Chapter 2

The Role of Modelling and Simulation

Introduction
• Simulation is now considered to be the third and extremely useful avenue of
scientific pursuit.
• It is used for:
 Gather insight
 Validate models
 Train humans
 Support statistical analyses
 Control real time processes
 Predict potential outcomes
 Test and Evaluate new systems/improvements to new systems
• Simulation functions under conditions of high uncertainty
• Points along this spectrum are defined by:
How much insight gathering is occurring
Uncertainty about the correctness of the simulation
Understanding of the phenomenon being simulated.
Types of simulation
• A simulation used in problem solving is generally believed to be a
proper representation of the phenomenon is was designed to
simulate.
• A simulation for gathering insight is by contrast more likely to be seen
as an incomplete and possibly incorrect model of phenomenon that is
being studied.
Types of simulation
• Problem solving simulations have little to no uncertainty associated with
them.
They can be used for years for the same system provided that the system
remains unchanged.
They are not tuned as they are believed to have low amounts of uncertainty
attached to them.
• Insight gathering simulations may have a great deal of uncertainty attached to
them.
It is seen as a part of the learning process, where the simulation itself will have
to be adapted to meet new requirements as insight is gained from previous
incarnations and uses of the simulation.
Using simulations to solve problems
• A simulation is considered to be valid for the context in which it is
being applied.
If a researcher is studying the factors trajectories and impact points of
munitions to support a certain study regarding ballistics. He has to use
a simulation which has been designed keeping those factors in mind.

If the designer designed the simulation for stationary firearms, the

researcher also has to use the same, otherwise the simulation can no
longer be expected to produce meaningful results.
Examples
• Ballistics computations
• Automobile stability under driving conditions
• Outcome of coin tossing
• Weather prediction
• Disease spread
• Cache architecture and processing performance
The questions we are asking
• What would happen if …?
• How will a …?
• Why would a …?
• Can a …?
• Does the …?
• Should we …?
Ballistics
• What: What firing velocity would be required to make a munition with certain physical
characteristics damage the weapon muzzle?
• How: How would a munition with certain physical characteristics and fired at a
selected velocity pass through the air?
• Why: Why would a munition with certain physical characteristics and fired at a given
velocity fly of course?
• Can: Can a munition with certain physical characteristics and fired at a given velocity
travel more than X kilometers?
• Does: Does the ballistics simulation ouput match the answers provided by
mathematical ballistics model?
• Should: Should we use a munition with certain physical characteristics to perform a
given task.
Coin tossing
• What: What are the most likely outcomes( sequence of heads and tails) of flipping a
fair coin 1000 times?
• How: How is it possible to observe 900 heads and 100 tails in 1000 flips of a fair coin?
• Why: Why in repeated sessions of 1000 flips of a fair coin is the outcome 900 heads
and 100 heads rarely observed?
• Can: Can 900 heads and 100 tails be observed in 1000 flips of a fair coin with a
probability greated than 0.1%?
• Does: Does the simulated distribution of possible outcomes for flipping a fair coin
match the expected distribution of outcomes?
• Should: Should we count on 1000 flips of a fair coin coming up heads 900 times and
tails 100 times to not occur?
Validate Models and Experiments
• When researching the behavior of a phenomenon a reasearcher does
two things:
 Creates analytical model
 Performs an experiment
For both cases, we can turn to simulation as a method of validating the
data or the model.
Let’s consider a model
Little’s Law: Expected length of a queue is always equal to the average
rate at which the server can process customer times the expected time
each customer will ultimately have to wait in the queue.
L = λW

L = mean queue length

λ = mean service rate
W = mean wait time
Experiments
• Physical data might be corrupted due to various experimental errors
• We use a simulation to check if the data are accurate.

Only possible if we have a model that represents the system.

Aleatory Uncertainty
• Some phenomenon are considered random because there is no better
way to express them.
• Random Variations.
• Also known as data uncertainty

A probability distribution is used.

Epistemic Uncertainty
• This type of uncertainty arises out of lack of knowledge about the
system.
• Structured uncertainty.
• Also known as model uncertainty
• Predictions vs Simulations
• What-if analyses
• Enhance Teaching and Education
Model Calibration
• It is performed in order to determine reasonable values for critical
simulation parameters.
• Until all confusions about the models validity and the uncertainties
are handled, a model is used to gather insights.
Gaining Insight
• Acquiring an understanding of the phenomenon of interest or of how
to simulate the phenomenon.
• It is the task of the simulation designer
• The user also gains insight but his insight is not instrumental in
improving the simulation.
What questions will the designer ask?
• What has the greatest influence?
• How will X and Y interact?
• Is there a way to make X happen?
• Why has unexpected behavior Z occurred?
• What new behaviors might emerge?

Which uncertainty are these questions trying to answer?

Lifetime of a simulation