14.01.2015 Views

k-Effective of the World - Nuclear Criticality Safety Division

k-Effective of the World - Nuclear Criticality Safety Division

k-Effective of the World - Nuclear Criticality Safety Division

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

LA-UR-11-03593<br />

Statistical Coverage Concerns in a<br />

Revised k-<strong>Effective</strong> <strong>of</strong> <strong>the</strong> <strong>World</strong><br />

Problem<br />

Brian Kiedrowski, Forrest Brown<br />

Los Alamos National Laboratory<br />

X-Computational Physics <strong>Division</strong><br />

1


Abstract<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

A revised version <strong>of</strong> k-<strong>Effective</strong> <strong>of</strong> <strong>the</strong> <strong>World</strong> is defined specifically to stress<br />

current Monte Carlo power iteration techniques. Results <strong>of</strong> numerous<br />

independent Monte Carlo calculations show that even for batch sizes that are<br />

typically considered reasonable (5-10K), incorrect results for k are obtained with<br />

non-trivial probability. The causes for what makes a problem prone to having<br />

<strong>the</strong>se difficulties are identified and advice for criticality safety practitioners is<br />

provided.<br />

2


Overview<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Review <strong>of</strong> k-<strong>Effective</strong> <strong>of</strong> <strong>the</strong> <strong>World</strong><br />

• The Revised Problem<br />

• Numerical Results<br />

• Analysis & Guidance<br />

3


X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

Review <strong>of</strong> k-<strong>Effective</strong> <strong>of</strong> <strong>the</strong> <strong>World</strong><br />

4


Problem Specification<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Proposed by E. Whitesides in 1971<br />

• Specification:<br />

– 9 x 9 x 9 array <strong>of</strong> Pu-239 spheres spaced 50 cm apart in<br />

vacuum<br />

– The array is surrounded by a thick water reflector and is<br />

subcritical<br />

– Replace central sphere with one that is exactly critical by<br />

itself<br />

– Final result should be supercritical<br />

5


Results<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• MCNP5-1.60 + ENDF/B-VII.0 data<br />

• For uniform array <strong>of</strong> identical spheres<br />

with surrounding water, sphere radii<br />

adjusted to r = 3.9 cm, so that<br />

Keff = .9328 ± .0002<br />

• Single bare sphere, r=4.928 cm,<br />

Keff = 1.0001 ± .0002<br />

• Whitesides' model problem:<br />

Replace center sphere <strong>of</strong> array<br />

by larger (critical) sphere<br />

Should be supercritical - is it <br />

6


Results: Then and Now<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Run problem multiple times with different random<br />

number sequence<br />

• KENO defaults in <strong>the</strong> 1970s:<br />

– 300 particles per cycle, skip 3, run 100 active cycles<br />

– Results yield k as being subcritical<br />

• Modern best practices:<br />

– Use 5000+ particles per cycle, analyze convergence<br />

diagnostics, start particles in each region<br />

– Results yield correct k as supercritical<br />

7


Results<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

Distribution <strong>of</strong> K eff for 200 replicas, various M = neutrons/cycle<br />

Frequency <strong>of</strong> Keff, for replicas<br />

M=200<br />

M=300<br />

M=10K<br />

M=500<br />

Keff<br />

8


Convergence<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

K eff vs cycle, various M<br />

M = neutrons/cycle<br />

H src vs cycle, various M<br />

M = neutrons/cycle<br />

K eff converges in<br />

75-100 cycles<br />

M = 10,000<br />

M = 10,000<br />

H src converges in<br />

100-150 cycles<br />

M = 5,000<br />

M = 5,000<br />

Must discard 150<br />

or more initial<br />

cycles<br />

M = 1,000<br />

M = 1,000<br />

Convergence<br />

depends on <strong>the</strong><br />

dominance ratio &<br />

source guess, NOT<br />

on neutrons/cycle<br />

M = 500<br />

M = 500<br />

100 200<br />

100 200<br />

Initial source guess = uniform sampling <strong>of</strong> points at sphere centers<br />

9


Results Explained<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Problem has 729 spheres, 300 per cycle not enough to<br />

sample problem space!<br />

– Problem <strong>of</strong> coverage<br />

• Small batch size leads to issues in normalization<br />

– Problem <strong>of</strong> non-conservative bias<br />

• About 150 cycles required to find a steady source<br />

– Problem <strong>of</strong> convergence<br />

• Modern best practices make this problem easy to solve<br />

10


X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

The Revised Problem<br />

11


A More Difficult Problem<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Simulate a flooded subcritical assembly storage vault<br />

• Flood <strong>the</strong> area between <strong>the</strong> spheres, reduce radii and<br />

distances between spheres to retain criticality<br />

– Decreases coupling between spheres<br />

• Place a thick layer <strong>of</strong> natural cadmium around <strong>the</strong><br />

central (most reactive) sphere<br />

– Creates an asymmetry in coupling between <strong>the</strong> most<br />

reactive region and <strong>the</strong> o<strong>the</strong>r region<br />

• These coupling characteristics create difficulties<br />

12


Problem Specifications<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Geometric Specifications:<br />

– Radius <strong>of</strong> non-central Pu-239 spheres: 3.75 cm<br />

– Radius <strong>of</strong> central Pu-239 sphere: 4.33 cm<br />

– Coating thickness: 0.5 cm<br />

– Center-to-center spacing: 20 cm<br />

• Material specifications:<br />

– Pure Pu-239, density: 20.0 g/cc<br />

– Natural cadmium, density: 8.65 g/cc<br />

– Pure water, density: 1.0 g/cc<br />

13


Problem Specification<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Source specification<br />

– Uniformly sample points in each sphere<br />

– 500 inactive cycles, 1000 total<br />

– Vary size <strong>of</strong> batch<br />

• <strong>Criticality</strong> results for variants:<br />

– Uniform array: keff = 0.95<br />

– Larger, coated sphere in infinite water bath: keff = 1.0<br />

– Composite system should be supercritical<br />

14


Problem Specification<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

15


X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

Numerical Results<br />

16


Reliability Tests<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Main Observation: MCNP (and any o<strong>the</strong>r package using<br />

power iteration) produces unreliable results<br />

– Sometimes correct (k > 1.0), sometimes wrong (k = 0.95)<br />

– Reliability increases with size <strong>of</strong> batch<br />

– A “good” starting source is important<br />

• Reliability is defined as <strong>the</strong> probability <strong>of</strong> getting <strong>the</strong><br />

correct answer<br />

– Determine with 100 independent MCNP trials with different<br />

random number sequences<br />

• Try to explain why…<br />

17


Probability Density <strong>of</strong> Results<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

0.35<br />

0.3<br />

0.25<br />

0.2<br />

0.15<br />

0.1<br />

0.05<br />

0<br />

5K<br />

20K<br />

10K<br />

18


Reliability Curve<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

1.0<br />

0.9<br />

Frequency <strong>of</strong> Incorrect k<br />

0.8<br />

0.7<br />

0.6<br />

0.5<br />

0.4<br />

0.3<br />

0.2<br />

0.1<br />

0.0<br />

0 5 10 15 20 25<br />

Batch Size (1000 neutrons/cycle)<br />

19


X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

Analysis & Guidance<br />

20


Discussion <strong>of</strong> Results<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Best practices<br />

– Batch size 5K or more may be too small<br />

– May require 100K+ for “difficult” problems<br />

• Probability <strong>of</strong> getting incorrect k decreases<br />

exponentially with batch size<br />

• No observable impact <strong>of</strong> bias<br />

– Mean value for “correct” peak has consistent average k<br />

– Problem is coverage not bias<br />

21


Discussion <strong>of</strong> Results<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Diagnostics <strong>of</strong>ten fail to detect coverage<br />

– Convergence in k plot is false<br />

– Shannon entropy <strong>of</strong>fers little to no indication<br />

– Possible to sample central sphere, but not initiate a sustaining<br />

fission chain<br />

• Prescribed source guess inadequate<br />

– Source guess points in central sphere guarantees<br />

convergence to correct k<br />

22


Recipe for a “Difficult” Problem<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Loosely-coupled regions<br />

• Varying importance<br />

– Most important region MUST be adequately sampled<br />

• Asymmetric coupling<br />

– Arises from material or geometric effects<br />

23


Asymmetric Coupling<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

24


Practical Guidance<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• Understand <strong>the</strong> problem<br />

– Exercise caution for problems with discrete zones<br />

– Identify most important region(s)<br />

• Good parameters defeats many problems<br />

– Large batch sizes (100K+) preferable<br />

– Source guess focus on most important region(s)<br />

• Analyze results closely<br />

– Check population tables to ensure adequate sampling<br />

– Does value <strong>of</strong> k match your intuition<br />

– For suspicious outcomes, run again with different random<br />

number seeds<br />

25


References<br />

X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

• X-5 MONTE CARLO TEAM, “MCNP – A<br />

General N-Particle Transport Code, Version<br />

5 – Volume I: Overview and Theory,” LA-<br />

UR-03-1987, Los Alamos National<br />

Laboratory (2003).<br />

• G.E. WHITESIDES, "Difficulty in Computing<br />

<strong>the</strong> k-effective <strong>of</strong> <strong>the</strong> <strong>World</strong>," Trans. Am.<br />

Nucl. Soc., 14, No. 2, 680 (1971).<br />

• F.B. BROWN, "Review <strong>of</strong> Best Practices for<br />

Monte Carlo <strong>Criticality</strong> Calculations", ANS<br />

NCSD-2009, Richland, WA, Sept 13-17<br />

(2009).<br />

• F.B. Brown, “’K-<strong>Effective</strong> Of The <strong>World</strong>’<br />

And O<strong>the</strong>r Concerns For Monte Carlo<br />

Eigenvalue Calculations”, SNA+MC-2010,<br />

Tokyo, Oct 17-20 [also, LA-UR-10-05548]<br />

• R.N. Blomquist, et al., "Source<br />

Convergence in <strong>Criticality</strong> <strong>Safety</strong> Analysis,<br />

Phase I: Results <strong>of</strong> Four Test Problems,"<br />

OECD <strong>Nuclear</strong> Energy Agency, OECD NEA<br />

No. 5431 (2006).<br />

• R.N. Blomquist, et al., "NEA Expert Group<br />

on Source Convergence Phase II:<br />

Guidance for <strong>Criticality</strong> Calculations", 8 th<br />

International International Conference on<br />

<strong>Criticality</strong> <strong>Safety</strong>, St. Petersburg, Russia,<br />

May 28 – June 1, 2007 (May 2007).<br />

• F.B. Brown, “Revisiting <strong>the</strong> ‘K-effective <strong>of</strong><br />

<strong>the</strong> <strong>World</strong>’ Problem”, Trans. Am. Nuc. Soc,<br />

102, June 2010, [also, LA-UR-10-00189]<br />

(2010).<br />

• F.B. Brown, B.C. Kiedrowski, J.S. Bull,<br />

"MCNP5-1.60 Release Notes", LA-UR-10-<br />

06235 (2010).<br />

26


X-Computational<br />

Physics Div.<br />

XCP-3, LANL<br />

Questions<br />

27

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!