See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328548568

Exact Transport Representations of the Classical and Nonclassical Simplified PN

Equations

Article  in  Journal of Computational and Theoretical Transport · October 2018

DOI: 10.1080/23324309.2018.1496938

CITATIONS READS
6 104

3 authors:

Ilker Makine Richard Vasques

Université Libre de Bruxelles The Ohio State University
2 PUBLICATIONS   6 CITATIONS    59 PUBLICATIONS   257 CITATIONS   

SEE PROFILE SEE PROFILE

R. N. Slaybaugh
University of California, Berkeley
57 PUBLICATIONS   568 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Ilker Makine on 18 December 2018.

The user has requested enhancement of the downloaded file.

25th International Conference on Transport Theory, Monterey, California, 16-20 October 2017

Exact Transport Representations of the Classical and Nonclassical

Simplified PN Equations
I. Makine1 , R. Vasques2 , and R.N. Slaybaugh3
1
ilker.makine@ulb.ac.be, Ecole Polytechnique de Bruxelles, ULB, avenue Franklin Roosevelt 50, 1050 Bruxelles,
Belgium. INSTN, centre CEA de Saclay, 91190 Gif-sur-Yvette, France
2
vasques.4@osu.edu, Department of Mechanical and Aerospace Engineering, The Ohio State University
201 W. 19th Avenue, Columbus, OH 43210
3
slaybaugh@berkeley.edu, Department of Nuclear Engineering, University of California, Berkeley
4151 Etcheverry Hall, Berkeley, CA 94720

Abstract - We show that the recently introduced nonclassical simplified PN equations can be rep-
resented exactly by a nonclassical transport equation. Moreover, we validate the theory by showing
that a Monte Carlo transport code sampling from the appropriate nonexponential free-path distri-
bution function reproduces the solutions of the classical and nonclassical simplified PN equations.
Numerical results are presented for 4 sets of problems in slab geometry.

1. Introduction

A nonclassical linear Boltzmann equation has been recently proposed [1,2] to address transport problems
in which the particle flux is not attenuated exponentially. The original motivation for this formulation came
from measurements of photon free-paths in atmospheric clouds, which could not be explained by classical
radiative transfer with exponential attenuation (cf. sections 5.1 and 8.3 in [3]). This theory has since been
extended and has found applications for neutron transport in reactor cores (cf. [4]) as well as image rendering
in computer graphics (cf. [5]).

The nonclassical theory requires the use of a memory variable, namely the free-path s, representing the
distance traveled by a particle since its previous interaction (birth or scattering). Assuming that scattering is
isotropic, the one-speed nonclassical transport equation with an isotropic internal source is written as [2]
 Z Z ∞ 
∂ψ δ(s) 0 0 0 0 0
+ Ω · ∇ψ + Σt (s)ψ = c Σt (s )ψ(x, Ω , s )ds dΩ + Q(x) . (1)
∂s 4π 4π 0

Here, ψ = ψ(x, Ω, s) represents the nonclassical angular flux, c is the scattering ratio, and Q(x) is the
source. The total cross section Σt is a function of s such that the free-path probability distribution function
Rs
Σt (s0 )ds0
p(s) = Σt (s)e− 0 (2)

does not have to be exponential.

Equation (1) is a generalization of the linear Boltzmann equation; classical transport is recovered when Σt
is independent of s. In that case, Eq. (2) becomes the exponential decay

p(s) = pc (s) := Σt e−Σt s , (3)

and Eq. (1) reduces to the well-known (classical) one-speed linear Boltzmann equation [2]
 Z 
1 0 0
Ω · ∇Ψ(x, Ω) + Σt Ψ(x, Ω) = c Σt Ψ(x, Ω )dΩ + Q(x) (4)
4π 4π
 I. Makine, R. Vasques, R.N. Slaybaugh

for the classical angular flux

Z ∞
Ψ(x, Ω) = ψ(x, Ω, s)ds. (5)
0

The simplified PN (SPN ) equations, first derived by Gelbard [6–8], have been shown to be a high-order
asymptotic approximation of the transport equation [9]. They are particularly attractive when addressing
problems in which the spatial and angular dependence of the angular flux is not severe. They are used
to improve the quality of transport physics in a diffusion model while avoiding the complexity of the full
PN (spherical-harmonics) or SN (discrete-ordinates) equations. Specifically, the SPN equations can be
implemented directly within a diffusion code, with numerical solutions frequently much more transport-like
than diffusion solutions. We refer the reader to [10] for a complete review on SPN theory.

It has been shown that certain cases in the hierarchy of the classical SPN equations (SP1 , SP2 , and SP3 ) can
be represented exactly by a nonclassical transport equation [11]. One can obtain explicit expressions for
the free-path distribution p(s) and the corresponding Σt (s) such that Eq. (1) can be converted to an integral
equation for the scalar flux
Z Z Z ∞
Φ(x) = Ψ(x, Ω)dΩ = ψ(x, Ω, s)dsdΩ (6)
4π 4π 0

that is identical to the integral formulation of the SPN equations. This result was later extended to include
the case of nonclassical diffusion [12].

In this work, the following original results are presented:

1. We demonstrate that the nonclassical simplified PN equations recently introduced in [13] are special
cases of the nonclassical Boltzmann equation (1). In particular, we derive explicit expressions for p(s)
and Σt (s) such that the nonclassical simplified P2 and P3 equations can be exactly represented by a
nonclassical transport equation. This fully generalizes the results introduced in [11,12]. (This result
has been shown for the nonclassical simplified P1 in [12].)

2. We show that the moments of the transport free-path distribution p(s) are approximated with increas-
ing accuracy as the order of the SPN equations increases. If p(s) is an exponential, even moments up
to 2N are exactly preserved.

3. We establish that the sampling of s from the probability function p(s) for this generalized results
can be explicitly performed in terms of a computed-generated random number. This allows us to use
Monte Carlo methods to solve these equations.

4. We present for the first time numerical simulations that validate the theory proposed in references
[11,12], as well as the generalized theory introduced here. We consider transport in slab geometry
and perform Monte Carlo simulations in which the free-paths are sampled from the appropriate non-
exponential distributions, demonstrating that they match the solutions obtained with both the classical
and the nonclassical forms of the simplified PN equations. This effectively shows that it is possible to
solve diffusion, SP2 , and SP3 problems using a nonclassical Monte Carlo transport method.

The remainder of the paper is organized as follows. In Section 2 we describe the classical and nonclassical
simplified PN equations. A sketch of the integral formulation for Eq. (1) is presented in Section 3. In

2/21
 Transport Representations of Simplified PN Equations

Section 4 we use Green’s function analysis to show that the nonclassical simplified PN equations can be
represented as nonclassical transport equations. Section 5 presents the numerical results that validate the
theory. The paper concludes with a brief discussion in Section 6.

2. Simplified PN Equations

The nonclassical simplified PN equations (Nc-SPN ) are a set of diffusion approximations to Eq. (1) derived
by Vasques and Slaybaugh using a high-order asymptotic expansion [13]. This asymptotic analysis requires
the first 2M raw moments of the free-path distribution p(s) to be finite in order to obtain the Nc-SPN
equations for N = M . Here, we will limit ourselves to presenting the Nc-SP1 , Nc-SP2 , and Nc-SP3
formulations for the nonclassical transport equation (1).

Let us define
Z ∞
m
sm p(s)ds


s := (7)
0

as the mth raw moment of the free-path distribution function p(s). The second-order Nc-SPN equations are
explicitly given as follows [13]:

(I) Nonclassical simplified P1 equation (Nc-SP1 ):

1 s2 2


1−c
−
 ∇ Φ(x) +
 Φ(x) = Q(x). (8)
6 s s

(II) Nonclassical simplified P2 equation (Nc-SP2 ):

1 s2 2

  

 
−
 ∇ Φ(x) + λ1 (1 − c)Φ(x) − s Q(x) + (9)
6 s
1 − c   

 1 − β1 (1 − c) Φ(x) = 1 − β1 (1 − c) Q(x),
s

where λ1 and β1 are constants given by

3 s4 1 s3



λ1 =
 −

 , (10a)
10 s2 2 3 s s2
1 s3


β1 =

2  − 1. (10b)
3 s s

(III) Nonclassical simplified P3 equations (Nc-SP3 ):

1 s2 2 

  
 1−c
−
 ∇ 1 + β1 (1 − c) Φ(x) + 2ν(x) +
 Φ(x) = Q(x), (11a)
6 s s

2  
1 s λ1 1 − β2 (1 − c)
−
 ∇2 Φ(x) + λ2 ν(x) +
 ν(x) = 0, (11b)
6 s 2 s

3/21
 I. Makine, R. Vasques, R.N. Slaybaugh

where λ2 and β2 are constants given by

"
2 #
9
5  27 s s6

3 
4 
10 s3



1 s s
λ2 =
2 
3 

 s −
 +3
 −
 , (12a)
10 s s − 9 s s4 5 21 s2 s2 3 s
"
2 #
1 10 s3 9
5
β2 =
2 
3 


 − s − 1, (12b)
10 s s − 9 s s4 3 s 5

and λ1 and β1 are given by Eqs. (10).

The Nc-SPN equations described in items (I)-(III) above represent

m  a generalization of the classical SPN
= s c := m!Σ−m

m
equations. If Σt is independent of s, then Eq. (3) holds and s t . Under these
circumstances, λ1 = 4/5, λ2 = 11/7, β1 = β2 = 0, and the Nc-SPN equations reduce to the classical
formulations as given in [9]:

(IV) Classical simplified P1 equation (SP1 ):

1 2
− ∇ Φ(x) + (1 − c)Σt Φ(x) = Q(x). (13)
3Σt
(V) Classical simplified P2 equation (SP2 ):
 
1 2 4 (1 − c)Σt Φ(x) − Q(x)
− ∇ Φ(x) + + (1 − c)Σt Φ(x) = Q(x). (14)
3Σt 5 Σt

(VI) Classical simplified P3 equations (SP3 ):

 
1 2
− ∇ Φ(x) + 2ν(x) + (1 − c)Σt Φ(x) = Q(x), (15a)
3Σt
 
1 2 2 11
− ∇ Φ(x) + ν(x) + Σt ν(x) = 0. (15b)
3Σt 5 7

The classical SPN formulations described in items (IV)-(VI) are asymptotic approximations of the classical
linear Boltzmann equation (4).

3. Integral Equation Formulation

In this section we sketch the derivation of the integral equation formulation for Eq. (1). A detailed derivation
can be found in [2].

Let S(x) be given by

Z Z ∞
S(x) = c Σt (s0 )ψ(x, Ω0 , s0 )ds0 dΩ0 + Q(x) = cf (x) + Q(x), (16a)
4π 0
where
Z ∞
f (x) = Σt (s0 )φ(x, s0 )ds0 = collision-rate density (16b)
Z0
φ(x, s) = ψ(x, Ω, s)dΩ = nonclassical scalar flux. (16c)
4π

4/21
 Transport Representations of Simplified PN Equations

Equation (1) can now be written as an initial value problem:

∂ψ
(x, Ω, s) + Ω · ∇ψ(x, Ω, s) + Σt (s)ψ(x, Ω, s) = 0, 0 < s, (17a)
∂s
S(x)
ψ(x, Ω, 0) = . (17b)
4π
Following the work in [2] and [11], we perform the following steps:

1. Calculate the solution of Eqs. (17) using the method of characteristics;

R R∞
2. Operate on this solution by 4π 0 Σt (s)(·)dsdΩ;

3. Perform a change of spatial variables, from the three-dimensional (3-D) spherical (Ω, s) to the 3-D
Cartesian x0 = x − sΩ.

This yields
p(|x0 − x|)
Z Z Z
f (x) = S(x0 ) dV 0 , (18)
4π|x0 − x|2
where p(|x0 − x|) is given by Eq. (2).

For classical transport, p(s) = pc (s) as given in Eq. (3), and

0 0
Σt e−Σt |x −x| 0 Σt e−Σt |x −x| 0
Z Z Z Z Z Z
0 0 0
f (x) = S(x ) dV = [cf (x ) + Q(x )] dV . (19)
4π|x0 − x|2 4π|x0 − x|2

This is the classical integral equation for the scalar flux obtained from Eq. (4). Sampling of s is given by
Z s
1 1
ξ= pc (s0 )ds0 = 1 − e−Σt s =⇒ s = − ln(1 − ξ) = − ln(ξ). (20)
0 Σt Σt

Table I presents expressions and numerical values of the first six moments of pc (s).

Moment General Expression Numerical Value if Σt = 1

1


s c Σt 1
2!

2
s c Σ2t
2
3!

3
s c Σ3t
6
4!

4
s c Σ4t
24
5!

5
s c Σ5t
120
6!

6
s c Σ6t
720

Table I: Moments of pc (s) (classical transport)

5/21
 I. Makine, R. Vasques, R.N. Slaybaugh

4. Exact Transport Representations of the Nc-SPN Equations

In this section we perform a Green’s function analysis for each of the Nc-SPN equations described in Sec-
tion 2. The goal is to convert these equations into an integral equation for the scalar flux Φ(x). By choosing
the appropriate p(s), the integral equation (18) becomes equivalent to the integral formulation obtained by
this Green’s function analysis. Therefore, using the corresponding Σt (s), Eq. (1) becomes an exact repre-
sentation of the Nc-SPN equations.

4.1. Nc-SP1 equation (diffusion)

This nonclassical result was originally presented in [12] and is included here for completeness. We define

−1
S(x) = c s Φ(x) + Q(x) (21)

and rewrite Eq. (8) as

−∇2 Φ(x) + α2 Φ(x) = α2 s S(x),


(22)
where α2 = 6/ s2 . The Green’s function for the operator (−∇2 + α2 ) on the left hand side of Eq. (22) is



0
e−α|x−x |
Gsp1 (|x − x0 |) = . (23)
4π|x − x0 |

Therefore, we can transform Eq. (22) into an integral equation for Φ(x) by taking
0
α s |x − x0 |e−α|x−x |
Z Z Z Z Z Z 2

0 0 0
2
S(x0 )dV 0 . (24)


Φ(x) = Gsp1 (|x − x |)α s S(x )dV =
4π|x − x0 |2


Bearing in mind that s represents the mean free path of a particle (i.e. the average distance between

−1
collisions), the collision-rate density can be written as f (x) = s Φ(x), such that
0
α2 |x − x0 |e−α|x−x |
Z Z Z
Φ(x)
f (x) =
 = S(x0 )dV 0 . (25)
s 4π|x − x0 |2

This result agrees with Eq. (18) if and only if

√ 2
6se− 6/<s >s
p(s) = psp1 (s) := α2 se−αs =
 . (26)
s2

R∞
It is easy to verify that psp1 (s) is always positive and that 0 psp1 (s)ds = 1, which proves that it is a
probability density function. The total cross section is given by

psp1 (s) α2 s
Σt (s) = Σt,sp1 (s) := R ∞ = . (27)
s psp1 (s0 )ds0 1 + αs

This shows that the nonclassical Eqs. (1) and (18) with p(s) = psp1 (s) and Σt (s) = Σt,sp1 (s) reproduce the
Nc-SP1 equation. We can calculate the mth raw moments of psp1 (s):
Z ∞ m+1 −√6/<s2 >s
m/2

m 6s e (m + 1)! s2
s sp1 :=
 ds = . (28)
0 s2 6m/2

6/21
 Transport Representations of Simplified PN Equations

If classical transport takes place, the moments in Table I hold and

√
psp1 (s) = 3Σ2t se− 3Σt s
, (29)

which is the probability function derived in [11] for the SP1 equation (13). The total cross section becomes

3Σ2t s
Σt,sp1 (s) = √ . (30)
1 + 3Σt s

Table II shows the nonclassical and classical expressions for the moments of psp1 (s), as well as numerical
values of the classical moments when Σt = 1. It can be seen from the general expressions that the second
moment of
the original transport p(s) is always exactly preserved. Comparing to the classical transport
moments sm c (see Table I), the first moment is slightly overestimated while the remaining higher-order
moments are underestimated.

Nonclassical Classical Numerical Value

Moment Expression Expression if Σt = 1

1/2
√2 s2 √2


s sp1 6 3Σt
1.1547
2

2
2
s sp1 s Σ2t
2

2 3/2
√4 √8 3

3
s sp1 6
s 3Σt
4.6188

10 2 2 40

4

s sp1 3 s 3Σ4t
13.3333
√

10 6 2 5/2 √80 5

5
s sp1 3 s 3Σt
46.1880

6 70

2 3 560
s sp1 3 s 3Σ6t
186.6667

Table II: Moments of psp1 (s) (diffusion)

Sampling of s is given by
√ √
     
Z s
−αs0 6 6
ξ= α2 s0 e ds0 = 1 − (1 + αs)e−αs = 1 − 1 +  q
  s exp − q
  s . (31)
0 s2 s2

Hence, q

1 s2
s = τ −1 (ξ) = √ τ −1 (ξ) (32)
α 6
where
τ (z) = (1 + z)e−z . (33)
In the classical case, this expression reduces to [11]
1
s= √ τ −1 (ξ). (34)
3Σt

7/21
 I. Makine, R. Vasques, R.N. Slaybaugh

4.2. Nc-SP2 equation

Using the definition in Eq. (21), we can rewrite Eq. (9) as

2
s Φ λ1
2  2
−
 ∇2 [(1 + λ1 )Φ] + [1 − β1 (1 − c)]
 = [1 − β1 (1 − c)] S − s ∇ S. (35)
6 s s 6

Let us define a new constant α such that

6 1
α2 =
2  [1 − β1 (1 − c)] . (36)
s 1 + λ1


s 1
Multiplying Eq. (35) by 6 2  1+λ

1
we obtain
s

λ1
 2
−∇2 Φ + α2 Φ = α2 s S −


s ∇ S, (37)
1 + λ1
which can be rewritten as
1 λ1

−∇2 + α2 Φ = α2 s S + s −∇2 + α2 S.
 


(38)
λ1 + 1 1 + λ1

The Green’s function associated with the operator −∇2 + α2 is

0
e−α|x−x |
Gsp2 (|x − x0 |) = . (39)
4π|x − x0 |
Using the Green’s function, we can write

α2 s

 Z Z Z
λ1

Φ(x) = Gsp2 SdV 0 + s S, (40)
1 + λ1 1 + λ1
and the collision-rate density is given by

α2
Z
Φ(x) λ1
f (x) =
 = Gsp2 SdV 0 + S. (41)
s 1 + λ1 1 + λ1

Note that the identity

Z ∞
δ(|x − x0 |)S(x0 ) 0
Z Z Z Z
1
S(x) = S(x + sΩ)δ(s)ds = δ(s)S(x + sΩ)dsdΩ = dV (42)
0 4π 4π 4π 4|x − x0 |2

holds, with x0 = x + sΩ, |x − x0 | = s, s2 dsdΩ = dV 0 . Therefore, we can rewrite the collision-rate density
as
0
α2 |x − x0 |e−α|x−x | δ(|x − x0 |)
Z Z Z Z Z Z
0 0 λ1
f (x) = S(x )dV + S(x0 )dV 0 . (43)
1 + λ1 4π|x − x0 |2 1 + λ1 4π|x − x0 |2
This result agrees with Eq. (18) if and only if

α2 se−αs λ1
p(s) = psp2 (s) := + δ(s). (44)
1 + λ1 1 + λ1

8/21
 Transport Representations of Simplified PN Equations

R∞
It is easy to verify that psp2 (s) is always positive and that 0 psp2 (s)ds = 1, proving that it is a probability
density function. We can determine the total cross section, which is written as
λ1
λ1 +1 δ(s) + α2 s
Σt (s) = Σt,sp2 (s) := . (45)
1 + αs

The nonclassical Eqs. (1) and (18) with p(s) = psp2 (s) and Σt (s) = Σt,sp2 (s) reproduce the Nc-SP2
equation. We can calculate the mth raw moments of psp2 (s):
Z ∞

m (m + 1)!
s sp2 := sm psp2 (s)ds = . (46)
0 αm (1 + λ1 )

If classical transport takes place, the moments in Table I hold and

25 2 −Σt √5/3s 4
psp2 (s) = Σ se + δ(s), (47)
27 t 9
which is the probability function derived in [11] for the SP2 equation (14). The total cross section becomes
4
9 δ(s) + 5 Σ2 s
Σt,sp2 (s) = p 3 t . (48)
1+ 5/3Σt s

Table III shows the nonclassical and classical expressions for the moments of psp2 (s), as well as numerical
values of the classical moments when Σt =
1. The classical moments of psp2 (s) give a more accurate
estimate of the classical transport moments sm c (Table I) than the ones obtained from psp1 (s) (Table II).
In particular, both the second and the fourth moments of the original transport p(s) are exactly preserved.
However, these moments are not generally preserved in the nonclassical case; for instance, the general
nonclassical second moment is conserved only if β1 = 0.

Nonclassical Classical Numerical Value

Moment Expression Expression if Σt = 1
q

2 s2 √
√ √ 20


s sp2 √ √
27Σt
0.8606
6 1+λ1 1−β1 (1−c)

2

2 s 2
s sp2 Σ2t
2
1 − β1 (1 − c)
√
2 3/2 √
√4 1+λ1 8 3

3
s sp2 √
3
s √
5Σ3t
6.1968
6 (1−β1 (1−c))

4 10(1+λ1 )
2 2 24
s sp2 3(1−β1 (1−c))2
s Σ4t
24
√ √
10 6(1+λ1 )3/2
2 5/2 144
√ 53

5
s sp2 3(1−β1 (1−c))5/2
s 5Σt
111.5419

6 70(1+λ1 )2
2 3 3024
s sp2 3(1−β1 (1−c))3
s 5Σ6t
604.8

Table III: Moments of psp2 (s)

9/21
 I. Makine, R. Vasques, R.N. Slaybaugh

Sampling of s is given by
Z s
λ1 1
psp2 (s0 )ds0 = 1 − (1 + αs)e−αs .


ξ= + (49)
0 1 + λ1 1 + λ1

Hence, (
λ1
0, for 0 < ξ < 1+λ1
s= 1 −1 λ1
, (50)
ατ ((λ1 + 1)(1 − ξ)), for ξ > 1+λ1

where τ (z) is defined in Eq. (33). In the classical case, this expression reduces to [11]
( 4
0, for 0 < ξ < 9
s= √ . (51)
√3 1 τ −1 9 4


5 Σt 5 (1 − ξ) , for ξ > 9

4.3. Nc-SP3 equations

Using the definition in Eq. (21), we can rewrite Eqs. (11) as

2
s 1
−
 ∇2 [1 + β1 (1 − c)] Φ + 2ν +
 Φ = S,
 
(52a)
6 s s

2
1 λ1 s2 2


λ2 s 2 1 − β2 (1 − c)
−
 ∇ ν+
 ν=
 ∇ Φ. (52b)
6 s s 2 6 s

This system can be rewritten as

2
s 1
−
 ∇2 [1 + β1 (1 − c)] Φ + 2ν +
 Φ = S,
 
(53a)
6 s s

2   " #
s λ1 1 − β (1 − c) 1 λ Φ
−
 λ2 − ∇2 ν +
2  ν=
1

 −S . (53b)
6 s 1 + β1 (1 − c) s 2 1 + β1 (1 − c) s

We need to find the Green’s functions associated to this system, which must satisfy

2
s 1 Φ
−
 ∇2 [1 + β1 (1 − c)] GΦ ν
 
sp3 + 2Gsp3 +
 Gsp3 = δ(x), (54a)
6 s s
GΦ

2  
s λ1 1 − β2 (1 − c) ν 1 λ1
−
 λ2 − 2 ν
∇ Gsp3 +
 Gsp3 −
sp3 (54b)
6 s 1 + β 1 (1 − c) s 2 1 + β1 (1 − c) s
1 λ1
=− δ(x).
2 1 + β1 (1 − c)

Here, GΦ ν
sp3 and Gsp3 are functions of r = |x| that enable Eqs. (53) to be written as
Z Z Z
0 0 0
Φ0 (x) = GΦ sp3 (|x − x |)S(x )dV , (55a)
Z Z Z
ν(x) = Gνsp3 (|x − x0 |)S(x0 )dV 0 . (55b)

10/21
 Transport Representations of Simplified PN Equations

Following the procedure presented in [11], we seek GΦ ν

sp3 (r) and Gsp3 (r) that satisfy

2
s 1 ∂ ∂  1 Φ
−
 2 r2 [1 + β1 (1 − c)] GΦ ν

sp3 + 2Gsp3 +
 Gsp3 = 0, (56a)
6 s r ∂r ∂r s

2  
s λ1 1 ∂ 2∂ ν 1 − β2 (1 − c) ν
−
 λ2 − 2
r Gsp3 +
 Gsp3 (56b)
6 s 1 + β 1 (1 − c) r ∂r ∂r s
" #
1 λ1 GΦ
sp3
−
 = 0;
2 1 + β1 (1 − c) s

with
" !#
∂GΦ ∂Gνsp3 ()

2
s 2 sp3 ()
−
 lim 4π [1 + β1 (1 − c)] +2 = 1, (57a)
6 s →0 ∂r ∂r
∂Gνsp3 ()

2    
s λ1 2 1 λ1
−
 λ2 − lim 4π =− . (57b)
6 s 1 + β1 (1 − c) →0 ∂r 2 1 + β1 (1 − c)

Specifically, we are looking for the Green’s functions of the form

e−αs
GΦ
sp3 = , (58a)
4πs
e−αs
Gνsp3 =a , (58b)
4πs
where α and a are constants to be determined.

Using that ∇2 GΦ 2 Φ 2 ν 2 ν
sp3 = α Gsp3 and ∇ Gsp3 = α Gsp3 , Eqs. (56) yield

2
s 
[1 + β1 (1 − c)] α2 + 2aα2 + 1 = 0,

− (59a)

62   
s λ1 1 λ1
− λ2 − α2 a + (1 − β2 (1 − c))a − = 0. (59b)
6 1 + β1 (1 − c) 2 1 + β1 (1 − c)

From Eq. (59b) we can determine a, which is given by

1 λ1
2 1+β1 (1−c)
a=
 . (60)
s2

λ1
(1 − β2 (1 − c)) − 6 λ2 − 1+β1 (1−c) α2

We can now write Eq. (59a) as

 

2 λ1
s  z1  2
− z1 +
  α + 1 = 0, (61)
6 s2 λ1

z2 − 6 λ2 − z1 α2

where

z1 = 1 + β1 (1 − c), (62a)
z2 = 1 − β2 (1 − c). (62b)

11/21
 I. Makine, R. Vasques, R.N. Slaybaugh

After some manipulations, we attain

2  !2  
2
s λ1 4 s
− z1 λ2 − α − (λ2 + z1 z2 ) α2 + z2 = 0. (63)
6 z1 6

The roots of this polynomial are given by

q
1 1
2 3 (λ2 + z1 z2 ) ± 9 (λ2 + z1 z2 )
2 − 49 (λ2 z1 − λ1 )z2
± 2


α =
2 2 . (64)
s 9 (λ2 z1 − λ1 )

Now we can revisit the expression for a; keeping only the non-complex solutions, we have
1 λ1
2 z1
a± =  q . (65)
3 1 1
z2 − 2z1 3 (λ2 + z1 z2 ) ± 9 (λ2 + z1 z2 )2 − 94 (λ2 z1 − λ1 )z2

Thus, we have two solutions of Eqs. (56): one for α+ and a+ , and the other for α− and a− . The general
solution is a linear combination of these two solutions:
! !
A + e−α+ s A − e−α− s
GΦsp3 (s) =
 +
 , (66a)
s 4πs s 4πs
! !
A + a+ e −α+ s A − a− e −α− s
Gνsp3 (s) =
 +
 , (66b)
s 4πs s 4πs

where A± are determined by Eqs. (57), such that

2
s
−λ1 3

s2
A+ a + + A− a − = , (67a)
λ2 z1 − λ1
 
1 + λ2 zλ11−λ1

2
+ − s
A +A = 6
2 . (67b)
z1 s

Let us define

2
0+ s +
A = A
2 , (68a)
3 s

2
0− − s
A = A
2 , (68b)
3 s
λ1
b= . (68c)
λ2 z1 − λ1
Then, Eqs. (67) become

A0+ a+ + A0− a− = −b, (69a)

0+ 0−
A +A = 2 (1 + b) /z1 . (69b)

12/21
 Transport Representations of Simplified PN Equations

Finally, we solve this system to obtain

2
− −
s 
A = γ , (70a)
s2

2
+ +
s 
A = γ , (70b)
s2
where
2a+ (1 + b)/z1 + b
γ− = 3 , (71a)
a+ − a−
2a− (1 + b)/z1 + b
γ+ = 3 . (71b)
a− − a+

The Green’s function we seek is given by


1 s h + −α+ s − −α− s
i
GΦ
sp3 (s) =
 γ e + γ e , (72)
4πs s2

and the collision-rate density can be written as

|x − x0 | h + −α+ |x−x0 |
Z Z Z
1 − −α− |x−x0 |
i
f (x) = γ e + γ e S(x0 )dV 0 . (73)
4π|x − x0 |2 s2



This result agrees with Eq. (18) if and only if

s h + −
i
p(s) = psp3 (s) :=
2  γ + e−α s + γ − e−α s . (74)
s
R∞
Although the calculation is not straightforward, psp3 (s) is always positive and 0 psp3 ds = 1, confirming
that it defines a probability density function. We can also determine the total cross section, which is written
as h i

s  γ + e−α+ s + γ − e−α− s
s2
Σt (s) = Σt,sp3 (s) :=     . (75)
γ + α+ +
2  e−α s + γ − α− +
2  e−α− s
1 s + 1 s
s s

The nonclassical Eqs. (1) and (18) with p(s) = psp3 (s) and Σt (s) = Σt,sp3 (s) reproduce the Nc-SP3
equations. We can calculate the mth raw moments of psp3 (s):
Z ∞ " #

m m (m + 1)! γ+ γ−
s sp3 := s psp3 (s)ds =
2  + . (76)
0 s (α+ )(m+2) (α− )(m+2)

If classical transport takes place, the moments in Table I hold and

psp3 (s) = Σ2t s 5.642025e−2.94134Σt s + 0.469086e−1.161256Σt s ,

(77)

which is the probability function derived in [11] for the SP3 equations (15). The total cross section becomes

5.642025(Σ2t s)e−2.94134Σt s + 0.469086(Σ2t s)e−1.161256Σt s

Σt,sp3 (s) = . (78)
5.642025( 1+2.94134Σ
8.651481
ts
)e −2.94134Σt s + 0.469086( 1+1.161256Σt s )e−1.161256Σt s
1.348515

13/21
 I. Makine, R. Vasques, R.N. Slaybaugh

Table IV presents the classical expressions for the moments of psp3 (s), as well as numerical values for
Σt = 1. The nonclassical expressions are not explicitly given, but can be obtained from Eq. (76); they
are consistent with the classical expressions. Comparing to the corresponding classical transport moments
in Table I, the moments are preserved more accurately than in the previous cases (SP1,2 ). In particular,
the second, fourth, and sixth moments are exactly preserved as expected. However, these moments are not
generally preserved in the nonclassical case.

Moment Classical Expression Numerical value if Σt = 1

1.042533


s sp3 Σt 1.0425
2

2
s sp3 Σ2t
2
5.94625

3
s sp3 Σ3t
5.9462
24

4
s sp3 Σ4t
24
120.734028

5
s sp3 Σ5t
120.7340
720

6
s sp3 Σ6t
720

Table IV: Moments of psp3 (s)

Sampling of s is given by

γ− h
Z s
γ+ h +
i −
i
ξ= p(s0 )ds0 = + 2 1 − (1 + α+ s)e−α s + − 2 1 − (1 + α− s)e−α s = F (s), (79)
0 (α ) (α )
such that
s = F −1 (ξ). (80)
In the classical case, this expression becomes
5.642025 
1 − (1 + 2.94134Σt s)e−2.94134Σt s

F (s) = (81)
8.651481
0.469086 
1 − (1 + 1.161256Σt s)e−1.161256Σt s .

+ (82)
1.348515

5. Numerical Results

This section gives numerical validation for the theoretical results presented in this paper, as well as in
references [11,12]. Specifically, we use a Monte Carlo (MC) transport code in which the free-paths were
sampled from the probability density functions derived in Section 4. The results of this MC transport code
are then compared against the results obtained by deterministically solving the corresponding classical or
nonclassical simplified PN equations. It is important to note that we are not concerned about obtaining the
correct transport solution of these problems. Our goal is to show that the nonclassical MC transport code
correctly reproduces the solutions of the deterministic simplified PN equations.

Due to the challenges of estimating the free-path distribution p(s) in multi-dimensional nonclassical sys-
tems, this task will be left for future work. In this paper, we consider only one-dimensional (1-D) slab

14/21
 Transport Representations of Simplified PN Equations

geometry transport. Two types of problems have been analyzed: (I) classical transport, and (II) nonclassi-
cal transport in an ensemble-averaged (homogenized) periodic random medium. For each of these cases,
we have simulated two different internal source configurations: (A) a homogeneous source throughout the
whole system (global source); and (B) a constant source located in a small region in the center of the system
(local source).

Sections 5.1 to 5.4 describe each set of problems and present the results obtained. All problems obey the
following simplifying assumptions:

• Transport takes place in slab geometry.

• Transport is monoenergetic and scattering is isotropic.

• The source Q emmits particles isotropically.

• There are no incoming particles through the boundaries of the system; that is, vacuum boundary
conditions.

5.1. Problem set I.A: classical transport, global source

Consider a slab with dimensions −50 ≤ x ≤ 50 composed of a homogeneous material with Σt = 1. We

assume that there is a homogeneous source Q = 1 throughout the whole system. Under these assumptions,
Eq. (3) holds and the classical SPN equations apply. In a diffusive system, it is expected that the scalar
fluxes obtained from solving SP1 [Eq. (13)], SP2 [Eq. (14)], and SP3 [Eqs. (15)] will converge to the same
value, which should approximate the scalar flux obtained from the solution of Eq. (4). As the system
becomes less diffusive, the scalar flux away from the boundaries should converge to the “infinite solution”
Φ(x) = Q/[(1 − c)Σt ].

This can be seen in Fig. 1, in which we present results of the simulations performed with the nonclassical
MC transport code. We sample the free-paths from Eq. (29) (MC SP1 ), Eq. (47) (MC SP2 ), and Eq. (77) (MC
SP3 ). For the diffusive system with c = 0.999, the SPN solutions converge to the same estimate of the scalar
flux (Fig. 1a). For the system with c = 0.2, the SPN solutions converge to Φ(x) = 1/(1 − 0.2) = 1.25, as
expected (Fig. 1b).

Figure 2 is a summary of the results obtained for this set of problems showing estimates for the scalar flux
at x = 0 for different values of c. The system varies from diffusive to absorbing, and the results of the MC
transport code (MC SP1,2,3 ) match those obtained by deterministically solving the SPN equations (SP1,2,3 ).
This is as expected; however, since the solutions of the different equations overlap, this set of results does
not provide by itself an appropriate validation of the theoretical predictions.

5.2. Problem set I.B: classical transport, local source

Consider a slab composed of a homogeneous material with Σt = 1. We assume a constant source defined
in a region of the system such that
(
1, for − 0.5 < x < 0.5
Q(x) = (83)
0, elsewhere.

15/21
 I. Makine, R. Vasques, R.N. Slaybaugh

(a) Flux for c = 0.999 (diffusive case) (b) Flux for c = 0.2 (nondiffusive case)

Figure 1: MC simulations for classical transport with global homogeneous source.

Figure 2: MC and deterministic estimates for the scalar flux (in log) at x = 0: problem set I.A

We allow the system dimensions to be as large as needed in order for the leakage to be negligible. This
means that dimensions will increase as c increases and the system becomes more diffusive.

As in the previous case, Eq. (3) holds and the classical SPN equations apply. If the system is diffusive, we
expect that the scalar fluxes will approximate the scalar flux obtained from the solution of Eq. (4). However,
as the system becomes less diffusive, the SPN equations should yield different results.

Figure 3 shows the results of the simulations performed with the nonclassical MC transport code. Once
again, for the diffusive system with c = 0.999, the SPN solutions converge to the same estimate of the
scalar flux (Fig. 3a). However, we can see that the SPN solutions yield different results for nondiffusive
systems as depicted in Fig. 3b.

Figure 4 is a summary of the results obtained for this set of problems. It shows estimates for the scalar flux
at x = 0 for different values of c. In particular, we see that the MC transport code (MC SP1,2,3 ) accurately
matches the solutions obtained by solving the SPN equations (SP1,2,3 ) deterministically. This validates the
theoretical predictions for the classical SPN transport representations originally introduced in [11].

16/21
 Transport Representations of Simplified PN Equations

(a) Flux for c = 0.999 (diffusive case) (b) Flux for c = 0.8 (nondiffusive case)

Figure 3: MC simulations for classical transport with constant local source

Figure 4: MC and deterministic estimates for the scalar flux (in log) at x = 0: problem set I.B

5.3. Problem set II.A: nonclassical transport, global source

We consider a random periodic system as depicted in Fig. 5. It represents a random segment of a periodic
medium with a period ` = 1.0, containing alternate layers of material 1 (solid) with thickness `1 = 0.5, and
material 2 (void) with thickness `2 = 0.5. Thus, the probability to find material i = 1, 2 in position x is
given by Pi = `i /` = 1/2.

Let us assume that this random periodic segment has dimensions −50 ≤ x ≤ 50, and that material 1 is a
homogeneous solid with Σt = 1. We also assume that there is a homogeneous source Q = 1 throughout the
whole system. Under these assumptions, the free-path distribution in the ensemble-averaged (homogenized)

Figure 5: Sketch of the random periodic medium

17/21
 I. Makine, R. Vasques, R.N. Slaybaugh

(a) Flux for c = 0.999 (diffusive case) (b) Flux for c = 0.2 (nondiffusive case)

Figure 6: MC simulations for nonclassical transport with global homogeneous source.

system is not and exponential. We can numerically estimate the moments of this homogenized angular-
dependent free-path distribution; these are given in Table V. For this class of problems we must apply the
Nc-SPN equations using the estimated moments given in Table V.


2
3
4
5
6
s s s s s s

2.000467 8.791506 56.152561 457.65052 4521.87604 52969.8811

Table V: Estimated moments of the free-path distribution in the homogenized random periodic medium

Similarly to the results in Section 5.1, it is expected that in a diffusive system the scalar fluxes obtained
from solving Nc-SP1 [Eq. (8)], Nc-SP2 [Eq. (9)], and Nc-SP3 [Eqs. (11)] will converge to the same value,
which should approximate the scalar flux obtained from the solution of Eq. (1). As the system becomes less
diffusive, the scalar flux away from the boundaries should converge to the volume-averaged infinite solution
Φ(x) = Q/[(1 − c)P1 Σt ].

This can be seen in Fig. 6, in which we present results of the simulations performed with the nonclassical
MC transport code. We sample the free-paths from Eq. (26) (MC SP1 ), Eq. (44) (MC SP2 ), and Eq. (74)
(MC SP3 ). For the diffusive system with c = 0.999, the Nc-SPN solutions converge to the same estimate
of the scalar flux (Fig. 6a). For the system with c = 0.2, the Nc-SPN solutions converge to Φ(x) =
1/[(1 − 0.2)0.5] = 2.5, as expected (Fig. 6b).

Figure 7 is a summary of the results obtained for this set of problems, showing estimates for the scalar flux
at x = 0 for different values of c. As in the classical case, the MC transport code (MC SP1,2,3 ) closely
agrees with the deterministic solutions of the Nc-SPN equations (SP1,2,3 ).

18/21
 Transport Representations of Simplified PN Equations

Figure 7: MC and deterministic estimates for the scalar flux (in log) at x = 0: problem set II.A

5.4. Problem set II.B: nonclassical transport, local source

Assume a slab composed of a random segment of the periodic medium described in Section 5.3. We consider
a constant source defined in a region of the system such that
(
1, for − 0.5 < x < 0.5
Q(x) = (84)
0, elsewhere.

As in Section 5.2, we allow the system dimensions to be as large as needed in order for the leakage to be
negligible.

We expect the Nc-SPN equations to yield similar results when the system is diffusive. This can be seen
in Fig. 8a. As c decreases, the solutions of the Nc-SPN equations differ, as depicted in Fig. 8b. Figure 9
presents a summary of the results obtained for this set of problems, in which it is clear that the MC transport
code (MC SP1,2,3 ) closely reproduces the deterministic solutions of the Nc-SPN equations (SP1,2,3 ).

We note that the MC estimates for the Nc-SP3 solutions present relative errors of 4%-5% for a few cases.
This is mainly due to the numerical and statistical errors introduced in the MC calculations by numerically
estimating the moments in Table V. Nevertheless, it is clear that the MC simulations predict the overall
behavior of the Nc-SP3 equations. These numerical results validate the original theory derived in this paper,
demonstrating that the nonclassical transport equation (1) can exactly represent the Nc-SPN equations.

6. Discussion

In this paper we have shown that the nonclassical simplified P1,2,3 equations (Nc-SP1,2,3 ) can be represented
exactly by the nonclassical transport equation (1). We derived an explicit expression for the free-path distri-
bution p(s) for each of these equations, and showed that these expressions are a generalization of the ones
previously obtained for the classical SP1,2,3 equations. We have shown that the moments of the transport
p(s) are approximated with increasing accuracy as the order N increases, with the even moments up to 2N
being preserved when p(s) is exponential.

19/21
 I. Makine, R. Vasques, R.N. Slaybaugh

(a) Flux for c = 0.999 (diffusive case) (b) Flux for c = 0.2 (nondiffusive case)

Figure 8: MC simulations for nonclassical transport with constant local source

Figure 9: MC and deterministic estimates for the scalar flux (in log) at x = 0: problem set II.B

Moreover, we present numerical simulations that validate our theoretical predictions as well as those pre-
sented in previous work [11,12]. We show that a nonclassical Monte Carlo transport code that samples
the free-paths from nonexponential distributions can accurately reproduce the deterministic solutions of the
Nc-SPN equations and the classical SPN equations. This has not been done before.

Although performed only in slab geometry, these numerical results pave the road to consistently simulate
these diffusion-based approximations using a Monte Carlo transport method. Furthermore, it hints to the
possibility of using these analytical formulations of p(s) to approximate the solutions of the nonclassical
Boltzmann equation in near-diffusive systems. Further work needs to be done to investigate how well this
approach performs in multi-dimensional nonclassical systems. However, this task must be left for future
work.

Acknowledgments

R. Vasques acknowledges support under award number NRC-HQ-84-15-G-0024 from the Nuclear Regu-
latory Commission. R.N. Slaybaugh acknowledges support under award number NRC-HQ-84-14-G-0052

20/21
 Transport Representations of Simplified PN Equations

from the Nuclear Regulatory Commission. The statements, findings, conclusions, and recommendations are
those of the authors and do not necessarily reflect the view of the U.S. Nuclear Regulatory Commission.

References

[1] E.W. Larsen, “A generalized Boltzmann equation for non-classical particle transport”, Proceedings
of the Joint International Topical Meeting on Mathematics & Computation and Supercomputing in
Nuclear Applications, Monterey, CA, Apr. 15-19 (2007).
[2] E.W. Larsen and R. Vasques, “A Generalized Linear Boltzmann Equation for Non-Classical Particle
Transport”, Journal of Quantitative Spectroscopy and Radiative Transfer, 112, pp. 619–631 (2011).
[3] A.B. Davis and A. Marshak, “Solar radiation transport in the cloudy atmosphere: A 3D perspective
on observations and climate impacts”, Reports on Progress in Physics, 73, 026801 (2010).
[4] R. Vasques and E.W. Larsen, “Non-classical particle transport with angular-dependent path-length
distributions. II: Application to pebble bed reactor cores”, Annals of Nuclear Energy, 70, pp. 301–
311 (2014).
[5] E. d’Eon, “Rigorous asymptotic and moment-preserving diffusion approximations for generalized
linear boltzmann transport in arbitrary dimension”, Transport Theory and Statistical Physics, 42, pp.
237–297 (2013).
[6] E.M. Gelbard, “Aplications of spherical harmonics method to reactor problems”, Technical Report
WAPD-BT-20, Bettis Atomic Power Laboratory (1960).
[7] E.M. Gelbard, “Simplified spherical harmonics equations and their use in shielding problems”, Tech-
nical Report WAPD-T-1182, Bettis Atomic Power Laboratory (1961).
[8] E.M. Gelbard, “Applications of simplified spherical harmonics equations in spherical geometry”,
Technical Report WAPD-TM-294, Bettis Atomic Power Laboratory (1962).
[9] E.W. Larsen, J.E. Morel, J.M. McGhee, “Asymptotic Derivation of the Simplified PN Equations”,
Proceedings of the ANS Topical Meeting on Mathematical Methods and Supercomputing in Nuclear
Applications, Karlsruhe, Germany, Apr. 19-23, 2 (1993).
[10] R.G. McClarren, “Theoretical aspects of the simplified Pn equations”, Transport Theory and Statis-
tical Physics, 39, pp. 73–109 (2011).
[11] M. Frank, K. Krycki, E.W. Larsen, R. Vasques, “The nonclassical Boltzmann equation and diffusion-
based approximations to the Boltzmann equation,” SIAM Journal on Applied Mathematics, 75,
pp.1329–1345 (2015).
[12] R. Vasques, “The nonclassical diffusion approximation to the nonclassical linear Boltzmann equa-
tion,” Applied Mathematics Letters, 53, pp. 63–68 (2016).
[13] R. Vasques and R.N. Slaybaugh, “Simplified PN Equations for Nonclassical Transport with Isotropic
Scattering,” Proceedings of the International Conference on Mathematics and Computational Meth-
ods Applied to Nuclear Science & Engineering - M&C 2017, Jeju, Korea, Apr. 16–20 (2017).

21/21
View publication stats