One major shortcoming in the area of computer simulation is in the background of code users and developers. The vast majority have strong theoretical and/or computer backgrounds, and very weak experience with experiments. When you apply a simulation code, you are performing an experiment. If you expect to go into this business, pay attention to the experimental training that you do get, and take time to read about the subject. Think about design of good controlled experiments, interpretation of data, and analysis or errors. Be prepared to use all of your theoretical and experimental background to interpret and justify the results of your computer simulation. The most important thing that you should learn from this class is to never trust the results of a computer simulation until you have checked it very carefully. This is in fact very good news for you. If results of computer simulations were foolproof, vendors, utilities and regulatory agencies could hire high school students to analyze reactor systems.

Early versions of RELAP included conservative assumptions to meet the US Nuclear Regulatory Commission's Appendix K requirements for licensing of reactors. Systems were modeled with 1-Dimensional flow equations which averaged over fairly large volumes of the reactors.

In the early 70's the Nuclear Regulatory Commission (NRC) became interested in "best estimate" (BE) calculations of reactor behavior in which Appendix K assumptions were replaced with the best knowledge of physical behavior. This path was followed because:

- NRC was worried that Appendix K is not always conservative;
- NRC wanted estimates of margins of safety;
- more knowledge leads to fewer unpleasant surprises.

Recent modifications to licensing rules permit use of BE codes for licensing purposes, provided the uncertainty of the code is properly quantified. It is very important to remember a best estimate is not always a good estimate.

The second major code family in this country (TRAC) began with a
research proposal to the NRC by Kay Lathrop and Bill Reed from Los
Alamos Scientific Laboratory (now LANL) in 1974. The product was meant
to include significantly more detail than RELAP, including 3-D flow
equations, but as a consequence of computational complexity, was to be
used for spot checks on RELAP results. Around 1980, TRAC split into PWR
and BWR versions, that looked very different to code users. Over
the years, TRAC became much faster than expected without loss of
detail,
and RELAP became significantly more detailed. As a result, the codes
evolved very similar in capabilities.

About 1996, the NRC realized that they had too many codes doing
about the same thing. To reduce maintenance costs, and the number
of people required to carry the expertise in all of these products,
they
decided to produce an analysis package combining the capabilities of
RELAP5, TRAC-PWR, TRAC-BWR, and a special purpose BWR code named
Ramona.
The result is a software package called the TRAC/RELAP Advanced
Computational Engine (TRACE).

By now you know that the standard theoretical model for fluid flow
is the Navier-Stokes equation set, and that heat conduction is modeled
with the Poisson Equation. These are partial differential
equations (PDEs) solved over some domain in space and time.
Regardless of the size of the domain, it is a continuum with an
infinite number of points at which values are available for the
solutions. Computers have only a finite amount of memory, so
solution methods are required that only evaluate solution variables at
a
finite number of locations in space and time. We will look at
details of the numerical solution procedure later. For now the
important thing to know is that the code user must choose how space is
broken up into a finite number of regions (often referred to as cells,
nodes, or volumes). The code will worry about sampling of time
with some guidance from you on the maximum and minimum permitted steps
forward in time from one solution to the next.. You will be engaged in
a
balancing act with this discretization of space. The more
(smaller) cells that you use in your system, the closer your answer
will
be to the correct solution of the PDEs. However, the more cells
that you use, the more computer time it will take to solve the problem.

TRACE and its predecessors use Finite Volume solution methods.
For now we will work with 1-D pipe flow. You are required
to
provide input that breaks each 1-D section into some number of volumes.
To fully describe your discretization, you provide the length
along the pipe center line of each of your volumes, the total
volume of each volume, and the area available to fluid flow at the ends
of each volume. This may seem like an over specification of
conditions to you, but permits "1-D" flow modeling in pipes with
a
variable cross-sectional area.