Q}_{60}$ had to be added: without it the calculation did not converge to
the same point in the collective space at small and large basis deformations.
\end{itemize}
\section{Conclusions}
The nuclear energy density functional theory is currently the only viable option
to achieve a microscopic description of nuclear fission. On-going development
of leadership class computing facilities all over the world offer a unique
opportunity to finally develop the nuclear DFT at very high precision. In this
work, we have discussed some of the numerical uncertainties associated with
implementations of DFT in the one-center harmonic oscillator basis. They clearly
point to the need of developing bases that are better adapted to the extreme
elongations characterizing the region near scission.
\section{Acknowledgments}
\label{sec:acknowledgments}
Discussions with W. Younes are warmly acknowledged. Support for this work
was provided in part through Scientific Discovery through Advanced Computing
(SciDAC) program funded by U.S. Department of Energy, Office of Science,
Advanced Scientific Computing Research and Nuclear Physics. It was partly
performed under the auspices of the US Department of Energy by the Lawrence
Livermore National Laboratory under Contract DE-AC52-07NA27344. Funding was
also provided by the United States Department of Energy Office of Science,
Nuclear Physics Program pursuant to Contract DE-AC52-07NA27344 Clause
B-9999, Clause H-9999 and the American Recovery and Reinvestment Act, Pub.
L. 111-5. An award of computer time was provided by the Innovative and Novel
Computational Impact on Theory and Experiment (INCITE) program. This
research used resources of the Oak Ridge Leadership Computing Facility
located in the Oak Ridge National Laboratory, which is supported by the
Office of Science of the Department of Energy under Contract DE-AC05-00OR22725.
It also used resources of the National Energy Research Scientific Computing
Center, which is supported by the Office of Science of the U.S. Department
of Energy under Contract No. DE-AC02-05CH11231.
\bibliographystyle{unsrt}
\bibliography{zakopane}
\end{document}
\begin{thebibliography}{10}
\bibitem{[Ber84]}
{J.-F. Berger, M. Girod, and D. Gogny, Nucl. Phys. A {\bf 428}, 23c (1984).}
\bibitem{[Ber89]}
{J.-F. Berger, M. Girod, and D. Gogny, Nucl. Phys. A {\bf 502}, 85c (1989).}
\bibitem{unedf}
{http://www.unedf.org}.
\bibitem{[Dob95]}
{J. Dobaczewski and J. Dudek, Phys. Rev. C {\bf 52}, 1827 (1995).}
\bibitem{[Dob96b]}
{J. Dobaczewski and J. Dudek, Acta Phys. Pol. {\bf B27}, 45 (1996).}
\bibitem{[Bar82]}
{J. Bartel, P. Quentin, M. Brack, C. Guet, H.-B. H\r{a}kansson, Nucl. Phys. A
{\bf 386}, 79 (1982).}
\bibitem{[Dob02]}
{J. Dobaczewski, W. Nazarewicz, and M. V. Stoitsov, Eur. Phys. J. A {\bf 15},
21 (2002).}
\bibitem{[War11]}
{M. Warda, and L.M. Robledo, Phys. Rev. C {\bf 84}, 044608 (2011).}
\bibitem{[War12]}
{M. Warda, and J.-L. Egido, Phys. Rev. C {\bf 86}, 014322 (2012).}
\bibitem{[Sto12]}
{M. Stoitsov, N. Schunck, M. Kortelainen, N. Michel, H.A. Nam, J. Sarich, and
S. Wild, nucl-th/arXiv:1210.1825 (2012).}
\bibitem{[Sch11]}
{N. Schunck, J. Dobaczewski, J. McDonnell, W. Satu{\l}a, J.A. Sheikh, A.
Staszczak, M. Stoitsov, P. Toivanen, Comp. Phys. Comm. {\bf 183}, 166
(2012).}
\end{thebibliography}