Presented by
Building Nuclei from the Ground Up: Nuclear Coupled-cluster Theory
David J. DeanOak Ridge National Laboratory
Nuclear Coupled-cluster Collaboration:T. Papenbrock, K. Roche, Oak Ridge National LaboratoryP. Piecuch, M. Wloch, J. Gour, Michigan State University
M. Hjorth-Jensen, Oslo
A. Schwenk, Triumf
Funding: DOE-NP, SciDAC, DOE-ASCR
2 Dean_NCCT_SC07
How do we describe nuclei we cannot measure?• Robust, predictive nuclear theory exists for structure and
reactions.
• Nuclear data needed to constrain theory.• Goal is the Hamiltonian and nuclear properties:
– Bare intra-nucleon Hamiltonian.– Energy density functional.
• Mission relevant to NP, NNSA.
• Half of all elements heavier than iron produced in r-process where limited (or no) experimental information exits.
• Nuclear reaction information relevant to NNSA and AFCI.
“Given a lump of nuclear material, what are its properties, and how does it interact?”
SupernovaSupernova
E0102-72.3E0102-72.3
3 Dean_NCCT_SC07
The Leadership Computing Facility effort will
• Enlarge ab-initio square to mass 100
• Enable initial global DFT calculations with restored symmetries
Pushing the nuclear boundaries
All Regions: Nuclear cross-section efforts (NNSA, SC/NP, Nuclear Energy)
Nuclear DFT effortNuclear DFT effortThermal properties regionsThermal properties regions
Nuclear Coupled Cluster effortNuclear Coupled Cluster effort
4 Dean_NCCT_SC07
Solved up to mass 12 with GFMC, converged mass 8 with diagonalization. We want to go much further!
Nuclear interactions: Cornerstone of the entire theoretical edifice
Real three-body interactionsderived from QCD-based
effective theories
Real three-body interactionsderived from QCD-based
effective theories
Method of Solution:Nuclear Coupled-Cluster Theory
Method of Solution:Nuclear Coupled-Cluster Theory
Depends on spin, angular momentum, and nucleon(proton and neutron) quantum numbers. Complicated interactions
Depends on spin, angular momentum, and nucleon(proton and neutron) quantum numbers. Complicated interactions
NNNji
jiAi
ii
VrrVM
H ++∇−
= ∑∑<=
),(2,1
22h
5 Dean_NCCT_SC07
• It boils down to a set of coupled, nonlinear algebraic equations (odd-shaped tensor-tensor multiply).
• Storage of both amplitudes and interactions is an issue as problems scale up.
• Largest problem so far: 40Ca with 10 million unknowns, 7 peta-ops to solve once(up to 10 runs per publishable result).
• Breakthrough science: Inclusion of 3-body force into CC formalism (6-D tensor)weakly bound and unbound nuclei.
Coupled-cluster theory: Ab initio in medium mass nuclei
exp Correlated ground-state
wave functionCorrelationoperator
Reference Slaterdeterminant
EnergyEnergy
Amplitude equationsAmplitude equations
… 321 TTTT
∑
∑
><
++
><
+
=
=
f
f
f
f
abij
ijbaabij
ai
iaai
aaaatT
aatT
εε
εε
2
1 THTE exp)exp(
0exp)exp( HTHT ab…ij…
ab…ij…
6 Dean_NCCT_SC07
exp(T)
T = T1 + T2 + T3 + …
R = excitation operator
POLYNOMIAL SCALING!! (good)
Early results
4 6 8 infty
Number of Oscillator Shells
-130
-125
-120
-115
-110
-105
-100
E =
(M
eV
)Wolch et al PRL 94, 24501 (2005)
E* (0+) = 19.8 MeV
E* (3-) = 12.0 MeV
*Eg.s. = –120.5 MeV
CCSDCR-CCSD(T)
Coupled cluster theory for nuclei
E = H = e-THeT
ab…H = 0ij…
RH = E*R
7 Dean_NCCT_SC07
Ab initio in medium mass nuclei
Fast convergence w
ith cluster rank
4HeCCSDCCSD(T)FY
-28
E(4 H
e) [
MeV
]
-30
-26
-24
-22
-20
-18
-16
2 4 5 6 10 12 14 16N=2n+1
16O
N = 6
N = 7
N = 8N = 9
N = 9N = 11N = 12
N = 13
-134
-136
-138
-140
-142
-144
E CC
SD(16
O) [
MeV
]
12 14 16 18 20 22 24ħ [MeV]
4He 16O 40Ca
E0
ECCSD
ECCSD(T)
-11.8
-17.1
-0.3
-60.2
-82.6
-5.4
-347.5
-143.7
-11.7
ECCSD(T)-29.2 -148.2 -502.9
Exact (FY) -29.19(5)
Error estimate: 1%<< 1% < 1%
Hagen et al., Phys. Rev. C 76,044305 (2007)
Hagen et al., Phys. Rev. C 76,044305 (2007)
40Ca
N = 3
N = 4
N = 5
N = 6N = 7
E CC
SD(40
Ca)
[MeV
] -400
-420
-440
-460
-480
-500
380
ħ [MeV]16 18 20 22 24 26 28 30 32
N = 8
1063 many-bodybasis states
8 Dean_NCCT_SC07
Hagen, Papenbrock, Dean, Schwenk, Nogga, Wloch, Piecuch, Phys. Rev. C 76, 034302 (2007)
Solution at CCSD and CCSD(T) levels involve roughly 67 more diagrams…
Challenge: Do we really need the full 3-body force, or just its
density dependent terms?
Challenge: Do we really need the full 3-body force, or just its
density dependent terms?
2-body only
0-body 3NF
1-body 3NF
2-body 3NF
Residual 3NF
1 2 3 4 5
/
CC
SD
10-4
10-3
10-2
100
10-1
estimated triples corrections
N
Inclusion of full TNF in CCSD: F-Y comparisons in 4He
<E>=-28.24 MeV +/- 0.1MeV (sys)
-28
-27
-26
-25
-24
-23
-22
EC
CS
D(T
) (M
eV
)
3 4 5 6N
+ +0 =
+ ++
+ ++
+ ++
+ ++
9 Dean_NCCT_SC07
Introduction of continuum basis states (Gamow, Berggren)
CorrelationdominatedSn=0
Sn
n ~ n
Open QSOpen QS
Closed QSClosed QS
Neutron number
En
erg
yCoupling of nuclear structure and reaction theory (microscopic treatment of open channels)
capturing states decaying states
L-
lm(k)
k1
k2 k3 k4
Re(k)
boun
d st
ates
antib
ound
sta
tes L+
10 Dean_NCCT_SC07
Ab initio weakly bound and unbound nuclei
Single-particle basis includesbound, resonant, non-resonantcontinuum, and scattering statesENORMOUS SPACES….almost 1k orbitals.1022 many-body basis states in 10He
He Chain Results(Hagen et al)
[feature article in Physics Today (November 2007)]
Challenge: Include 3-body forceChallenge: Include 3-body force
Gamov states capture the halo structure of drip-line nuclei
4 5 6 7 8
r [fm]
10-2
10-1
100
r2
(r)
[fm
-1]
G-HF basis
HO-HF basis
11 Dean_NCCT_SC07
System of non-linear coupled algebraic equations solve by iteration n = number of neutrons and protons N = number of basis states Solution tensor memory (N-n)**2*n**2
Interaction tensor memory N**4
Operations count scaling O(n**2*N**4) O(n**4*N**4) with 3-body O(n**3*N**5) at CCSDT
Solution of coupled-cluster equation
•Many such terms exist.•Cast into a matrix-matrix multiply algorithm.
•Parallel issue: block sizes of V and t.
•Many such terms exist.•Cast into a matrix-matrix multiply algorithm.
•Parallel issue: block sizes of V and t.
Basic numerical operation:Basic numerical operation:
tnew (ab, ij) = V (kl, cd)told (cd, ij)told (ab, kl)k,l=1, n
c,d=n+1,N
12 Dean_NCCT_SC07
Partial sumt2 reside on each processor
Memory distribution across processors
V(ab, c1, d1)t2 partial sum
V(ab, c1, d2)t2 partial sum V(ab, c2, d2)
t2 partial sumV(ab, c2, d1)
t2 partial sum
V(ab, c1, d1)
t2 partial sumV(ab, c1, d2)
t2 partial sumV(a, b, c2, d2)
t2 partial sumV(ab, c2, d1)
t2 partial sum
Code parallelism
Global reduce (sum) t2, distribute
t2(ab, ij) = V (kl,cd)tij tklkl <fcd >f
cd ab
13 Dean_NCCT_SC07
Future direction
Current algorithm scales to 1K processors with about 20% efficiency. Attacking problems in mass 40 region is doable with current code.
Develop algorithm that spreads both the 2-body matrix elements and the CC amplitudes (in collaboration with Ken Roche) Enables nuclei in the mass 100 region and should scale to 100K processors (under way).
Designing further parallel algorithms that calculate nuclear properties to calculate densities and electromagnetic transition amplitudes.
Eventual time-dependent CC for fission dynamics.
14 Dean_NCCT_SC07
Contact
David J. DeanPhysical Sciences DirectorateNuclear Theory(865) [email protected]
14 Dean_NCCT_SC07
References:
Dean and Hjorth-Jensen, PRC 69, 054320 (2004); Kowalski, Dean, Hjorth-Jensen, Papenbrock, Piecuch,PRL 92, 132501 (2004); Wloch, Dean, Gour, Hjorth-Jensen, Papenbrock, Piecuch, PRL 94, 21501 (2005);Gour, Piecuch, Wloc, Hjorth-Jensen, Dean, PRC (2006); Hagen, Dean, Hjorth-Jensen, Papenbrock, PLB (2007);
Hagen, Dean, Hjorth-Jensen, Papenbrock, Schwenk, PRC 76, 044305 (2007); Hagen, Papenbrock, Dean, Schwenk, Nogga, Wloch, Piecuch, PRC 76, 034302(2007); Dean, Physics Today (November 2007)