EPJ Web of Conferences 146, 12008 (2017) DOI: 10.

1051/epjconf/201714612008
ND2016

An analytic approach to probability tables for

the unresolved resonance region
David Brown1,a and Toshihiko Kawano2
1
National Nuclear Data Center, Brookhaven National Laboratory, Upton NY, USA
2
T-2, Los Alamos National Laboratory, Los Alamos NM, USA

Abstract. The Unresolved Resonance Region (URR) connects the fast neutron region with the Resolved
Resonance Region (RRR). The URR is problematic since resonances are not resolvable experimentally yet
the fluctuations in the neutron cross sections play a discernible and technologically important role: the URR in
a typical nucleus is in the 100 keV – 2 MeV window where the typical fission spectrum peaks. The URR also
represents the transition between R-matrix theory used to described isolated resonances and Hauser-Feshbach
theory which accurately describes the average cross sections. In practice, only average or systematic features
of the resonances in the URR are known and are tabulated in evaluations in a nuclear data library such as
ENDF/B-VII.1. Codes such as AMPX and NJOY can compute the probability distribution of the cross section
in the URR under some assumptions using Monte Carlo realizations of sets of resonances. These probability
distributions are stored in the so-called PURR tables. In our work, we begin to develop a scheme for computing
the covariance of the cross section probability distribution analytically. Our approach offers the possibility of
defining the limits of applicability of Hauser-Feshbach theory and suggests a way to calculate PURR tables
directly from systematics for nuclei whose RRR is unknown, provided one makes appropriate assumptions
about the shape of the cross section probability distribution.

1. Introduction can compute self-shielding factors, Bondarenko factors,

cross section bands, etc..
For neutron induced reactions below 20 MeV incident
Processing codes adopt a Monte Carlo approach to
energy, the Unresolved Resonance Region (URR) connects
computing probability distributions. They generate an
the fast neutron region with the Resolved Resonance
ensemble of “resonance ladders”, namely a simulated set
Region (RRR). In the RRR, R-matrix theory is used
of levels for the compound nucleus populated by the
to describe the shape and correlations between resolved
incident neutron accompanied by the resonance widths
resonances in the cross sections. In the fast region,
of these states. These simulated resonances are generated
Hauser-Feshbach theory with the Width Fluctuation
using the average parameters in the ENDF formatted files.
Correction (WFC) accurately describes the cross sections.
These realizations of simulated resonances are used to
In between, the URR is problematic: here the resonances
reconstruct a sample cross sections and from the ensemble
are not resolvable experimentally yet the fluctuations
of reconstructed cross sections, the probability distribution
in the neutron cross sections play a discernible and
of the cross section is determined. These Monte-Carlo
technologically important role. Indeed, the URR in a
approaches are cumbersome and time consuming, often
typical nucleus is in the 100 keV – 2 MeV window, where
dominating the time taken to process a single evaluation. In
the typical fission spectrum peaks.
this contribution, we ask whether it is possible to compute
In practice, only average or systematic features of
the PDF x, ℘(σx |E), analytically?
the resonances in the URR are known and these are
Our proposed scheme is straightforward, we assume
tabulated in evaluations in a nuclear data library such as
that the cross section PDF can be approximated with some
ENDF/B-VII.1 [1]. With judicious application of nuclear
suitably shaped multivariate PDF, say a multi-variate log-
reaction phenomenology, one can compute the average
normal, Poisson or χ 2 distribution, as these are consistent
cross sections in the URR region (see pp. 343-350 of the
with the constraint that σx > 0. We then compute
σx 
ENDF Format manual [2]). In fact, one can go further
and cov(σx , σ y ) =
σx σ y  −
σx 
σ y  and convert these to
than just the average cross section. Several processing
a form appropriate for our assumed PDF.
codes, notably NJOY [3], AMPX [4], CALENDF [5]
For this scheme to work we must compute the
and GRUCON [6], offer one the ability to compute the
average cross section and the cross section covariance.
probability distribution function (PDF) for the total cross
The average cross section is already known in both the
section at a fixed incident energy ℘(σ(n,tot) |E) as well
URR [2] and from the Hauser-Feshbach formula with the
as the conditional probabilities for the neutron induced
Width Fluctuation Correction (WFC). Surprisingly, both
reactions (n, x), ℘(σ(n,x) |σ(n,tot) , E). From these ℘’s, one
URR and the Hauser-Feshbach theory with the WFC use
same basic theory, but with practical differences. Both
a e-mail: dbrown@bnl.gov assume have narrow isolated resonances so that the weak

c The Authors, published by EDP Sciences. This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0
(http://creativecommons.org/licenses/by/4.0/).
EPJ Web of Conferences 146, 12008 (2017) DOI: 10.1051/epjconf/201714612008
ND2016

coupling approximation is valid and we may treat the The SLBW reaction cross section has poles in both the
resonances with the Single Level Breit-Wigner (SLBW) upper and lower half-planes of the complex energy surface.
approximation. Both approaches assume the average cross Therefore we will consider a semi-circular contour integral
section may be computed with an ensemble average of with the contour closing in the upper half plane, enclosing
resonances, assuming that the resonance widths follow a all the poles in the cross section above the real axis.
Porter-Thomas (χ 2 ) distribution. In the URR, the degree- Performing this integral we easily arrive at
of-freedom ν of the χ 2 distribution is specified by the
evaluator. In the WFC of the Hauser-Feshbach theory, ν 2πga
aµ bµ

σab  ≈ (6)
is determined from a phenomenological prescription such Eka2 µ µ
as that in Ref. [7]. The actual cross section average is then
given essentially by the Hauser-Feshbach equation in both We assume that the number of resonances is large and that
cases but the details of the averaging procedure differ in we may replace the sum over resonances with an ensemble
practice. In the URR, the MC2-II algorithm [8] is used average. So, we replace
while in the fast region the WFC is used as given e.g., by  
Gruppelaar [9]. Given that both use essentially the same
aµ bµ E a b
≈ . (7)
approach, we should be able to make the fast and average
µ
µ D 
URR cross sections agree. More importantly, we should be
able to use this common approach as a guide for computing The ensemble average

· is an average over all possible
the covariance. values of the widths, which are assumed to have a Porter-
Thomas or χ 2 distribution with ν degrees of freedom,
2. The average cross section ℘ P T (x|ν). For a = b:
We begin by reviewing the origins of the Width Fluctuation    ∞

a b a
Correction (WFC) for the compound nuclear cross section. = da ℘ P T νa
 0 a
This discussion motivates our approach to the cross section  ∞

correlation
σab σcd  in Sect. 3. b

× db ℘ P T νb
Here we follow the derivation in Ref. [9]. Assuming 0 b
that we have narrow isolated resonances,   D  E   ∞

c a b
where  is a typical resonance width, D is the mean × dc ℘ P T νc
c 
resonance spacing and E is the typical energy scale of c=a,b 0
measurements. In this case the Single Level Breit Wigner (8)
(SLBW) approximation is valid and we have the total
channel cross section of and for a = b:
   2   ∞

4πgc
cµ   a a
σc = 2 sin φc +
2
ψµ cos 2φc +χµ sin 2φc = da ℘ PT
νa
kc µ µ  0 a
 ∞ 
2 (9)
(1) c c
and the partial channel reaction cross section of × dc ℘ PT
νc
c=a 0
c 
4πga
aµ bµ
σab = ψµ . (2) with
ka2 µ µ2
 ∞

c
Here the profile functions ψ and χ are

c  = dc ℘ PT
νc c =  c . (10)
0 c
µ2 /4 (E − E 0 )µ /2
ψµ + iχµ = +i · The integrals in Eqs. (8)–(9), can be done numerically,
(E − E 0 )2 + µ2 /4 (E − E 0 )2 + µ2 /4 as is done in most processing codes (actually using the
(3) same MC2 -II algorithm) or performed analytically follow
Gruppelaar [9]. Using the fact that
These channel cross sections must of course be added  ∞  ∞
appropriately to compute the total, capture, elastic or any 1

−t
other reaction cross section. = dte = dte− c c t (11)
 0 0
Consider now the energy average of a reaction cross
section in the SLBW approximation: and the result
 E+E/2  ∞

1 c

σab  = d E  σab (E  ) (4) Jn (t,  c , νc ) = dc ℘ P T νc (c )n e−c t
E E−E/2 0 c

n νc
−(n+νc /2)
If we assume that we have narrow isolated resonances, then 2 c ( 2 + n) 2t c
= 1 + ,
E  µ and νc ( ν2c ) νc
(12)
4πga
aµ bµ

σab  ≈
Eka2 µ 4 we obtain the usual
 ∞  
1 a b a b
× d E (5) = Wab . (13)
−∞ (E − E 0 ) 2 +  2 /4
µ  

2
EPJ Web of Conferences 146, 12008 (2017) DOI: 10.1051/epjconf/201714612008
ND2016

Here Wab is the WFC factor: with t = t1 + t2 + t3 and  = δa f + δb f + δc f + δd f +

ν f /2. We can simplify this result further first rewriting
2 the t’s in cylindrical coordinates and integrating out the
Wab = 1 + δab
νa angular dependence. In analogy with the WFC factor
 ∞  2 c

−(δac +δbc +νc /2) (14) above, we define Wabcd in terms of the integration result:
× dx 1+ x . 3  
0 νc   a b c d
c Wabcd =
a b c d 3
With these, we obtain the energy average reaction cross

 ∞
2δab 2δcd x2 
section = 1+ 1+ dx
2πga  a  b νa νc 0 2 f

σab  = 2 Wab . (15)
ka D 
−
2 f
Since we are in the weak coupling limit, we have Tc ≈ × 1+ x (22)
νf
2π  c /D and we recover the usual Hauser-Feshbach
equation: This is an intriguing result as this Wabcd factor looks
ga Ta Tb very much like the WFC factor used in the average cross

σab  = 2  Wab . (16) section.
ka c Tc
We can now write down the covariance between the
cross sections very compactly. First
3. Cross section correlations
π D Wabcd
Now consider the cross section correlation
σab σcd . Again
σab σcd  =
σab 
σcd 
 Wab Wcd
we assume that we have narrow isolated resonances and as (23)
before, the energy average in the cross section correlation 2π 2 Wabcd
=
σab 
σcd 
can be approximated T Wab Wcd


using Tc = 2π c /D. Therefore, in this model the cross
1 4πga 4πgc

σab σcd  ≈ section covariance is given by
E ka2 kc2 2



 ∞ aµ bµ /4 cov(σab , σcd ) =
σab 
σcd 
2π Wabcd
− 1 . (24)
×  T Wab Wcd
−∞ (E − E µ ) +µ /4
2 2
µν
 We note that there is nothing in this expression or the
cν dν /4 expressions for Wab or Wabcd that keep the covariance
× dE  (17)
(E  − E ν )2 +ν2 /4 positive. This indicates that this model has a limited range
of applicability that we will investigate in the next section.
Again, this has poles in both the upper and lower half
planes so we perform. the contour integration as before. 4. The behavior of fluctuation factors
Since we have assumed narrow isolated resonances, we
To understand the behavior of both the WFC factor Wab
find
and the WFC-like factor Wabcd , we consider a system with

π/4 4πga 4πgc
aµ bµ cµ dµ a four neutron channels with identical widths, a typical

σab σcd  ≈ number active in the URR of a typical nucleus, and one
E ka2 kc2 µ3
µ gamma channel with Tγ = 10−4 . We use the Kawano-
(18) Talou systematics for ν [7]. Figure 1 shows the factors as a
As before, we replace the sum over resonances with an function of neutron transmission coefficient.
ensemble average Raw plots of the factors are not very instructive.
  Therefore, consider

aµ bµ cµ dµ E a b c d √ the standard deviation of the cross
≈ (19) section: σab = cov(σab , σab ). In Fig. 2 we plot
µ
µ3 D 3 σab /
σab  as a function of Tn in this example. Clearly
the fluctuations in the cross sections are very large for
Similar to before, we use small Tc and diminish rapidly as Tc increases. We also note
 ∞  ∞  ∞ that the gamma channel, with its small Tγ exhibits relative
1
= dt 1 dt 2 dt3 e−(t1 +t2 +t3 ) (20) fluctuations an order of magnitude larger than the neutron
3 0 0 0 channels.
We now investigate how the model behaves as
and assume Porter-Thomas distributions with degree of the number of channels increases at fixed Tn = 0.2.
freedom ν. We arrive at We know from Eq. (24) that the covariance has the
 

possibility of going negative. In Fig. 3 we plot the relative
a b c d 2δab 2δcd
= a b c d 1 + 1+ variance of the elastic cross section (σaa /
σaa )2 =
3 νa νc
 ∞  ∞  ∞ cov(σaa , σaa )/
σaa 2 for elastic scattering as a function

× dt1 dt2 dt3 of the number of open neutron channels. We see as the
0 0 0 f
number of channels reaches becomes large the variance

− drops below zero, indicating a breakdown of the model.
2t f This in a way is no surprise as we are deeply in the strong
× 1+ (21)
νf coupling limit.

3
EPJ Web of Conferences 146, 12008 (2017)
ND2016
Figure 1. A comparison of the WFC and WFC-like factors Wab
and Wabcd as a function of neutron transmission coefficient. As
described in the text, this calculation has four neutron channels
and one gamma channel and Tγ = 10−4 . Here all channels are
assumed to be neutron channels unless otherwise stated (e.g.,
channel “b” for capture).
Figure 2. The relative variance of the reaction cross sections as
a function of neutron transmission coefficient. Here all channels
are assumed to be neutron channels unless otherwise stated (e.g.
channel “b” for capture).
5. Outlook
We have outlined the beginnings of a scheme to
compute the compound nuclear cross section probability
distribution function in the URR region. Our approach
offers the possibility of defining the limits of applicability
DOI: 10.1051/epjconf/201714612008
Figure 3. Plot of the relative variance of the elastic scattering
cross section as a function of the number of neutron channels. In
the inset, we show the region at high channel number, showing
the breakdown of the approach in the strong coupling limit.
of Hauser-Feshbach theory and suggests a way to
calculate PURR tables directly from systematics for
nuclei whose RRR is unknown, provided one makes
appropriate assumptions about the shape of the cross
section probability distribution. This scheme should also
be applicable to unstable nuclei where no resonances
can be measured in practical experiments. To complete
this program, we must fully characterize the region of
applicability and test it against other probability table
generations tools.
References
[1] M.B. Chadwick, M. Herman, P. Obložinský, et al.,
Nucl. Data Sheets 112, 2887–2996 (2011)
[2] A. Trkov, M. Herman, D.A. Brown, CSEWG
Document ENDF-102, BNL Report BNL-90365-2009
Rev.2 (2011)
[3] R.E. MacFarlane, A.C. Kahler, Nucl. Data Sheets 111,
2739–2890 (2010); R.E. MacFarlane, A.C. Kahler,
LANL Report LA-UR-12-27079 (2012)
[4] M.E. Dunn, L.C. Leal, Nucl. Sci. Eng. 148, 30–42
(2004)
[5] P. Ribon, J.M. Maillard, CEA Report CEA-N-2485
(1986); P. Ribon, Proc. Seminar NJOY and THEMIS,
Saclay, France, June 20–21, 1989, OECD Data Bank
(1989)
[6] V.V. Sinitsa, A.A. Rineiski, V.V. Tebin, OECD/NEA/
Subgroup 38 Meeting, NEA Headquarters, Paris,
France May 12–16 (2014)
[7] T. Kawano, P. Talou, Nucl. Data Sheets 118, 183–186
(2014)
[8] Ed. by D. Cacuci, Handbook of Nuclear Engineering
(Springer, 2010)
[9] H. Gruppelaar, G. Reffo, Nucl. Sci. Eng. 62, 756–763
(1976)