Eigenvalue problem for a multiplying system discretized into N cells: ... to use (0)F instead of (n)F, to obtain higher eigenpairs. ..... F Sn is equivalent to Sn = 1 kn.
ANS Winter Meeting Las Vegas, NV, Nov. 610, 2016
The Iterated Fission Matrix Manuele Auero, Yishu Qiu, Massimiliano Fratoni University of California, Berkeley
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Outline
Introduction:
The Fission Matrix method & `Eigenmode' perturbation theory
How we obtain higher eigentriplets via Monte Carlo transport (...and why we need them)
The Iterated Fission Matrix
Combining Fission Matrix and Iterated Fission Probability
Asymptotically exact eigenvalues w/ nite mesh size (...yet another trick with latent generations)
Conclusions and ongoing work:
A `matrix-free' approach to the Fission Matrix method
(Petrov-)Galerkin eigenvalue method on Krylov subspaces (...it's all about projection) Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Eigenvalue problem for a multiplying system discretized into N cells:
1 F · Sn kn
Sn = Sn,I =
N 1 X FI ←J · Sn,J kn J=1
k0 and
S0 : fundamental eigenpair
FI ←J : average eective number of ssion neutrons produced in cell I from a neutron born in cell J
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method
Two main sources of error: Statistical uncertainty (not discussed here)
Spatial discretization (studied in this work)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Estimates for discretized F are usually obtained in criticality source Monte Carlo simulations:
(0) FI ←J
=
R
R
r 0 ∈VI
r ∈VJ
F(r 0 ← r ) · S0 (r ) · dr dr 0 R
S0 (r ) · dr
r ∈VJ
S0 (r ) is the (continuous) fundamental ssion source distribution estimated via Monte Carlo (0)
F represents a consistent discretization for the solution of: S0 = Manuele Auero UC Berkeley
1 F · S0 k0 The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes Consistent discretization of F(r 0 ← r ) for higher eigenmodes:
R (n) FI ←J
=
R
r 0 ∈VI r ∈VJ
F(r 0 ← r ) · Sn (r ) · dr dr 0 R
Sn (r ) · dr
r ∈VJ
Unfortunately, Because to use
(0)
Sn (r ) is not available!
lim
VI ,VJ →zero
(n) FI ←J
→
(0) FI ←J ,
it is common practice
F instead of (n) F , to obtain higher eigenpairs.
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes Case study: UAM GENIII PWR MOX core (2D)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes
k1 /k0 ' 0.995
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes
k2 /k0 ' 0.995
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes
k3 /k0 ' 0.985
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes
k4 /k0 ' 0.984
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
The Fission Matrix method Higher eigenmodes
k5 /k0 ' 0.972
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory The eect of a perturbation x on the ssion source distribution S0 is projected onto the set of eigenmodes S = S1 , S2 , S3 · · ·
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory The eect of a perturbation x on the ssion source distribution S0 is projected onto the set of eigenmodes S = S1 , S2 , S3 · · ·
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory The eect of a perturbation x on the ssion source distribution S0 is projected onto the set of eigenmodes S = S1 , S2 , S3 · · ·
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory The eect of a perturbation x on the ssion source distribution S0 is projected onto the set of eigenmodes S = S1 , S2 , S3 · · ·
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
dF can be eciently calculated via Monte Carlo (X)GPT for dx several perturbed parameters x in a single Serpent run (collisionhistory approach)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory The eect of a perturbation x on the ssion source distribution S0 is projected onto the set of eigenmodes S = S1 , S2 , S3 · · ·
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
Continuous XGPT estimators for S†i
Manuele Auero UC Berkeley
T
dF S0 dx
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory The eect of a perturbation x on the ssion source distribution S0 is projected onto the set of eigenmodes S = S1 , S2 , S3 · · ·
d S0 = dx
∞ X
Si
†
Si ·
i=1
Td
F
S0
dx k0 − ki
Today's problem
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory Perturbation of
239
Pu resonance parameters -- Γf @0.2956 eV
Cross sections sensitivities (3% Γf relative perturbation)
XS sensitivity [-]
0.5 0.0 -0.5
MT102 MT18
-1.0 4
Reference XS [b]
10
MT102 MT18
3
10
2
10
1
10
0
10
-7
-6
10
10
-5
10
Energy [MeV]
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
`Eigenmode' Perturbation Theory
d S0 = dx
∞ X i=1
Si
†
Si ·
Td
F
S0
dx k0 − ki
1 · 10−3 error on ki → 10% 20% error on the projection coecient Very accurate estimates for higher eigenvalues ki are required (target accuracy . 10−4 )
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
Accuracy of the Fission Matrix estimates for k1
k1 /k0 ' 0.995
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
Accuracy of the Fission Matrix estimates for k1
k1 /k0 ' 0.995
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
Accuracy of the Fission Matrix estimates for k3
k3 /k0 ' 0.985
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
Accuracy of the Fission Matrix estimates for k3
k3 /k0 ' 0.985
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
Accuracy of the Fission Matrix estimates for k5
k5 /k0 ' 0.972
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Fission Matrix method `Eigenmode' Perturbation Theory Accuracy of the Fission Matrix estimates for ki
Accuracy of the Fission Matrix estimates for k5
k5 /k0 ' 0.972
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
The Iterated Fission Matrix (IFM) If kn is a real, positive eigenvalue,
Sn =
1 · kn
F Sn is equivalent to Sn =
1
kn
`
·
`
F Sn
The main idea:
Replace the ssion kernel F(r 0 ← r ) with the `th iterated ssion kernel (`) F(r 0 ← r ), to produce the `th IFM New MC estimators based on
Iterated Fission Probability
` = number of latent generations (as in IFP)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
`Iterated' ssion kernel (self-convolution of the ssion kernel)
F(r 0 ← r )
P FI ←J =
current gen.
wnew · δ(rnew ∈I ) · δ(rprogenitor ∈J)
P prev. gen.
wprogenitor · δ(rprogenitor ∈J)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
`Iterated' ssion kernel (self-convolution of the ssion kernel) (2)
F(r 0 ← r ) = (F ∗ F) (r 0 ← r ) ≡ F(r 0 ← r 00 ) · F(r 00 ← r ) · dr 00 R
V
P (2)
FI ←J =
current gen.
wnew · δ(rnew ∈I ) · δ(rprogenitor ∈J)
P gen. −2
wprogenitor · δ(rprogenitor ∈J)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
`Iterated' ssion kernel (self-convolution of the ssion kernel) (3)
F(r 0 ← r ) ≡
s
F(r 0 ← r 00 ) · F(r 000 ← r 00 ) · F(r 00 ← r ) · dr 00 dr 000
V
P (3)
FI ←J =
current gen.
wnew · δ(rnew ∈I ) · δ(rprogenitor ∈J)
P gen. −3
wprogenitor · δ(rprogenitor ∈J)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Calculating the Iterated Fission Matrix The Iterated Fission Matrix
R (`) (0) FI ←J
=
r 0 ∈V
R I
(`) (0) (`)
F can be derived as:
F(r 0 ← r ) · S0 (r ) · dr dr 0
r ∈VJ
R
S0 (r ) · dr
r ∈VJ
Analog estimator for the numerator in criticality simulation: number of ssion neutrons in cell I with the `th -generation progenitor tagged with cell-label J Propagation of information to descendants as in IFP
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Discretization errors Because of discretization errors, (`) FI ←J VI ,VJ →zero (n)
lim
(`) (0)
F
6=
→
(`) (0) FI ←J
→
(0) FI ←J
(0)
F
`
converges faster than lim
VI ,VJ →zero
(n) FI ←J
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Convergence of the Iterated Fission Matrix for k1
k1 /k0 ' 0.995 1.000 0.995
Dominance ratio k 1/k 0
0.990 0.985 0.980 0.975 0.970
l =1 l =2 l =6 l =11 l =16
0.965 0.960 0.955 0.950 10 0
10 1
10 2
10 3
10 4
Mesh size
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Convergence of the Iterated Fission Matrix for k1
k1 /k0 ' 0.995 10-1 l =1 l =2 l =6 l =11 l =16
Error
10-2
10-3
10-4 0 10
10
1
10
2
10
3
10
4
Mesh size
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Convergence of the Iterated Fission Matrix for k5
k1 /k0 ' 0.972 k5/k0
−2
Error
10
−3
10
g=0 g=1 g=5 g=10 g=15
−4
10
1
10
2
3
10
10 mesh size
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Convergence of the Iterated Fission Matrix...
WHY?
The main assumption behind the Fission Matrix Method is that any neutron born in the cell J has the same probability of producing a neutron in the cell I,
at the next generation
The main assumption behind the Iterated Fission Matrix is that any neutron born in the cell J has the same probability of producing a neutron in the cell I, after ` generations
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
The Iterated Fission Matrix Spatial convergence of the Iterated Fission Matrix Implementation in SerpentOpenFOAM
Implementation in SerpentOpenFOAM OpenFOAM: C++ Finite-Volume multiphysics toolkit SerpentOpenFOAM internal coupling (OF as library)
OpenFOAM classes for unstructured mesh + multiphysics coupling (CFD/Porous Media CFD)
lapacke C wrapper for lapack FORTRAN linear algebra package arpack++ C++ wrapper for arpack FORTRAN library for large sparse (generalized) eigenvalue problems via ARNOLDI method
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Summary The problem
The calculation of higher eigenpairs is often performed adopting the Fission Matrix Method Some application (e.g., perturbation theory) requires very accurate estimates for the eigenvalues ki (∼tens of pcm) FM needs a very ne spatial discretization to achieve accurate ki estimates (might be prohibitive in 3D cases)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Summary Our solution The Iterated Fission Matrix (IFM) is a combination of the Fission Matrix and the Iterated Fission Probability The implementation∗ requires the propagation of cell indexes to descendant ∗
straightforward implementation for MC codes w/ Fission Matrix and
Iterated Fission Probability
IFM showed to converge much faster (in the considered case) → smaller meshes
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Summary Our solution The Iterated Fission Matrix (IFM) is a combination of the Fission Matrix and the Iterated Fission Probability The implementation∗ requires the propagation of cell indexes to descendant ∗
straightforward implementation for MC codes w/ Fission Matrix and
Iterated Fission Probability
IFM showed to converge much faster (in the considered case) → smaller meshes Behavior of statistical errors should be investigated! Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Ongoing work Applications
Applications involving IFM & MC Perturbation Theory: Nuclear data uncertainty propagation for power distribution in large and loosely-coupled reactors Convergence acceleration of coupled Monte CarloCFD multiphysics simulations via Newton iterations (submitted to M&C2017)
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
ARNOLDI iteration for the eigenvalue problem The adoption of ARNOLDI iteration in the SerpentOpenFOAM implementation ensures fast and cheap (CPU & memory) solution of the eigenvalue problem The discretized ssion matrix F is sparse and large If F is N × N , only m N eigenmodes are of interest Full decomposition of F is expensive (CPU & memory) Arnoldi method operates on a Krylov subspace of F 2 3 k K = span v, Fv, F v, F v , · · · , F v Only matrix-vector multiplications, it never operates on F
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Matrix-free approach to the Fission Matrix Method Question Why should we spend resources building a huge matrix F if we never use it? Answer uhhh... maybe we shouldn't! Idea Let's use a method that doesn't need to access elements of F and let it operate directly on the ssion kernel F(r 0 ← r ) What? (Petrov-)Galerkin projection and Krylov subspace expansion on random ssion source distributions
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Fission kernel Petrov-Galerkin projection
Trial subspace: Φ = {φ1 (r ), φ2 (r ), φ3 (r ), · · · , φm (r )} Test subspace: Ψ = {ψ1 (r ), ψ2 (r ), ψ3 (r ), · · · , ψm (r )}
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Fission kernel Petrov-Galerkin projection
Trial subspace: Φ = {φ1 (r ), φ2 (r ), φ3 (r ), · · · , φm (r )} Test subspace: Ψ = {ψ1 (r ), ψ2 (r ), ψ3 (r ), · · · , ψm (r )} Z
Z
aij = F (ψi , φj ) =
ψi (r 0 ) F(r 0 ← r ) φj (r ) · dr dr 0
r 0 ∈V r ∈V
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Fission kernel Petrov-Galerkin projection
Trial subspace: Φ = {φ1 (r ), φ2 (r ), φ3 (r ), · · · , φm (r )} Test subspace: Ψ = {ψ1 (r ), ψ2 (r ), ψ3 (r ), · · · , ψm (r )} Z
Z
aij = F (ψi , φj ) =
ψi (r 0 ) F(r 0 ← r ) φj (r ) · dr dr 0
r 0 ∈V r ∈V
Analog estimator for all the aij in a single MC run:∗ P aij =
current gen.
P prev. gen.
∗
wnew · ψi (rnew ) · φj (rprogenitor )
wprogenitor · φj (rprogenitor ) · φj (rprogenitor )
Correction for S0 (r ) required
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Fission kernel Petrov-Galerkin projection
Trial subspace: Φ = {φ1 (r ), φ2 (r ), φ3 (r ), · · · , φm (r )} Test subspace: Ψ = {ψ1 (r ), ψ2 (r ), ψ3 (r ), · · · , ψm (r )} Z
Z
aij = F (ψi , φj ) =
ψi (r 0 ) F(r 0 ← r ) φj (r ) · dr dr 0
r 0 ∈V r ∈V
Z bij = hψi , φj i =
ψi (r ) φj (r ) · dr r ∈V
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Fission kernel Petrov-Galerkin projection
Trial subspace: Φ = {φ1 (r ), φ2 (r ), φ3 (r ), · · · , φm (r )} Test subspace: Ψ = {ψ1 (r ), ψ2 (r ), ψ3 (r ), · · · , ψm (r )} Z
Z
aij = F (ψi , φj ) =
ψi (r 0 ) F(r 0 ← r ) φj (r ) · dr dr 0
r 0 ∈V r ∈V
Z bij = hψi , φj i =
ψi (r ) φj (r ) · dr r ∈V
A = F (Φ, Ψ) ;
Manuele Auero UC Berkeley
B = hΦ, Ψi
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Fission kernel Petrov-Galerkin projection
Trial subspace: Φ = {φ1 (r ), φ2 (r ), φ3 (r ), · · · , φm (r )} Test subspace: Ψ = {ψ1 (r ), ψ2 (r ), ψ3 (r ), · · · , ψm (r )} Z
Z
aij = F (ψi , φj ) =
ψi (r 0 ) F(r 0 ← r ) φj (r ) · dr dr 0
r 0 ∈V r ∈V
Z bij = hψi , φj i =
ψi (r ) φj (r ) · dr r ∈V
A = F (Φ, Ψ) ;
B = hΦ, Ψi
Reduced (m × m) generalized eigenvalue problem: A y = λB y Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Reduced Eigenvalue problem If λn and
ny
are the nth eigenpair of
A y = λB y,
eigenvalues and eigenmodes of the full problem
Sn = k1n F Sn
can be approximated as:
kn ' λn ;
Sn (r ) ' Φ n y =
m X
yn,h · φh (r )
h=1
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Reduced Eigenvalue problem If λn and
ny
are the nth eigenpair of
A y = λB y,
eigenvalues and eigenmodes of the full problem
Sn = k1n F Sn
can be approximated as:
kn ' λn ;
Sn (r ) ' Φ n y =
m X
yn,h · φh (r )
h=1
If the subspaces Φ and Ψ are not good enough...
Krylov subspace expansion via IFP (in the same run): Φ = {Φ, F Φ, (F ∗ F) Φ, (F ∗ F ∗ F) Φ} Ψ = Ψ, F† Ψ, F† ∗ F† Ψ, F† ∗ F† ∗ F† Ψ Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
First test: 2D homogeneous molten salt mixture
50 orthogonal random vectors (red noise):
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
First test: 2D homogeneous molten salt mixture
50 orthogonal random vectors (red noise):
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Results
10th forward eigenmode 1-G diusion Galerkin
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Results
13th forward eigenmode 1-G diusion Galerkin
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Results
23th forward eigenmode 1-G diusion
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Results
23th forward eigenmode 1-G diusion Galerkin
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Results
23th forward eigenmode Krylov-Galerk. 1-G diusion Galerkin
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
Thank you for the attention!
Questions? Suggestions?
Manuele Auero UC Berkeley
The Iterated Fission Matrix
aaa
Introduction The Iterated Fission Matrix Conclusions and ongoing work
Conclusions Current applications (Petrov-)Galerkin method for the continuous ssion kernel
`Eigenmode' Perturbation Theory Sn =
1 · F Sn kn
S†n =
and
T 1 · F S†n kn
The eigenmodes are bi-orthogonal and normalized such that: T
S† S = I
S†0 S†1 .. .
S†n
S0 S1 · · ·
Manuele Auero UC Berkeley
1 0 ··· 0 1 · · · Sn . . .. = . .. .. 0 0 ···
The Iterated Fission Matrix
0 0 .. .
1
aaa