Title: Numerical methods in the KIKO3D threedimensional reactor dynamics code
1Numerical methods in the KIKO3D three-dimensional
reactor dynamics code
KFKI Atomic Energy Research Institute
19th International Conference on Transport Theory
István Panka, András Keresztúri, Csaba J.
Hegedus panka_at_sunserv.kfki.hu
2KFKI Atomic Energy Research Institute
- INTRODUCTION
- Derivation of the nodal kinetic equations
- Numerical solution of the steady-state problem
- Numerical solution of the time dependent problem
- Generalized Minimal Residual (GMRES) method
- BiConjugate Gradient Stabilized (Bi-CGSTAB)
method - Examples
- Conclusions
-
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
3KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations
- The KIKO3D code 1,2,3
- developed by KFKI AEKI,
- solves the 2 group diffusion equations in
homogenized fuel assembly geometry with a special
advanced nodal method, where generalized response
matrices of the time dependent problem are
introduced, - the unknowns are the scalar flux integrals on
the boundaries, - the time dependent nodal equations are solved by
using the Improved Quasi Static factorization
method
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
4KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations (cont.)
- The time dependent neutron balance in each node
is written into the following form
(1a)
(1b)
where
comprises the terms of the static neutronic
equation,
is the fission operator,
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
5KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations (cont.)
is the two-group scalar flux,
and
are the delayed neutron fractions,
are the precursor densities,
is the fission spectrum.
As a result of the applied advanced nodal method
the time dependent nodal equation for the whole
reactor can be derived, where the unknowns are
the time dependent amplitudes
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
6KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations (cont.)
(2a)
(2b)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
7KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations (cont.)
The factorization of the improved quasi static
(IQS) method is introduced as
(3)
where A(t) is the amplitude function, and f(t) is
the shape function changing slowly with the time.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
8KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations (cont.)
Finally we get for the shape factor equation and
for the precursor equation
(4a)
(4b)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
9KFKI Atomic Energy Research Institute
- Derivation of the nodal kinetic equations (cont.)
And the point kinetic equations
(5a)
(5b)
where W are given by
Equations (4a), (4b), (5a) and (5b) form a close
set to be solved.
and
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
10KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
In the steady-state case we have to solve
(6)
There are cases when this equation do not fulfill
with the input data.
We introduce an artificial parameter being
usually called static reactivity in the reactor
physics in order that Eq. (6) should be
satisfied
(7)
where
This equation is called the static neutronic
diffusion equation at the beginning of the
transient.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
11KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
Now the mathematical problem given in the above
equation can be written as
(8)
This is a special eigenvalue problem, where
is a large sparse matrix depending on the
parameter
The task is that one has to tune so that the
smallest eigenvalue of Eq. (8) equals 0. The
physical problem is such that 0 is the smallest
eigenvalue and the adjusted equals the
static reactivity.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
12KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
In one iteration cycle the applied solution
method can be summarized in four steps.
- Step 1. Fix the value of parameter (initially
it is 0) and calculate
- Step 2. Calculate the left and right eigenvectors
which belong to the smallest eigenvalue of
- Step 3. Estimate the smallest eigenvalue by the
generalized Rayleigh quotient
(9)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
13KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
- Step 4. Calculate the correction term by a linear
approximation of
(10)
We refine the perturbed reactivity as
(11)
These steps together are repeated until the
parameter is less than a given tolerance.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
14KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
- Step 2 is the most time consuming step.
- Some methods for solving the smallest
eigentriple problem will be surveyed here.
- In order to use of methods which result in the
largest eigenvalue and eigenvectors we introduce
an equivalent problem
(12)
Now we are looking for the largest eigenvalue and
eigenvectors of A.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
15KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
Power iteration
(13)
where
is the estimated largest eigenvalue in step i.
- Advantages
- simple
- trusted
- Disadvantage
- its convergence can be slow (convergence is
governed by the ratio )
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
16KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
Variants of power iteration (faster
convergence) Split matrix into two parts and
rearrange the eigenvalue problem
(14)
from which one gets the iteration
(15)
- If A1 is the diagonal part of A then we have the
power Jacobi iteration - If A1 is the (diagonal lower triangular) part
of A then we have the power Gauss-Seidel
iteration
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
17KFKI Atomic Energy Research Institute
- Numerical solution of the steady-state problem
(cont.)
- Advantages
- faster than simple power iteration
- the algorithm can be arranged that the left and
right eigenvectors are calculated by the same
sweep
Power iteration and power Gauss-Seidel iteration
are applied in KIKO3D code.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
18KFKI Atomic Energy Research Institute
- Numerical solution of the time dependent problem
- The shape factor equation (4a) and the equations
for the precursor density (4b) are solved so
called macro step by macro step and the point
kinetic equations (5a, 5b) are solved micro step
by micro step. - Between two macro steps point kinetic equations
are solved because the shape factor equation must
be calculated only rarely taking into account
that it depends on time slowly.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
19KFKI Atomic Energy Research Institute
- Numerical solution of the time dependent problem
(cont.)
- For the numerical solution of point kinetic
equations Runge-Kutta methods are usually used. - Investigations have been shown that these
equations result in a stiff differential equation
system for which in our solution a generalized
Runge-Kutta third and fourth order method are
applied as given by Kaps and Rentrop 4. - Practically, the solution of point kinetic
equations is much faster than the solution of
shape-factor equation.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
20KFKI Atomic Energy Research Institute
- Numerical solution of the time dependent problem
(cont.)
The shape factor equation using backward
differential scheme
(16)
The shape factor equation can be brought in each
macro time step into the following form
(17)
Matrix A is usually non-symmetric so we have to
apply such techniques which are applicable to
solve that problem.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
21KFKI Atomic Energy Research Institute
- Numerical solution of the time dependent problem
(cont.)
- In the old methodology it was dealt with standard
large sparse techniques, where Gauss-Seidel
preconditioning and a GMRES- type solver were
applied. - According to the goal accelerating the large
sparse matrix equation solution - in the new
procedure the BiConjugate Gradient Stabilized
(Bi-CGSTAB) was selected and built into the
KIKO3D code. - It must be mentioned that in both cases we used
Gauss-Seidel preconditioning in the KIKO3D code
so that in fact one has to solve
(18)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
22KFKI Atomic Energy Research Institute
- Generalized Minimal Residual (GMRES) method
- Applying the modified Gram-Schmidt
orhtogonalization to the Krylov-subspace(
) one can build
an orthonormal basis of that. - This procedure is called Arnoldis method, which
brings matrix A with an orthonormal similarity
transform to an upper Hessenberg matrix. - Matrices of upper Hessenberg form have zeros
below the lower codiagonal positions, i. e. - In the literature there are two variants to
realize the GMRES method -
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
23KFKI Atomic Energy Research Institute
- Generalized Minimal Residual (GMRES) method
(cont.)
- A GMRES method was published by Saad and Schults
for solving linear system, which minimizes the
Euclidean norm of the residual vector in the
spanned space in each iteration cycle. - This algorithm computes a sequence of orthogonal
vectors and combine these through a least-square
solve and update. - In this procedure Saad and Schults have brought
the above mentioned Hessenberg matrix to an upper
triangular matrix.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
24KFKI Atomic Energy Research Institute
- Generalized Minimal Residual (GMRES) method
(cont.)
- On the other hand Walker and Zhou 5 published
a simpler variant of GMRES - Our method corresponds to this latter one. This
algorithm bring matrix A at once into the form of
an upper triangular matrix so that programming
and using of that is easier.
Denote the residual vector belonging to the
vector
(19)
Collect this vectors into matrix R
(20)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
25KFKI Atomic Energy Research Institute
- Generalized Minimal Residual (GMRES) method
(cont.)
and create the QR factorization of the matrix AR
and denote H the matrix R in QR
(21)
Introduce the so called ATA orthogonal vectors
(22)
from which it follows
(23)
Now the following statement can be said
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
26KFKI Atomic Energy Research Institute
- Generalized Minimal Residual (GMRES) method
(cont.)
- Let an approximated solution of
and In this case the vector
is such that the Euclidean norm of
residual vector is
minimal in the space spanned by the
sub-orthogonal vectors - It can be seen that this subspace is the same as
the space spanned by residual vectors - The algorithm can be made so that we do not need
to store the vectors -
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
27KFKI Atomic Energy Research Institute
- Generalized Minimal Residual (GMRES) method
(cont.)
In the KIKO3D code the error of the solution are
estimated in Chebyshev norm as
(24)
where hjk comes from the modified Gram-Schmidt
othogonalization algorithm applied to the vectors
qk .
- This is not an exact estimation for this problem.
Here we have only assumed that each residual
vector is in fact a ATA vector and then in first
step the above mentioned estimation can be
derived.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
28KFKI Atomic Energy Research Institute
- BiConjugate Gradient Stabilized (Bi-CGSTAB)
method (cont.)
- According to the goal we have investigated other
Krylov-subspace for the solution of - These were Conjugate Gradient (CG), Biconjugate
Gradient (BiCG) and BiConjugate Gradient
Stabilized (Bi-CGSTAB). - Additionally, Broyden and Boschetti pointed out
that BiCG type algorithms can be effective if
matrix is diagonally dominant. In our
investigated cases (Examples) this assumption was
not always realized but even so the Bi-CGSTAB
algorithm worked in most cases well.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
29KFKI Atomic Energy Research Institute
- BiConjugate Gradient Stabilized (Bi-CGSTAB)
method (cont.)
- Conjugate Gradient (CG)
- the method is well known and studied
- orthogonality relations
- applicable if matrix A is symmetric and positive
definite - it is also suitable for non-symmetric matrices
by introducing the normal equations - a lot of disadvantages because of the squaring of
the condition number
(25)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
30KFKI Atomic Energy Research Institute
- BiConjugate Gradient Stabilized (Bi-CGSTAB)
method (cont.)
- Bi-CG method proposed by Fletcher
- biorthogonality relations are satisfied
- it can be used for non-symmetric matrices
- the convergence of Bi-CG can become irregular or
it can stagnate and may also break down
(26)
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
31KFKI Atomic Energy Research Institute
- BiConjugate Gradient Stabilized (Bi-CGSTAB)
method (cont.)
- Bi-CGSTAB method proposed by Van der Vorst 6
- it can be derived from Bi-CG algorithm and from
GMRES algorithm without recurrence - the method is such that locally a residual vector
is minimized - there is no matrix-vector product with transpose
matrix - it can be also used for non-symmetric matrices
- it has smoother convergence behaviour than Bi-CG
or another transpose-free variant Conjugate
Gradient Squared.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
32KFKI Atomic Energy Research Institute
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
33KFKI Atomic Energy Research Institute
- BiConjugate Gradient Stabilized (Bi-CGSTAB)
method (cont.)
Taken into account the above discussed arguments
the Bi-CGSTAB algorithm has been built in the
KIKO3D where the error of the solution is
estimated by the following term
(27)
- This is a rough estimation of the real error and
its effectiveness must be tested. - However, it must be mentioned that according to
the literature it can not be considered a
satisfactory estimation being generally effective
for the real error in case of using Bi-CGSTAB
algorithm.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
34KFKI Atomic Energy Research Institute
- The verification of Bi-CGSTAB and the validation
of KIKO3D using Bi-CGSTAB have been performed. - 5 cases will be presented
- The convergence behaviours of the respective
algorithms including the speed of convergence
were tested. - Additional information to Case 1 to 4 in the
VVER-440 core there are 349 assemblies and the
lattice pitch is 14.7 cm. Each node corresponds
to a part of the axially divided assembly, which
is of height 25 cm. The dimension of the matrix
A was 1540815408 with 423216 non-zero elements. -
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
35KFKI Atomic Energy Research Institute
- Case 1
- In a given macro step (at the beginning of the
transient) of a rod ejection transient the linear
equation system was extracted. - This equation system was solved by both GMRES and
Bi-CGSTAB algorithms. - Reference solution was also prepared and compared
to the estimated solutions in Chebyshev norm.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
36KFKI Atomic Energy Research Institute
Fig. 1 Estimated relative errors depending on
CPU time in Case 1.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
37KFKI Atomic Energy Research Institute
Fig. 2 Estimated and real relative errors
depending on time in Case 1.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
38KFKI Atomic Energy Research Institute
- Case 2
- Experiences have showed that the speed of the
numerical solution of a given equation system can
be slower when the physical system is far from
the steady-state. - In such a macro step of the above mentioned
transient the same investigations have been
carried out.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
39KFKI Atomic Energy Research Institute
Fig. 3 Estimated relative errors depending on
CPU time in Case 2.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
40KFKI Atomic Energy Research Institute
Fig. 4 Estimated and real relative errors
depending on time in Case 2.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
41KFKI Atomic Energy Research Institute
- Case 3
- An uncertainty analysis 7 have been performed
in case of a control rod ejection accident, where
the input parameters were varied and a lot of
(100) computer running were performed. - In one case of 100 runs, where physically extreme
input data are applied, the KIKO3D code using
Bi-CGSTAB algorithm (KIKO3D-Bi-CGSTAB) was not
able to achieve the solution.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
42KFKI Atomic Energy Research Institute
- The KIKO3D code using GMRES algorithm with a
dimension of 12 (henceforth KIKO3D-GMRES)
calculated the solution but it was usually (the
convergence problems appeared in each macro step)
not able to reach the given prescribed relative
error and the results could not interpreted
physically. - In the first macro step of this transient the
linear equation system was extracted and solved
with GMRES and Bi-CGSTAB algorithms
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
43KFKI Atomic Energy Research Institute
Fig. 5 Estimated relative errors depending on
CPU time in Case 3.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
44KFKI Atomic Energy Research Institute
Fig. 6 Estimated relative errors depending on
the number of iterations in Case 3.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
45KFKI Atomic Energy Research Institute
- Case 4
- The calculation of the full transient given in
Case 3 is considered. - performed firstly only by KIKO3D-GMRES (number
of the iterations at the static problem 26 ITE) - Later the convergence problems were solved by
iterating the steady state solution better
(number of the iterations at the static problem
35 ITE).
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
46KFKI Atomic Energy Research Institute
Fig 7a Nuclear power, given in core averaged
linear heat rate in Case 4.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
47KFKI Atomic Energy Research Institute
Fig 7b Relative deviation of the nuclear powers
calculated by KIKO3D-GMRES and KIKO3D-Bi-CGSTAB
in Case 4.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
48KFKI Atomic Energy Research Institute
Fig 8a Reactivity in Case 4.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
49KFKI Atomic Energy Research Institute
Fig 8b Deviation of the reactivities calculated
by KIKO3D-GMRES and KIKO3D-Bi-CGSTAB in Case 4.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
50KFKI Atomic Energy Research Institute
Fig 9 Elapsed time in macro step in Case 4.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
51KFKI Atomic Energy Research Institute
- Case 5
- The validation of the KIKO3D code using Bi-CGSTAB
algorithm was carried out in the case of a
Control Rod Withdrawal kinetic process measured
on a critical facility corresponding to a
VVER-1000 core 8. - In the VVER-100 core there are 163 assemblies and
the hexagonal assembly lattice pitch is 23.6cm. - The dimension of the matrix A was 1095110951
with 152743 non-zero elements.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
52KFKI Atomic Energy Research Institute
- The induced change of the local power density was
measured in some fuel assemblies by micro fission
chambers at different heights - In the VVER-100 core there are 163 assemblies and
the hexagonal assembly lattice pitch is 23.6cm. - The reactivities of the reactimeters PIR1 and
PIR2 were determined by inverse point kinetics
from the signals of two out-core ionisation
chambers KNK-56, placed at opposite sides of the
radial core edge. - Good agreement was found between the calculated
and measured relative powers at detector
positions
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
53KFKI Atomic Energy Research Institute
Fig 10a Reactivity obtained from inverse kinetic
calculations (used KIKO3D results) and reactivity
obtained from the signal of ionization chambers
in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
54KFKI Atomic Energy Research Institute
Fig 10b Deviation of the reactivities obtained
from inverse kinetic calculations used
KIKO3D-GMRES and KIKO3D-Bi-CGSTAB results in Case
5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
55KFKI Atomic Energy Research Institute
Fig 11a Reactivity in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
56KFKI Atomic Energy Research Institute
Fig 11b Deviation of the reactivities calculated
by KIKO3D-GMRES and KIKO3D-Bi-CGSTAB in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
57KFKI Atomic Energy Research Institute
Fig 12a Measured and calculated relative powers
at detector position 71H in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
58KFKI Atomic Energy Research Institute
Fig 12b Relative deviation of the relative
powers at detector position 71H calculated by
KIKO3D-GMRES and KIKO3D-Bi-CGSTAB in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
59KFKI Atomic Energy Research Institute
Fig 13a Measured and calculated relative powers
at detector position 126L in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
60KFKI Atomic Energy Research Institute
Fig 13b Relative deviation of the relative
powers at detector position 126L calculated by
KIKO3D-GMRES and KIKO3D-Bi-CGSTAB in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
61KFKI Atomic Energy Research Institute
Fig 14 Elapsed time in macro step in Case 5.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
62KFKI Atomic Energy Research Institute
- The estimated relative error is very close to the
real relative error in case of preconditioned
Bi-CGSTAB but the real relative error is
conservatively overestimated by the estimated
relative error (approximately with 2 order of
magnitude) in case of preconditioned GMRES. - The real relative errors of the solutions
obtained by GMRES or Bi-CGSTAB algorithms are in
fact closer together in a given time than the
estimated relative errors - In case of transients the main conclusions were
that the KIKO3D code using Bi-CGSTAB algorithm
converges to the solution safely and it is 7-12
faster than the KIKO3D code using GMRES
algorithm.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
63KFKI Atomic Energy Research Institute
Some references
1 A. Keresztúri, L. Jakab A Nodal Method for
Solving the Time Depending Diffusion Equation in
the IQS Approximation, Proc. of the first
Symposium of AER, Re, September, 1991 2 A.
Keresztúri KIKO3D - a three-dimensional kinetics
code for VVER-440. Transactions of the ANS Winter
Meeting, Washington, 1994 3 A. Keresztúri, Gy.
Hegyi, Cs. Maráczy, I. Panka, M. Telbisz, I.
Trosztel and Cs. Hegedus, "Development and
validation of the three-dimensional dynamic code
- KIKO3D", Annals of Nuclear Energy 30 (2003) pp.
93-120 4 Kaps, P., Rentrop, P. Generalized
Runge-Kutta methods of order four with stepsize
control for stiff ordinary differential
equations, Numerische Mathematik, 33 (1979) pp.
55-68 5 H. F. Walker and L. Zhou A Simpler
GMRES, Numerical Linear Algebra with
Applications, 1 (1994) pp. 571-581.
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005
64KFKI Atomic Energy Research Institute
6 H. A. van der Vorst Bi-CGSTAB A fast and
smoothly converging variant of Bi-CG for the
solution of non-symmetric problems, SIAM J. Sci.
Stat. Comput., 13 (1992) pp. 631-645. 7 I.
Panka "Uncertainty Analysis for Control Rod
Ejection Accidents Simulated by KIKO3D/TRABCO
Code System", International Conference Nuclear
Energy for New Europe 2004, Portoro, Slovenia,
September, 2004 8 Mittag, S. Grundmann, U.
Weiß, F.-P. Petkov, P.T. Kaloinen, E.
Keresztúri, A. Panka, I. Kuchin, A. Ionov, V.
Powney, D. Neutron-kinetic code validation
against measurements in the Moscow V-1000
zero-power facility, Nuclear Engineering and
Design 235 (2005) pp. 485-506
19th International Conference on Transport Theory
Budapest, Hungary, July 24-30, 2005